D5.1: MAC approaches with FBMC and simulation results (first version)

Transcription

1 Grant Agreement No.: Research and Innovation action Call Identifier: H2020-ICT quality of Service Provision and capacity Expansion through Extended-DSA for 5G D5.1: MAC approaches with FBMC and simulation results (first version) Version: v1.0 Deliverable type Dissemination level Report PU (Public) Due date 30/06/2016 Submission date 14/07/2016 Lead editors Authors Reviewers Work package, Task WP5, T5.1, T5.2 Keywords Klaus Moessner & Haeyoung Lee (UNIS) Benoit Miscopein, Rida El Chall (CEA); Bismark Okyere (Intel); Klaus Moessner, Haeyoung Lee, Marcin Filo (UNIS), Salva Díaz, Oscar Carrasco, Federico Miatton (Sistelbanda) Keith Briggs (BT), Valerio Frascolla (Intel), Anastasius Gavras (Eurescom) 5G, edsa, MAC, small cell Abstract One of the research objectives of the SPEED-5G project is the development of a new MAC layer that firstly exploits the flexibility in terms of resource fragmentation and access, e.g., leveraging on the benefit offered by the introduction of the FBMC technology; and secondly facilitate the efficient resource allocation across a number of co-existing heterogeneous radio access technologies. This is done in order to more flexibly exploit radio spectrum which can be handled under different licensing regimes. This deliverable contains a detailed description of MAC strategies for small cells, which are capable of the efficient use of a range of bands having different characteristics.

2 Document revision history Version Date Description of change Contributor(s) Document creation UNIS Structure of chapter 3 on Multi-Rat Access CEA Materials of section 4.1, 4.3 and 5.3 in. CEA, UNIS Materials of chapter 2 on small cell requirements and Chapter 4 on Mac definition in Updates of chapter 3 and restructure (separation of MAC architecture and a specific MAC design) Materials of section 6.2 (preliminary simulation results for extended suburban scenarios) in Sistelbanda CEA Sistelbanda Updates on chapter 2 and 4 Sistelbanda Updates of chapter 3 for the edsa concept CEA Updates of section 4.2 on UE MAC sub-blocks Intel Updates of section 5.3 and 6.3 on TDD MAC UNIS Sistelbanda final contributions now in. Sistelbanda Updates of section 5.1 on FBMC MAC design CEA Updates on formatting and add abstract, introduction and conclusion UNIS Updates on chapter 2, 4.1, 5.2, 5.3, 6.2, and 6.3 Sistelbanda, UNIS Contents of section 6.1 in, Updates on UE MAC CEA, Intel Review feedback BT, Intel, Eurescom Final editing UNIS Disclaimer This report contains material which is the copyright of certain SPEED-5G Consortium Parties and may not be reproduced or copied without permission. All SPEED-5G Consortium Parties have agreed to publication of this report, the content of which is licensed under a Creative Commons Attribution-NonCommercial-NoDerivs 3.0 Unported License 1. Neither the SPEED-5G Consortium Parties nor the European Commission warrant that the information contained in the Deliverable is capable of use, or that use of the information is free from risk, and accept no liability for loss or damage suffered by any person using the information. Copyright notice SPEED-5G Consortium Parties Acknowledgment The authors would like to thank all contributors and reviewers of this deliverable SPEED-5G Consortium Parties Page 2 of 82

3 Executive Summary SPEED-5G s main objective is to investigate and develop technologies that address the well-known challenges of predicted growth in mobile connections and traffic volume by successfully addressing the lack of dynamic control across wireless network resources, which is leading to unbalance spectrum loads and a perceived capacity bottleneck. In this perspective, SPEED-5G has defined the concept of enhanced dynamic spectrum access (edsa) with the scenario where a large number of small cells can access co-existing heterogeneous radio access technologies and heterogeneous spectrum bands (of different spectrum regimes) simultaneously. To implement the edsa concept, three key functions are identified: 1) dynamic channel selection, ii) interference mitigation by load balancing, and iii) capacity maximization by smart offloading. For the access to non-contiguous spectrum from multiple bands of different regimes, FBMC (Filter Bank Multicarrier) which is robust to synchronization errors [1] has been chosen for the physical layer. Considering this context, the deliverable provides a MAC description for the SPEED-5G project. Firstly, the requirements of the small cell (SC) in SPEED-5G are identified based on investigation of the general SC characteristics and requirements. Besides of the traditional SC requirements, additional requirements are identified 1) to split the control plan from the user plan, 2) to support a network cloud, and 3) to support the network virtualization, considering the SPEED-5G use cases. Then, the high-level MAC requirements are provided, including management of contention-based and noncontention-based random access control scheme to exploit (lightly) licensed spectrum and unlicensed spectrum together, and capability to support FBMC. With identification of a set of SPEED- 5G MAC features, a block diagram describing the SPEED-5G MAC together with its control and data plane interfaces, building blocks and service access points has been defined and explained. Finally, based on these requirements and definitions, the MAC protocols have been developed which can exploit the flexibility and scalability for the scenarios of SPEED-5G and can be operated on the various spectrum regimes considering the virtualized network architecture. In addition, to cope with the case that the SC cannot rely on the virtualized architecture and the information of spectrum characteristics is not available, a self-organizing MAC algorithm is also designed and described. This deliverable reports on the first version of the MAC design and the preliminary results of the MAC protocols. More results and sophisticated outcomes of the simulation campaigns along with improved MAC protocols will be included in the next deliverable D SPEED-5G Consortium Parties Page 3 of 82

4 Table of Contents Executive Summary... 3 Table of Contents... 4 List of Figures... 6 List of Tables... 7 Abbreviations Introduction High-level requirements and MAC vision Small Cell categories Small Cell use-cases Small Cell requirements Main MAC layer features SPEED-5G MAC and protocol stack vision Summary edsa concept development and MAC functional description edsa use-cases and overall block diagram Scenario 1: Dynamic channel selection Scenario 2: Interference mitigation by load balancing Scenario 3: Capacity maximization by traffic smart offloading Overall block diagram MAC functions and interfaces for edsa support MAC function and IOs Licensed spectrum domain Unlicensed spectrum domain MAC interface with RRM and upper layers MAC interface with physical layer Summary An edsa-capable MAC design Small cell MAC design MAC interfaces MAC entities UE MAC design Summary Design of MAC functions for edsa FBMC MAC operating in unlicensed spectrum Physical layer assumptions SPEED-5G Consortium Parties Page 4 of 82

5 5.1.2 MAC design for IoT traffic operating in the 868 MHz band MAC design for broadband traffic operation in the 5 GHz band Broadband wireless access in virtual-reality offices and realistic extended suburban HetNet scenarios Scenario characteristics Carrier aggregation Quality of service Enhanced spectrum usage IoT support Decentralized dynamic TDD MAC for ultra-dense networks Summary MAC evaluation UL/DL offload on shared spectrum In realistic traffic patterns Evaluation methodology Simulation results and analysis Broadband wireless access in realistic extended suburban and virtual reality office HetNet scenarios Evaluation methodology Simulation results and analysis Decentralized dynamic TDD MAC for ultra dense networks Evaluation methodology Simulation results and analysis Summary Conclusion and future steps References SPEED-5G Consortium Parties Page 5 of 82

6 List of Figures Figure 1: SPEED-5G MAC Layer from cell side Figure 2: SPEED-5G functional block diagram for edsa support Figure 3: MAC interfaces and entities from cell side Figure 4: Scalable and Flexible 5G UE MAC Architecture Figure 5: OQAM pre-processing in FBMC transmission chain Figure 6: FBMC modulator and demodulator Figure 7: Power spectral density of OFDM, UFMC and FBMC-OQAM with several values of K Figure 8: Superframe structure Figure 9: Basic mode for NC notification Figure 10: NC refinement 1: wake-up acknowledgment Figure 11: NC refinement Figure 12: 5 GHz band allocation (ETSI) Figure 13: LBT timing example [22] Figure 14: Proposed frame structure for broadband access in 5 GHz Figure 15: CA options Figure 16: Frame on licensed band using FBMC on licensed band Figure 17: Frame with whitespaces for control data using FBMC on licensed band Figure 18: Virtual home layout Figure 19: Extended Suburban HetNet Scenario Figure 20: Virtual office layout Figure 21: HeNBs installed indoors and every two houses at 2.6 GHz Figure 22: Three outdoor enbs cells at 800 MHz Figure 23: Cumulative Distribution Function (CDF) of the achieved throughput per user Figure 24: CDF of the served latency per user Figure 25: CDF of the served throughput per user Figure 26: CDF of the achieved latency per user Figure 27: Regular/Planned cluster layout Figure 28: Comparison of network capacity performance of the proposed MAC and Wi-Fi Figure 29: The impact of varying BW by aggregation on the number of users in outage Figure 30: Performance of dynamic channel selection mechanism SPEED-5G Consortium Parties Page 6 of 82

7 List of Tables Table 1: MAC function location and link direction association Table 2: Interface from RRM and upper layer to MAC Table 3: Interface from MAC to RRM, upper layer and MAC functions Table 4: Bidirectional interface between MAC and PHY layers Table 5: Narrowband modulation and coding schemes Table 6: Narrowband FBMC parameters Table 7: Modulation and coding schemes of FBMC PHY for 10 MHz and 20 MHz bands Table 8: Broadband FBMC parameters Table 9: Uplink and Downlink Traffic patterns Table 10: IoT device classes Table 11: IoT MAC parameters Table 12: ETSI Transmission parameters in 5 GHz band Table 13: LBT procedure parameters Table 14: FBMC broadband access frame parameters Table 15: Main simulation parameters Table 16: LTE Simulation Parameters Table 17: Main simulation parameters for unlicensed band use on virtual office scenario Table 18: Additional simulation parameters for the subset of the extended suburban under consideration Table 19: Additional simulation parameters for the virtual office under consideration Table 20: A set of main simulation parameters SPEED-5G Consortium Parties Page 7 of 82

8 Abbreviations ACK ACL ACLR A-MPDU A-MSDU AP BLER BP BPSK BS ASA CA CAP CAPEX CCA C-IoT CoMP CP-OFDM CQI crrm CSAT CSI CSMA/CA cson CTS D2D DCI DFS DL DRX DSA dson DTP Acknowledgement Adjacent Channel Leakage Adjacent Channel Leakage Ratio Aggregated-MAC PDU Aggregated-MAC SDU Access Point Block Error Rate Beacon Period Binary Phase Shift Keying Base Station Authorized Shared Access Carrier Aggregation Contention Access Period Capital Expenditure Clear Channel Assessment Cellular IoT Coordinated Multipoint Cyclic Prefix based Orthogonal Frequency Division Multiplexing Channel Quality Indicator Centralized RRM Carrier Sensing Adaptive Transmission Channel State Information Carrier Sense Multiple Access with Collision Avoidance centralized Self-Organizing Network (SON) Clear To Send Device-to-Device Data Control Indicator Dynamic Frequency Selection Downlink Discontinuous Reception Dynamic Spectrum Access distributed Self-Organizing Networks Data Transfer Period SPEED-5G Consortium Parties Page 8 of 82

9 ED edsa embb enb ER FBMC FFT FS-FBMC GBR GTP HARQ HeNB HeMS HetNet HO I/O IoT IPSec KPI LAA LBT LTE-U LUT LWA MAC MAR MAS MCS MIMO miot MOCN MPAP NC OAM&P OFDM Energy Detection extended Dynamic Spectrum Access Extreme Mobile BroadBand enodeb Exception Report Filter Bank Multicarrier Fast Fourier Transform Frequency Spreading-FBMC Guaranteed Bit Rate GPRS Tunnelling Protocol Hybrid Automatic Repeat Request Home enodeb HeNB Management System Heterogeneous Network Handover Input and Output Internet of Thing Internet Protocol Security Key Performance Indicator Licensed Assisted Access Listen Before Talk LTE-Unlicensed Lookup Table LTE/WLAN link aggregation Medium Access Control Mobile Autonomous Reporting Medium Access Slot Modulation and Coding Scheme Multiple Input Multiple Output Massive Internet of Thing Multi-Operator Core Network Multi-Purpose Access Point Network Command Operations, Administration, Maintenance and Provisioning Orthogonal Frequency Division Multiplexing SPEED-5G Consortium Parties Page 9 of 82

10 OPEX OQAM PAM PAPR PDCP PDU PER PF PHY PR QCI QoE QoS QPSK RA RACH RAN RAT RB ReTx RLC RR RRC RRM RTS SAP SC SCF SC-FDMA SDR SDU SIMO SON SRS TB Operational Expenditure Offset Quadrature Amplitude Modulation (QAM) Pulse Amplitude Modulation Peak-to-Average Power Ratio Packet Data Convergence Protocol Protocol Data Unit Packet Error Rate Proportional Fair Physical Layer Periodic Report QoS Class Identifier Quality of Experience Quality of Service Quadrature Phase Shift Keying Random Access Random Access Channel Radio Access Network Radio Access Technology Resource Blocks Retransmission Radio Link Control Round Robin Radio Resource Control Radio Resource Management Request To Send Service Access Point Small Cell Small Cell Forum Single Carrier FDMA Software Defined Radio Service Data Unit Single Input Multiple Output Self-Organizing Network Sounding Reference Signal Transport Block SPEED-5G Consortium Parties Page 10 of 82

11 TCP TTI TVWS Tx UDN UFMC UL URC WSN Transmission Control Protocol Time Transmission Interval TV White Space Transmission Ultra-Dense Network Universal Filtered Multicarrier Uplink Ultra-Reliable Communication Wireless Sensor Network SPEED-5G Consortium Parties Page 11 of 82

12 1 Introduction SPEED-5G s main objective is to investigate technologies to consider the well-known challenges of growth in mobile connections and traffic volume by addressing the lack of dynamic control across wireless network resources, which is leading to unbalance spectrum loads and a perceived capacity bottleneck. In order to solve such problems, SPEED-5G focuses on extended dynamic spectrum access (edsa) in the scenario where a large number of small cells (SCs) can operate across co-existing heterogeneous radio access technologies and use heterogeneous spectrum bands simultaneously. To implement the edsa concept, the following key functions are identified: i) dynamic channel selection, ii) interference mitigation by load balancing, and iii) capacity maximization by smart offloading. This deliverable has the purpose of describing the medium access control (MAC) approach designed to support operation in, and exploitation of, shared spectrum resources in licensed, unlicensed and lightly-licensed bands in dense environment scenarios. In this regard, the MAC layer is as a component of the SPEED-5G virtualized protocol stack defined in the deliverable D3.2 [2]. In particular, the support of several radio access technologies (RATs) and the ability to combine several carriers in both uplink and downlink are key aspects of the SPEED-5G MAC layer. At stake in the system definition is the way MAC layer and radio resource management (RRM) functions interact, which is a key part of the SPEED-5G work. Indeed, it governs how the strategies of capacity maximization implemented in RRM can be translated into MAC configuration parameters, as well as what kinds of information are forwarded from MAC to RRM. It is worth noting that the MAC layer is assumed to be configured by the virtualized RRM algorithms but the case when the link with the virtualized architecture does not exist shall be considered as well. Indeed, in this situation, the MAC layer has no means to either manage the interference coming from the neighbouring cells or to get information on the available spectrum bands. This may lead to a jeopardised performance unless a self-optimizing MAC should be selected as a back-up configuration. Another important part of the SPEED-5G project is the assumption that the MAC layer can exploit the FBMC (Filter Bank Multicarrier) physical layer (PHY). Due to a sharp spectral localisation, the FBMC PHY shows extremely interesting features for the MAC design such as the ability to exploit fragmented spectrum or a relaxed synchronization among multiple users. Thus, the MAC algorithms are designed with the assumption that the FBMC PHY is available and can be used on the various spectrum regimes including unlicensed spectrum where traffic can be efficiently offloaded The deliverable is structured as follows: Chapter 1 is the introduction, while Chapter 2 discusses the MAC layer features and requirements for SPEED-5G at a high level based on investigation of the general SC characteristics and requirements. Chapter 3 explains a functional description of the MAC layer supporting multi-rats and multi-bands to implement the edsa concept in SPEED-5G. The overall block diagram of the SPEED-5G MAC is included along with description of each functions and interfaces between MAC and the neighbouring layers. Chapter 4 provides the MAC architecture designed for SPEED-5G from both cell side and user side. Moreover, it is described how the scalable and flexible MAC functional description, capable of coordinating the use of radio resource across heterogeneous RATs and multiple bands can be depicted in a way amendable to implementation in real devices. In Chapter 5, the MAC algorithms that can be operated with a FBMC PHY and can be used as one RAT on the various spectrum regimes are described. The virtualized architecture defined in the project is also considered in designing the MAC algorithms. In addition, a self-organizing MAC algorithm is proposed which is capable of operating autonomously without the information of SPEED-5G Consortium Parties Page 12 of 82

13 spectrum characteristics to cover the case when the SC cannot reply on the virtualized architecture. The preliminary results on all proposed MAC algorithms are included along with simulation parameters in Chapter 6. Finally, Chapter 7 elaborates on the conclusions of the deliverable SPEED-5G Consortium Parties Page 13 of 82

14 2 High-level requirements and MAC vision As a general introduction, this chapter first summarizes the small cell (SC) characteristics and requirements. In a second step, the MAC-layer features and the envisioned MAC for SPEED-5G and its characteristics are listed. 2.1 Small Cell categories There are many ways to categorize SCs; the most common is that which is based on coverage area. Following this convention and ordered from more to less area of coverage, the SC types are: micro cell, pico cell and femto cell. According to the current state of the art [1], their main characteristics are like follows: The micro cell is mainly an outdoor cell designed to have a short-range coverage compared to the macro cell. The main objective of this cell is to enhance the macro coverage serving indoor and outdoor devices. Depending on the scenario, the coverage ranges from 2 km if the cell is an outdoor rural, to above 400 meters when in dense urban environment. Some of the relevant micro cells usages are, for instance, to cover black-hole areas where the macro cells do not provide the minimum required signal strength, to improve the final user performance, or to support a very localized spot of data traffic data originated, for instance, in shopping streets. These micro cells are intended to support a high number of connected devices. The pico cell is a low-power equipment used in public indoor areas as stations or shopping malls or outdoor places like a stadium where, during the event, a huge amount of traffic is generated in a short period of time. In any case, they are mainly indoors and their main goal is to cover specific spots where the resources provided by the macro cells are not sufficient. The expected covered range is around 200 to 500 meters. They are expected to support a medium to high number of devices connected simultaneously to the network. The femto cell is a low-cost equipment mainly designed to operate in home environments supporting a small number of static devices. They also may be used as a relay when located inside cars, trains or buses. In these cases, they act as genuine cells for the devices inside the vehicle, but they use the wireless link to macro or micro cells as a backhaul. The relay is considered a special device for macro and micro cells. The area of coverage is about a maximum of 300m, even shorter in home environments in order to avoid interferences. SPEED-5G goes beyond this SC categorization including new SC types that support the proposed use cases and the proposed market drivers [4]. The SPEED-5G architecture and the MAC layer support the traditional SC categorization ensuring the backwards compatibility but also are able to support the novelties introduced by the project: The master cell and phantom SCs is a concept introduced by DOCOMO. The main idea is to split the control plane from the user plane. The master cell sends both planes as it is done nowadays. On the other hand, the phantom SC sends only the data plane. Using this solution, the device has to be connected, as minimum, to two different cells. As the master cell sends the control data, it has to be visible for as many devices as possible. Considering this requirement, the best solution to deploy the master cell is to use a macro cell. This solution may be used over the Extreme Mobile BroadBand (embb) use case. A cloud of SCs is a novel concept used to support massive Internet of Things (IoTs) deployments. The idea is that the IoT device sends data to the cloud instead of sending it to a single SC. This procedure reduces the signalling data into the network. The periods without IoTs transmission may be used to send more user data or to mitigate interferences. The cell protocol stack has to be ready to reject duplicated packets and the architecture has to SPEED-5G Consortium Parties Page 14 of 82

15 support to work as a network cloud. The virtual cell is a cell whose part of its stack has been virtualized. Each layer of the stack may be virtualized depending on the fronthaul, the SC requirements or the equipment to perform the virtualization. The SC has to support reconfigurations following the Software Defined Radio (SDR) being considered a Multi-Purpose Access Point (MPAP) [5]. This means that the same SC can work into the licensed, unlicensed or lightly licensed bands depending on the scenario requirements. 2.2 Small Cell use-cases The identification of the SC requirements mainly depends on the actual use-case where they will be deployed. In order to define a common framework, the Small Cell Forum (SCF) identified and categorized the most important SC use cases. They cover the most common functionalities and scenarios where the macro cell resources or coverage are not sufficient. The main use-cases are listed below, each one being characterised by the deployment conditions or the kind of load it has to deal with: The home use case [6] has cells designed to work into residential environments, deployed inside the houses and therefore without any controlled deployment. The main goal of this use case is to deploy a SC in each home, so the number of SCs expected is huge. They are expected to work in an individual way since the operator has no control on where the SC is located. Due to the huge number of SCs expected, the low number of connected devices and assuming that it will be mostly indoor, the equipment has to be as cheap as possible. For all these reasons, the most suitable cells are the femto cells. Nowadays, most common SCs are the WiFi Access Points (APs) working on unlicensed bands while LTE SCs are expected to be deployed in The LTE SCs will use licensed bands and thus support end-to-end Quality of Service (QoS) and Quality of Experience (QoE). The enterprise use case [7] is ready to be deployed on enterprises, hospitals or government buildings. In the same way as the home SC, most of the proposed scenarios are indoor, but instead of being designed to support a small number of devices, the network is ready to support a medium number of devices generating more traffic than home devices. These cells are deployed by the operator and, considering the area of coverage more than one SC will be required. The most suitable SCs to be deployed on these scenarios are pico cells and femto cells. While the expected devices in the home use case are mostly static, the devices here will have to support mobility feature. As this use case has more than one cell with mobile devices, the handover (HO) procedures between the SCs have to be optimized, ensuring reliability on the network. In this use case, the number of deployed SCs is scenario dependant. Additionally, most of these scenarios require end-to-end QoS/QoE thus, the deployments will be done considering a licensed band and Dynamic Spectrum Access (DSA) techniques. Nowadays, these use cases are deployed using WiFi APs for data services without consider voice or ultra-reliable services. The urban use case [8] is deployed by operators in order to support public locations and events like stadiums, stations or concerts. These SCs may be deployed permanently or on demand, depending on the requirements and the location but, in any case, the number of SCs deployed is low. In order to achieve all these requirements, most of the SCs are to be micro or pico cells. The reason to deploy these types of SCs is that a huge number of devices should be supported and the power transmission and computational resources are higher than femto cells. Depending on the environment (static or mobile) the mobility support is a feature SCs may have to implement. The rural use case is designed to cover rural areas like farms, solar fields, fruit fields or a few houses. These SCs cover areas where the macro cell coverage is not optimal and then, it is not able to support the generated traffic. Depending on the area of coverage of the spot the SPEED-5G Consortium Parties Page 15 of 82

16 required SC will be a micro cell, pico cell or femto cell. While a couple of femto cells can cover a single farm or field and can deal with the number of devices and they generated traffic, the area of two or three houses has to be covered by a microcell, even if these houses have a femto cell inside as described into the home use case. The remote use case is the one that is deployed on emergency scenarios where the first responders require reliability and security on their communications. This kind of ephemeral networks cannot be predicted by operators, and are deployed when needed and removed when not. Considering an emergency scenario composed by policeman and fireman carrying a lot of sensor where it is required an ultra-reliable and ultra-secured communication, the most suitable SCs are micro cells or pico cells. The mobile relay use case is considered when a SC s backhaul is wireless. The devices are connected to the relay which provides the service to devices in a transparent way. The relay is the only element which is connected to the traditional wireless network. This use case is mostly for SCs located into public transports or cars or for device-to-device communications. Compared to this, the use cases proposed by SPEED-5G [4] are different since they are based on the kind of traffic they induce. As a consequence, a SPEED-5G use-case may be covered by one or a combination of SCF use-cases. For instance, The Massive IoT (miot) Communication use case in SPEED-5G distinguishes two IoT categories: high-end IoT, and low-end IoT, with a focus on the latter. The low-end IoT devices may be deployed indoor and outdoor; thus any of the SCF use cases has to support them. The Extreme Broadband (embb) Wireless use case is a mixture of domestic, enterprise and public access in indoor and outdoor urban areas. These SCs have to be ready to support different traffic types as well as Radio Resource Management (RRM) reconfigurations and Self-Organizing Networks (SONs) modifications. The High-Speed Mobility use case is mostly related with mobile relay covering device-todevice (D2D) as well as a train or a bus with a SC which serves the devices inside the vehicle. It is highly linked to the mobile relay use case from SCF. The Ultra-Reliable Communication use case depends on the service, but not on the scenario, meaning that the SCF use-cases have to support this kind of traffic. 2.3 Small Cell requirements The requirements for SC deployment are related to the SC type and the supported uses cases. The list below shows the SC requirements identified in SPEED-5G in WP3 deliverables. The Management System is common to every SC type and use case. The 3GPP series 32 defines the concepts and standardizes the requirements, procedures, interfaces and information models for operations, administration, maintenance and provisioning (OAM&P) [9]-[12]. The protocol and the data model to exchange information and commands between the Home enodeb Management System (HeMS) and each SC are defined by the Broadband Forum in the Technical Reports TR-69 [13] and TR-196 [14] respectively. The main advantage of using an HeMS is that it allows remotely configuring and managing a multi-vendor SC network. SON should be a mandatory requirement in home, enterprise and remote deployments. For a home use case, SON allows a better management of interference generated by the huge number of SCs it comes with. In an enterprise or a remote deployment, SON may ensure the required ultra-reliability. SON solutions for the rural use case are less critical than for the other use cases due to the low SCs density. SON functions may be done in a distributed (dson), centralized (cson) or hybrid (distributed and centralized) ways. The SCF has defined a SON API [15] for interoperability procedures in order to support different vendor SPEED-5G Consortium Parties Page 16 of 82

17 algorithms. Virtualization refers to the abstraction of the different communication layers of the SC protocol stack. It provides multiples advantages to the operators since, for instance, it improves Capital Expenditure (CAPEX) and Operational Expenditure (OPEX) costs, allows hiding device mobility or improves the spectrum usage by allowing a true joint scheduling. The Multi-Operator Core Network (MOCN) [16] allows sharing the physical SC between multiple-operators. Thus, a single cell is connected with more than one core. The shared resource into the SC may be equal or asymmetric depending on the operator s agreement. Then, since the number of resources is configurable, the resource allocation configuration has to be dynamically configurable. This solution reduces the operator cost since, in several scenarios like the home one, devices from different operator may share the same SC. This feature is currently provided by operators and is known as radio access network (RAN)- sharing [17]. edsa is one of the main proposition of the SPEED-5G project for enhancing the licensed, lightly licensed and unlicensed bands management. To support this feature, SCs have to be able to configure the required functionalities on the devices and to support the mechanisms and procedures like an efficient implementation of the carrier aggregation (CA) feature or the usage of the lightly licensed or unlicensed spectrum band in an efficient way. The main goal is that the SC has to be able to dynamically use the most appropriate bands without interfere other cells and systems. In order to achieve this, the SC has to store the results of sensing the spectrum. Multiple Radio Access Technologies (RATs) support is a basic feature for supporting edsa. CA may be done in the same RAT using an extended, or a new band, but also may be done using another RAT depending on traffic characteristics and constraints. Multiple services have to be supported in a 5G network and consequently, different traffic types are to be simultaneously handled. The QoS Class Identifier (QCI) is defined to categorize different traffics by its type (granted or not granted), the traffic priority, the maximum end-to-end packet delay, and the packet loss rate. Each RAT has its own QCI requirements and they are used by the MAC schedulers and the RRM algorithms for an optimal resource usage depending on the multiple services that they have to support. Key Performance Indicators (KPIs) are used as the metric by which it is possible to indicate the network performance based on individual SC reports. The reported values are focussed on the accessibility, retainability, integrity, availability, or mobility. It is 3GPP, series 32 for GSM, UMTS and LTE, which standardizes some KPIs, defining how they have to be measured and how the final value has to be obtained. The main idea of these standardized KPIs is to have the same procedures for every supplier, thus avoiding misunderstandings. In addition to the standard values, each operator or supplier may have its own KPIs based on its own type of network deployment. The division between control and user plane allows supporting scenarios with phantom cells. Natively, the proposed architecture for managing the SCs supports this division and the SC has to be capable of sending and receiving user control and user data from the Control Plane GW and the Forwarding Plane GW respectively. Massive IoT support and its traffic require avoiding the central core due to the huge number of devices. This kind of device sends very low payload packets, but the high number of devices can make very high the final traffic load generated into the network. For that reason, the SPEED-5G architecture has to be able to forward the IoT data to large data centres instead of using the central core as a gateway SPEED-5G Consortium Parties Page 17 of 82

18 2.4 Main MAC layer features The MAC layer is responsible of multiplexing and de-multiplexing logical channels into transport channels and vice versa. On the receiving side, the MAC layer recovers MAC service data units (SDUs), while on the transmitting side, it allocates MAC protocol data units (PDUs) into physical radio resources known as Transport Blocks (TBs), as well as defining the Modulation and Coding Scheme (MCS) in order to maximize the channel capacity. 3GPP [18] defines a set of MAC functions which are required for performing accurate and comparable simulations. These simulation functions are listed here: Multiplexing of MAC SDUs from one or different logical channels onto TB to be delivered to the physical layer on transport channels De-multiplexing of MAC SDUs from one or different logical channels from received TBs delivered from the physical layer on transport channels Scheduling information reporting, like Channel Quality Indicator (CQI) reporting Error correction through the Hybrid Automatic Repeat Request (HARQ) Priority handling between UEs by means of different dynamic scheduling algorithms Priority handling between logical channels of one MAC entity Transport format selection Logical channel prioritisation Transport format selection In addition to these very common MAC functions, we derive from deliverable D3.2 [2], some additional high-level functions and requirements for the SPEED-5G MAC layer: Management of contention-based and non-contention-based Random Access Channel (RACH) schemes. The MAC layer should manage different random access schemes for supporting different use cases, such as embb on licensed and unlicensed spectrum or miot, and the ability to pre-empt and prioritise random access requests, regulating the way the users camp on to the network, including the contention resolution strategy handling various licensing regimes Capability to handle real-time reconfiguration in order to allow using a dynamic control regime Capability to handle lightly licensed and unlicensed where scheduling co-operation between neighbours may not be present and contention-based access must be managed. This may require control messages such as Request To Send (RTS) / Clear To Send (CTS) Priority handling between new transmission and retransmission depending on the logical channel and the QoS/QoE Co-operation with the scheduler to provide uplink and downlink grant requests at the required QoS Support of power-saving techniques such as Discontinuous Reception (DRX) Support of different mechanisms for allowing narrowband transmissions from IoT devices Capability to support different waveforms, such as Orthogonal Frequency Division Multiplexing (OFDM) and Filter Bank Multicarrier (FBMC) SPEED-5G Consortium Parties Page 18 of 82

19 The idenfied MAC functionalities are mapped into the following Table 1. MAC function cell UE downlink uplink Mapping between logical channels and transport channels Multiplexing De-multiplexing Error correction through HARQ X X X X X X X X X X X X X X X X X X X X Transport Format Selection X X X Priority handling between UEs X X X Priority handling between logical channels of one MAC entity X X X Logical Channel prioritisation X X Scheduling information reporting X X Radio Resource Selection Table 1: MAC function location and link direction association X 2.5 SPEED-5G MAC and protocol stack vision The key value of SPEED-5G is to support edsa, and consequently the defined protocol stack has to be capable of managing multiple edsa techniques simultaneously per band, and allocating the RAT in an optimal way. The goal of this section is to describe how the SPEEG-5G MAC layer can be managed by a protocol stack which is mapped on a virtualized architecture, like it is reported in [2]. Chapter 3 provides the functional description of the MAC and details how the RRM and the MAC layers work together. Then, Chapter 4 describes how it can be implemented in real systems, refining the required functionalities of each building block. These two sections are the basis for understanding how edsa is managed and supported into the network system. Figure 1 below represents the edsa-capable protocol stack developed in the SPEED-5G project. The whole protocol stack is configured and managed by a virtualized centralized RRM (crrm) entity which deals with multi-user, multi-cell and multi-connection network capacity issues. The crrm objectives are the maximization of the system spectral efficiency across the different available frequency bands while being the main signalling interface between of the 5G small cells. The proposed functional split being set at the Packet Data Convergence Protocol (PDCP) level (in charge of both the control and data planes), allows having a common security framework for all the RATs of a cell. In order to natively support carrier aggregation on different possible RATs, the user plane includes a Multi-Path-TCP layer, for scheduling TCP services over different radio bearers in parallel. These TCP paths are embedded in different GTP tunnels in order to re-direct these services to one or more RATs of a cell or of multiple cells, in order to support Coordinated Multipoint (CoMP) and virtual multi-cell Multi Input Multi Output (MIMO) schemes. The control plane is handled by the 5G-RRC function which manages the control procedures of the supported RATs SPEED-5G Consortium Parties Page 19 of 82

20 Figure 1: SPEED-5G MAC Layer from cell side Figure 1 highlights how SPEED-5G MAC layer can handle multiple RATs and be configured by the higher 5G-Radio Resource Control (RRC) layer. The proposed MAC layer includes several different instances or parallel threads (depending on the retained virtualisation option) of RAT-dependent MAC functions, such as the resource allocation scheduling algorithms or the HARQ management. The main idea is to provide enough flexibility to add new RAT and provide backwards compatibility towards legacy RATs, such as LTE and WiFi. Each scheduler is not univocally associated to one or more MAC-Adaption Layer (5G Radio Link Control (RLC)) logical channels and it is important to note that a 5G-RLC can be potentially connected to different schedulers, for instance connected to an LTE and an LAA (Licensed Assisted Access) scheduler. The MAC layer has to be highly flexible and scalable, which can be obtained thanks to different functional splits in the virtualisation process. Although virtualization is (respectively not) suited to processes which are idle most of the time (resp. which require large computational resources), the most suitable virtualization split for the air interface protocol stack actually depends on the physical fronthaul characteristics., which constitutes the most common the bottleneck in up-to-date architectures. Depending on the stack deployment, the MAC may be virtualized or not. More details about this may be found at [4]. When the MAC is not virtualized, the deployed MAC contains the whole set of functions and the centralized Radio Resource Management (crrm) activates or deactivates them. In this case, flexibility is achieved by crrm (re)configuration procedures. When the MAC is virtualized, flexibility is also achieved by the SDR paradigm that allows the replacement of some functions of the protocol stack on the fly. New trends require new functionalities on the MAC. The proposed SPEED-5G MAC is also able to support channel coexistence in order to properly execute edsa. The channel coexistence is mainly for lightly licensed and unlicensed spectrum access, since the licensed spectrum is tightly regulated. For this reason, channel coexistence is mainly used for CA over lightly licensed spectrum access or access to unlicensed bands. The channel coexistence functionality is responsible for triggering the scheduling procedure ensuring the seamless usage of shared resources. Considering current technologies as WiFi, channel coexistence is responsible for knowing when a new transmission has to be initiated. Then, using this module, compatibility with current technologies is ensured SPEED-5G Consortium Parties Page 20 of 82

21 Furthermore, this functionality is ready to support, for instance, the required Licensed Assisted Access (LAA) for UL which is still under 3GPP study. Another important requirement is supporting MOCN where more than one mobile operator works over the SC and consequently, sharing the wireless resources. The proposed MAC supports it natively, since it has a scheduler per RAT and the scheduling policies are configurable. The wireless resources are finite and shared between operators, for that reason, the network has to be capable of establishing fairness policies. How the scheduler is configured is out of scope for this deliverable. It is important to note that, some of the optimal parameter of the edsa techniques should be shared between operators as they are global for the cell and the optimal ones for the network. On the other hand, there are parameters which each operator may modify depending on its own criteria. Although the proposed SPEED-5G MAC is decentralized, the design also allows the deployment of a centralized MAC supporting the well-known basic features for 5G networks such as massive MIMO, CoMP, or cloud-ran functions. 2.6 Summary This introductive chapter presents the background considered in the SPEED-5G project and which has been used for the MAC design. It gives an overview of the SCs categories and the requirements we have defined, derived from the use-cases elaborated in the WP3. It presents the protocol stack and MAC vision in SPEED-5G, in order to support the edsa feature. Based on this high level introduction, the next chapter presents the result of a project-wise work dedicated to the definition of the edsa concept which relies on a top-down approach. In particular, the next chapter digs into the SPEED-5G functional block diagram by defining the MAC functions which are required implementing the edsa mechanism and how MAC and RRM are working together SPEED-5G Consortium Parties Page 21 of 82

22 3 edsa concept development and MAC functional description This chapter is the overall functional description of the SPEED-5G MAC layer. It comes from a topdown approach which has been adopted in the project in order to define the concept of edsa, which has been brought into focus in the proposal. This concept development has been based on the elaboration of realistic and representative use-cases of edsa which have been described and represented by means of a set of functions [2]. In the same deliverable, an overall block diagram of the SPEED-5G system has been determined, composed of functions that are classified depending on whether they relate to MAC or RRM. The deliverable D4.1 [21] provides a description of the RRMrelated functions and their respective inputs and outputs (I/Os). This chapter takes the edsa use-cases as an input and intends to provide a functional description of the SPEED-5G multi-rat capable RAT which emerges as one of the main outcome of the project. MAC functions are detailed along with their I/Os and the interfaces between MAC and the neighbouring layers are highlighted. 3.1 edsa use-cases and overall block diagram This section is a summary of a chapter of the deliverable D3.2 [2] where the three scenarios which have been selected to illustrate the edsa concept are extensively described Scenario 1: Dynamic channel selection An initial situation is envisaged where multiple SCs are managed by a single network operator. The SCs are conveying a mix of traffic, that corresponds to the traffic in the SPEED-5G use-cases that are Extreme Mobile Broadband (xmbb), Ultra-Reliable Communications (URC), massive IoT (miot) and high-speed mobility. There would be an assumed variable amount of interference, degrading the QoS. A decision to select a band is made by the RRM, from a combination of licensed and unlicensed spectrum available, and possibly also to choose a different RAT depending on the context and the ability of the user terminal. The aims would be to meet some specified QoS requirement to allow cell selection as one mechanism to allow opportunistic sharing as an option, similar to TV white space (TVWS) to be backwards-compatible with existing LAA and LTE-Unlicensed (LTE-U) standards Scenario 2: Interference mitigation by load balancing This scenario relies on a group of neighbouring SCs managed by the same operator and sharing the same licensed spectrum. These SCs convey a mix of traffic composed of broadband, ultra-reliable communications, and IoT. Due to the high traffic load and the dense deployment of SCs, the level of interference may be too high, leading to an excessive degradation of the QoS/QoE. A centralized RRM controller decides to trigger a load balancing process. Based on the characteristics of the different available shared bands (bandwidth, central frequency, path loss, and regulations) and the estimation of the interference of these bands, part of the data traffic is steered to some nonlicensed bands. As far as control traffic is concerned, one assumes it is kept on the licensed spectrum. Regarding the offloaded traffic, at the MAC layer side, a suitable channel is selected for each traffic type and/or for each specified band. This can be done either by relying on prior available sensing measurements, or by resorting on the sensing capabilities of the small cells. Additionally, a MAC frame format is configured, using either a default configuration provided by the RRM part or through an adaptive MAC frame configuration SPEED-5G Consortium Parties Page 22 of 82

23 3.1.3 Scenario 3: Capacity maximization by traffic smart offloading Compared to the previous load balancing scenario, this smart offloading scenario relies on a single cell, which can be controlled in an equivalent manner by a centralized or a local RRM entity. We assume here that the RRM is mainly implemented in a centralized virtualized entity. Compared to scenario 2, here we consider that each cell needs to notify its own traffic load and send this report to the centralized entity which controls overall traffic load of each cell. This means that the centralized RRM monitors the capacity performance of the small cell, as well as the traffic load in the queues. In the same way as for load balancing, RRM decides what kind of traffic can be offloaded on a nonlicensed/lightly licensed/tv white spectrum band. Traffic with the most stringent QoS requirements is allocated on licensed spectrum and traffic with non-stringent packet error rate (PER) and latency requirements can be moved onto a supplemental carrier on unlicensed spectrum. Predetermined association rules (traffic class vs. spectrum band) could be established: e.g., high data rate (QoS: 2.6 GHz, Non-QoS: 5 GHz, 2.3 GHz (Authorized Shared Access, ASA), 700 MHz (TVWSs). RRM identifies the candidate spectrum chunk, providing at least maximum bandwidth or transmit power to the MAC (scheduler). The channel where the communication will be established within the band is selected by the MAC. This scenario can be extended via the concept of Device-to-Device (D2D) communication and demonstrate how the capacity of a small cell can be improved by considering the supplemental carrier chosen used by D2D in non-licensed/lightly licensed/tv white space spectrum. D2D can operate anywhere in the cell coverage and increase the capacity of the small cell by utilization supplemental carrier. We consider LTE-U/LAA kind of mechanism for D2D to co-exist with other RATs inside the cell. D2D communication in licensed/unlicensed bands is under operator control which offloads the traffic on supplemental carrier Overall block diagram Based on the analysis of the three edsa scenarios, a set of required functions have been extracted and classified depending whether they are related to MAC or RRM. This work, performed in WP3 and reported in deliverable D3.2 [2] has arrived at a SPEED-5G functional block diagram which is shown in Figure 2. D3.2 provides a complete description of the overall block diagram. In order to define the RRM capabilities, D4.1 gives a detailed view on the RRM sub-part, providing descriptions of each block along with respective inputs and outputs [21]. The following section provides the same description, applied on the MAC functions appearing in Figure MAC functions and interfaces for edsa support This section intends to drill down into the detailed description of each function of the MAC sub-part represented in Figure 2. This will make it possible to extract the list of inputs and outputs of each functional block as well as to define the interfaces of the MAC layer with the neighbouring layers (RRM and PHY layer) SPEED-5G Consortium Parties Page 23 of 82

24 Figure 2: SPEED-5G functional block diagram for edsa support MAC function and IOs MAC Controller This function is the entry point of the MAC layer, from an upper layer perspective. It dispatches the traffic in the queues either to the licensed or the unlicensed part of the access, depending on the RRM decision. More particularly, it takes the queues as an input and orients the control traffic to licensed spectrum access and the offloaded data traffic to the unlicensed spectrum access. It also provides the required control parameter to the MAC. Inputs From RLC: traffic queues (which are identified with QoS objectives) From RRM sub-part: o o From Load Balancing/Offloading Control : Offload pattern (e.g., LAA or LTE-U or LTE/WLAN link aggregation (LWA), or MuLTEfire) From RAT/Spectrum Selection/Aggregation : Selected unlicensed band with regulation rules Its sensed interference level (e.g., interference temperature) Forbidden channels, if any Maximum allocatable bandwidth, power per RAT and/or offloaded traffic type SPEED-5G Consortium Parties Page 24 of 82

25 Outputs Default MAC frame configuration Licensed band configuration and maximum allocatable resources Schedulers configurations Default parameters. For instance, Clear Channel Assessment (CCA) thresholds for channel assessment in the presence of coexisting systems (e.g., threshold for LTE detection, threshold for WIFI activity detection) To the licensed spectrum operation part: o o o Scheduler configuration Per logical channel, amount of data to be mapped in the next frame Band configuration and maximum allocatable resources (bandwidth, power) To the unlicensed spectrum operation part o o o o to Channel Selection : Selected unlicensed band (with regulation framework) Sensed interference level (e.g., interference temperature), Forbidden channels, if any Maximum allocatable bandwidth to MAC Frame design : MAC frame configuration to Interference-aware Contention Scheduler Maximum allocatable bandwidth, power per RAT, and/or offloaded traffic type To CCA/CSAT Default parameters. For instance, CCA thresholds for channel assessment in the presence of coexisting systems (e.g., Wifi) Licensed spectrum domain The function of the Non-Contention Scheduler block is responsible for allocating resources into transport blocks, depending on the load, service objectives and the instantaneous channel conditions of the active UEs. Common schedulers for LTE systems are for instance RR and PF in frequency and/or time domains. The scheduler has an interface with the user-plane upper layers to get the status of the queues and with the HARQ process, since retransmission of packets will be considered as transport blocks to be scheduled together with new packets. Inputs Outputs From RRM via MAC Controller : scheduler configuration, frame configuration, band configuration with maximum number of resource blocks, maximum transmit power From User plane upper layer : load with QoS objectives From HARQ : retransmission control From PHY : CQI updates from connected devices SPEED-5G Consortium Parties Page 25 of 82

26 To User plane upper layers : per logical channel, amount of data to be mapped in the next frame Unlicensed spectrum domain Channel Selection Compared to the licensed spectrum domain where the channel is selected by the RRM sub-part, we assume that for the unlicensed spectrum domain, the band is selected by the RRM whereas the channel is selected by the MAC sub-part, mainly due to the higher level of reactivity required by the unlicensed nature of the spectrum. The RRM sub-part can recommend a channel based on past experienced, predefined settings, or autonomous decision; but the recommendation can be ignored by the MAC, unless co-operation between base stations (BSs) is decided at the RRM level. At the MAC level, the channel selection consists in identifying in a given spectrum band (e.g. the 5 GHz band) what will be the actual spectrum chunk(s) where the traffic can be transmitted. The channel selection shall comply with the regulation, such as in the 5 GHz band, where a channel may be selected following the Dynamic Frequency Selection (DFS) rule specified in [22]. The channel selection relies on a sensing capability which can be implemented in many ways, like energy detection, waveform detection and protocol signalling packet recognition. This detection function is beyond the scope of SPEED-5G. Inputs Outputs From RRM via MAC Controller o o o o o Selected unlicensed band (with regulation framework) Sensed interference level (e.g., interference temperature) Forbidden channels, if any Maximum allocatable bandwidth Default frame configuration To Interference-aware Contention scheduler o Channel numbers to use for mapping PDUs on physical resource blocks CCA-CSAT This block performs a critical function which consists in implementing proper means to co-exist with other radio systems sharing the same band where the SPEED-5G device operates. In particular, this function is responsible of sensing the channel to assess whether it is idle or busy so that a transmission could be deferred if it happens that the channel is occupied by another system. For instance, this function is mandatory within the European regulatory framework for the 5 GHz band since it requires a Listen-Before-Talk (LBT) procedure. In a contention-based view, this function is therefore responsible of triggering the non-licensed scheduler once the channel has been deemed as idle. In the particular case of LTE-U, this block implements the Carrier Sense Adaptive Transmission (CSAT) [23] which is the interference management between LTE and Wifi in the unlicensed spectrum, as it has been specified by the LTE-U forum SPEED-5G Consortium Parties Page 26 of 82

27 Inputs Outputs From User plane higher layers : Ready to transmit notification From RRM via MAC Controller : Detection thresholds for coexistence (default or optimized thresholds) To Interference-aware Contention Scheduler : channel idle flag To Spectrum Manager and Inter RAT cooperation : Wideband SINR Contention based scheduler This function is invoked to decide the amount of data to be allocated per device for achieving the required QoS, and to map the MAC PDUs to be sent to the physical layer, once a transmit opportunity is obtained (e.g., the channel is idle after a CCA). It mainly depends on frame structure, the load available at upper plane higher layer level, and CQIs. As in the licensed spectrum domain, the scheduler can be implemented following many algorithms, which is beyond the scope of this deliverable. Inputs Outputs From RRM via MAC Controller : scheduler configuration From user plane upper layer : load with QoS objectives From HARQ : retransmission control From Channel Selection : band configuration with maximum number of resource blocks, maximum transmit power From MAC Frame Design : default (or optimized) frame structure From PHY: CQI updates from connected devices To User plane upper layers : per logical channel, amount of data to be mapped in the next frame MAC Frame Design This function is responsible of providing a suitable MAC frame structure to the scheduler. By a suitable frame structure is meant one which is able to manage interference level or to adapt to the current load. This block is part of the smart access of SPEED-5G, since depending on the experienced conditions, the MAC frame will be modified so that a target objective (e.g., capacity, energy, interference level) can be optimized. For instance, the balance between contention-based and schedule-based access can be modified to expand the schedule-based access when the interference in a shared band is high; the complementary operation (i.e., increasing the duration of the contention-based access within the frame) is also possible in the case of transmitting moderate traffic in a sparsely crowded band. In the same way, this function can take the decision of modifying the TTI duration, according to the interference and load conditions. The optimized MAC frame configuration is provided by the RRM sub-part (namely Inter RAT cooperation ). But when this function is not available at RRM level and/or the SC is operating in an autonomous mode, this block is able to take this kind of decisions to optimize the MAC frame SPEED-5G Consortium Parties Page 27 of 82

28 Inputs From RRM (inter RAT cooperation) via MAC controller: Frame structure If centralized RRM is not available o From KPIs : CQIs and performance indicators like block error rate (BLER) o From User plane upper layers : load with QoS objectives o From Spectrum manager : Wideband SINR measured on the band and channel Outputs To unlicensed spectrum scheduler : frame structure PHY Adaptation Layer This function has one main task, which is transforming MAC SDUs into MAC PDUs to be sent to the PHY. In particular, the transmission of a packet requires changing a MAC SDU into a PDU, by adding the MAC header (preamble, start and end of frame). It also acts as a kind of API between the MAC and the PHY layers, as it provides means to send bidirectional commands like MAC PDUs or measurements done at the PHY level (i.e., wideband SINR, CCA, and CQIs). Inputs Outputs From User plane higher layers : MAC SDUs From RRM or MAC: PHY measurements commands (CCA, wideband SINR) From PHY: Measurement reports, erroneous or correct packet reception indication, received PDUs To PHY: PDUs to transmit, HARQ retransmission control information, measurements commands To CCA-CSAT : channel idleness status To scheduler: erroneous or correct packet reception indication, CQIs updates To User plane upper layer : MAC SDUs MAC interface with RRM and upper layers From the previous subsections, a function description of the interface of the MAC layer with RRM sub-part and entities located in upper layers can be derived, so that it can be further elaborated in the following sections when the implementation is reported. By functional description is meant the elaboration of a list of the different signals and messages which are exchanged between the MAC and RRM entities, user-plane upper layers (e.g., RLC) and SPEED-5G virtualized entities like the spectrum manager SPEED-5G Consortium Parties Page 28 of 82

29 MAC Function item function/entity originator domain MAC controller Offload pattern RAT/spectrum selection/aggregation Licensed spectrum scheduler Channel selection CCA-CSAT Unlicensed spectrum scheduler MAC frame design PHY adaptation layer Schedulers (licensed and unlicensed domains) Scheduler configuration, Amount of data scheduled in the next frame, Maximum allocatable resources (bandwidth, power), default frame configuration Optimized frame configuration Selected unlicensed band (with regulation framework) with interference level, forbidden channels, maximum bandwidth, default frame configuration Detection thresholds for coexistence Scheduler configuration, band configuration and max allocable resource Frame configuration Opt. CQIs and BLERs Load with Qos requirements Wideband SINR RRM sensing commands MAC SDUs Queues with QoS requirements RAT/spectrum selection/aggregation Inter RAT cooperation RAT/spectrum selection/aggregation RAT/spectrum selection/aggregation (default) Inter RAT cooperation (optimized) RAT/spectrum selection/aggregation (default) RAT/spectrum selection/aggregation (default) Inter RAT cooperation (optimized) KPIs database RLC Spectrum manager RAT/Spectrum selection/aggregation RLC RLC CCA-CSAT Ready to transmit RLC PHY adaptation layer MAC SDUs RLC RRM User plane upper layers Table 2: Interface from RRM and upper layer to MAC SPEED-5G Consortium Parties Page 29 of 82

30 MAC Function item function/entity destination domain CCA-CSAT Wideband SINR Inter RAT cooperation Schedulers (licensed and unlicensed domains) Schedulers (licensed and unlicensed domains) Queues status Data scheduled in the next frame RAT/spectrum selection/aggregation CCA-CSAT Wideband SINR Spectrum manager PHY adaptation layer Channel selection MAC SDUs Channel (central frequency and bandwidth / channel numbers if any) to use for scheduling RLC RLC Unlicensed spectrum scheduler CCA-CSAT Channel idle notification Unlicensed spectrum scheduler Frame design Frame configuration Unlicensed spectrum scheduler PHY adaptation layer HARQ indication CQIs updates Unlicensed spectrum scheduler Licensed and unlicensed spectrum schedulers RRM Upper layers MAC Table 3: Interface from MAC to RRM, upper layer and MAC functions MAC interface with physical layer In the same way than the previous section, this section explains the interface between MAC and PHY layers. direction MAC function item MAC to PHY PHY adaptation layer PHY adaptation layer PHY adaptation layer PDUs to transmit Measurement commands PHY to MAC PHY adaptation layer Received PDUs Schedulers CCA-CSAT HARQ retransmission control information CQI updates from connected devices, Packet error indicator Channel idleness status Wideband SINR Table 4: Bidirectional interface between MAC and PHY layers SPEED-5G Consortium Parties Page 30 of 82

31 3.3 Summary In this chapter, the SPEED-5G proposal of edsa has been detailed from the MAC perspective, following a top-down approach adopted in the project. Indeed, edsa use-cases are identified and detailed in WP3, whereas the RRM components are described in WP4. This chapter collects the description of the MAC functions involved in the selected edsa use-cases, giving their respective inputs and outputs. Also a functional description of the interfaces of the MAC layer has been derived, in order to prepare the implementation work that will be addressed later on in the project. The following chapter shows how the functional block diagram (see Figure 2) can be modified so that it can fit with the SPEED-5G implementation from the SC and UE perspectives SPEED-5G Consortium Parties Page 31 of 82

32 4 An edsa-capable MAC design This chapter explains how the scalable and flexible MAC functional description presented in the previous chapter (capable of coordinating the use of radio resource across heterogeneous RATs and multiple bands) can be depicted in a way amenable to implementation in real devices. A first implementation for SCs is presented showing how the functions of chapter 3 can be grouped and interfaced one with another and with neighbouring layers as well. In addition, the MAC architecture of the UE side is provided, showing how it differs from the one of a SC. 4.1 Small cell MAC design One of the main requisites of the MAC layer is supporting a high flexibility and scalability. The SPEED- 5G MAC is highly flexible and reconfigurable since the proposed entities may be deployed on-thefly depending on the scenario but may also have different features depending on the end-user requirements and the underlying PHY RAT. Scalability is achieved providing support for deploying several scheduler instances on the fly, linked to the same PHY or to different PHYs for optimizing the existing radio resources of a given cell or for dynamically managing the overall capacity of the SC network. Also, this can be done while providing real-time access and data integrity to the different scheduler entities ensuring thus service continuity and QoE. Figure 3 shows the required software modules for the implementation of the SPEED-5G MAC layer together with the way they are interconnected. The messages exchanged between the MAC and the other layers use Service Access Points (SAPs) for both the control and the data planes. The following is a short introduction to the entities contained in Figure 3. The entities and the interfaces will be detailed later in this subsection. The MAC contains as many schedulers as the number of supported RATs and SPEED-5G technology aims to coordinate schedulers of different cells in order to support edsa operation in licensed, unlicensed and lightly licensed RATs. A scheduler ( Scheduler RAT ) is composed of the (scheduling) Algorithms, QoS, HARQ, ReTx (retransmission) and Tx (transmission) modules. Providing a brief module introduction, the QoS module contains the QoS policies that have to be applied; the scheduling decisions include the assigned resources to each scheduled UE each TTI, the HARQ stores each transmitted Transport Block; the set of UEs with data to be retransmitted each TTI is provided to the scheduler by the ReTx module; the set of UEs with new data to be transmitted is provided to the scheduler each TTI by the Tx module; the Algorithms module provides, each TTI, the radio resources and the bands assigned to each UE. The Tx/Rx entity is responsible of the (De)multiplexing logical channel and (dis)assembling upper layer PDU packets, and also executes the channel coexistence procedures for lightly licensed or unlicensed spectrum access procedures. The KPI Collector entity stores the defined KPIs for further simulation analysis while the Sensed Data entity receives the sensing measures done for lightly and unlicensed bands. The MAC proposed in SPEED-5G covers licensed, lightly licensed and unlicensed spectrum operation modes. It is important to note that not all the described entities require being deployed. Some of them, as for instance the QoS module of Scheduler RAT, are optional. As a result, a specific, efficient and flexible MAC is used depending on the end-user services SPEED-5G Consortium Parties Page 32 of 82

33 4.1.1 MAC interfaces Figure 3: MAC interfaces and entities from cell side The interfaces define how and which data is exchanged by the instantiated entities. The interfaces are divided between general MAC configuration and specific MAC configuration for scheduling procedures. The nomenclature reserves the initial C on the interface name for control messages. Instead, interfaces without the initial C transmit data. Not that the foundations of these messages are defined in [20]. CMAC SAP: This interface contains the control messages for a complete MAC configuration, including the configuration of the KPI Collector, the Sensed Data or the Scheduler RAT modules. This interface allows initializing, updating, or removing any of the configured functionalities. MAC SAP: This interface is responsible to send and receive data frames from upper layers. The required messages and their content require being defined. PHY SAP: The MAC layer is connected with the PHY by the PHY SAP interface. For DL traffic, the final MAC frames to be sent to the UEs are transmitted through the PHY SAP interface. Furthermore, the scheduling decision is required on the PHY in order to map, for instance, the assigned resource block to physical resource blocks or to apply the defined channel modulation. For UL traffic, it is the PHY which sends to MAC, for instance, the received UE PDUs, or the sensed channel data, to the MAC layer. CPHY SAP: It is responsible for sending the configuration data to the PHY including all the required information to perform edsa. CSCHED SAP: It is the interface for sending control data for Controller Configuration SPEED-5G Consortium Parties Page 33 of 82

34 and Controller Scheduler RAT. This interface is used to configure and update the required scheduling parameters of the cell, the UEs parameters, the supported schedulers and QoS characteristics. Furthermore, the configuration associated to UEs or QoS may be released by RRM or OAM when required. Other configuration parameters as transmission power or transmission bandwidth are also provided by the CSCHED SAP interface. All the modifiable parameters have to be identified and set into the message structure. At least, messages for setting/updating, requesting and removing data are required. SCHED SAP: This interface contains the information used by the scheduling algorithm. Then, the messages should contain the triggering data, the user-data buffers, the UE reports as, for instance, the CQI measures. The required information has to be identified for DL and for UL transmissions MAC entities KPI Collector: it stores the required MAC KPIs identified. These KPIs are sent to other layers, in order to provide the required inputs for their internal procedures and algorithms. In order to do that, the MAC layer defines a new KPI interface. The way the data is retrieved or sent is out of scope of this document. Sensed Data: it contains the sensing data retrieved for the lightly and unlicensed spectrum band. This information is used by the scheduler to allocate radio resources on these bands but it is also sent to upper layers when required for cell coordination procedures. The sensed data contains the result of the different edsa techniques used. Configuration: it contains the common and specific configuration parameters related to the cell, the UE and the scheduler. The cell, UE and scheduler are configured by the RRM or the OAM, depending on the situation and the scenario. For instance, the initial manual configuration is done by the OAM while the auto-configuration is done by the RRM algorithms that are checking the reported KPIs. The configuration data has to be easily accessible and maintain the integrity for scalability purposes. o o o Cell Configuration: This module contains all the parameters related to the cell and used to configure the cell functionalities. The common parameters are, for instance, the supported RATs whereas the specific parameters per RAT refer, for example, to the frequencies supported by each RAT. UE Configuration: This module keeps the characteristics of the connected UEs and, depending on the MAC, the characteristic of historical connected UEs. This information is shared with RRM in order to properly configure the MAC. This information is used by the MAC in order to schedule each UE. Scheduler Configuration: This module contains information about the supported schedulers and common configuration parameters. Besides, each RAT may have a specific scheduler that has its own scheduling algorithms supporting specific parameters. This entity also contains specific data per RAT. The scheduler of each RAT and its configuration parameters are set by the RRM entity. Controller: This entity works as a dispatcher of the control messages. It is responsible of the message routing, where the input messages are sent to the appropriate functional block while, the output messages are sent to the appropriate external layer. The Controller entity also contains the parsers in order to translate the received messages into internal data structures. Then, the internal structures may be shared between different MAC entities. On the other hand, it contains the output parsers. The output parsers are used when a trigger for sending a message is received. Then, it reads from the internal structures and sends the message according to the defined API between the source and the target layer. Scheduler RAT: it contains all the required functionalities performing the physical radio SPEED-5G Consortium Parties Page 34 of 82

35 resources allocation per UE. The MAC design is flexible allowing optional entities such as QoS or HARQ. The final deployment is based on the end-user service so, for instance, the MAC layer may not include the QoS and the HARQ modules since they are specifically designed for supporting non-critical sensor networks. In order to be scalable, the HARQ, QoS, ReTx and Tx modules have to be accessible in real time, allowing multiple instances of the same MAC layer ensuring data integrity. The HARQ, ReTx and Tx modules provide the UEs that should be scheduled each TTI. The provided UEs do not have the same priority to be scheduled so, the higher priority ones are the UEs provided by the HARQ module while the lower priority ones are the one the provided by the Tx module. o o o o o HARQ: The main responsible of the retransmission mechanism. It keeps previous scheduling decisions in order to reduce the system latency when a retransmission is required. The HARQ module is divided into slots, each one storing a new transmitted packet. Then, the HARQ has several slots, depending on the cell configuration, for each connected UE. Therefore, when the scheduler has the complete data to be transmitted for each UE and, before this data is sent to the PHY, it is stored into the HARQ. When the stored data is not to be transmitted, it is replaced with the new transmitted data. QoS: This module is responsible to ensure the QoS defined for each service type. For instance, LTE standardizes several QCI defining the traffic priority, the maximum packet delay and the allowed packet losses. Then, each traffic type is associated to a single QCI. This module is the responsible to identify the UEs that, for QoS reasons, have to be scheduled each TTI achieving the minimum quality requirements. ReTx: This module provides to the Algorithms module all the UEs that, each TTI, have a pending retransmission. The specific data to be retransmitted has to be previously stored into the HARQ module. Tx: This module provides to the Algorithms module all the UEs with new data to be transmitted. Algorithms: This module is divided into Time Allocation and Frequency Allocation algorithms. The Time Allocation algorithm decides whether there are any UEs that are to be scheduled in the next available time resource. More specifically, in the case of licensed bands, this module decides which UEs will be scheduled in the next available TTIs. In the case of unlicensed or lightly licensed bands there is no notion of TTIs due to the non-deterministic transmission scheme and the use of LBT; therefore this module uses information available from the Sensed Data entity to know when a transmit opportunity is available and be aware about the time when users can be scheduled. The UE selection is done by considering the users QoS, as well as the ReTx and the Tx data. Once the prioritization of users is done and the set of UEs to be scheduled are selected, the Frequency Allocation" algorithm decides the specific frequency resources assigned to each UE. The frequency resources may be specific resource blocks like in LTE but also available channels or technologies like WiFi. This module is able to allocate UEs on licensed, lightly or unlicensed bands, depending on the configuration provided by the RRM entity. Also, the module includes the required algorithms for DL and UL independently. Tx/Rx: This entity is responsible for triggering the Scheduler RAT entity. The trigger may be raised each 1ms, as is done on LTE, or when the channel is free to be used in lightly licensed or unlicensed bands. This entity is also responsible to multiplex different logical channels and (dis)assemble SDUs received from upper layers o o (De)Multiplex: It performs the (de)multiplex of the logical channels including control and traffic logical channels creating the final package to be send to PHY. (Dis)Assembly: The assembly procedures are responsible to assemble the upper SPEED-5G Consortium Parties Page 35 of 82

36 o layer MAC SDUs into the MAC PDUs to be sent to the PHY. This block is responsible to add the MAC headers. Channel Coexistence: its main function is to raise the scheduling trigger when the shared channel is suitable to be used without interfering with transmissions of other cells. The trigger order is based on different techniques as, for instance, the Clear Channel Assessment (CCA), the Carrier Sensing Adaptive Transmission (CSAT) techniques for channel coexistence used for perform LAA or LTE-U. 4.2 UE MAC design A scalable and flexible MAC that is able to coordinate the use of radio resources across heterogeneous RATs has been described above for the SC. Going forward, there is also the need to look into a futuristic scalable, flexible, and unified 5G UE MAC that is able to support all the services and use cases described in D3.2 [2]. From the spectrum usage perspective, Speed-5G proposes a novel and flexible 5G device MAC in a unified framework. The envisioned scalable and flexible 5G UE MAC architecture is shown in Figure 4. The proposed MAC can operate in licensed, lightly-licensed and unlicensed spectrum. As each RAT has a specific PHY, part of the MAC flexibility is achieved by managing multi-rat depending on the configuration. The PHY is out of the scope of SPEED-5G and for that reason, Figure 4 shows a unified PHY block. Upper Layers (Control Plane) Upper Layers (User Plane) Data Flow Control Flow UE MAC QoS Policies Control Unit Spectrum Sensing Control Sensed Data MAC configuration Coexistence CCA/CSAT Lightly-Licensed Tx Opportuny Scheduler TX Buffer (De-)Multiplexer HARQ Process Others Rx Buffer PDU Filtering Random Access Control Scheduler Tx/Rx Frame Aggregation Frame (De-)Fragmentation PHY Figure 4: Scalable and Flexible 5G UE MAC Architecture The SC and the UE must be synchronized in functionalities and therefore the description of the common sub-blocks in the SC MAC at section 4.1 holds true for the UE MAC. The following gives some brief description of the sub-blocks in Licensed and Unlicensed Blocks in Figure 4 that differ in SPEED-5G Consortium Parties Page 36 of 82

37 functionalities to the ones described in section 4.1 for the SC MAC Licensed Spectrum Block The operation of this block is under the control of the SC, as it is currently in legacy LTE-A. Depending on the availability of licensed bands and carriers, this block can be configured to address traffic queues of stringent QoS; e.g., non-delay tolerant. Transmit opportunities (Tx-Op) through Unlicensed/Lightly-licensed Spectrum Block can supplement CA in UL of licensed carriers and unlicensed carriers a similar concept to LAA-UL. Scheduler: It is worth mentioning that the scheduler at the UE MAC is not the same in functionality to the scheduler at the SC MAC. In the Licensed Spectrum Block in the UE MAC, the scheduler is similar to the one currently standardized in LTE-A for UL [24]. To ensure fairness and at the same time handle traffic of stringent QoS, the traffic queue from Upper layers (user plane) will have different priorities assigned, as well as bit rate at which the data in those queues have to be transmitted. The creation of MAC PDUs will be based on this priority configuration. Those traffic queues (or logical channel in LTE) with high priority will be served first and according to their configured bit rate, and the procedure will continue until the max size of the MAC PDU (or max Transport block size in LTE) is achieved. This ensures that if queues with high priority do not have any data available to transmit in UL, the next high priority queue can be served, avoiding any starvation. Random Access Control: Most of the time, the UE will remain in IDLE mode in order to save battery. When it has data pending to be transmitted, and there is no dedicated radio resources assigned to it, the device initiates a Random Access (RA) procedure in order to be connected to the network. In LTE-A, it is referred to as RACH procedure. A UE that does not have any dedicated radio resources to use to send data in UL, would first select from among a number of preambles and transmit this randomly in the shared medium. It is possible that other UEs would select the same preamble and in this case, two or more UEs transmitting the same preamble would result in collision and would lead to the SC not responding to any of those UEs. Whenever there is a collision, the SC will initiate a back-off procedure and the affected UEs will try to transmit their preambles again, most likely at different time slots to avoid further collisions. The waiting time between preambles (re-)transmission is configured by the SC. SC can also configure the UE with a preamble, one that the SC is sure will not collide with other shared preambles transmitted from other UEs. This can give priority to certain UEs or in most cases prioritize certain transmissions against others. PDU Filtering: This filtering functionality in DL will aid D2D communication [24]. The impact on the UE MAC design will be further investigated in the next deliverable, D5.2. Unlicensed/Lightly-Licensed Spectrum Block This block is used to handle transmission and reception of traffic through the unlicensed or lightlylicensed spectrum. It facilitates the offloading edsa techniques at the RRM and MAC at the SC. Additionally, it is used to augment spectrum aggregation at the Licensed Spectrum Block by detecting for Tx-Op in the unlicensed spectrum, whilst ensuring fair sharing of this spectrum and coexistence with other devices. Scheduler: The scheduler in Unlicensed/Lightly-Licensed Spectrum Block as opposed to the one in the Licensed Spectrum Block is rather different. Frame aggregation is a technique introduced in n [25] that can be used to combine a number of frames into a larger frame and transmit this at a Tx-Op. Frame aggregation could be either Aggregated-MAC SDU (A-MSDU) or Aggregated-MAC PDU (A-MPDU). Each one of these aggregated schemes could best serve traffic with certain QoS. For instance, A-MSDU can best serve traffic with tighter delay requirements and at the same do not require that every aggregated SDU is acknowledged, e.g., in the VoIP use case; whilst A-MPDU can be used for traffic of best-effort QoS such as video streaming. The Scheduler module in the Unlicensed/Lightly-Licensed SPEED-5G Consortium Parties Page 37 of 82

38 Spectrum Block uses the QoS of the different traffic flows from upper layers together with the spectrum conditions to decide which aggregation scheme to use at each Tx-Op. Spectrum Manager Block Some of the edsa techniques proposed by Speed-5G require spectrum sensing to efficiently utilize the lightly-licensed and unlicensed spectrum and fairly co-exist with primary users. This spectrum sensing is not limited to only SC. IoT devices using unlicensed and lightly-licensed spectrum should autonomously decide when to access the spectrum, and they have to make autonomous decision from the sensing data stored or configurations from SC based on sensing data forward by the UE. Sensing algorithms and limited processing of the sensing data and storage of the sensed data will be handled by Spectrum Manager Block. 4.3 Summary This chapter provides a first description of how SPEED-5G MAC framework can be translated into an implementable version for both SCs and UEs. In particular, it presents how the MAC functional blocks can be configured to that multiple RATs and schedulers can be jointly used to implement edsa in real devices. Compared to the list of functions defined in chapter 3, additional features like KPIs or spectrum sensing measurements collectors which are required to facilitate the communication with the RRM have been defined. Also, the framework revisits the MAC functional block diagram by grouping and refining the different blocks in order to arrive at a description which clarifies the a user plane and control plane interfaces. The next chapter is dedicated to the design of MAC functions which are required to implement edsa; in other words, the documents defines so far the MAC framework and the next chapter details some of the algorithms which are composing it SPEED-5G Consortium Parties Page 38 of 82

39 5 Design of MAC functions for edsa This chapter explains the proposed MAC algorithms operating on both unlicensed and licensed spectrum in the small cell environment. For the MAC algorithm design, the heterogeneous traffic type and the spectrum characteristics (i.e., licensed vs. unlicensed) are considered to implement the edsa concept. The MAC algorithms have been designed with the assumption that a FBMC PHY layer is available and can be used as one RAT on the various spectrum regimes. All the MAC algorithms are designed assuming they are linked with the SPEED-5G virtualized architecture defined in [2]. In addition, a self-organizing MAC algorithm capable of operating autonomously without the information of spectrum characteristics is also included to cover the case when the small cell can t rely on the SPEED-5G virtualized architecture. 5.1 FBMC MAC operating in unlicensed spectrum This section reports the design and orchestration of blocks involved in the non-licensed path of the MAC layer; referring to Figure 2, it particularly focuses on the MAC Frame design, the CCA/CSAT and the Interference-aware contention scheduler functions. More precisely, they correspond to the access of a radio bearer on an unlicensed band, assuming a FBMC PHY. This section starts with the assumptions made on the PHY layer and describes how the access is managed for two kind of traffic type: miot and xmbb. The two access schemes are validated on a system level simulator using synthetic and realistic scenarios. Also, MAC layer is implemented in Task 5.3 on a custom HW platform where FBMC is planned to be ported. Compared to the list if functions mentioned above, the PHY adaptation layer is implemented to act as an interface with the FBMC PHY Physical layer assumptions FBMC general principles FBMC was introduced in the 1960s by [24][27]. It is a multi-carrier modulation, but unlike Cyclic Prefix based Orthogonal Frequency Division Multiplexing (CP-OFDM) where the sinc kernel filter ensures subcarrier orthogonality with respect to the FFT operation, FBMC uses a prototype filters that filters each subcarrier. A proper design of the prototype filter enables to trade time and frequency localization, and thus to control adjacent carrier leakage ratio (ACLR). Because the prototype filter s response spreads each subcarrier over several adjacent subcarriers, another dimension is used to restore orthogonality. In the frame of this project, the offset QAM (OQAM) approach is considered. OQAM consists in a complex to real conversion where real and imaginary parts of each complex symbol are multiplexed in consecutive time samples into Pulse Amplitude Modulation (PAM) symbols (Figure 5). In order for this pre-processing not to impact data rate, the PAM symbols are up-sampled by a factor of 2. Figure 5: OQAM pre-processing in FBMC transmission chain SPEED-5G Consortium Parties Page 39 of 82

40 Then, the output real numbers are multiplied by an offset QAM sequence to form a new complex symbol: h k,l = ( 1) kl (j) k+l e k,l. These symbols are consequently filtered through a polyphase network G(z). The FBMC modulator and demodulator are shown in Figure 6. Figure 6: FBMC modulator and demodulator The depth K of the prototype filter governs the level of adjacent channel leakage (ACL), like shown in Figure 7, where FBMC waveform frequency localisation can be compared, for several values of K, with genuine (CP-)OFDM and another waveform called Universal Filtered Multicarrier (UFMC) [46]. On this figure, on can observe that the leakage on the adjacent channel is as high as -13 dbc for OFDM, where as it can be as low as -60 dbc for FBMC for K=4. Figure 7: Power spectral density of OFDM, UFMC and FBMC-OQAM with several values of K. Compared to OFDM, this excellent spectral localisation of FBMC signal implies a set of characteristics a MAC layer can exploit to optimize the medium access: Relaxed synchronicity: multi-user communications in CP-OFDM require a very tight synchronisation to avoid inter-carrier interference. Indeed simultaneous transmissions in contiguous bands have to respect a timing misalignment smaller than the cyclic prefix length SPEED-5G Consortium Parties Page 40 of 82

41 In comparison, FBMC can gracefully tolerate asynchronous communications, because of the very small of adjacent carrier interference. In terms of MAC design, this means that UL FBMC communications could be set-up in such a way that devices can transmit on contiguous bands, without being synchronised. Reduced guard band: Thanks to its very low ACLR, it has been shown in [28] that FBMC can tolerate a very small guard band, compared to other multi-carrier waveforms as one intercarrier spacing suffices. The MAC layer can thus manage in an efficient way the allocation of (thin) resource blocks, even if they are fragmented and spread in the band. When coupled with the previous feature, this offers the capability of efficiently using the band. Relaxed uplink power control: Along the same idea of very low ACLR, the UL signal in contiguous bands can be transmitted without a strict power control. Indeed, provided that ACLR is more than 60 db, two simultaneous signals on contiguous channels can arrive on a SC with a very large difference of UL received power. More details about the FBMC modulation and the filter coefficients can be found in [29], whereas implementation issues are presented in [30]. The remainder of the section presents first the PHY options we have selected and presents the MAC design for both IoT and broadband traffic operating in unlicensed spectrum Narrowband PHY description This section provides various aspects of PHY specifications for IoT devices operating in the 868 MHz frequency band based on a FBMC scheme. It is worth mentioning that this work is not addressing the lowest end of IoT devices, requiring a very low data rate (e.g., lower than 1 kbps) with an extremely simple channel access scheme. Considered spectrum band The operating band of the network is the 868 MHz ISM band, in which different systems may exist. In this shared band a low channel bandwidth is generally available; therefore the devices can operate in this band while respecting the regulatory requirements [31]. In this work, we consider that the whole system will be operated on a 3 MHz wide carrier, located in the MHz band. In this band, the regulatory requirements specify that the end-devices may transmit on any available channel at any time through respecting a maximum transmit power of 25 mw and a duty cycle < 1%. Channel coding A convolutional code with constraint length 7 and 1/2 coding rate (generator polynomials are 171 and 133 in octal) is used. Tail bits of length 8 are added at the end of the input stream to force the encoder to its all 0 initial state. For other code rate (R = 2/3, 3/4, 5/6), puncturing can be performed based on the mother convolutional code of rate 1/2. Modulation and coding scheme The binary input is converted into complex symbols using Gray-coded constellation mappings. The supported modulations include BPSK, QPSK, 16-QAM and 64-QAM. Various MCS considered in the IEEE standard are used as listed in Table 5 [47]. The amount of information that can be sent within a slot depends on the slot duration and the MCS used. The available MCS ranges from MCS0 to MCS7 and their corresponding data rates range from 86 kbps to 874 kbps (for a packet payload size of bytes). In case of beacon, control and command frames; the MCS 1 (QPSK, coding rate 1/2) should be used. Data and corresponding acknowledgement (ACK) frames must be transmitted using the same MCS (the one specified for data frame) SPEED-5G Consortium Parties Page 41 of 82

42 MCS modulation coding rate data rate (Kbps) 3 0 BPSK 1/2 [86-90] 1 QPSK 1/2 [ ] 2 QPSK ¾ [ ] 3 16-QAM 1/2 [ ] 4 16-QAM 3/4 [ ] 5 64-QAM 2/3 [ ] 6 64-QAM 3/4 [ ] 7 64-QAM 5/6 [ ] Table 5: Narrowband modulation and coding schemes Modulation parameters The maximum number of subcarriers (N) is set to 64 with an inter-carrier spacing f of 60 khz. The stream of complex symbols is divided into groups of N c complex symbols, where N c the number of non-null modulated subcarriers (active subcarriers). FBMC modulation is then applied at the transmitter and at the receiver, where a preprocessing step (complex-real conversion), a filtering step and a sum-overlap step are carried out. The transmitted time of FBMC signal depends on the overlapping factor (K), the number of FBMC symbols (N symb ) and the number of subcarriers (N) : T signal = (KN + (2N symb 1) N 2 )/F s, where F s is the sampling frequency which is equals to F s = N f = 3.84 MHz. Table 6 shows the parameters for FBMC modulation retained in the case of IoT traffic operating in the 868 MHz band. Parameters value / sub-band value/ total Band N N c 3 3*12 [-18, 17] f 60 khz 60 khz K 4 4 Bandwidth B 180 khz 2,16 MHz F s = N f 3.84 MHz 3.84 MHz Table 6: Narrowband FBMC parameters Broadband FBMC physical layer description This PHY mode can be used in any spectral chunk where 10 MHz and 20 MHz carriers can be allocated; for instance 5 GHz, 3.5 GHz, 2.3 GHz bands could be suited for this configuration. Note that the TVWS scenario is a particular case, due to the sharing with TV broadcasting signal characteristics. Indeed, channelization shall be similar to the TV multiplex bandwidth, meaning 8 MHz band in Europe and 6 MHz in the US. The interested reader can refer to [1] to get a detailed specification of the IEEE standard, which is based on an 8 MHz FBMC physical layer. This subsection provides the specification of a 10 MHz and 20 MHz configuration of the FBMC physical layer, which is assumed for broadband communications. Modulation and coding scheme: The binary input is converted into complex symbols using Gray-coded constellation mappings and the 3 data rate depends on the packet payload size. It has been computed for bytes SPEED-5G Consortium Parties Page 42 of 82

43 supported modulations are QPSK, 16-QAM, 64-QAM and 256 QAM. The channel coding process relies on a set of two possible coding options which are turbo or convolutional coding of rate 1/3, like for the LTE physical layer in downlink [32]. After the channel coding stage, rate adaptation is applied, leading to a wider range of actual coding rates [32]; we assume that rate adaptation leads to coding rates of 1/3, 1/2, 2/3, 3/4 and 5/6. Modulation and coding scheme is adapted during the communication so that throughput of devices is maximized when the channel conditions they experience are good. In the same way it is done in LTE, this is made possible by means of measuring the SINR on reference signals and reporting the CQI to the SC. The definition of reference signals are beyond the scope of this document (one can refer to [32] to get a description of the LTE procedure). Assumption is made that data packets are turbo-coded whereas control (short) packets (beacon, CQI updates, transmission requests) are coded using the convolutional coder. Table 7 shows the different MCS available for 10 MHz and 20 MHz bands. Because of the effect of the prototype filter (K factor), the actual throughput of a FBMC link depends on the number of bits the packet contains. In Table 7, we give the asymptotical throughput obtained for a packet of an infinite length. MCS modulation coding rate efficiency throughput 10 MHz (Mbps) QPSK QPSK QPSK QPSK QPSK 16-QAM 16-QAM 16-QAM 16-QAM 16-QAM 64-QAM 64-QAM 64-QAM 64-QAM 64-QAM 256 QAM 257 QAM 258 QAM 259 QAM 260 QAM 1/3 1/2 2/3 3/4 5/6 1/3 1/2 2/3 3/4 5/6 1/3 1/2 2/3 3/4 5/6 1/3 1/2 2/3 3/4 5/ throughput 20 MHz (Mbps) Table 7: Modulation and coding schemes of FBMC PHY for 10 MHz and 20 MHz bands Modulation parameters In this work, we assume to rely on a particular implementation of FBMC called Frequency Spreading- FBMC (FS-FBMC) [30]. Performance study shows that this implementation allows reaching the same level of performance of the UL PHY of LTE (based on Single Carrier (SC)-FDMA) even when the number of sub-carriers is reduced by a factor of 4. In other word, if the 10 MHz option of LTE is based on 1024 sub-carriers and an inter-carrier spacing of 15 khz, an equivalent FBMC PHY would be based on a 256 carriers with a 60 khz inter carrier spacing. As a consequence of this subcarrier number reduction, the Peak-to-Average Power Ratio (PAPR) obtained with the FBMC-256 is similar to the SC-OFDM, meaning that we can use the same SPEED-5G Consortium Parties Page 43 of 82

44 modulation scheme for both the UL and the DL, which is another very interesting feature of FBMC for MAC design. parameters 10 MHz band 20 MHz band N N c f 60 khz 60 khz K 4 4 Bandwidth B 9 MHz 18 MHz F s = N f MHz MHz Table 8: Broadband FBMC parameters MAC design for IoT traffic operating in the 868 MHz band This section outlines the main PHY and MAC assumptions considered for IoT systems operating in the 868 MHz band. In particular, PHY assumptions based on FBMC techniques are first described. Traffic patterns and device classes are next presented to support the requirements of IoT systems. Then MAC layer design is considered in order to support the different device classes Traffic patterns and device classes In this section, traffic patterns derived from 3GPP cellular-iot (C-IoT) are presented followed by the device classes defined to meet the traffic applications. Traffic patterns In order to determine the traffic patterns that govern the MAC design, contributions of the C-IoT Working Group in GERAN have been extensively analysed. For instance, four different application traffic models are described to reflect the traffic characteristics of applications expected to be supported and which are the baseline of performance assessment of the proposed PHY/MAC specifications. For details on the different modes, see [33]. These modes are: UL: Mobile Autonomous Reporting (MAR) exception report (ER), e.g., alarms, UL: MAR periodic report (PR), DL: Network command (NC), DL: Software update / reconfiguration model. In the following, we consider only the three first ones. The characteristics of the considered traffic in the uplink and downlink directions are summarized in Table 9. For system capacity evaluations, MAR periodic reports and network commands are considered. In this traffic model, the split of devices between MAR periodic and network command is MAR periodic (80%) and network command (20%). The uplink traffic consists of periodic reports, exception reports and the responses to the network commands, while the downlink traffic consists of network commands and the acknowledgments for PRs and ERs SPEED-5G Consortium Parties Page 44 of 82

45 downlink traffic uplink traffic D5.1: MAC approaches with FBMC and simulation results (first version) events MAR periodic reports (PRs) (e.g., gas/water/electric smart metering reports, Smart agriculture Smart environment) (80% of devices) MAR exception reports (ERs) (e.g., alarm detectors, tamper notifications, power failure notification from smart meters) frequency (reports per device) 1 per day 1 per 2 hours 1 per hour 1 per 30 min Every few months or every years device percentage 40 % 40 % 15 % 5 % payload bytes latency 100 % 20 bytes 10 s Response to NCs same as NCs 50 % of NCs bytes Network commands (NCs) 1 per day 40 % (e.g., command to switch on the 1 per 2 hours 40 % lights, trigger the device to send an 1 per hour 15 % uplink report) 1 per 30 min 5 % (20 % of devices) Ack for PRs same as PRs 50 % of PRs 0 bytes Not critical (a few minutes) 5% resp. <1s 20 bytes Depends on command (from 10 min to 1 sec) Ack for ERs same as ERs 100 % of ERs 0 bytes Table 9: Uplink and Downlink Traffic patterns The periodic reporting mode pertains to sensor-based devices, which are expected to send periodic measurement reports. Periods can vary from a few minutes up to 1 day. 3GPP GERAN considers a minimum period of 30 minutes [33]; 3GPP RAN considers a minimum period of 5 minutes (1 min optional). Typical examples of such devices are environmental monitoring sensors (1 packet up to each 15 minutes) or gas smart metering (1 packet every 12h). Generally speaking the lower is the transmission rate the more stringent is the energy constraint, leading to maximum expected lifetime of 10 years on a 5 Wh battery. Assumption is made that 50 % of PRs require an acknowledgment. Compared to PR, exception reports are event-driven packets emissions on the device side. ERs typically occur when the value of the sensed physical quantity exceeds a given threshold, implying a kind of alert-type emission. Provided this type of report corresponds to a form of extraordinary behaviour, assumption is made that ER require an acknowledgment. In addition, GERAN includes a maximum latency ERs have to comply with. The maximum latency value is set to 10 s. According to [33], 20% of the traffic can be for network commands. These DL packets are short packets (20 bytes payload) sent with the same packet generation rate compared to PRs (from 1h to a day). Assumption is made that 50% of NCs induce a response on the device side. NC response is a packet of the same length order than a PR. The latency to response to the NC is a critical point, 5 % of the devices are assumed to respond with a delay less than 1 s SPEED-5G Consortium Parties Page 45 of 82

46 Device classes Depending on the applications and requirements of IoT traffic, the devices are classified into 3 main categories denoted as class 0 to class 2 (C p, p = 0,1,2). Table 10 summarizes the characteristics and communication capabilities for each class. class 0 device class 1 device class 2 device characteristics Uplink traffic (Periodic reports) Packet arrival rate λ p (Reporting interval T RI ) Number of repetitions = [ 0 2] (default 1) Uplink traffic (Periodic and exception reports) with an Ack. Packet arrival rate λ p (Reporting interval T RI ) Ack timeout = 2 superframe durations Max number of retransmission = 3 Uplink traffic (Periodic and exception reports, Response for NC) Downlink traffic (NCs) Packet arrival rate λ p (Reporting interval T RI ) Ack timeout, Max number of retransmission Wake-up Time T w Response for NC : yes or no (50 % in TR ) Max latency for sending response: 1s (5%), 1 minute (20%), 10 minutes (75%) Table 10: IoT device classes All IoT devices are expected to perform periodic reporting. Periodic reporting can be modeled by a Poisson arrival process with arrival rate λ p = 1 T RI, where T RI denotes the duration of the reporting interval. This reporting interval ranges typically from few minutes to 1 day and must be specified for each device. Exception reporting is supported by class 1 and class 2 devices. Meanwhile class 2 devices are expected to receive network commands. The devices in class 0 support only uplink traffic. They send periodic reports to the SC without requiring any acknowledgment. This type of device is not aware if a collision occurs or if the packet is correctly received. Therefore the device can transmit the packet with redundancy in order to increase the probability of successful receiving the reports. The number of packet repetitions can be set between 0 and 2 with a default value of 1. The devices in class 1 perform periodic and exception reporting. This class of devices requires an acknowledgement of the transmitted packet. If no acknowledgment is received within the specified period (Ack timeout), the packet is retransmitted by the device until successful transmission (Ack received) or when a maximum number of retransmission is reached. Class 2 devices are more critical in IoT network. This type of device is supposed to be able to receive network command from the SC. Provided IoT devices are particularly constrained from a power consumption perspective, they are in a sleeping mode most of the time. This means that the radio is switched off most of time, implying that the management of the downlink is particularly tricky. To overcome this issue, we propose that the device wakes up periodically for a short period of time T W, then listens to beacon to determine if there is any traffic (network command) from the SC. If a response to the network command is required, the maximum latency for sending response to NC must be specified for such device. The wake-up time and the latency are in general highly related to each other, the lower is the latency to sending the response, the shorter is the wake-up time of the device SPEED-5G Consortium Parties Page 46 of 82

47 MAC design supporting the different device classes The MAC layer design has been derived based on ECMA-392 standard that specifies a contentionbased and reservation-based channel access mechanisms along with others functionalities [34]. In particular, the MAC layer has been simplified and modified to support multi-band access. The main functionalities retained for IoT traffics are: A reservation-based channel access mechanism, A contention-based channel access mechanism. A network operates in master-slave mode. In the master-slave mode, a device is designated as master (also referred to as network coordinator or SC in case of cellular context) and others are associated with the master as slaves. The master coordinates channel access in the master-slave mode. Communication is normally established between slave devices and the master device. The SC transmits beacon frames to provide the basic timing for the network and to carry reservation and scheduling information for accessing the medium. These beacon frames allow advertising the SC presence, and synchronizing the devices. The reservation-based channel access mechanism consists to dynamically allocate a set of resources in a contention-free period based on a previous negotiation. An efficient scheduling mechanism must be therefore developed for flexible resource utilization. Meanwhile, the contention-based channel access mechanism allows devices to contend in order to get access to the medium; the device can either send a reservation request or a data packet. The MAC layer design must support also packet retransmission mechanism, power-saving mode and network command modes. The superframe The basic timing structure for frame exchange is a superframe. The superframe duration is specified as msuperframeduration. The superframe is composed of 512 medium access slots (MASs), where each MAS duration is mmasduration (Figure 8). As shown on Figure 8, each superframe starts with a beacon period (BP), which extends over one MAS. A recurring superframe consists of a beacon period (BP), a data transfer period (DTP) and an inactive period. The DTP is divided into contention-free period (CFP) and contention access period (CAP). The length of CFP and CAP is adjustable and depends on the network traffic type and density. e.g., CFP can be used in the case of small portion of devices, while CAP can support a large number of devices. This allows a scalable and flexible structure for a dense heterogeneous IoT network. It is worth mentioning that the beacon frame is transmitted over the total band with MCS1, while the data and others command/control frames are transmitted in sub-bands during DTP. We assume that the devices are able to decode the beacon and to subsequently obtain the information over all the sub-bands by performing fast Fourier transform (FFT) operation on the received beacon frame SPEED-5G Consortium Parties Page 47 of 82

48 Superframe m-1 Superframe m Superframe m+1 Fr e q u 12 Su b- ba nd s BP.. Medium Access Slots (MASs) mmasduration Data Transfer period (1,2 ms) CFP (UP) CFP (DL) CAP (UP) Slots.. Inactive Time 1 slot [0-12] slots [0-3] slots [0-512] slots (Ajustable) Figure 8: Superframe structure Beacon frame The beacon frame should include all the information for network operation: Reservation resource for uplink traffic, Reservation resource for downlink traffic, Frame structure Delimitation of the contention access period. The beacon frame should also specify the mode used to send network commands as it will be later described. The information element IE (CFP IE and CAP IE) are considered to describe the resource reserved for each scheduled device within CFP and the resource retained for CAP. CFP IE is used to announce the reserved MASs and their corresponding bands. Meanwhile, CAP IE identifies the MASs in which contention access transmissions are allowed. The beacon payload size is variable in function of the number of scheduled devices and traffic type and density. The maximum number of scheduled devices that can be supported in the beacon period (1.2 ms) is up to 40 devices. Each one can have up to 5 slots for transmission. The maximum allowed slots in CFP is 15 slots that can be split into 12 slots for uplink traffic and 3 slots for downlink traffic. In the case of network command traffic using the basic mode, the maximum number of devices that can be supported is up to 12. Medium Access modes Depending on the application requirements, the medium can be accessed in different ways: During the BP, the SC sends only beacon frames, During DTP, o o In CFP, devices participating in the reservation send frames, where the scheduling may be negotiated within a reservation request. In CAP, the devices contend for access to the medium to send data traffic (report) or to send reservation request to the SC trying to reserve a dedicated resources for data transmissions in the next superframe. Based on the received request and resources SPEED-5G Consortium Parties Page 48 of 82

49 availability, the SC notifies the corresponding devices about the dedicated resources in the beacon frame in the next superframe. During inactive period, all devices keep silent. In particular, the operation access for the downlink and uplink cases are detailed in the following sections. Uplink medium access Periodic reporting: when a device has a report to send, the device first wakes up at the corresponding reporting interval, and then it performs a beacon synchronization and contends to access the medium. A multiband Carrier Sense Multiple Access with Collision Avoidance (CSMA/CA) algorithm can be used to contend to the medium for channel access both in time and frequency domains [35]. The packet emission can be done along the two following options: Request for a transmission grant (sensitive data): During the CAP, the device transmits a transmission grant request. If there are sufficient resources to allocate the request, the small cell schedules the transmission in the next superframe, and uses the beacon to notify the device accordingly. If there is no notification in the beacon (no resource available or the reservation request is denied), the device shall retransmit the grant request in the CAP with the highest priority. If an acknowledgment is required, the SC may reserve a dedicated slot for the acknowledgment in the CFP (DL). Contention-based access (insensitive data): in this case, the device transmits the packet during the CAP. If an acknowledgment is required, we assume that there exits independent DL acknowledgement slots at the end of the CAP such that the SC transmits Ack packets without interrupting the UL data packets. Exception report: Assuming this kind of traffic will be very infrequent, using the CAP in the superframe is a reasonable idea. ER transmission requires then the following operations: Device synchronises on the small cell beacon, Multiband CSMA-CA transmission on CAP period, All ER packets are supposed to be ACKed by the small cell. Downlink medium access Network commands: a key issue of DL for energy constrained devices is how to cope with device dormancy states and to make sure that the DL can be set up, having a system reactive enough to send a command within a minimum latency, if required. We propose to rely on an approach which is close to preamble-sampling MAC protocol for energy-efficient wireless sensor networks (WSN) like [36]. In this WSN protocol class, the devices sleep for most of the time, except during periodic short instants (T ch ) when they wake up and sense the channel. If the channel is sensed as idle, nodes are going back to sleep mode. If a device senses the channel as busy, it keeps listening until a packet is decoded. In preamble sampling protocols, all devices share the same channel sensing period but are not synchronized. Subsequently, a device which wants to communicate with another one has to use a wake-up preamble prior to the data payload, which must be at least as long as T ch Refinements like the ability to learn the wake-up schedule of a device to reduce the duration of the preamble or the ability for a receiving device to notify the sender that it has awoken (i.e., the preamble is no longer useful) exist in the literature. In the DL specification of our MAC protocol related to the transmission of network commands, we are reusing the preamble sampling concept that puts the burden of a packet transmission on the emitter, allowing the receiving mode to manage its power consumption by means of period channel checks (see below). An important input for specifying this check interval is the expected reactivity of the system: what is the maximum latency for a node to receive NC? This maximum latency depends SPEED-5G Consortium Parties Page 49 of 82

50 on the application and may range from 1s to 10 min. In order to meet the maximum latency requirement, each IoT device (able to receive network commands) has to be configured so that it wakes up periodically for a short period, synchronizes on the beacon and checks whether a NC is pending. For each node, this wake-up period, called T w in our protocol, is set to the maximum tolerable latency that comes with application requirement. Basic mode Assume SC has a network command for Device #i which has a wake-up period of T w. The SC will include in the beacon the field that indicates that a NC is pending for device #i and this notification will be repeated in a number of beacons such as it covers a duration as long as T w. The notification includes a timer (countdown) telling when the NC is actually emitted in the future (as shown in Figure 9). The device wakes up when the timer reaches zero to resynchronize to the beacon and receive the NC. If the NC asks for a response, resource in the next superframe can be scheduled and indicated in the NC notification or NC itself. Figure 9: Basic mode for NC notification NC refinement 1: wake-up acknowledgment The basic mode consists in repeating the NC notification in the beacon over a period of time longer than the device wake-up period. On average, half of these repetitions are useless since the intended device will be awake. As in [37], the device can interrupt the NC repetition sequence by sending an ACK message in the CAP following the beacon which he manages to synchronise on. Thanks to this ACK message, the device triggers the emission of the NC in the following frame (Figure 10). NC refinement 2 As in [38], SC can learn the device s wake-up schedule. Accordingly, the SC can reduce the duration of the preamble, to start the preamble emission just before the device wake-up instant. A margin has to be applied, taking into account the possible clock drift since the last time the device woke up (T θ ). This mode requires that devices have to inform (piggyback with an ACK or a response) the SC about their up-to-date wake-up schedules. If the SC does not have a recent wake-up schedule update (T θ >T w ), the basic mode is used SPEED-5G Consortium Parties Page 50 of 82

51 Figure 10: NC refinement 1: wake-up acknowledgment Figure 11: NC refinement 2 MAC parameters Table 11 lists the value for the MAC sub-layer parameters. MAC parameters value msuperframeduration ms mmasduration 1.2 ms mmascount 512 mbeaconslotduration 1.2 ms (1MAS) mmaxframepayloadsize 0 bits 1600 bits (0 bytes- 200 bytes) CWmin 16 CWmax 1024 SIFS 10 μs DIFS SIFS + 2SlotTime SlotTime 40 μs Table 11: IoT MAC parameters SPEED-5G Consortium Parties Page 51 of 82

52 The medium access slot duration of 1.2 ms has been considered in the superframe in order to support different packet sizes while respecting the duty cycle of each device ( < 1%). It is worth noting that the total packet size is made up of payload size and header overhead of 20 bytes. Therefore, the time required to transmit the smallest packet size of 20 bytes using MCS 1 is 1 ms which falls within one slot. On the other hand, the corresponding time to transmit a packet payload size of 100 Bytes is 5.5 ms and 1.15 ms using MCS1 and MCS7, respectively. These times correspond to 5 slots and 1 slot, respectively Discussion of power consumption models In this section, we discuss the power consumption models of IoT devices which are energy constrained. Generally, power consumption depends on many factors such as the transmission time, transmit power level, receiving time, sleep mode and idle mode duration. Since most IoT devices are assumed to have a long battery life which may ranges from hours to 10 years, a low power consumption model has to be specified for each device. Four operation modes can be distinguished: Tx state, Rx state, idle state and sleep state. The Tx state is the most energy consuming one, in which the microcontroller is running and the radio is in Tx mode. In the Rx state, the microcontroller is running and the radio is in Rx mode. The Rx state is critical since the device spends a lot of time to listen to the medium. In idle state, the microcontroller is running but the radio is off. Meanwhile, in sleeping state, the microcontroller is in sleep mode and the radio is off. Power consumption in downlink direction is more critical than in uplink direction. Indeed in downlink direction, the challenge is to transmit network command from the SC to some devices without requiring that the devices continuously listen to the channel. The preamble sampling technique is proposed in this context (see section ), where the device regularly listens to the radio channel for a short duration to check if it has a notification to receive. A key question arises here: which power consumption model will be more efficient? Two cases can be distinguished: 1) When the device is always synchronized with the beacon and listen to all beacons or 2) when the device is in sleeping mode for most of the time (multiple of the superframes) and then wakes up asynchronously, listen to the channel and synchronizes with the beacon? In the first case, it is assumed that the device wakes up and listens to all beacons. The ratio of listen period to superframe period is equal to 1.2/ = 1/512. Note that, the device clock may have drifted so the device wakes up too late. Therefore, to guarantee that the device is well synchronized with the beacon, a guard period just before the beacon period has to be considered taking into account a possible drift. For simplicity, this guard period is not considered in the analysis. (A maximum guard time of 100 μs can be considered). In the second case, the device does not listen to all beacon frames, it listens to the medium at its wake-up period, then synchronizes with the beacon. The time required to listen to the medium is critical and the worst case arises when the device wakes up and the beacon has been just sent, therefore the device must listen to the medium for at least one superframe duration. The best case scenario is obviously that the device wakes up just before the beacon. The baseline listen time we consider in this analysis is the average listen time, which is of half superframe duration. Let T th be the wake-up period threshold, and T sensing the duration required to listen to the medium and to synchronize on a beacon; the ratio of listening duration over wake-up period corresponds to T sensing T th. In order find the minimum wake-up period that leads to a better energy efficiency compared to synchronize on all the beacons, one should determine T th which provides the same power consumption efficiency of the two considered cases, the ratios must be equal: T sensing T th = 1/512, it allows that the wake-up period threshold is equal to T th = s = 5.24 min and T th = s = 2.6 min in the case of worst-case sensing and reasonable case sensing, respectively. Therefore if the wake-up period of the device is less than the threshold (T th = 5 min or 2.5 min ),it is more efficient to the device to synchronize and listen to all beacons SPEED-5G Consortium Parties Page 52 of 82

53 Otherwise, the wake-up (sleeping) period of the device may be adjusted depending on traffic pattern. The device can also exchange sleep schedule and next wake-up period with the SC. This sampling schedule can be for example embedded into the network command response or in an Ack. The SC must in this case maintain a table of the device addresses and their corresponding wake-up periods. Using this table, the SC starts to send the notification just before the wake-up period by taking into considerations the clock drifts, reducing the number of NC notifications repetitions. The power state transition for different IoT devices can be summarized as follows: If a device is expecting to receive network command and must listen to all beacon frames, then the device shall wake up for a guard time prior to the beacon period in every superframe to receive the beacon. Then the device may go into sleep mode and wakes up prior to the start of the reserved resource indicated in the beacon to receive the NC. After receiving the NC, the device may go to sleep mode again. If a response to NC is required, then the device shall wake up again prior to the reserved resource indicted to send the response. If the device is in sleep mode for most of the time, it shall wake up at least one superframe and shall synchronize to the beacon during this superframe. If it has a notification to receive NC (with a countdown field set to the number of remaining superframes before sending the NC), the device may go into sleep state and shall reduce the value of the countdown for each superframe duration. If this value reaches zero, the device shall wake-up prior to resynchronize to the beacon and receive NC SPEED-5G Consortium Parties Page 53 of 82

54 5150MHz 5350MHz 5470MHz 5725MHz 5855MHz 5875MHz 5925MHz D5.1: MAC approaches with FBMC and simulation results (first version) MAC design for broadband traffic operation in the 5 GHz band This section deals with the design of a MAC protocol for the transmission of broadband traffic in the 5 GHz band. With regard to the edsa concept development addressed in the project (see section 3.1.4, Figure 2), this contribution describes how FBMC physical layer can be handled to offload some traffic on unlicensed spectrum. The way this traffic can be coordinated in some way with traffic sent on licensed resource is not part of this contribution. This is however a key item that should be addressed in the course of the project, so that an LAA-like access based on FBMC could be showcased Regulatory framework ETSI has defined a set of conditions devices shall respect when accessing the so-called 5 GHz band, which is represented in Figure 12. We assume that only the spectrum allocated to RLAN and FWA can be used in our case, with a specific restriction for MHz which can be used for indoor scenarios only. WAS/RLAN WAS/RLAN FWA 200MHz 255MHz 150MHz ITS 70MHz Figure 12: 5 GHz band allocation (ETSI) In terms of bandwidth, maximum transmit power, Table 12 shows the limits imposed by ETSI different channel bandwidths, knowing that a minimum 5 MHz channel is required. Note that we assume, for the sake of simplicity that the Transmit Power Control (TPC) mechanism is not be implemented, reducing the output power by a factor of 2 when channel is picked in [ ] and [ ] MHz. freq. range (MHz) max mean EIRP (dbm) max mean EIRP density (dbm/mhz) comment WAS/RLAN MHz and 40 MHz channels FWA MHz channels MHz channels Transmit Power Control (TPC): TPC ensures an average reduction in the aggregated transmission power by at least 3 db (5 db for FWA) compared with the maximum permitted transmission power. TCP is not required for channels within the band MHz. Without TPC, the highest permissible average EIRP (density) are reduced by 3 db. Table 12: ETSI Transmission parameters in 5 GHz band As well, ETSI specifies that Dynamic Frequency Selection (DFS) is required in the bands where devices have to coexist with primary users which are radar devices; the considered bands are MHz and MHz. This mechanism ensures that primary users can be detected and avoided; also it provides a means to distribute the channel selection in a near-uniform manner in the band SPEED-5G Consortium Parties Page 54 of 82

55 The implementation of DFS is considered as beyond the scope of this contribution, which focuses on how the access is handled once the channel is selected. Together with DFS, a Listen-Before-Talk (LBT) mechanism is mandatory, to assess whether the channel is occupied or not. ETSI describes two modes which are frame based equipment and load based equipment ; assumption is made that we use the first option in this work. The clear channel assessment (CCA, see Figure 2) is done using an energy detector and shall respect a minimum duration. Table 13 gives details on how LBT procedure should be applied and Figure 13 depicts an example of timing for the frame based equipment mode. For further details on DFS and LBT, the interesting reader can refer to [22]. parameter requirement comment Clear Channel Minimum 18 μs Assessment time Channel Occupancy time Minimum 1 ms, maximum10 ms Idle period Minimum 5% of channel occupancy time Fixed frame period Equals to Channel Occupancy time + Idle Period Short control signaling transmission time Maximum duty cycle of 5% within an observation period of 50 ms CCA Energy detection threshold (0 dbi antenna) If EIRP=23dBm at transmitter Threshold -73 dbm/mhz Otherwise (for a different transmit power level denoted PH) Threshold = -73(dBm/MHz) + 23(dBm) PH(dBm) Table 13: LBT procedure parameters Part of Channel occupancy time For WAS/RLAN Figure 13: LBT timing example [22] Clearly, the LBT mechanism, as shown on Figure 12, makes it possible to turn this regulatory rule into frame-based TDD access where the small cell can trigger the scheduling of a frame composed of uplink and downlink slots upon the detection of an idle channel. Depending whether the channel is idle or not, the frame will be sent or refrained. The channel occupancy time, as exposed in Table 13, can vary from 1 ms to 10 ms; the frame length can be mapped on this value which is a typical RRM decision depending on the interference level and the load to be sent on the unlicensed spectrum SPEED-5G Consortium Parties Page 55 of 82

56 Frame description and access mode The baseline of the frame structure relies on the ECMA 392 standard ([39]), which is very close from the IEEE standard, and therefore from the MAC frame we developed for the IoT case. This is a beacon-enabled frame, composed of slots that could be allocated to uplink or downlink communications, depending on the load in the network. The decision of the number of slots allocated to uplink and downlink is made by the scheduler, according to the traffic load. This is beyond the scope of this work and we assume a PF scheduler. From this standard, we keep the frame structure based on a beacon and time-slots, the frame being composed of A scheduled access period with an uplink and a downlink part. The slots of each portion spans over the whole channel, meaning that the scheduler can allocate resource blocks (RBs) based on a number of slots on a number of sub-channels (subcarriers), depending on the traffic type and CQI. In the uplink part there is no need to synchronise the devices which transmit on adjacent channels since FBMC support this asynchronicity; a guard time in slots has to be provisioned so that there is no temporal leakage from one slot to the next one. The length of the scheduled period and the beacon is actually the Channel Occupancy Time of Figure 13. A contention access period located at the end of the frame and composed of a reduced number of slots. The CAP is optional and shall be activated in the case there is no licensed carrier to use as a primary carrier. In this case, the CAP is used to let UEs transmit sporadic and non QoS demanding traffic, CQI updates and traffic requests. In the other case, requests and CQI updates will be transmitted on the primary carrier. An Idle period at the end of the frame. This is required by the European regulation which mandates that at least 5% of the frame length shall be idle. This period can be advantageously used by devices to sense the channel and feed the spectrum manager with sensing report accordingly. Figure 14: Proposed frame structure for broadband access in 5 GHz SPEED-5G Consortium Parties Page 56 of 82

57 Frame parameters are reported in Table 14. MAC parameters value mmasduration 500 µs mmascount Variable mbeaconslotduration 1 ms (2MAS) mmaxframepayloadsize Variable CWmin 16 CWmax 1024 SIFS 10 us DIFS SIFS + 2SlotTime SlotTime 40 μs Table 14: FBMC broadband access frame parameters According to the regulation, the Fixed Frame length triggers the CCA period of the small cell. If the small cell gets an indication of a busy channel after a CCA, it shall not transmit a beacon during the next Fixed Frame Period. In the case of using the 5 GHz band as a supplemental carrier, the CCA timing is advertised on the primary carrier operated in licensed spectrum. This would allow devices to manage the activation and de-activation of the 5 GHz transceiver and therefore manage the power consumption. In the case the 5 GHz is used with no licensed anchor, devices have to seek for beacons with no means to optimize their power consumption, except provisioning the possibility of relying on DRX if it turns out that a beacon is not received Frame structure optimization Referring to section 3 and to Figure 2 in particular, this section gives an overview of how RRM can modify the frame structure presented on Figure 14, according to the traffic load or interference conditions. This configuration is provided as a default configuration by the RAT/Spectrum selection/aggregation, which could be optimized by the Inter RAT cooperation. Optimizations may include Scheduled periods durations management so that the frame can be tuned to support a given traffic profile (mainly downlink vs. balanced traffic patterns). Alternatively, the frame can be mostly made by the CAP in the case there is only best effort traffic to convey, although this case does not appear very relevant. Fixed Frame Period reduction: if the channel is shared with other systems and the interference level is high, the RRM would reduce the Frame duration, implying to give CCA a dominant role. CCA duration extension: when coexisting with another system, the RRM may give more priority to it by extending the CCA duration. This extension also provides for richer sensing information to report to the spectrum manager. These optimizations proposal have to be refined during the course of WP5 and integrated in simulations so that their impact could be assessed SPEED-5G Consortium Parties Page 57 of 82

58 5.2 Broadband wireless access in virtual-reality offices and realistic extended suburban HetNet scenarios This section describes the required DL and UL MAC functionalities in order to support broadband wireless access over the virtual office and the extended suburban heterogeneous network (HetNet) scenarios. The previous sections describe how edsa is one of the main SPEED-5G enablers and its influence in order to solve different network lacks related with broadband wireless scenarios. In these scenarios, the RRM edsa algorithms are used, for instance, to decide which data type is offloaded from a congested spectrum band. How this is done and the required data to do it depend on the specific edsa technique and are defined in section 3 and in [2] and [21] Scenario characteristics The main goal of the MAC layer in these scenarios is to support as much traffic as possible with a defined QoS. Nowadays, commercial networks experiment a constant increase of its spectrum demand and, for that reason, one of the main goals of 5G is to support 1000 times more traffic relative to current values. In order to treat with this challenge, SPEED-5G project proposes the use of the edsa techniques which efficiently uses the available spectrum bands considering licensed, lightly licensed, and unlicensed bands. The use of these bands is done depending on, for instance, the traffic type, the UE characteristics or the instantaneous spectrum usage. Since the channel characteristics are non-static, the proposed MAC layer is ready to be (re)configured in real time for an optimal spectrum band usage. The UEs expected in these scenarios are mostly smartphones performing voice and data calls. In addition, IoTs are also expected located close to the UEs in this scenario. Most of the IoTs are wearables, monitoring the users and sending periodically data burst. The fact that several kinds of UEs are simulated, provide complete information about the MAC behaviour. The simulated UEs are mostly static or pedestrian so it is not required to define new channels for these scenarios. Vehicular and D2D simulations are out of the scope. Due to the characteristics of both environments, energy consumption is not a real issue and, for that reason, it is also out of the scope. The traffic type depends on the UE category. In this context, the UE category does not mean the standard UE category. In a simulation environment, the UE category is referred to the traffic type characteristics. For instance, one category is a UE doing only voice calls; other categories are UEs performing voice and data calls, or UEs doing burst traffic, and the different combinations of the mentioned categories. Different UEs in the simulation support different bands, having UEs that support just one band to several ones. This characteristic is important in order to know how the proposed MAC layer is capable of differentiating them and assigning resources only on the supported bands. The control plane data uses an ideal channel. The simulation goal is to test the MAC-layer behaviour, but not the specific procedures when the control data is not received by the target. For that reason, the control data plane is always received with an ACK. The following points describe the most relevant functionalities required to perform Broadband Wireless Access: carrier aggregation, QoS and enhanced spectrum usage Carrier aggregation The MAC layer in these scenarios requires the RRM configuration indicating the band where CA has to be used, the traffic type that can be offloaded or a specific MAC scheduler configuration to deal with different traffic types per RAT. Figure 15 shows several CA options where, the Frequency Allocation module of the MAC block diagram is responsible to allocate resources per UE into resources of a specific frequency or, using the concept of channel depending on the RAT. Option 1 and 2 do CA into the same band in a contiguous or non-contiguous frequency. In option 3, the CA is SPEED-5G Consortium Parties Page 58 of 82

59 done on another band. For instance, option 3 may combine two different RATs, one operating into a licensed band as LTE (red square) and another one operating into an unlicensed band as WiFi (blue square). This is a simple example due to the fact that option 3 may operate completely over the licensed band, as for instance using LTE on band 3 and 7. Scheduler Algorithms Time Allocation Freq Allocation Figure 15: CA options LTE networks configure the CA bands to be used in DL and UL, but SPEED-5G aspires to provide an enhanced usage of the spectrum. In that sense, configuring the bands to be used, as it is done in current technologies, is not enough. The main goal is, as previously mentioned, to configure the band and the traffic type to be scheduled on each band, depending on the QoS and the transmission RAT. How the cell is configured is not a big issue since the crrm and the protocol stack are linked together or are implemented in the same machine, but the UE configuration will require an enhancement on the control messages. New information, at least a mapping between the QCI or the RAT and the band, is required. The MAC Controller entity of Figure 2 is the responsible to configure the licensed and unlicensed schedulers to appropriately execute CA. In addition, the different functionalities of the unlicensed block are configured with thresholds, frames, and channel selection, which have to be configured. The Channel Selection block uses the reported physical data of the configured band to appropriately configure the MAC for operating into the most suitable band s channel. In parallel, the CCA/CSAT block is responsible to analyse the channel sensed values and, based on the configured thresholds, triggers the scheduling process. The previous procedures for licensed and unlicensed bands are valid in any case. The CA case shows how both bands work in a collaborative way into the same MAC layer Quality of service Frequency Frequency Band Intra-band contiguous aggregation Intra-band non-contiguous aggregation Inter-band aggregation The QoS has to be achieved by the SPEED-5G MAC in the same way that is achieved in current technologies (e.g., the voice traffic). It does not use a huge amount of resources; the packet size is lower than 1 kbits each 20 ms; but it is fully prioritized into the network being categorized as Guaranteed Bit Rate (GBR). For more information about how QoS is managed, for instance in LTE, [40] describes the 3GPP standardized values for each traffic type defined in Release 13. The QoS is achieved by traffic categorization and by channel estimation, adapting the transmission to the channel condition. The wireless channels are, by definition, non-static and therefore, the proposed MACs have the mechanisms to estimate the channel quality for DL and UL communications. The way the channel estimation is done into licensed and unlicensed bands are perfectly defined in current literature and should be defined on each RAN. That is, no further research is required in this field. Figure 3 shows how the QoS is managed into the Scheduler entity. The proposed scheduler design SPEED-5G Consortium Parties Page 59 of 82

60 is valid for every RAT activating or deactivating each sub-block. For that reason, the MAC Controller entity of Figure 2 should configure this module for licensed and unlicensed bands by using the RRM messages Enhanced spectrum usage The overall edsa objective is the optimal usage of the spectrum resources. New control information to add in the protocol overhead has been identified to support the proposed enhanced carrier aggregation scheme. In order to avoid control reservation areas or extra cell UE notifications indicating new control data, an algorithm capable of synchronizing the cell and the UE for sending control data has been defined. The new algorithm avoids extra computation on the UE side since the control data region and sending period is negotiated at the UE connection stage. A patent related to this algorithm is pending; concrete details will be provided in the next deliverable, D5.2. SPEED-5G proposes the usage of FBMC where the control region contains pilot and synchronization symbols as well as whitespaces. The idea is to use these whitespaces to transmit the control messages leaving the data region fully free to be used exclusively for the user-data plane. The number of whitespaces and when they are available in FBMC is fully linked with the PHY but, as the MAC and PHY works in a collaborative way, the MAC layer also knows the whitespaces and when they appear. For that reason, is possible to know when the control data may be transmitted and the amount of data capable to be transmitted. On licensed spectrum, the number of pilot symbols, called P, is variable and depends on the channel status. The space between pilot symbols used to transmit the synchronization symbols, is also variable, called D. Ideally, P has to be as small as possible and D, on the contrary, as big as possible. The details about P and D may be found in Figure 17. Figure 16 shows how the control region and the TTI may vary. The variability of these parameters is also under a patent process and for that reason, the complete procedure will be provided in the following deliverables, D5.2. Licensed Band Configurable Time Interval (RRM) Configurable Time Interval (RRM) freq Control Data Control Data Configurable number of symbols (RRM) time Figure 16: Frame on licensed band using FBMC on licensed band Making a zoom, Figure 17 shows the pilot, the synchronization symbols and the whitespaces. The tasks on future deliverables for licensed band focus on checking if the pilot and synchronization symbols do not disturb the control plane data located into the whitespaces, and the optimal configuration of P and D values allows having as much resources for the user-plane data as possible. The D value will be studied in order to know if it can be variable into the same frame SPEED-5G Consortium Parties Page 60 of 82

61 Figure 17: Frame with whitespaces for control data using FBMC on licensed band Further research is required in order to know the reconfiguration period. Depending on this period, the functionality to obtain P and D may be delegated to a centralized RRM or it has to be done into the MAC. How the P and D are notified to the UEs is out of scope as well as how the entity who measures them notifies the values to configuration layers like RRC in LTE. When P and D are calculated into RRM, it is the MAC Controller entity of Figure 2 that configures the scheduler to appropriately schedule the control messages into the whitespaces of the control region. The final P and D are fully related with the total amount of user data that can be delivered. Then, each configuration implies, for instance, a new LUT to be used by the scheduler. When these parameters are calculated by the MAC layer into the licensed block, the MAC Controller is in charge of notifying to upper layers the new configuration IoT support Due to the scenario characteristics, most of the IoTs devices may be categorized as wearable. The market of wearables which track human habits is becoming popular. Nowadays, wearables are linked to the mobile device which is responsible to upload the data to the cloud. SPEED-5G envisions natively supporting wearables which directly send data to the network. For instance, in addition to the novel proposal described on section 5.1, current technologies as ZigBee [41] should be natively supported in the proposed MAC ensuring backwards compatibility. ZigBee is a wireless high level protocol used to send a secured low throughput transmission with a low battery consumption. In any case, the controller entity (Figure 3) of the broadband MAC and its interfaces with upper and lower layers have to be ready to support any kinds of IoT configuration and communications SPEED-5G Consortium Parties Page 61 of 82

62 5.3 Decentralized dynamic TDD MAC for ultra-dense networks This section describes the MAC approach in order to support broadband wireless access over the ultra-dense networks (UDNs). The UDNs, seen as a key enabler of the next-generation mobile networks, are expected to be installed without traditional RF planning and proper site selection in many cases. Lack of such proper planning of the network deployment can cause excessive inter-cell interference in the UDNs. In addition, as SCs (i.e., constituents of the UDNs) are foreseen to operate in licensed as well as unlicensed spectrum bands, inter-system and inter-tier interference becomes a major concern. A small size of cells in ultra-dense networks may result in a large growth of signalling regarding the mobility management. Considering such challenges, a new MAC protocol is designed to enable efficient operation in scenarios characterized by dense planned and unplanned deployments of SC base stations operating on licensed as well as unlicensed spectrum bands. In some extend, this MAC approach can be seen as a default configuration of SCs, which could be used in the case when the connection with the virtualized edge RRM functions is not available (see [2]). In this situation, the self-optimizing feature of this approach is an effective way to cope with load and interference experience in ultra-dense networks. The proposed MAC protocol supports the edsa concept by allowing contiguous and non-contiguous channel aggregation, as well as seamless handovers between licensed, unlicensed and lightly licensed spectrum. In addition, this MAC introduces an autonomous mechanism for efficient monitoring of different channels which provide necessary information for proper assessment of channel status (while limiting the power consumption). Thus, dynamic channel selection becomes the core of the proposed MAC, with the main task of preventing QoS degradation resulting from high level of interference and simplifying network planning. Moreover, channel aggregation is also utilized to mitigate interference. The operation of the proposed MAC can be described considering Figure 2. Without any external information on spectrum characteristics, the proposed MAC uses channel selection operation and adopts the contention-based scheduler for both licensed and unlicensed bands, differently with two proposed MAC protocols included in previous sections. In the proposed MAC, licensed and unlicensed/lightly-licensed spectrum are all treated by the same mechanism. The following features have been integrated into the new MAC design: Cluster-based architecture (higher and lower MAC) Time-slotted access Multi-channel operation Dynamic channel selection and carrier aggregation based on spectrum sensing Decentralized decision-making Waveform-independent Further design details of the proposed MAC are currently subject to an ongoing patent application and will be disclosed in due course, within a future deliverable D SPEED-5G Consortium Parties Page 62 of 82

63 5.4 Summary This chapter reviews the design of some MAC functions, which are involved in the edsa capable framework of chapter 4. More specifically, it considers the management of 2 RATs in a carrier aggregation mechanism which is depicted in chapter 3 as enabler for capacity maximization and interference management scenarios. The work reported on this topic, which assumes a FBMC PHY, will be further detailed once a patent application is done. Also, a channel access scheme which exploits the FBMC characteristics is described. It has been designed to allow the offload of broadband and IoT traffic on unlicensed bands (respectively 5 GHz and 868 MHz bands). These two MAC functions have been designed under the assumption that RRM provides some configuration elements like the selected band where traffic can be offloaded or the default MAC frame in the unlicensed spectrum carrier. The interface between MAC and RRM has to be further refined to make it amenable to implementation in task T5.3. Finally this chapter also presents a MAC protocol which has been designed in the case the SC cannot rely on a connection with the virtualized architecture where the centralized RRM is located. In this situation, the MAC layer doesn t get any configuration from RRM and should be able of self-optimization to deal with the interference level which characterises ultra-dense networks SPEED-5G Consortium Parties Page 63 of 82

64 6 MAC evaluation This section introduces the system-level simulation of the MAC designs described in Chapter 5. For each scenario, the simulation assumptions (e.g., parameters, layouts, models, performance metrics and etc.), analysis of the simulations results and conclusions are presented. 6.1 UL/DL offload on shared spectrum In realistic traffic patterns This section reports on the work performed on the simulation of MAC protocol for offloading IoT traffic on 868 MHz band, as depicted on section It has been developed on a system-level simulator on which a realistic home environment has been ported. This subsection describes the evaluation methodology, namely detailing the network parameter, layout and models that have been chosen. The deployment scenarios on shared spectrum defined in Speed-5G consider two use cases: indoor scenario and outdoor scenario. The main purpose of such scenario is to analyse the traffic load that can be offloaded on shared spectrum using the SPEED-5G MAC layer based on FBMC PHY, with respect to interference context. In this deliverable, only the indoor scenario is representative of a future home environment or virtual office scenarios for IoT traffic. For future home scenario, we consider the layout shown in Figure 18, which is derived from virtual office layout presented in [42]. The outdoor scenario corresponds to broadband use case and will be investigated in D5.2 scheduled in December Evaluation methodology Figure 18: Virtual home layout In this section, the home indoor deployment scenario is considered with the layout specified in Figure 18. The entire scenario consists of a building having a total area of 45m by 20m. In particular, the building consists of several apartments of area 15 m by 10 m, each composed of many rooms (6 rooms). In this baseline scenario, we consider one SC per apartment composed of a base station and N = 40 IoT devices, whose distances from the base station are uniformly randomly distributed. In this scenario, a bidirectional communication is generally supported; however the UL communication from an end device to the base station is expected to be the predominant traffic In the simulations, we are restricted to UL traffic with periodic reports sent within the CAP period. The simulations are run using the WSnet system-level simulator with Mobile Autonomous Reporting (MAR) periodic traffic model. The purpose is to evaluate the performance of the proposed MAC design to support SC traffics on 868 MHz shared band, in which different systems may exist SPEED-5G Consortium Parties Page 64 of 82

65 Therefore interference between devices in the SC and also with others systems must be considered to make the transmission possible. To model the interference, we randomly drop in the adjacent apartments a number of devices which send periodic reports to occupied the channel and interfere with the reports sent by the devices in the considered SC. Table 15 presents the main parameters for this indoor deployment scenario for IoT traffic in home context which are in line with [33] and [45]. simulation parameters indoor scenario (baseline) Layout One small cell (one single apartment). See Figure 18. Frequency band 868 MHz Bandwidth 3 MHz [ ] Number of sub-bands 12 Sub-band bandwidth 180 khz Distance-depth path loss model Mix of ITU InH [referring to Table B in [33] [45] LoS L P = log 10 (d) (d km, f = 868 MHz) NLoS L P = log 10 (d) Penetration loss (Only internal wall penetration loss is considered Propagation channel model Doppler spread (Mobility) Shadowing std. deviation Interference/noise Antenna configuration Antenna Pattern BS Noise Figure UE Noise Figure Thermal Noise Figure Rx Antenna gain of BS + connector loss Rx Antenna gain of UE Number of UEs NLoS2 L P = log 10 (d) = Wi p Wi = the loss in internal walls = [4-10]dB uniformly distributed p = the number of penetrated internal walls (0-3) TU (ITU InH) 1 Hz ( Stationary IoT devices, speed = 0 km/h) Los (3 db), NLoS (4dB), NLoS2 (8dB) Sensitivity (baseline model) Interference model: drop a large number of devices in adjacent small cell to generate period traffic that interfere with the traffic of the small cell of interest. SC: N Tx = 1, N Rx = 2 IoT device: N Tx = 1, N Rx = 1 2D Omni-directional (baseline) 3 db 5 db -174 dbm/hz 5 dbi 0 dbi 40 UEs per small cell SPEED-5G Consortium Parties Page 65 of 82

66 simulation parameters UE dropping for each network Traffic model indoor scenario (baseline) All UEs randomly dropped and within coverage of the small cell in the unlicensed band Uplink traffic produced by low end IoT devices sending periodic reports in uplink traffic. DL/UP traffic produce by sensors/actuators is not considered. Table 15: Main simulation parameters For UL traffic models requirements, the following performance metrics are considered for the evaluation of MAC protocol: latency, packet delivery ratio (percentage of the transmitted packets) or percentage of dropped packets, average number of packet retransmissions, power consumption throughput. The latency for UL periodic reporting is evaluated, as specified in 3GPP cellular IoT, by including the time for device to synchronize to the network and the time for an access attempt from the device till the time to successfully receive the UL application layer payload at the SC. The capacity metric is defined as spectral efficiency in number of reports/200 khz/hour. Low power consumption of devices is required to enable long battery life. The devices may have a battery life ranging from hours to 10 years. The purpose of energy consumption analysis is to calculate the achievable battery life for an IoT device using a specific candidate solution. A 5 Wh battery capacity should be assumed, without consideration of battery leakage impact since this depends on battery technology Simulation results and analysis The simulation for this scenario is still under development and the results and analysis will be shared in the next WP5 deliverable, D SPEED-5G Consortium Parties Page 66 of 82

67 6.2 Broadband wireless access in realistic extended suburban and virtual reality office HetNet scenarios This section shows the preliminary results for the extended suburban scenario and the virtual office. The extended suburban results are focus on LTE as the base results to compare the SPEED-5G proposals. The virtual office results are focussed on channel coexistence, evaluating the SOTA and comparing, in the following deliverables, it with the project proposals for coexistence Evaluation methodology Extended suburban The Extended Suburban HetNet scenario has been defined in Speed-5G for the purpose of investing the performance of a highly typical European suburban area, which is also common in the suburbs of some major US cities as well as in some regions in Asia. This scenario is very meaningful in order to evaluate the 5G HetNet performance in both space and time in such a typical European Suburban area. The Extended Suburban HetNet scenario is comprised of a mixture of residential areas, green spaces and a single large commercial area. In the following, the typical elements that form part of the Extended Suburban HetNet scenario are listed and briefly described. Residential areas, represented by blocks of houses equally spaced in the x and y direction. A big commercial area, i.e., the typical area used up by a shopping mall. Recreational areas, including parks and green spaces. Streets, which act as the dividers between the above areas. The characteristics of this scenario provides a high flexibility due to it contains several areas of interest that can be investigated individually, making it particularly suitable to perform different studies using the same base scenario. The entire scenario is shown in Figure 19, which occupies an area of roughly 700 m by 600 m. The red rectangle highlights the area used to perform the initial simulations, whose results are provided in the following sections. Figure 19: Extended Suburban HetNet Scenario SPEED-5G Consortium Parties Page 67 of 82

68 The residential areas modelled by the Extended Suburban HetNet scenario are comprised of blocks of houses, separated by streets parallel to the houses. Each block is composed of four houses and each house is equipped with a backyard, according to the typical building structure of houses in suburban areas. This scenario also models the pavements installed between the streets and the houses themselves. In this baseline scenario, one small cell home enodeb (HeNB) is installed every two houses, and a single three-sector macro enodeb (enb) base station is installed in the centre of the big shopping area, on top of the roof at the height specified in next section. The orientation of each of the three sectors is also specified in the following section. The main simulation parameters for the Extended Suburban HetNet scenario are summarized in the following Table 16. LTE Simulation Parameter Distance-dependent path loss Small cell layer: Indoor scenario 3D distance between BS and UE is applied Macro cell layer: Outdoor scenario Propagation model ITU P.1238 Propagation ITU P.1411 Propagation Wall penetration loss Standard deviation of the normal distribution used to calculate shadowing Mobility model of UEs 7 db for residential buildings (concrete walls with windows) 15 db for commercial buildings (concrete walls without windows) Indoor nodes: 8 db Internal walls: 5 db Stationary Outdoor nodes: 8 db External walls: 5 db Total BS transmit power 21 dbm for 20 MHz 46 dbm for 20 MHz UE power class Bandwidth 21 dbm MHz (FDD) Antenna pattern 2D Omni-directional Three-sector directional antenna with antenna orientations = {0, 120, -120} degrees Antenna Height: 3.5 m 35 m UE antenna Height 1.5 m BS Noise Figure 5 db 5 db UE Noise Figure Antenna gain of BS including connector loss Antenna gain of UE 7 db 0 dbi 17 dbi 0 dbi Number of UEs 5 UEs per small cell. Ratio of Small cell/macro cell UEs: 80/20. UE dropping for each network Minimum distance (Inter-site 2D distance) Data traffic model The users attached to the Small cell layer are mostly indoors, while for the macro layer they are composed of a mixture of outdoor and indoor UEs, randomly spread over the whole simulation area. Randomly dropped 18 m 700 m Full buffer UDP SPEED-5G Consortium Parties Page 68 of 82

69 LTE Simulation Parameter Thermal noise level Number of antennas UE receiver Small cell layer: Indoor scenario 174 dbm/hz 2x2 MMSE Frequency Band 2.6 GHz 800 MHz Scheduler Virtual Office Proportional Fair both in Frequency and Time domains Table 16: LTE Simulation Parameters Macro cell layer: Outdoor scenario While the results for the Extended Suburban scenario are focused on LTE simulations and its results, the virtual office initial simulations results are focus on multi-rat coexistence. The simulation is done over the layout is the proposed at [42] by the 3GPP. Using this kind of standard layout, the obtained results are easily comparable with other studies. This scenario considers two operators deploying four SCs each in the single-floor building. The SCs of each operator are equally spaced and centred along a building (120 m by 50 m). The distance between the UEs and SCs is random. Figure 20 shows the scenario layout. The preliminary results are focus on the LAA and WiFi coexistence. Operator A Operator B 50m 120m Figure 20: Virtual office layout Table 17 summarizes the main parameters used to perform the virtual office simulations. It is important to note that this simulation does not include the macro layer so all the UEs are connected to the indoor small cells. System bandwidth per carrier 20 MHz Frequency Band 5.18 GHz Number of carriers 1 for DL SC TX power 18 dbm UE TX power 18 dbm Distance-dependent path loss IEEE 80211ax Indoor Small BSS Hotspot [43][44] Antenna pattern 2D Omni-directional is baseline; directional antenna is not precluded Antenna Height: 6 m SPEED-5G Consortium Parties Page 69 of 82

70 UE antenna Height Antenna gain + connector loss Antenna gain of UE Fast fading channel between enb and UE Number of UEs UE dropping per network Minimum distance Cell - UE Traffic model SC noise figure UE noise figure UE speed Cell selection criteria UE Bandwidth 1.5 m 5 dbi 0 dbi ITU InH 5 UEs per cell per operator Randomly dropped 3 m Constant Bit Rate UDP 400 kbps 5 db 9 db 0 km/h For LAA UEs, cell selection is based on RSRP in the unlicensed band. For WiFi STAs, cell selection is based on RSS (Received signal power strength) of WiFi APs. RSS threshold is -82 dbm. 20 MHz for LAA and WiFi Table 17: Main simulation parameters for unlicensed band use on virtual office scenario Simulation results and analysis Extended Suburban The simulation area whose results are provided in this section refers to the area highlighted by the red rectangle drawn in Figure 19. Besides the simulation parameters listed in Table 16, some additional ones that are relevant to this particular simulation are provided in Table 18. The preliminary simulations are done using the standard LTE RAT only. The results are obtained without using any of the studied edsa techniques. The main idea is to have these results as the base to compare the network improvement when SPPED-5G proposals are applied. additional LTE simulations parameters small cell layer Downlink EARFCN macro cell layer Number of cells 36 3 (equivalent to a single threesector macro enb base station) Number of users 180 (indoor users randomly deployed within the whole set of houses representing the residential areas) Closed Subscriber Group Closed (same group for all cells) Open Duration of simulation 10 s Number of independent seeds (mixture of indoor and outdoor users randomly dropped over the whole simulation area) Table 18: Additional simulation parameters for the subset of the extended suburban under consideration Figure 21 and Figure 22 show the pathloss into the selected area for the small cell and the macro SPEED-5G Consortium Parties Page 70 of 82

71 layer respectively. These images capture only a snapshot of the radio propagation; which represents only one precise time instant of the shadowing process that the radio signals undergo. Additionally, the strange salt-and-pepper effect is also related to the particular sampling frequency that was used to generate these images, which in turn affects the relatively low resolution of these images. small cell layer macro cell layer Figure 21: HeNBs installed indoors and every two houses at 2.6 GHz. Figure 22: Three outdoor enbs cells at 800 MHz In Figure 23 and Figure 24, the simulation results for the subset of the Extended Suburban scenario under consideration are provided. In order to obtain these results, 24 independent simulations have been performed and combined. The focus of these initial results is on throughput and latency, for both the small cell layer and the macro cell layer. Delay Throughput Figure 23: Cumulative Distribution Function (CDF) of the achieved throughput per user Figure 24: CDF of the served latency per user The dropping of the users in the scenario is random and the area depends on the desired cell to be attached, therefore for each independent drop the different users will be placed at different locations and will therefore experience different SINR levels. Since the simulation focus is set on the SC and the macro layers delay and throughput evaluation and, not in HetNet SON functions which try to mitigate interferences, each layer has a different frequency. The macro cells operate in the 800 MHz frequency band while the SCs cells are in 2.6 GHz frequency band, avoiding interference over these two heterogeneous layers. On the other hand, nodes belonging to the same layer are transmitting in the same frequency band and consequently, it will indeed interfere with each other SPEED-5G Consortium Parties Page 71 of 82

72 Virtual Office The preliminary results are obtained configuring two different RATs, one RAT per operator. The operator A is configured to use LAA while the operator B uses WiFi. The parameter modified on the simulations is the Energy Detection (ED) threshold used in LBT. This parameter is modified from -52dBm to -82dBm in steps of 10dB. Other parameters remain equal for each simulation. The following parameters, 19, are related to how the simulation is done, instead of to the simulated scenario itself. additional LTE simulations parameters Downlink EARFCN Number of cells Number of users Duration of simulation 4 per operator 20 per operator 20 s LBT Energy Detection Thresholds -52, -62, -82 Number of independent seeds 72 small cell layer Table 19: Additional simulation parameters for the virtual office under consideration The cumulative distribution function (CDF) throughput and the CDF latency per user results are shown in Figure 25 and Figure 26 respectively and demonstrate how the ED variation affects. Figure 25: CDF of the served throughput per user Lower ED thresholds imply that LAA will transmit more frequently since the restrictions for doing it are relaxed. The consequence is, that as the ED threshold is lower (as much of time the channel is occupied by LAA), and thus WiFi has less opportunities to find a free channel in which to transmit. Figure 25 shows the LAA and WiFi CDF throughput for different ED Thresholds. It is possible to check the expected behavior and how, a -82 dbm threshold makes that all throughput of the LAA users is transmitted while the 30 % of the WiFi users are not reaching even 300 kbps. On the other hand, when the network is configured to use a high threshold, -52 dbm, the network coexistence is achieved due to the fact that the LAA and WiFi users throughputs are almost the same SPEED-5G Consortium Parties Page 72 of 82

73 Figure 26: CDF of the achieved latency per user In Figure 26 is possible to obtain two different conclusions. The first one is that, when the system has more capacity to serve the offered traffic, the latency decreases. LAA and WiFi have lower latencies with lower and higher ED thresholds respectively since more throughput is achieved. The second conclusion is that LAA and WiFi have different latencies. When the ED threshold is set to -52 dbm, the coexistence in the network is achieved but, even that, a 10% of WiFi users may undergo latencies higher than 150 ms while the maximum LAA latencies does not reach 100 ms. The provided results in this section focus only on LTE only, without CA or SON procedures. As an extension of these results, the designed MAC will be integrated into the simulator the SPEED-5G proposals based on RRM algorithms and the novel MAC. In that sense, current simulations can be used as the base to evaluate the impact on the performance. Specifically, in future deliverable D5.2, the performance for the same layout, but having one SC installed per house, instead every two, will be investigated in the extended suburban scenario SPEED-5G Consortium Parties Page 73 of 82

74 6.3 Decentralized dynamic TDD MAC for ultra dense networks Evaluation methodology The proposed MAC in this section focuses on the broadband wireless use case in the scenario to provide full SC outdoor coverage. To do this, high SC densities per km 2 is assumed to have inter-site distances between 20 m to 130 m. Different deployment strategies are considered which include regular (i.e., hex-grid like) deployment of SCs and irregular (i.e., random) deployment. Regular deployment of clusters of SCs is also investigated (such deployments assume that location of clusters of SCs in the simulation scenario is regular whilst the deployment of SCs within the cluster can be either regular or irregular, see Figure 27). Regular/planned small cell layout within a cluster area Irregular/unplanned small cell layout within a cluster area Figure 27: Regular/Planned cluster layout By using the ns-3 platform, the proposed MAC has been simulated on a system level considering realistic channel models and wrap-around to minimize border effects. Thus, variable number of rings to enable simulation of different cell overlaps is considered. A subset of parameters for deployment scenarios considered in this simulation is presented in the following table which are in line with the 3GPP [42] SPEED-5G Consortium Parties Page 74 of 82

75 Parameter Network layout Wrap-around STA/AP height STA distribution Path loss model Shadow fading model Fast fading model Mobility Traffic model Traffic type Value Regular (Hexagonal) grid, Irregular (Random) will be considered in the next deliverable. Yes (variable number of rings) 1.5 m / 10 m Random/uniform distribution ITU-R UMi based model sigma = 3dB (LOS) / 4dB (NLOS), decorr = 10m (LOS) / 13m (NLOS), correlation between arbitrary pairs of links Not considered Not considered Full buffer (saturated model) Elastic (TCP New Reno), Non-elastic (UDP) Packet size (size of the packet 1500 bytes transmitted on the air interface, i.e., with MAC, IP and TCP overheads) (Application layer packet size: 1424 bytes) AP density Ptx 462/km 2 (ISD=50m) -55 dbm/hz Simulation results and analysis Table 20: A set of main simulation parameters This section presents and discusses initial results of proposed MAC performance compared to the IEEE standard MAC. Further evaluation results against LTE MAC will be added into a future deliverable, D5.2. Firstly, Figure 28 shows the comparison of network capacity supported by Wi-Fi and the new proposed MAC. The channel of 20 MHz is assumed to be available for WiFi while 10 channels of 2 MHz are assumed available for the proposed MAC. Three cases of channel aggregation are shown: 1) 10% BW per cell only 10% BW of overall BW (i.e., 1 channel among 10 channels) is used for communication, 2) 20% BW per cell 2 channels are exploited by aggregation, and 3) 40% BW per cell 4 channels with aggregation. By exploiting inherent interference diversity, the new proposed MAC adopting the dynamic channel selection can increase the network capacity compared to legacy Wi-Fi SPEED-5G Consortium Parties Page 75 of 82

76 New Wifi Figure 28: Comparison of network capacity performance of the proposed MAC and Wi-Fi Figure 29 shows the proposed MAC algorithm has enough flexibility in resource allocation leading to less outage compared to WiFi. Since the proposed MAC dynamically select channels of good channel quality, it can reduce the number of users in outage even for the wide bandwidth use. Figure 29: The impact of varying BW by aggregation on the number of users in outage In Figure 30, the performance of the proposed MAC is evaluated for various traffic QoS requirements in various SINR environments. For a given required SINR target level, it is shown that the proposed MAC can guarantee the required SINR level by dynamically changing carriers of better SNRs SPEED-5G Consortium Parties Page 76 of 82

77 Figure 30: Performance of dynamic channel selection mechanism The preliminary results depicted in this section show that the proposed MAC adopting dynamic spectrum aggregation helps users to use better channel with higher SNRs, thus achieving an increased network capacity. In addition, when it uses more bandwidth by aggregation, it has flexibility in resource allocation, leading to less performance less. While only some of the preliminary results are included in this section, sophisticated results will be included in a future deliverable SPEED-5G Consortium Parties Page 77 of 82