WiMAX System-Level Evaluation Methodology

Transcription

1 WiMAX System-Level Evaluation Methodology Raj Jain Professor of Computer Science and Engineering Washington University in Saint Louis AATG Plenary Meeting, San Diego, CA Wednesday, July 12, 2006 Copyright 2004, 2005 WiMAX Forum WiMAX Forum Internal Use Only

2 Overview WiMAX Model Components Link-Level vs System-Level Models PHY Abstraction System Simulation Assumption Traffic Models MAC layer model NS-2 Modeling Activities WiMAX Forum Internal Use Only Slide 2

3 Goal Quantitative proof of WiMAX superiority Marketing vs Engineering: Qualitative vs Quantitative Carriers Need: Capacity Planning Performance Optimization Operational Guidelines Users Need: Operational Guidelines Vendors need: Performance impact of various features WiMAX Forum Internal Use Only Slide 3

4 Goals (Cont) WiMAX Real-world applications Workload should exercise bottleneck Wireless Mobile Internet Contents Distribution over WiMAX/Metro-WiFi Particularly sensitive to QoS Issues not clearly understood Competition with 3GPP Very thorough analysis available WiMAX Forum Internal Use Only Slide 4

5 Sample Questions: Capacity Planning: How many VOIP users can I support with 2 levels of diffserv, packet header suppression, 2x2 MIMO? What is the average response time for a typical web user in a car traveling at 60 km/hr,? Performance Optimization: What is the optimal packet size for video streaming? Operational Guidelines: Should we use 2 levels of differentiated services? How much more we gain by 4 levels? Performance impact of various features: 1x2 vs 2x2 antenna HARQ ARQ WiMAX Forum Internal Use Only Slide 5

6 WiMAX Model Components Applications (VOIP, VoD, Remote Backup, ) Workload Characteristics, QoS Requirements Transport and IP Layers (TCP/UDP, IP, RTP, ) TCP/IP Parameters: MTU Size, Buffers, MAC Layer (ARQ, Burst Allocation, FEC, ) Interference from other systems, Physical Layer (Coding, Antenna, AAS, OFDM, ) Topography (Height, Cell size, Customer density, ) Abstraction WiMAX Forum Internal Use Only Slide 6

7 Key Specs of Each Module This document defines the following for each module: Workload: Input characteristics Configuration: User controlled Configuration Parameters Service Provider Controlled Configuration parameters Manufacturer Controlled Configuration Parameters Features/Algorithms to be studied/compared Metrics: Throughput, Delay, Jitter, Availability, Assumptions WiMAX Forum Internal Use Only Slide 7

8 Participating Forum members Service Providers ATT, Bell Canada, Sprint, Verizon Equipment vendors Lucent Huawei Alvarion eaccess ArrayComm Others Intel Venturi Wireless This is an evolving list with expression of interest from a number of other members WiMAX Forum Internal Use Only Slide 8

9 Link-Level vs System-Level Models Link-Level: Goal: Study different signal transmission and reception schemes Single Link Single Cell Single Base Station Emphasis on PHY Some MAC Application Transport MAC PHY System-Level: Goals: Application Level Performance Multiple users Multi-Cells Multiple Base Stations Large # of subscribers Emphasis on All Layers => PHY abstraction Application Transport MAC PHY [AT&T] WiMAX Forum Internal Use Only Slide 9

10 System Simulation Approach Base Station Target cell [3GPP] WiMAX Forum Internal Use Only Slide 10

11 System Simulation Approach Simulate multiple WiMAX cells Model different applications with different levels of penetration Simulate application traffic streams; use realistic traffic models Distribute user session randomly among the cells Utilize neighboring cell traffic to create interference in the center cell Abstract PHY to a table/graph mapping physical condition to Block Error Rate (BLER) Apply generic MAC scheduler and MAC layer interfacing with PHY abstraction No link level simulation WiMAX Forum Internal Use Only Slide 11

12 System Simulation Diagram [Alvarion] WiMAX Forum Internal Use Only Slide 12

13 Sample Link-Level Simulation Results [AT&T] WiMAX Forum Internal Use Only Slide 13

14 Modeling Bulk Path Loss COST-231 Hata path loss model for urban macrocell PL(d,θ) = α 0 +10α 1 log 10 (d)+s+g(θ) α 0 = 345, at 19 GHz α 1 = 35 S denotes shadow fading random variable, lognormal with standard dev 8 db To sectors of same cell: 05 correlation To sectors of different cells: uncorrelated G(θ) denotes directional antenna gain Propagation Model Parameters Log-Normal Shadowing SD (σs) COST 231 Suburban 8 db BS shadowing correlation 05 Penetration Loss 10 db Value [ArrayCom] WiMAX Forum Internal Use Only Slide 14

15 Modulation Constrained Shannon s Capacity Spectral Efficiency (bps/hz) Shannon Capacity Modulation Constrained Shannon Capacity 16QAM WiMAX Modulation Constrained Shannon Capacity (-3dB) 16QAM Es/No Shannon s Capacity: C = B log 2 (1+SNR) Theoretical max Need to adjust for actual coding used, which in WiMAX varies with SNR Shannon s Capacity 16 QAM Capacity 16 QAM Capacity 3dB WiMAX The FEC of WiMAX Works with 3 db of the constrained Shannon Capacity [AT&T] WiMAX Forum Internal Use Only Slide 15

16 Channel Models Use standard fading models to time-correlated fading samples for consecutive frames Spatial Channel Model (SCM) (3GPP) ITU with spatial correlation matrix Find the frequency-domain coefficients for a representative set of tones for each slot allocated One tone is sufficient if all the tones fall within the coherence bandwidth of the channel Multiple tones are needed if the tones are distributed across the band (PUSC) Interference from all active inter-cell users Model permutation exactly, or Model permutation as randomly distributed Freq [ArrayCom] WiMAX Forum Internal Use Only Slide 16

17 PHY Abstraction: Inner vs Outer Receiver BER, BLER, FER Random Information Sequence Transmitter MIMO Channel Receiver AWGN Inner Receiver Outer Receiver Inner Receiver: Per sub-carrier processing N tx x N rx x N s OFDMA => subcarriers Outer Receiver: Per channel processing FEC, Decoder [AT&T] WiMAX Forum Internal Use Only Slide 17

18 Outer Layer Abstraction: L = F(C, MCS) L = Link layer performance indicator C = Channel quality indicator MCS = Modulation and coding schemes (MCS) Link layer performance indicators: Packet (burst) error rate (PER) Burst Error Rate (BUER) Block-error Rate (BLER) Bit error rate (BER) Symbol error rate (SER) Slot error rate (SLER) and etc Open Issue: PER vs BUER vs BLER Need member contributions [Intel] WiMAX Forum Internal Use Only Slide 18

19 Combining SINR Channel Quality Indicator C = fn(sinr) Problem: Combine SINR for n subcarriers to a single SINR for the channel SINR eff = f -1 {1/n Σ f(sinr k )} Possible Solutions: Average SINR MIC EESM MIM ECRM WiMAX Forum Internal Use Only Slide 19

20 Average SINR Mapping Function = Identity: F(SINR k ) = SINR k Effective SINR: SINR eff = 1/n Σ SINR k Different tones in the broadband channel undergo different amounts of (correlated) fading Averaging the SINR on these tones does not give a true SINR Not Recommended WiMAX Forum Internal Use Only Slide 20

21 MIC Mean Instantaneous Capacity Mapping Function = Capacity: F(SINR k ) = log 2 (1+SINR k ) Capacity of the k th subcarrier (Shannon s Capacity): C k = Log 2 (1+SINR k ) Mean Instantaneous Capacity: MIC = 1/N Σ C k SINR eff = 2 MIC - 1 Channel can be modeled as a AWGN channel with MIC Read Packet error rate (PER) from a pre-computed table PER = f(packet Length,MIC, MCS) MCS = Modulation and coding schemes (MCS) Issue: Should we use constrained Shannon capacity in C k? This method has been included in an annex [Intel] WiMAX Forum Internal Use Only Slide 21

22 ESM Effective SINR Mapping An alternative implementation of MIC Tables use SINR eff in place of MIC Channel can be modeled as a AWGN channel with SINR eff Read Packet error rate (PER) from a pre-computed table PER = f(packet Length,SINR eff, MCS) MCS = Modulation and coding schemes (MCS) SINR eff = 2 1/N Σ log2(1+sinrk) 1 This is also included in the annex [Intel] WiMAX Forum Internal Use Only Slide 22

23 EESM Exponential Effective SINR Mapping Mapping Function: Exponential F(SINR k )=-e -SINRk/β % β is adjusted to match the environment Effective SINR: SINR eff = -β ln {1/N Σ -e -SINRk/β } EESM from each transmission is used and the mapping table s dimensionality goes up with the number of HARQ transmissions Requires β-training Beta are determined from pre-runs Seek contributions from vendors Also utilizing BUPT to supply results from link runs Recommended Default Method WiMAX Forum Internal Use Only Slide 23

24 ECRM Effective Code Rate Method Instead of computing effective SINR, compute effective code rate ECRM-1: ECRM-2: R eff =R 0 (1/N Σ SINR k2 )/(1/N Σ SINR k ) 2 Reff = R 0 [1/N Σ log 2 (1+β SINR k )]/[1/N Σ log(1+β SINR k ] 2 R 0 is the transmitted code rate, and R eff is the effective code rate [Lucent] WiMAX Forum Internal Use Only Slide 24

25 Multiple Antennas Desired User Bulk Pathloss Model Channel model FFT Channel Matrix BF/MIMO Receiver Algorithm SINR per tone Bulk Pathloss Model Channel model FFT Per-tone Interference Allocation Equivalent SINR (EESM/MIC) AGWN SNR Bulk Pathloss Model Channel model FFT LUT (SNR, PDU size) Bulk Pathloss Model Interfering Users Channel model Recommend starting with SIMO or 2x2 [ArrayCom/Intel] FFT PDU Success/ Error PER Bernoulli Toss (p) WiMAX Forum Internal Use Only Slide 25

26 MAC layer model MAC Layer Model Components: Scheduler HARQ Power Control Resource contention QoS Advanced topics AAS Admission control Handover control frame builder interface How packets are split into frames Packet Header Suppression (PHS) Queuing of Frames Receiving ACKs Determining Frame drop event based on simulated BLER (from PHY abstraction) WiMAX Forum Internal Use Only Slide 26

27 Joint AATG & MTG Activity: PHY-MAC Abstraction for OFDMA Help on defining the right level of abstraction Help from MTG: Guidance and model reviews Utilize a Exponential Effective SIR Mapping (EESM) type of method to reduce complexity Meeting to collect ideas and come out with a common method WiMAX Forum Internal Use Only Slide 27

28 Sample System Simulation Parameters Parameters Number of 3-Sector Cells 19 Operating Frequency Duplex Channel Bandwidth BS-to-BS Distance 2500 MHz TDD 10 MHz 28 km Value Minimum Mobile-to-BS Distance 36 m Antenna Pattern 70 (-3 db) with 20 db front-toback ratio BS Height 32 m Mobile Terminal Height 15 m BS Antenna Gain 15 dbi MS Antenna Gain -1 dbi BS Maximum PA Power 43 dbm Mobile Terminal Maximum PA Power 23 dbm # of BS TX/RX Antenna 1/2/4 # of MT TX/RX Antenna 2 BS Noise Figure 4 db MS Noise Figure 7 db [WiMAXWP] WiMAX Forum Internal Use Only Slide 28

29 Sample OFDMA Parameters Parameters Values System Channel Bandwidth (MHz) 10 Sampling Frequency (F p in MHz) 112 FFT Size (N FFT) 1024 Sub-Carrier Frequency Spacing 1094 khz Useful Symbol Time (T b = 1/f) 914 us Guard Time (T g =T b /8) 114 us OFDMA Symbol Duration (T s = T b + T g ) 1029 us Frame duration 5 ms Number of OFDMA Symbols 48 Null Sub-carriers 184 DL PUSC Pilot Sub-carriers 120 Data Sub-carriers 720 Sub-channels 30 Null Sub-carriers 184 UL PUSC Pilot Sub-carriers 280 Data Sub-carriers 560 Sub-channels 35 [WiMAXWP] WiMAX Forum Internal Use Only Slide 29

30 Mobile WiMAX Configuration Parameters Parameters Value Cell Configuration 3 Sectors/Cell Frequency Reuse 1/3/1 Users/Sector 10 Traffic Type Full Buffer Channel Estimation Idle PHY Abstraction EESM [i] Scheduler Proprietary Proportional Fair Link Adaptation Realistic with delay feedback Antenna Configuration 1x2, 2x2 DL Alamouti STC, VSM MIMO Support UL Collaborative SM MIMO Switch Adaptive STC/VSM switch HARQ CC, 3 Retransmissions Coding CTC Frame Overhead 11 OFDM Symbols (7 DL, 3 UL, 1 TTG) Data Symbols per Frame 37 A 28:9 DL/UL Partition B 22:5 i 3GPP TSG-RAN-1, "Effective SIR Computation for OFDM System-Level Simulations," R , Meeting #35, Lisbon, Portugal, November 2003 [WiMAXWP] WiMAX Forum Internal Use Only Slide 30

31 Results from Example Simulation Run Cases Antenna SIMO Link DL: 28 data symbols UL: 9 data symbols Sector Throughput Spectral Efficiency DL: 22 data symbols UL: 15 data symbols Sector Throughput Spectral Efficiency DL 88 Mbps 12 bps/hz 66 Mbps 11 bps/hz UL 138 Mbps 055 bps/hz 22 Mbps 059 bps/hz MIMO DL 1360 Mbps 187 bps/hz 1063 Mbps 176 bps/hz UL 183 Mbps 073 bps/hz 274 Mbps 083 bps/hz These are MAC-level results Application-level results and issues are discussed next Issue: Collaborative spatial multiplexing modeling [WiMAXWP] WiMAX Forum Internal Use Only Slide 31

32 Application Layer Module Instant Messaging Streaming Video Mobile TV Video Conferencing FTP Telemetry VOIP Music Gaming VPN Background Datacasting Web Access Goal: To understand the effect of application specific options: VOIP: Codec, customer usage, connection duration Gaming: Types of games, user interactivity Video: Frequency and lengths of video Workload: Number of users, Application Mix, Usage Pattern WiMAX Forum Internal Use Only Slide 32

33 Application Layer Module Configuration Parameters: User: Type of usage, Usage frequency Service Provider: Number of servers, Charging policies, services Manufacturer: Capacity of servers Metrics: Number of customers supported Traffic Pattern Burst interval statistics, Burst size statistics, Maximum burst size, Features: Required delay tolerances, bit rates required for each application WiMAX Forum Internal Use Only Slide 33

34 Application Traffic Models Full buffer Advantages: Simple, Easily reproducible results, packets/sec Disadvantages: Not representative of real applications, cannot estimate user capacity, delay characteristics Not recommended for application level system simulations Real queue models HTTP FTP Non-Real-time TV WiMAX Forum Internal Use Only Slide 34

35 Traffic Models (HTTP) Instances of packet arrival at base station reading time A packet call First packet of the session A session Last packet of the session packet call packet call N d D pc (Reading Time) main object embedded objects WiMAX Forum Internal Use Only Slide 35

36 Traffic Models (FTP) Packet calls D pc Packets of file 1 Packets of file 2 Packets of file 3 WiMAX Forum Internal Use Only Slide 36

37 Traffic Models (NRTV) Video Streaming Session (= simulation time) time T 2T (K-1)T 0 KT T B (Buffering Window) D C (Packet Coding Delay) Packet Size WiMAX Forum Internal Use Only Slide 37

38 Transport Layer Module Traffic Shaping Buffer Allocation Congestion Avoidance ECN FRR DiffServ Classification Retransmission IPv4 IPv6 SACK MPLS Goal: To understand the effect of WiMAX specific TCP Optimizations Traffic Shaping, prioritization, buffer allocation, Guidelines for use of TCP options/algorithms WiMAX Forum Internal Use Only Slide 38

39 Transport Layer Module (Cont) Workload: Burst arrival pattern Configuration Parameters: User: TCP/UDP/IP implementations Service Provider: Prioritization Manufacturer: Set of scheduling, shaping algorithms, level of drop preferences, Metrics: Goodput, Delay, Jitter, Loss rate Features: DiffServ, Traffic shaping, WiMAX Forum Internal Use Only Slide 39

40 The current Plan Define a uniform methodology for WiMAX system level simulation Adapt 3GPP and 3GPP2 documents Universities to create WiMAX simulations and produce data Rensaler Polytechnic Institute (RPI ): Simulating T1 replacement service over WiMAX links; estimate optimized parameter set Information and Communication University (ICU) Korea : Compare WiBro, HSDPA vs WiMAX configurations for VoIP, Selected TCP applications Beijing University of Posts and Telecommunications (BUPT): PHY abstractions, Link simulation outputs for system simulation Clemson: DOCSIS MAC simulation Washington University in Saint Louis (WUSTL): Methodology and Scheduler, HARQ, Power Control Publish results that can be used by MWG WiMAX Forum Internal Use Only Slide 40

41 Current Draft: Table of Contents 1 INTRODUCTION, 4 2 SYSTEM MODELING ARCHITECTURE, 15 3 APPLICATION TRAFFIC MODELS, 24 4 PROTOCOL LAYER MODULES, 28 5 MAC LAYER ABSTRACTION, 28 6 PHY MODEM ABSTRACTION FOR SYSTEM MODEL, 28 ANNEX A: CHANNEL MODELS FOR SYSTEM MODELING, 31 ANNEX B : PHY ABSTRACTIONS FOR WIMAX SYSTEM MODELING, 66 ANNEX D : ENHANCED MAC ABSTRACTION REQUIREMENT, 70 ANNEX E NS2 FRAMEWORK COMMON MODULES, 73 ANNEX F: SYSTEM MODELING RESULTS, 74 APPENDIX G: SIMULATION DESIGN DECISIONS AND ISSUES LIVING LIST, 75 REFERENCES, 80 WiMAX Forum Internal Use Only Slide 41

42 NS-2 Modeling Activities Baseline NS-2 code RPI is developing the code RPI is documenting their NS-2 model Will have a code version control platform for sharing the NS-2 code Modeling S/W Architecture Document including the definitions of various APIs Intel is helping put this together APIs for scheduler and the OFDMA PHY abstraction HARQ modeling and Error indication MAC to PHY API based on MAC PDU Scheduler to use Subchannel level SINR (vs subcarrier level) WiMAX Forum Internal Use Only Slide 42

43 Acknowledgement Contributions from the following companies have been used: AT&T Intel Arraycom Telsima Lucent Alvarion Venturi Wireless WiMAX Forum Internal Use Only Slide 43

44 References 3GPP, TS25-892, OFDM Modeling 3GPP, TS25-996, SCM for MIMO 3GPP2, CR1002-0, CDMA2000 Evaluation Methodology, Rev 0, December 10, 2004 SCM-111, SCM Text V23, Spatial Channel Model AHG [WiMAXWP] WiMAX Forum, Mobile WiMAX Part I: A Technical Overview and Performance Evaluation, February 2006 [AT&T] Arun Ghosh, Link and System Level Simulations of WiMAX Network, AATG Interim, June 2006 [Alvarion] Z Roth, AATG Interim, June 2006 [Intel] Jie Hui and V Muthaiah, MIC for PHY Abstraction, AATG Teleconference, June 2006 [Arraycom] Arvind, Network Simulation Methodology for Multiple-Antenna WiMAX Systems, AATG Interim, June 2006 [Lucent] Larry Ozarow, Link Layer Abstraction Methods for OFDM Systems, AATG Interim, June 2006 WiMAX Forum Internal Use Only Slide 44