Performance Evaluation of a MAC Protocol for Radio over Fiber Wireless LAN operating in the 60-GHz band

Transcription

1 Performance Evaluation of a Protocol for Radio over Fiber Wireless LAN operating in the 6-GHz band Hong Bong Kim, Adam Wolisz Telecommunication Networks Group Department of Electrical and Computer Engineering Technical University of Berlin Sekr FT- Einsteinufer 87 Berlin Germany {hbkim, wolisz}@ee.tu-berlin.de Abstract Wireless networks using radio over fiber (RoF) technology operating in millimeter-wave (mm-wave) bands have been suggested as promising solutions to meet increasing user bandwidth and mobility demands. Due to the high penetration loss of mm-wave band signal, a wireless LAN (WLAN) based on this technology has properties quite different from those of conventional WLAN systems. That is, every room in a building should have at least one base station (BS). Thus, a challenging problem lies in the medium access control () protocol design so that it can support QoS requirements as well as a fast and easy handover. A protocol (Chess Board Protocol) based on frequency switching (FS) codes has been proposed by the authors considering the situation [7]. In this paper performance evaluation results for simple six variants of it are described and discussed. I. INTRODUCTION In order to meet modern ever increasing user bandwidth and mobility demands, a wireless network based on radio over fiber (RoF) technology operating in millimeter-wave (mmwave) bands has been suggested as a promising solution. In this network mm-wave signal is transmitted over optical fiber between the control station (CS) and base stations (BSs), and the BSs serve as access points for mobile hosts (MHs). Due to the high penetration loss of mm-wave band signal, a wireless LAN (WLAN) based on this technology has properties quite different from conventional WLAN systems operating in.4 or GHz bands. That is, every room in a building should have at least one BS. Therefore, simple and cost-effective BS will be a key to the success of the system. Recent research in this field has been focused on such components operating in mm-wave bands [] [6]. Especially, the 6 GHz band is of much interest since a massive amount of license-free spectrum has been allocated with a worldwide overlab of GHz (9 6 GHz) []. A medium access control () protocol (Chess Board protocol) featuring fast and easy handover and QoS support has been proposed by the authors [7], which is based on This work was supported in part by the Federal German Ministry of Education and Research within the TransiNet project. frequency switching (FS) codes. Adjacent picocells employ orthogonal FS codes to avoid possible co-channel interference. This mechanism allows a MH to stay tuned to its frequency during handover, which is a major characteristic of the proposed protocol. In this paper a simulation study for simple six variants of the Chess Board protocol is described. The paper begins with a brief description of the Chess Board protocol in section II. In section III simulation results are shown and discussed, and section IV summarizes the work. II. CHESS BOARD PROTOCOL DESCRIPTION A. Basic Operations A brief description of the Chess Board protocol [7] is given in this section. The simple structure of the BS and 6-GHz wave characteristics leads to a centralized network architecture with many picocells, where most of the BS functions of conventional WLANs are shifted to the CS (Fig. ). By subdividing the total system bandwidth, M (frequency) channels are obtained, where M channels (f,f,...f M ) are used for down-link transmission and the other M channels (f M+,f M+,...f M ) for up-link transmission. In addition, the time axis is also subdivided into time slots of equal length and M time slots are grouped into a frame. The MH is assigned a pair of channels (f i,f M+i ), i =,,...,M and a pair of time slots (t k mod M,t k+ mod M ), k =,,... for down- and up-link communication, respectively. Only after having received a permit from the down-link channel f i during the time slo k mod M, the MH may transmit up-link packets over the up-link channel f M+i during the nexime slot t k+ mod M. Every BS supports all channels, but each of them is used in the proper time slot. Fig. shows an example FS patterns for down- and up-link, respectively, when M is five. During every frame time, each of the M time slots and M channels is utilized once and only once. Adjacent picocells must not use identical FS patterns to avoid possible co-channel interference. One FS pattern used by one picocell can be reused by other picocells that are sufficiently separated from one another to

2 Optical Fiber Next ch Permit CS BS BS BS DOWN addr. P Data Reserv. Piggyback Down-link Up-link t 4 f f f f t 4 fm+ fm+ fm+ Picocell Picocell Picocell t 4 f f f f t 4 fm+ fm+ fm+ frame (t f ) slot (t s ) t 4 f f f f t 4 fm+ fm+ fm+ Fig.. A radio over fiber wireless LAN system operating in the 6 GHz band. Each room (picocell) has its own base station (BS) and BSs are connected to the control station (CS) using optical fiber. Adjacent picocells have different frequency switching (FS) patterns to avoid co-channel interference. Up-link FS pattern is obtained by shifting (delaying) the corresponding down-link FS pattern by one time slot. This figure shows an example FS patterns when the number of channels (M) is five. avoid interference. For proper operations using FS patterns, we assume thahe system is synchronous. Fig. shows down- and up-link slot formats and essentially, they have the same format. The down-link slot begins with a address indicating the destination of the slot. This address is followed by a permission field, which authorizes transmission of the MH specified in the address in the following up-link slot. The next field is for down-link payload, destined to the MH specified by the address. The last field consists of another address and reservation result, which indicate whether the request for bandwidth from the addressed MH is successfully confirmed or not. The up-link slot constitutes two parts. The first one is used for up-link data transmission consisting of MH s address, piggyback field, and payload field. The second part is for reservation used only by MHs that have not yet succeeded in reservation. If the piggyback field is set, it means that MH still has more data to send and requests assignment of one more slot. Non-setting this field indicates that its transmit buffer is empty. Each payload field (both up-link and downlink) allows packing several small packets or fragmentation for a large size packet. To request a permit for up-link transmission on some channel, the MH sends a requeso the CS using the reservation field in any slot in this channel. B. Mobility Support As long as a MH remains tuned to a certain channel (frequency pair) and time slot pair it can expeche time UP Fig.. time addr. Packet P Data... Packet Reserv. guard time Slot formats for the down- and up-link data transmission. instances at which down-link slots arrive. However, when it moves into an adjacent picocell using a different FS pattern, the MH will receive the down-link slot in an unexpected time instance. Thus, the MH can easily realize within at most a single frame time (t f = Mt s ) that it moved into another picocell (i.e., the number of channels (M) and the slot size (t s ) determines the minimum handover latency). Furthermore, using the following slot for the up-link transmission the MH can and should make a reservation in the new pair of time slots. Note thahe CS knows thahe MH with this address has been transmitting previously in a different picocell, so it is possible a) to treat such request with higher priority than requests for new reservations and b) release any possible reservation in the old picocell. C. Six Variants of Chess Board Protocol Six variants of the Chess Board protocol, classified into two groups, are considered in the paper as shown in Table I. In the first group (group A), MHs are assumed noo have a capability to change channels during operation, whereas MHs in group B are assumed to be able to change channels. A is the simplest among six variants, in which the CS allows a single channel to be used by only one MH, and it has no queue for requests. When the channel is being used, further requests are blocked. In particular, if a MH moves into another picocell in which its channel is already used, there is no way to continue transmission. A is similar with the only difference thahe CS does enqueue requests for busy channels rather than rejecting the request. When the channel is released the CS assigns io the MH ahe head of the queue. In interleaved usage of slots in a single channel by several MHs is performed. Slots are assigned using a simple roundrobin fashion, thereby assuring equal portion of the capacity to each candidate MH. In group B protocols MHs are assumed to be able to change channels. So the CS attempts to find capacity in other channels, not only in the channel on which the request has been issued. In B when a request from a MH arrives ahe BS, the CS investigates each channel if it is reserved or not. If a channel is free it assigns the channel to the MH by sending the channel number. When all channels are reserved the request is rejected. A queue for requesting MHs is maintained in B, and as soon as a channel is released it is allocated to the MH ahe head of the queue. is similar to B with an exception that when

3 the number of requesting MHs is greater than the number of channels, all the channels are shared by the MHs in a roundrobin fashion. In this case whenever the CS transmits a permit to a MH it must also inform the MH of the next channel number in the permit field of downlink slot (see Fig. ). It can be expected that group B protocols will outperform group A protocols ahe expense of increased complexity. m m m m BS MH III. NUMERICAL RESULTS A. Simulation Scenario and Assumptions Since two parameters of the Chess Board protocol (the number of channels and slot length) play an important role in determining handover latency and channel data rate (see [7]), in our simulation study emphasis is placed on the effects of the two parameters on system performance and performance comparison of six variants of the protocol. Because of the centralized nature of the system down-link transmission is so simple that only up-link performance is considered. The indoor environments for simulation includes four picocells and 4 MHs (Fig. ). MHs are assumed to be freely moving across boundaries between picocells according to random waypoint mobility model. Some assumptions for simulations are as follows: After having sent a request a MH receives the reservation result in one frame time from the down-link channel. The capacity of a BS (BT BS ) is. The channel data rate (BT ch ) is equal to BT BS /M O, wherem is the number of channels and O is overhead including address, permit field, reservation field and guard time. Message traffic consists of three packets: 4 bytes (4%), 76 bytes (%), and bytes (%) [8]. Since Chess Board protocol allows for packing small packets into a data slot or fragmenting a large packet into several data slots, different size of packets frequently found in widearea Interneraffic are used to investigate the effect of the slot size on system performance. Interarrival time between packets is exponentially distributed. Channel is perfect (error-free). p-persistent algorithm is used for sending a request packet. MHs use harmonic backoff algorithm (attempting probabilities, /, /,...) in computing their probabilities to transmit a request. Initial channel distribution by MHs is uniform. Other assumptions and parameters for simulation are summarized in Table II. Each simulation was run for at least sec (simulated time) including warm-up phase of sec. 9 % confidence intervals were calculated for the mean packet delay and normalized throughput whose variations from the sample mean were less than % for all results. As a simulation tool CSIM 8 was used. Fig.. Indoor environment for simulation consisting of four picocells and 4 mobile hosts. TABLE I SIX VARIANTS OF THE CHESS BOARD PROTOCOL Channel change Request when the during operation channel is used Group A blocked A NO queued A shared B blocked if not B YES queued if not B shared if not B. Delay Performance The mean packet delay is the average time spent by a packet from the instant it is generated till its transmission is complete. Fig. 4 and show the mean packet delay of group A and B protocols when the number of channels is five and the slot size is bytes, respectively. In Fig. 4 as the traffic load increases, A and perform better than A. Similarly in Fig. B and are better than B as the traffic load grows. We can also see that group B protocols highly outperform group A protocols because of the group B s trunking ability. Since the similar trends are observed with different number of channels and slot size, from now on we will only consider and protocols. The impact of the slot size in is shown in Fig. 6 indicating smaller slot size is better than larger one. However, TABLE II SUMMARY OF THE SIMULATION PARAMETERS Picocells 4 Guard time bytes MHs 4 FCS 4 bytes BS Capacity Buffer Length Mbytes adrs 6 bytes Mobility Model Random Waypoint Permit field byte Speed of MH. m/s Reservation slot 6 bytes Simulated Time sec Flag 8 bytes Statistics Collection after sec

4 A A slot = byte A A Traffic load (Mbps) Fig. 4. The mean packet delay of group A s when M is five and slot size is bytes B B B B. slot = byte Traffic load (Mbps) Fig.. The mean packet delay of group B s when M is five and slot size is bytes. due to the fixed size overhead too small size may cause larger delay under heavy traffic load. In order to investigate the effect of the number of channels on delay performance, simulations were run with a fixed-size slot and various number of channels. Fig. 7 shows that smaller number of channels is beneficial in terms of the mean packet delay, due to the fachahe frame time is directly proportional to the number of channels (t f = M t s ). An interesting fact is observed when the traffic load is Mbps. When M is 9, each of the 7 channels is shared by two MHs and the other two channels are shared by three MHs respectively, since channel distribution of MHs is assumed uniform and the total number of MHs is 4. If the three MHs using the same channel come together in the same picocell their total traffic load (9 Mbps) becomes greater than the channel data rate (8. Mbps /9), thus resulting in high mean packet delay. Similar arguments can be applied for M=7,8. Notice that if the time duration thahe three MHs stay in the same picocell is so long that MH s buffer is not enough to store overflowed packets there must be packet loss. However, as the MH has enough buffer in our simulation compared to the time duration no packet loss is observed when the traffic load per MH is less than or equal to for. Whereas when M is each channel is shared by only two MHs. Thus, if two MHs using the same channel stay in the same picocell their total traffic load (6 Mbps) is smaller than the channel data rate (7.7 /). That is the reason why the mean packet delay is lower when M is than when M is 9. Therefore, we can see thahe mean packet delay in depends not only on the number of channels and slot size but also on the channel distribution of MHs. The impact of slot size of is shown in Fig. 8. Just as in smaller slot size is advantageous in delay performance but due to the fixed-size overhead, small slot is not always beneficial. Fig. 9 shows how the number of channels influences the delay performance. With lighraffic delay grows along with the number of channels since frame time is proportional to it. As traffic load increases the delay has a minimum when M is around ten. Note that in our simulation scenario the average number of MHs in a picocell is ten. So when M is each MH is assigned its own channel resulting in rare 4 Mbps Mbps 4 4 Slot size (bytes) Fig. 6. The mean packet delay of when M is five and the slot size is bytes. collision and minimum reservation delay. If M is over ten the system has more channels than MHs on the average, thereby increasing frame time and wasting of bandwidth. C. Throughput Performance Throughput (normalized) is defined as the ratio between aggregate traffic transferred from MHs to BSs and the total system capacity. In Chess Board protocol throughput is affected by the channel data rate since the maximum bandwidth a MH can use is limited by it. Fig. and show throughput results of and, respectively. When M =throughput is limited by MH s traffic load as each channel is shared by multiple MHs (two MHs on the average) and the channel data Mbps Mbps slot = byte Number of channels Fig. 7. The mean packet delay of when the slot size is bytes and M is.

5 Mbps Mbps 4 4 Slot size (bytes) Fig. 8. Mean packet delay of when M is five and the slot size is bytes. Normalized throughput M=,s= M=,s= M=,s= M=,s= M=,s= M=,s= Traffic load per MH (Mbps) Fig.. Throughput of protocol with different number of channels and slot sizes. Mbps Mbps slot = byte Number of channels Fig. 9. Mean packet delay of when the slot size is bytes and M is. rate is about Mbps ( /). On the other hand, when M =throughput is restricted mainly by channel data rate, which is about Mbps in this case. Furthermore, larger slot size is favorable in terms of throughput although the effect is not noticeable. From the poinhe line is no longer linear packet loss begins to occur. It can also be seen that highly outperforms. IV. CONCLUSION The wireless LAN environment using radio over fiber technology operating in the millimeter-wave (mm-wave) bands imposes quite different requirements on the system design as compared to the conventional WLANs. Since the high Normalized throughput M=,s= M=,s= M=,s= M=,s= M=,s= M=,s= Traffic load per MH (Mbps) Fig.. Throughput of protocol with different number of channels and slot sizes. penetration loss of mm-wave signal many BSs should be employed to cover indoor areas. In such network with high number of small cells, the issue of mobility management has a very special significance. A protocol, called Chess Board protocol, featuring fast and easy handover and QoS support has been proposed in [7]. In this paper six variants of the Chess Board protocol were considered and their performance has been evaluated by a simulation study. Simulation results have shown that group B protocols which are assumed to have a capacity to change channels during operation highly outperform group A protocols where a fixed channel is assumed for each MH. Delay performance of both of them depends on the slot size and the number of channels; moreover, in group A it relies also on channel distribution of MHs. Smaller slot size and smaller number of channels are beneficial in delay performance with lighraffic load; however, that is not always true as traffic load grows. On the other hand larger slot size and smaller number of channels are in favor of throughput performance. REFERENCES [] P. Smulders, Exploiting the 6 GHz Band for Local Wireless Multimedia Access: Prospects and Future Directions, IEEE Commun. Mag. pp. 4 47, Jan.. [] K. Kitayama, et. al, An Approach to Single Optical Component Antenna Base Stations for Broad-Band Millimeter-Wave Fiber-Radio Access Systems, IEEE Trans. Microwave Theory Tech., vol. 48, pp , Dec.. [] T. Kuri, et. al, 6-GHz-Band Full-Duplex Radio-On-Fiber System Using Two-RF-Port Electroabsorption Transceiver, IEEE Photon. Technol. Lett., vol., no. 4, pp. 49 4, Apr.. [4] K. Kitayama, K. Ikeda, T. Kuri, A. Stöhr, and Y. Takahashi, Full-duplex demonstration of single electroabsorption transceiver basestation for mmwave fiber-radio systems, Microwave Photon. MWP, pp. 7 76,. [] G. Grosskopf, D. Rohde, and R. Eggemann, Mbit/s Data Transmission at 6 GHz Using an Optically Steered Antenna, ECOC, Muenchen, Germany, vol., pp. 4, Sep.. [6] R.-P. Braun, G. Grosskopf, D. Rohde, and F. Schmidt, Low-Phase- Noise Millimeter-Wave Generation at 64 GHz and Data Transmission Using Optical Sideband Injection Locking, IEEE Photon. Technol. Lett., vol., no., pp. 78 7, May 998. [7] H. B. Kim, H. Woesner and A. Wolisz, A Medium Access Control Protocol for Radio over Fiber Wireless LAN operating in the 6-GHz Band, in Proc. th European Personal Mobile Commun. Conf., pp. 4 8, Apr.. [8] K. Thompson, G. J. Miller and R. Wilder, Wide-Area Internet Traffic Patterns and Characteristics, IEEE Network, pp., Nov. 997.