Copyright Notice

c©2009 IEEE. Personal use of this material is permitted. However, permissionto reprint/republish this material for advertising or promotional purposes or forcreating new collective works for resale or redistribution to servers or lists, or toreuse any copyrighted component of this work in other works must be obtainedfrom the IEEE.

A New Cross-Layer based Dynamic Resource Allocator for IEEE 802.16e Networks

Alberto Nascimento Departamento de Matemática e Engenharias

Universidade da Madeira Funchal, Portugal

ajn@uma.pt

Jonathan Rodriguez Instituto de Telecomunicações

Aveiro, Portugal jonathan@av.it.pt

Abstract— A new Dynamic Resource Allocator (DRA) architecture, based on a cross-layer framework, by using the control and signal channels implemented in the IEEE802.16e standard, is proposed. Commonly and a new designed packet schedulers are used in the performance evaluation of the DRA, by means of System Level Simulations under a realistic network layout, conducted on a proprietary C++ platform, fully compatible with the IEEE802.16e standard. Results are presented for the full queue and 3GPP Near Real Time Video (NRTV) traffic models and for the ITU PedB, PedA and VehA channel models for mobile speeds of 3 and 30 km/h. A newly designed scheduler based on the normalized (to the average) Signal to Interference plus Noise Ratio (SINR) is considered and its performance compared against the opportunistic Max CI and fair Round Robin.

Keywords- IEEE 802.16e, Mobile WiMAX, Dynamic Rsource Allocator, Cross-Layer Design, Packet Schedulers, Full Queue Traffic Model, System Level Simulatiors)

I. INTRODUCTION (HEADING 1) In the last few years IEEE802.16 based Mobile WiMAX

[1] system has gained much attention with the acceptance as one of the 3rd Generation (3G) communications specifications. Mobile WiMAX is a broadband wireless solution that enables the convergence of mobile and fixed broadband networks through a common wide area broadband radio access technology and flexible network architecture. The salient features supported by Mobile WiMAX include high data rate, guaranteed QoS, scalability and so on [2]. Technologies such as Hybrid Automatic Repeat Request (HARQ), Space Time Coding (STC), Adaptive Antenna Systems (AAS) and Multiple-Input-Multiple-Output (MIMO) have been added to support mobile environments and to improve broadband access speed.

Mobile WiMAX adopts the Orthogonal Frequency Division Multiple Access (OFDMA) technology for improved multipath performance and efficient spectrum allocation. It is particularly suitable for adaptive transmission and resource allocation due to existence of parallel sub-channels in frequency domain. A key principle of adaptive resource allocation is to exploit the inherent system diversities in various system domains through the intelligent management of the allocation and access of users to the resources available.

In the last few years researches have proposed a new paradigm for the layered architecture for communications: the cross-layer design (CLD) framework. In a general way, the CLD paradigm refers to protocol design in which the dependence between protocol layers is actively exploited [3]. In this paper, a flexible DRA architecture framework, specifically designed for a cross-layer-based architecture under Mobile WiMAX network is proposed. It implements a cross-layer architecture model by use of the control channels specifically designed for the fast exchange of signalling messages for ARQ and efficient channel estimation and feedback [4]. The DRA exploits the inherent diversities in Mobile MAC layer to deliver seamless transport of homogenous traffic in a cost-effective manner. Central to the performance of the DRA is the type of scheduler implemented. This paper is organized as follows: section 2 presents in detail our proposed cross-layer based DRA architecture for IEEE 802.16e networks and the schedulers implemented into the DRA. Section 3 describes the procedures followed in the definition of the interface between physical and system level simulations. Section 4 presents the DRA performance. Section 5 concludes the paper.

II. DYNAMIC RESOURCE ALLOCATOR (DRA) Most research available in the literature is based on rather

simplistic analysis on the performance of WiMAX networks, under simplistic channel and traffic models, and mostly for a single cell scenario. Few proposals were made for the true evaluation of the system level performance of a WiMAX network under a realistic cellular layout, channel and traffic models. And even fewer elaborate on the interface issue between link and system levels [5-7]. In this work the design and implementation of the DRA was validated in a specific tailored Mobile WiMAX system level platform, implemented according to the WiMAX Forum system profile [8]. Common schedulers were implemented jointly with the proposed DRA in order to evaluate the performance of the network and simulation results are compared against theoretical values.

A. Architecure The DRA architecture comprises the following modules: (i) Packet Scheduler, (ii) Resource Allocator, (iii) Hybrid Automatic Repeat Request (HARQ) and (iv) Link Adaptation

978-1-4244-3302-5/09/$25.00 ©2009 IEEE

Authorized licensed use limited to: UNIVERSIDADE DE AVEIRO. Downloaded on December 16, 2009 at 08:21 from IEEE Xplore. Restrictions apply.

(Adaptive Modulation and Coding – AMC). It is illustrated in Figure 1.

Figure 1. Dynamic Resource Allocator architecture with constituent modules

In each transmission time interval (TTI) the DRA allocates radio resources in the TDD radio frame, according to the priority list from the scheduler. Whenever a user is selected for transmission, the DRA withdraws the required data bits from the respective queue and fills up a data burst in the resource space. The set of packets constitute a MAC-PDU and is filled with padding bits if necessary. In what follows the DRA procedures are described, bearing in mind downlink communication.

B. Link Adaptation (LA) At the beginning of each TTI the LA module computes the proper Modulation and Coding Scheme (MCS), according to the Channel Quality Information (CQI) reported in the uplink CPICH channel. For each burst created in the resource allocation space the MCS scheme to be used is chosen in equation (1).

[ ])1(maxarg iiMCSi

BLERRiset

−=∈

(1)

Where setMCS represents the set of modulation and coding schemes, iR is the throughput achieved for the selected MCS scheme and iBLER is the predicted Block Error Rate (BLER), as given by equation (2)

( ))(),()( ,)( nMCSnfnBLER kCQIk

ik γ= (2)

The BLER is a function of the predicted CQI, CQIk ,γ .

The DRA selects the most spectrally efficient MCS, i.e., the one that maximizes the achievable bit rate and at the same time complies with the expected BLER under the desired threshold level. The CQI is estimated by each mobile station attached to the cell, by means of the SINR value of the OFDM symbol carried in the frame’s preamble. The CQI is reported back to the base station via the CQICH control channel on the uplink sub-frame, on a frame-by-frame basis.

C. Asynchronous Hybrid Automatic Repeat Request (HARQ)

HARQ procedures are used in error recovery. Each transmitted MAC PDU is mapped into one of the available HARQ processes for each user and is in charge of the transmission and re-transmissions until it is successfully received. Each HARQ process is associated to one buffer in the mobile station for storing the result of the combination of successive versions of the same MAC PDU. In the simulations the mobile station combines all versions of the MAC PDU using a Chase Combiner. Retransmissions of the same MAC PDU keep the original MCS scheme used in the first transmission attempt. HARQ buffers are freed when the radio block is successfully received or when the maximum number of allowed transmission attempts, attemptsN , has been achieved. An ACK control channel in the uplink sub-frame is implicitly assigned to the HARQ process and is used in the transmission of ACK or NACK feedback messages reporting the status of packet reception at the mobile station receiver.

D. Scheduler Dynamic scheduling is based on different metrics reported from other layers in the protocol stack, in agreement with the cross-layer architecture paradigm. The metrics used in the computation of the priority metrics are: ⋅ CQI reports sent in every uplink sub-frame. Such

channel-sensitive scheduling techniques exploit multi-user diversity, resulting in the so-called “Multi-User Diversity Gain” [9].

⋅ The user average data transmission rate. ⋅ Amount of information in the each user’s buffer (in bits). ⋅ Number of transmission attempts. The following schedulers were implemented and plugged into the DRA architecture framework: Max C/I (CI) The CI scheduler opportunistically assigns resources to the user with the highest channel gain by scheduling, at the beginning of each TTI, the mobile with the best channel quality indicator among the set of active mobiles.

The user selection is given by equation (3): ( )

{ },...2,1,0),(maxarg

,...,1==

∈nnCQInk i

Ki (3)

Where )(nCQIi is the CQI reported by the thi mobile in the thn frame period.

Round Robin (RR) This is a classic time division multiplexing-based algorithm where the delay between successive transmissions to the same user is fixed and equal for all. The scheduler gives priority to each user in a sequential way without taking into consideration any user QoS requirements. Max C/I over Average C/I (AvgCI) This scheduler compensates the influence of the mean SINR on the instantaneous SINR measured by each user. The mean SINR depends on the antenna gains, path-loss and shadowing, and is highly influenced by the distance from the mobile to

Authorized licensed use limited to: UNIVERSIDADE DE AVEIRO. Downloaded on December 16, 2009 at 08:21 from IEEE Xplore. Restrictions apply.

base station. Mobiles on the cell edge have a much smaller mean SINR due to the distance from the cell centre and due to the inter-cell interference, which results in starvation on the transmission opportunities access. The mobiles are ranked according to the scheduling rule given by equation (4).

( ){ }

,...2,1,0,)()(

maxarg,...,1

==∈

nnCQInCQI

nki

i

Ki (4)

Where )(nCQI i is the SINR from user i , averaged over a period of avT . In the simulations avgT =1.5s

E. Resource Map Definition The IEEE802.16e defines the slot as the minimum resource allocation granularity in the radio frame. The size of the slot depends on the type of sub-channelization used and on the type of connection (uplink or downlink).

As the standard does not specify any algorithm for resource allocation (this task is left for manufactures proposals differentiation), in the proposed DRA module, the resource allocation space was defined as a matrix of sub-channels per OFDM symbols in which the minimum granularity of resource allocation is the Radio Access Unit (RAU). Each RAU is a container with dimensions 10 sub-channels per 6 symbols, corresponding to 30 slots, transmitting with the same MCS scheme. With a focus on downlink TDD operation, it is assumed that out of the 48 OFDM symbols available in the TDD radio frame 35 symbols are assigned for the DL sub-frame, of which a fixed control overhead of 5 symbols is assumed for the preamble, FCH, DL-MAP and UL-MAP and the remaining 30 symbols are used for data transmission. As each RAU comprises 30 slots, with 30 symbols available for data transmission there are (30/2)*30 = 450 slots per frame and 450/30 = 15 RAUs per DL sub-frame. Each RAU is large enough to accommodate a MPDU size of 6480 bits using the 3/4 coded 64QAM MCS scheme. This corresponds to a peak bit rate of 1.296Mbps per RAU or 19.44Mbps per frame.

It is worth mentioning that a smaller RAU size will result in higher overhead because more users can be assigned resources in the frame (depending on the offered load per user). Meanwhile, a larger RAU size will result in higher error rates because the BLER must be computed for all RAUs used in the transmission of the MPDU, assuming a large burst of data is transmitted. Therefore the size of the RAU is a trade-off between efficiency in data allocation per resource, overhead (in terms of DL/UL-MAP size and padding bits to fill each RAU) and the probability of error in the decoding of each resource. The adequate size depends on the type of traffic model being supported.

In the beginning of each TDD frame the DRA assigns free RAUs from the resource map. Whenever a user is selected by the DRA, the DRA withdraws the selected IP packets from the respective queue or from the respective HARQ process in case of retransmission.

F. Resource Allocation Procedure Two lists of scheduling users are created: the high and the low priority lists. The high priority list is filled with users assigned to HARQ processes attempting retransmissions and for which the priority timer is set. A retransmission is attempted whenever an error occurs in decoding a RAU inside a burst. Whenever a burst of data is received with error, after the first transmission attempt is performed, a discard timer is set. If the burst is not received with success before the expiration of this timer the HARQ process is put into the non-active state and resources are freed for the allocation of a new data. Users in both high and low priority lists are prioritized according to the same scheduling algorithm and to the same metrics. Users in the high priority list are given higher priority in detriment of the users in the low priority list.

The following steps are performed until one of the following three conditions is not satisfied anymore: (i) the list of priorities is empty, (ii) there is no more information to be transmitted (iii) or there are no more resource units for allocation. Step 1: - The scheduler elaborates the list of mobiles selected for transmission. Only mobiles whose CQI is higher than a given Admission Threshold are considered in the metric computation. This parameter is used for the efficient utilization of the set of resources, by reducing the number of errors due to bad channel quality. Step 2: The BS computes the power assigned per resource unit, RUtP , , in function of the remaining transmit power budget, budgetP , and the number of available resource units

budgetN :

budgetbudgetRUt NPP /, = (5)

Step 3: The best MCS scheme selectedMCS is selected according to the predicted SINR:

RUtpredpred xPCQISINR ,= (6)

The predicted SINR, is computed according to the CQI value reported by the mobile station on the uplink sub-frame:

predCQI .

Step 4: The power per resource RUtP , is computed. The power assigned to each resource depends on the reported CQI. If the selected MCS scheme is the most robust one and the expected BLER is larger than the threshold for this service,

threspred BLERBLER > , than a larger transmit power may be allocated to each resource unit. The new value must be such that the expected SINR, predSINR , equals the threshold:

thresSINR : ( )max,,, ,/min RUtthreshRUt PCQISINRP = (7)

Where budgetRUt PP =max,, . The value of the threshold

threshSINR is computed in order to make the predicted BLER to be equal to the expected threshold for the service. The new

Authorized licensed use limited to: UNIVERSIDADE DE AVEIRO. Downloaded on December 16, 2009 at 08:21 from IEEE Xplore. Restrictions apply.

value of the power to be assigned to each resource unit can not be greater than the remaining power, budgetP . Step 4: The maximum number of resource units, RUN , available for allocation depends on the computed power per resource unit , RUtP , :

RUtbudgetRU PPN ,/= (8)

Step 5: The number of bits carried in each resource unit is deduced from the selected MCS scheme and the available number of resource units:

RURUletransmitab xSizeNN = (9) Step 6: The number of bits to be transmitted depends on the number queueN of bits waiting in the BS’s queue:

),min( letransmitabqueuetransmit NNN = (10)

Step 7: The number of resource units to be allocated is computed:

⎥⎥

⎤⎢⎢

⎡=

RU

transmittransmitRU Size

NN , (11)

Step 8: The RA map is updated by allocating transmitRUN , resources to the given mobile with the selected MCS scheme,

selectedMCS and the computed transmit power per resource unit, RUtP , . Step 9: The BS updates the mobiles list as well as the remaining power available for transmission, budgetP , and the

amount of resource units still available for allocation , budgetN . The power available for data transmission in the downlink was assumed as uniformly distributed among the whole set of RAUs in the TDD DL sub-frame. The power assigned to each RAU, comprising the burst assigned for data transmission, for each user, depends on the amount of power available for allocation and on the amount of RAUs allocated to the user. Also, although the schedule and resource allocation is performed in a 2D frame, as the PUSC sub-channelization mode spreads the set of sub-carriers composing each sub-channel across the whole spectrum of each OFDM symbol all sub-channels behave similarly and the position of each RAU in the frame is affected only by the allocation of co-localized RAUs in interfering neighbouring cells.

III. LINK LEVEL INTERFACE System level performance evaluation is based on a separation among system protocol layers, where RRM (Radio Resource Management) algorithms, performing at the TTI granularity, assume the physical layer as a “black box”. This performance simulation strategy allows obtaining results in an acceptable time frame with minimal complexity using Look-Up Tables (LUTs). The interfaces between PHY and MAC layers are based on the radio link quality reflected by the SINR ratio reported for each data sub-carrier in each sub-channel. According to the scenarios used in the simulations the Actual Value Interface is assumed.

TABLE I. SIMULATION SETUP CONFIGURATION

Parameter Value BS Maximum Transmission Power

43dBm

Propagation model Kmin R),(log6.371.128 10 RL +=

Penetration Loss 10dB Log-Normal and Shadowing Correlation

Standard Deviation = 8dB; 0.5 for sectors of different BSs and 1 for sectors of the same BS

Channel Model ITU PedB, PedA and VehA Number of Users 72, 120 Traffic Models Full Queue and 3GPP Near Real

Time Video Duplex Mode TDD Operating Frequency 2.5GHz Bandwidth and FFT size 10MHz; 1024 sub-carriers Frame Duration 5ms Number of OFDM Symbols in DL

35

Preamble, FCH, DL/UL MAP overhead

5 symbols

Sub-channelization DL-PUSC Burst Size 10 sub-channels x 6 symbols (15

resources) MS speed 3 and 30 km/h (moving average filter length)

1.5s

CQIT (CQI feedback delay)

2 frame periods

Discard Timer and Priority lengths

15 and 3 frame periods respectively

ACK/NACK feedback 1 frame period in CQICH UL sub-frame

retN Maximum Number of Retransmissions

4

Number of HARQ processes per MS

4

BLER Threshold for Link Adaptation

10%

There are several approaches to map OFDM link performance in a system level simulator. The motivation is to map the per-subcarrier symbol SINR computed for each mobile to a single effective scalar value using an appropriate mapping function. In this work the approach followed is based on the Exponential Effective SIR Mapping (EESM). In EESM mapping the set of coded symbols SINRs are mapped into a single value named the Effective SINR value. This value is used to match AWGN LUTs resulting from Link Level Simulations. The EESM [10] expression determines how the effective SINR is obtained from the multiple SINRs on different sub-carriers:

⎟⎟⎟

⎠

⎞

⎜⎜⎜

⎝

⎛−= ∑

=

−N

k

SINR

eff

k

eN

SINR1

1ln ββ (12)

Where β is a correction parameter used to adapt the formula to the different types of scenarios used in the simulations (different MCS schemes, MIMO techniques and channel modes). It must be optimized from link-level simulations for each type of modulation and coding rate combination used.

Authorized licensed use limited to: UNIVERSIDADE DE AVEIRO. Downloaded on December 16, 2009 at 08:21 from IEEE Xplore. Restrictions apply.

IV. DRA PERFORMANCE

A. Simulation Assumptions The system level simulation methodology was based on a combined dynamic snapshot approach, where each simulation is composed by a set of runs. In each simulation run users are uniformly distributed along the 3 sectors of the central BS. Neighbouring cells are assumed as transmitting with full power at full load. They contribute with inter-cell interference only. Path loss values, user positions and random shadowing fading are drawn and held constant throughout the whole run, whilst fast fading is simulated every frame period. At the beginning of each run (new snapshot) new user positions and shadowing values are re-computed. Simulations were conducted for a Single-Input Single-Output (SISO) transmission scheme with a simplified Maximum Ratio Combiner (MRC) receiver. Channel quality is estimated with no error and there is no error in the feedback of the CQI report in the CQICH channel and on the ACK/NACK report in the ACK channel as well. Simulation results are provided for full buffer and 3GPP Near Real Time Video with 64kbps data rate (NRTV64) traffic models with 1x3x1 frequency reuse for downlink connection only.

B. Performance Evaluation Figures 2 and 3 are, respectively, the plots of the CDF of the user service throughput and of the system throughput for the three schedulers, under the full queue traffic model.

Figure 2. CDF of the User Average Service Throughput

System Throughput - PedB - 3Km/h

02468

101214

OTAThroughput

OTA3GPP ServiceThroughput

Avg. OfferedLoad

Traf

fic (M

bps)

AvgCIRRCI

Figure 3. System Throughput

As can be seen from both plots the AvgCI scheduler is the one with best performance among the three proposed schedulers as it manages to provide transmission opportunities for users in the edge as well as closer to the BS. This is because the normalization of the instantaneous SINR by the mean SINR

results a symmetrical channel among all users, which compensates the differences among average SINR values. According to the plot of the CDF of the user service throughput in figure 2, as could be expected, the CI scheduler is totally unfair, although resulting in highest service rates, and the RR is the most fair one, but with the lower rates among three schedulers. The Avg CI results in a good compromise between fairness and service rate. The CI scheduler results in much better system throughput for the full queue traffic model. Therefore it provides a good estimate of the system capacity. However, this is achieved for a higher percentage of unsatisfied users, as can be seen from the heavy tail distribution in the plot of figure 2, associated to users with bad channel quality in the border of the cell. The difference between the average offered load and the service throughput is lower for the CI scheduler because users in the edge of the cell are not scheduled for transmission. This is confirmed from the plot of the MCS distribution in figure 4. The CI scheduler achieves the highest MCS schemes and the AvgCI results in a rather symmetric distribution along the set of MCS schemes, because it manages to service users closer to the cell centre (with higher MCS schemes), as well as further away from the cell centre (with more robust and lower MCS schemes due to bad channel quality). The CI scheduler schedules most of the transmissions with the highest MCS schemes because of its opportunistic nature.

Mcs Distribution

0,0

0,1

0,2

0,3

0,4

0,5

0,6

MCS0MCS1

MCS2MCS3

MCS4MCS5

MCS6MCS7

MCS8

Per

cent

age Avg CI

RRCI

Figure 4. MCS Distribution

Figure 5 is the CDF plot of the user service throughput for the AvgCI scheduler for the three ITU channel models used in the simulations. As expected both PedB and PedA channel models, with 3km/h mobile speed, result in significant better user service throughputs when compared to the VehA channel model with 30km/h mobile speed. This is because the system throughput is significantly higher with slow varying channels, as for channels exhibiting fast variations the scheduling decisions and the selection of the proper MCS scheme do not accurately match the actual conditions encountered during transmission. Also, the PedB channel has a larger delay spread and hence a smaller coherence bandwidth than the PedA channel. This means that with the PedB channel different sub-carriers will be affected differently along the spectrum of the OFDM symbol, while with the PedA channel model all sub-carriers will be roughly affected the same way, i.e. the channel is roughly flat for the PedA channel. As PUSC sub-channelization mode randomly assigns sub-carriers along the symbol spectrum, the fact that some sub-carriers will be more attenuated than others with the PedB channel model, result in

Authorized licensed use limited to: UNIVERSIDADE DE AVEIRO. Downloaded on December 16, 2009 at 08:21 from IEEE Xplore. Restrictions apply.

smaller values of the compressed EESM SINR value obtained from the CQI in the preamble of the frame. As all sub-carriers are roughly attenuated by the same value with the PedA channel model the EESM SINR value will be higher for this channel model. As can be seen, the PedA channel model result in the best performance in terms of system throughput among all three channel modes. In order to overcome the bad performance achieved with the VehA channel mode with 30km/h a different link to system level interface, based on average channel values should be used as look-up tables, or more robust MCS schemes could be employed.

Figure 5. CDF of user service throughput for the AvgCI

scheduler

Figure 6. System Throughput

Figure 7. CDF of User Service Throughput for NRTV users with

PedB and 3km/h

Figure 6 is the plot of the CDF of the user service throughput for the 3 schedulers under the 3GPP NRTV with 64kbps data rate, for 120 users and the PedB channel model with 3km/h. As

can be seen, about 65% of the users are not serviced with the CI scheduler and the service throughput for the 5% tile of users, which corresponds to those users in the edge of the cell, is equal to 30 kbps. The RR scheduler corresponds to roughly 5 kbps for the same 5% tile point. On the other hand both AvgCI and CI schedulers produce the same service throughput of roughly 60 kbps for the 90% tile of users, which are the ones closer to the base station. Among the three schedulers the AvgCI presents the best performance mainly for users in the edge of the cell.

Figure 6 is the plot of the CDF for the average packet delay under the same conditions. It can be confirmed that the AvgCI scheduler presents the best performance in terms of packet delays as 90% of the users have an average packet delay lower than 15 seconds while the RR and CI result in average packet delays of roughly 40 and 70 seconds respectively.

V. CONCLUSIONS In this paper a new DRA architecture, implemented according to the cross-layer functionality set available in the IEEE802.16e specification, is proposed. Three schedulers: CI, RR and AvgCI, are plugged into the DRA to evaluate its performance under a Full Queue Traffic Model. According to system level simulations it was verified the superior performance of the AvgCI scheduler and the feasibility o the cross-layer architecture.

VI. ACKNOWLEDGMENT This work was carried out in the scope of the WHERE project that is supported by the European Commission in the framework of FP7 ICT-2007-1-1.1 with contract no 217033.

VII. REFERENCES [1] IEEE802.16e (2005) [IEEE Std 802.16e, “Part 16: Air Interface for

Fixed and Mobile Broadband Wireless Access Systems”. [2] WiMAX Forum, Mobile WiMAX – Part I: A Technical Overview and

Performance Evaluation, Aug. 2006. [3] V. Srivastava, “Cross-Layer Design: A Survey and the Road Ahead”,

IEEE Communications Magazine, Dec. 2005. [4] Kwon T, Lee H, Choi S, Kim J, Cho D., “Design and Implementation of

a Simulator Based on a Cross-Layer Protocol Between MAC and PHY Layers in a WiBro Compatible IEEE802.16e OFDMA System”, IEEE Comm. Mag., pp. 136-146, Dec. 2005

[5] Balachandran K. et al, “Design and Analysis of an IEEE 802.16e-Based OFDMA Communication System”, Bell Labs Technology Journal, 11(4), pp. 53-73.

[6] Wang F, Ghosh A, Sankaran C, Fleming P, “WiMAX Overview and System Performance”, IEEE VTC-2006, pp. 1-5, Fall.

[7] Wang F, Ghosh A, Sankaran C, Benes S, “WiMAX System Performance with Multiple Transmit and Multiple Receive Antennas”, IEEE VTC-2007, pp 2807-81, Spring.Hoymann C, “Analysis and Performance Evaluation of the OFDM-based Metropolitan Area Network IEEE802.16”, Computer Networks 49(3), pp. 341-363.

[8] “WiMAX System Evaluation Methodology Version 1.0”, WiMAX Forum 2007.

[9] Viswanath P, Tse D, “Opportunistic beamforming using dumb antennas”, IEEE Transactions on Information Theory, June 2002.

[10] Ericsson, “Effective SNR Mapping for Modelling Frame Error Rates in Multiple-State Channels”, 3GPP2-C30-200340429-010.

Authorized licensed use limited to: UNIVERSIDADE DE AVEIRO. Downloaded on December 16, 2009 at 08:21 from IEEE Xplore. Restrictions apply.