Mobile Netw Appl (2008) 13:385–397 DOI 10.1007/s11036-008-0065-1

Dynamic Resource Allocation Architecture for IEEE802.16e: Design and Performance Analysis A. Nascimento & J. Rodriguez & S. Mumtaz & A. Gameiro & C. Politis

Published online: 2 July 2008 # Springer Science + Business Media, LLC 2008

Abstract Mobile communications has witnessed an exponential increase in the amount of users, services and applications. WiMAX (Worldwide Interoperability for Microwave Access) targets to provide broadband connectivity to wide area coverage, both in fixed and in mobile environments, as well as in the provision of QoS constraints of those applications and services envisioned for Next Generation Networks (NGNs), which results in significant design challenges in the MAC (Medium Access Control) to provide the seamless transport of heterogeneous traffic in a cost-effective manner. This paper proposes a Dynamic Resource Allocation (DRA) architecture for the IEEE802.16e broadband wireless system (also known as Mobile WiMAX) that can provide operators the flexibility to deliver broadband traffic with high spectral efficiency. The proposed DRA architecture framework encompasses scheduler, Link Adaptation (LA), Resource Allocation (RA) and Hybrid Automated Repeat Request (HARQ) functional blocks which interwork seamlessly. The performance of the DRA was evaluated using commonly A. Nascimento Departamento de Matemática e Engenharias, Universidade da Madeira, 9000-390 Funchal, Portugal A. Nascimento (*) : J. Rodriguez : S. Mumtaz : A. Gameiro Instituto Telec./Uni. Aveiro, 3810-193 Aveiro, Portugal e-mail: [email protected] J. Rodriguez e-mail: [email protected] C. Politis WiMM group, Kingston University, London KT1 2EE, UK e-mail: [email protected]

deployed scheduling policies: Max C/I, Proportional Fair and Round Robin schemes. Simulation results show that the proposed DRA scheme has the capacity to provide enhanced coverage and QoS provisioning for the area networks (WANs), such as the ones envisioned for B3G mobile wireless networks. Keywords IEEE802.16e . dynamic resource allocation . WiMAX . scheduling

1 Introduction During the last years the technological achievements in microelectronics, namely in the nanotechnology field, opened up the possibility of integrating highly complex and energy efficient signal processing devices within the mobile handset. This opened the door for new applications in terms of the portability of applications and services originally designed for the fixed Internet, plus the development of new ones specifically for the mobile environment. In fact, the mobile market suggested an exponential increase in the penetration of users, services and applications that led to the implementation of a new generation of mobile wireless networks and devices. While 3G cellular networks were mainly designed for voice and low bit-rate data services, multimedia applications (video, audio and data) are expected to be the main drivers for the mobile wireless networks of the near future (beyond 3G wireless networks—B3G). Multimedia applications place highly demanding constraints on both high data rate transmission and Quality of Service demands. Moreover, in contrast to fixed networks, the mobile radio channel is a very aggressive medium for the transmission of information, specifically for the envisaged scenario by the ITU2000 that

386

aims to address 100 Mbps under high mobility conditions. [1]. As a consequence, complex and innovative technologies in the field of signal processing have to be applied in the transmission chain in order to overcome the limitations imposed by such an aggressive medium. Although not being a new technology, Orthogonal Frequency Division Multiplexing (OFDM) is a modulation scheme which presents great benefits if applied to such environments [2]. Essentially, OFDM splits an input serial bit stream of information into parallel bit streams with much lower bit rate. The increase in the transmitted symbol duration over the radio channel results in a more robust signal that is more tolerant to delay spread resulting from multipath propagation and alleviates the constraints of the mobile radio channel on Broadband Wireless Networks (BWA). This simple idea avoids the use of such complex equalizers and RAKE receivers used in WCDMA transmission technologies, and presents OFDM as a strong candidate for adoption in B3G networks [2]. The performance of B3G networks is highly influenced by the efficiency in the utilization of the available radio resources (i.e. power, channels, time-slots and spatial beams in WiMAX IEEE802.16e). In addition, because radio spectrum is at a premium for wireless and cellular operators, it is important to investigate techniques that can maximize spectral efficiency to ensure cost-effective services are passed on to the end-user. As such, the systems envisaged for B3G networks must adopt efficient strategies for managing radio resources for heterogeneous packetbased services. Dynamic resource allocation (DRA) has been adopted as typical means of adapting the transport channels transmission capability according to the variability in traffic conditions, so that several users can share a common channel with high utility. Further efficiency can be achieved by exploiting cross-layer information from the PHY layer as well as from higher layers (i.e. application layer), to provide efficient mapping of data onto radio resources available within the Medium Access Control (MAC) layer of Wireless protocols [3]. Cross-layer results in the optimization of system performance (namely in terms of system throughput) while at the same time satisfies the applications QoS demands [4]. The emergence of Worldwide Interoperability for Microwave Access (WiMAX) [5] reflects a significant step towards a wireless networking ambient as it brings data services to the subscriber on the move, and provides a step closer to a fully data driven environment. Initially designed as a system for the provision of wireless broadband internet access for small business and residential fixed access users IEEE802.16-2004 [6], the requirement for wireless mobility has led to the IEEE802.16e [7], also known as Mobile WiMAX. Technologies such as Hybrid Automated Repeat Request (HARQ), Space Time Coding (STC), Advanced

Mobile Netw Appl (2008) 13:385–397

Antenna Systems (AAS), Multiple Input Multiple Output (MIMO) and Spatial Division Multiple Access (SDMA) have been added to support mobile environments and to improve the broadband access speed [7, 9]. Two of the most promising attributes offered by WiMAX is the potential for OFDM to overcome the restrictions of the radio channel due to multipath propagation, namely in non-line-of-sight (NLOS) scenarios and the use of Orthogonal Frequency Division Multiplexing (OFDMA) which provides an efficient multiple access scheme for the efficient use of radio spectrum. This latter issue is particularly suitable for adaptive transmission and resource allocation due to the existence of parallel sub-channels in the 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 particular Mobile WiMAX can provide three distinct types of resources/degrees of freedom [8]: frequency sub-channels in the frequency domain, OFDM symbols in the time domain, and additionally spatial beams in the space domain through the use of MIMO antennas. Additional sub channelization options have been added to Mobile WiMAX to improve operating conditions. In this article we propose a flexible Dynamic Resource Allocation (DRA) architecture framework specifically envisioned for the WiMAX standard, which can exploit its inherent features to deliver seamless transport of heterogeneous traffic in a cost-effective manner. The proposed DRA implements a cross-layer architecture model by the use of the functionalities which were specified in the standard, such as control channels for the fast exchange of signaling messages for the HARQ module and for fast and efficient channel estimation and feedback [3]. The individual modules, which constitute the DRA, and the way in which they interact under the proposed cross-layer protocol model, are presented in detail and its performance analyzed by means of simulations. Central to the performance of the DRA is the type of scheduler which is implemented. The scheduler was intentionally left out from the specifications as a means to provide differentiation among equipment manufacturers and operators. In this article the performance of the proposed DRA is evaluated under the implementation of three commonly used schedulers: Max CI, Proportional Fair and Round Robin. This article is organized as follows: Section 2 presents related work in the field of WiMAX system level simulation methodologies and current architectures for resource allocation and scheduling; Section 3 describes the WiMAX system models for system level analysis and scenarios, Section 4 presents the proposed DRA architecture framework and its constituent modules; Section 5 describes the core of the DRA scheme, namely the approach followed in

Mobile Netw Appl (2008) 13:385–397

our simulation platform concerning the implementation of the Resource Map of the WiMAX system; Section 6 presents the simulation results for the proposed DRA architecture, and the conclusions in Section 7.

2 Related work Dynamic Resource Allocation and Scheduling is one of the most fundamental problems in wireless networks. In particular, although significant work has been realized regarding the performance of WiMAX networks, pertaining to the scheduling of Real Time (RT) and Non-Real-Time (NRT) service flows in both fixed (IEEE802.16-2004) and Mobile (IEEE802.16e) cases, very few of the different approaches deal with DRA architectures which can be implemented in real WiMAX networks. Also, the performance of many of these algorithms is based under simpler network layouts and analytical models. In [10] the authors investigate the performance of a WiMAX network based on link and system level simulations. Some DRA methods for the basic WiMAX system profile are proposed and their performance investigated. The authors also make some recommendations on design and deployment configurations. Standard scheduling algorithms such as the Maximum C/I (Max CI) and Proportional Fairness (PF), coupled with simple full queue traffic model are used in the analysis of the performance of the DRAs. In [11] a cross-layer architecture framework for performance improvement of WiMAX compatible WiBro system in Korea is presented and investigated. The framework architecture is based on the exchange of signaling information between Physical (PHY) and Medium Access Control (MAC) layers, by means of the control channels available in the standard. Figure 1 WiMAX frame structure

387

In [3] the authors analyze the performance of a prototype for a WiMAX IEEE802.16e network and analyze its performance under a variety of traffic models for commonly services envisaged for Next Generation Networks (NGN), such as web browsing (HTTP) and FTP. System and Link level simulations were conducted and the performance of the network was evaluated and compared to the performance of 3GPP High Speed Downlink Packet Access (HSDPA) system, by means of typical metrics such as system and user throughput as well as spectral efficiency. In [12] the performance of Mobile WiMAX for different scenarios such as the use of Multiple Input Multiple Output (MIMO) antennas for the downlink (DL) and uplink (UL) are investigated. System level simulations were used to determine the throughput performance in DL and UL of proposed WiMAX network architecture for full buffer and HTTP traffic models. In particular the authors investigate on the coverage efficiency of the different scenarios considered for the control channels in the MAC frame. A comprehensive set of simulations are conducted for different kinds of channel models. In [13] the authors are concerned in particular with the performance of WiMAX networks using MIMO schemes. Again system and link level simulations were used to infer about the performance of WiMAX networks under such scenarios via performance metrics such as sector and user throughput, spectral efficiency and network quality, measured as the estimated packet error rate (PER). In [14], the system performance for the IEEE802.162004 standard is given, a.k.a. Fixed WiMAX. The influence of configuration parameters such as packet fragmentation, the amount of padding bits and the size of the payload of each MAC Protocol Dat Unit (PDU) packet on system performance are analyzed with recommendations for optimum system parameters.

388

To the best of the authors knowledge most of the approaches available in the literature, for investigating WiMAX system performance under realistic scenarios do not consider and/or investigate scheduling algorithms specifically designed for the provision of Quality of Service (QoS). Although the authors in [3] elaborate on a generic scheduling algorithm for QoS provisioning via the implementation of Utility Function principle [15, 16], no consideration is given to DRA design for RT services. Also the work available in the literature does not consider the specifics of the system behavior when Hybrid Automated Repeat Request (HARQ) protocol is implemented. 3 WiMAX system level modeling In this section we address the steps followed in the modeling of WiMAX system level aspects. The proposed DRA configuration is described in greater detail.

Mobile Netw Appl (2008) 13:385–397

Adjacent sub-carrier allocation Sub-carriers are contiguously allocated and associated in small chunks called bins. This is a frequency specific sub-channelization scheme with an intrinsic gain called multi-user frequency allocation gain. This channel mode is very efficient in scenarios where the radio channel is stable for many radio frames, typically in scenarios of standing still mobiles and with small inter-cell interference. In this article we focus primarily on the so-called Full Usage Sub-carrier Channelization (FUSC) channel mode. In FUSC mode every sub-channel is available for allocation in the cell. The data sub-carriers are randomly allocated along the whole spectrum of the FFT. In the FUSC mode the small unit of allocation is the slot, equal to one subchannel in frequency domain by one OFDM symbol in the time domain. A group of slots defines a burst in the radio frame. A burst can be assigned to more than one user provided the same modulation and coding scheme is followed in the transmission of all packets allocated to it.

3.1 WiMAX frame structure Figure 1 illustrates a particular implementation of the IEEE802.16e TDD frame structure for DL-PUSC (Partial Used Sub channelization Scheme). Although Mobile WiMAX supports both Frequency Division Duplexing (FDD) and Time Division Duplexing (TDD), only the TDD mode is being supported by the system profiles designed by the WiMAX forum for equipment compliance and interoperability. Also TDD offers some advantages over the FDD mode such as the support of asymmetrical data rates in UL and DL and also fast estimation of the DL channel in the UL transmission. In the frame, resources are available in two domains: frequency (sub-carriers) and time (OFDM symbols). The OFDM symbols available for data transmission and the sub-carriers constitute distinct logical slots. Different types of logical channels are available, depending on the type of channel model used. The standard [7] defines two basic types of generic channel models: channel models that are intrinsically diverse in frequency called diversity subcarrier permutation modes and the channel modes that make use of the frequency selective nature of the radio channel: Diversity sub-carrier permutation Sub-carriers are assigned to each logical sub-channel by pseudo-randomly distribution across the whole set of data sub-carriers available for data transmission. The frequency diversity of the DSP subchannelization scheme is effective for moving mobiles with fast changing radio channels (i.e. with time coherence of a few radio frames).

3.2 SINR modeling In the simulation a wideband SISO channel model is implemented by a six tapped delay model, according to the Ped. B 3 km/h channel model [17]. The narrowband fading channel is generated by a Jake’s model where the carrier frequency and the speed are used to define the statistics of the fading according to [18]. Slow fading is modeled according to a log normal distribution and spatial shadowing correlation between mobiles and Base Station (BS) is implemented. The shadowing SHj(x, y) in dB between one Mobile Station (MS) at position (x, y) and BS is given by: pffiffiffiffiffiffiffi  SH j ðx; yÞ ¼ 0:5 F0 ðx; yÞ þ Fj ðx; yÞ ð1Þ Where F0 (x, y) and Fj (x, y) are spatial functions following a Gaussian distribution with zero mean and standard deviation (σ) in dB, generated using the method described in [19]. They have a spatial correlation given by: Rðd Þ ¼ exp½ ln ð2Þ DecorreLength  d

ð2Þ

Where d is the distance between two points in the network layout and DecorrLength is the shadowing de-correlation length in meters. Using the aforementioned channel models, the SINR (Signal-to-Interference-plus-Noise Ratio) of each OFDM sub-carrier is computed according to the approach followed in [20]: gk ¼

Ior Nused   Hk Ioc þ N0 Nd þ PDR  Np

ð3Þ

Mobile Netw Appl (2008) 13:385–397

389

Where p represents the multi-path index for the Ped. B channel model Ak is the amplitude value corresponding to the long-term average power for the pth path of this same channel model fk is the relative frequency offset of the kth sub-carrier of the specific FUSC sub-channel and Tp is the relative time delay of the pth path. 3.3 Link level interface In the receiver, post-processing of the signals received from the serving and interfering BSs is performed. For a SingleInput-Single-Output (SISO) architecture and a matched filter in the receiver, the received signal at the kth subcarrier for the target user is computed according to:

Figure 2 BLER vs. SNR dashes AWGN, inverted triangle EESM)

Where Nused is the total number of sub-carriers, PDR is the Pilot-to-Data sub-carriers power ratio and Nd is the number of data sub-carriers per OFDM symbol, for the FUSC channel mode. No is the receiver thermal noise power and Ioc is the other-cell interference power density, assumed spatially and temporarily uncorrelated. It is assumed that neighboring cells transmit with maximum power, i.e. with full load. The gain of the kth sub-carrier is computed according to the recommendations of [21] as follows: 2 N paths  X  jqp e2pfk Tp  Hk ¼  Mp Ap e    p¼1

Figure 3 WiMAX dynamic resource allocation module (DRA)

ð4Þ

Y ð0Þ ½k  ¼

qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð0Þ ð0Þ Pslot Ploss H ð0Þ ½k X ð0Þ þ ½k  N qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi X ð jÞ ð jÞ þ Pslot Ploss H ð jÞ ½k X ð jÞ ½k  þ N ð0Þ ½k 

ð5Þ

j¼1 ð0Þ

Where N is the number of interferers, Pslot is the transmit ð jÞ power per slot for the jth cell, Ploss is the distance dependent path loss, including shadowing and antenna gains/losses, H(j)[k] is the channel gain for the desired MS from the jth cell and for the kth sub-carrier, X(j)[k] is the transmitted symbol by the jth cell at the kth sub-carrier, N(0)[k] is the thermal noise at the received sub-carrier, modeled as AWGN with zero mean and variance s 2n . The transmission of a coded block over different sets of sub-carriers results in a number of SINR measures that

390

Mobile Netw Appl (2008) 13:385–397

equals the number of sub-carriers sets, which can be quite high. Hence, data compression is mandatory. The coded symbol SINR in sub-carrier k is given by:

ð0Þ

SINR ½k  ¼

N P j¼1

2 ð0Þ ð0Þ  Pslot Ploss H ð0Þ ½k  2 ð jÞ ð jÞ Pslot Ploss jH ð jÞ ½k j þs 2n

ð6Þ

  ðiÞ BLERk ðnÞ ¼ f g k;CQI ðnÞ þ g Lk ; MCS k ðnÞ

The set of coded symbols SINRs are mapped onto a single value named the Effective SINR value. This value can be used to match AWGN Look-Up Tables (LUTs). The EESM [22] expression determines how the effective SINR is obtained from the multiple SINR’s on the different subcarriers:

SINReff

SINR P  b p 1 X ¼ b ln e P p¼1

scheme, and BLERi is the predicted BLER for the MCS scheme, which is a function of the predicted Channel Quality Information (γk, CQI) reported from each MS and on the accumulated SINR value on the Chase Combiner for L1 P g k;CQI ðn  t Þ), as previous transmission attempts (g Lk ¼ t¼1 follows:

! ð7Þ

Where the parameter β is to be optimized for every link mode (MCS) based on link level simulation results. Figure 2 shows that the EESM approach results in BLER points very close to the ones obtained by simulation over an AWGN channel, from link level simulations.

ð9Þ

The DRA selects the most spectrally efficient MCS, i.e., the one that maximizes the achievable bit rate and at the same time keeps the predicted Block Error Rate (BLER) under the desired threshold level (10% in the simulations). In order to perform adaptive modulation and coding, Channel State Information (CSI) must be measured and reported by each MS by means of a feedback channel (for FDD duplexing mode) or by means of the UL transmissions (for TDD duplexing mode). We assume the CSI is estimated by each MS attached to the cell by means of the SINR value of the OFDM symbol carried in the frame’s preamble. We also assume that the CSI is reported back to the BS via the CQICH control channel on the UL subframe, on a frame-by-frame basis. Due to the intrinsic nature of FUSC sub-channelization mode we assume the same CSI value for each one of the 16 sub-channels in the frame. 4.2 Asynchronous hybrid automatic repeat request

4 DRA architecture The proposed DRA scheme is composed of the following modules: Packet Scheduler, Hybrid Automated Repeat Request (HARQ) with Chase Combiner and Link Adaptation. It is summarized in Fig. 3. 4.1 Link adaptation At the beginning of each MAC frame the choice of the proper transmission mode (i.e. Modulation and Coding Scheme used—MCS) is defined by the Link Adaptation Module. We have considered 10 MCS schemes encompassing QPSK, 16QAM and 64QAM modulation schemes and the Convolution encoder according to the profiles envisioned by the WiMAX forum. For each selected user in each burst, the MCS scheme to be used is chosen according to the following decision rule: i ¼ arg max ½Ri ð1  BLERi Þ i2MCS set

ð8Þ

Where MCSset represents the set of modulation and coding schemes, Ri is the throughput achieved for the MCS

The number of slots used in the transmission of the packet depends on its size and on the type of modulation and coding scheme used which defines the size of each individual slot in the MAC frame and is an output of the Link Adaptation. The amount of slots used in the transmission of the packet is named a burst and it is assumed each burst is assigned only one single packet. The packet is concatenated with padding bits, if necessary, processed and mapped onto the burst. IEEE802.16e specifies HARQ procedures for error recovery where information resulting from multiples transmissions of the same packet is softly combined by the Chase Combiner. Each MS has 4 simultaneous HARQ processes for transmission of information. Each HARQ process is in charge of the transmission and re-transmission of a single packet, until this packet is successfully received, and is associated to one buffer in the MS to store the result of the combination of successive versions of the same packet. On the reception of each version of the packet, the MS combines this current version with previous ones of the same packet using Chase Combining. Once an HARQ process has been selected for transmission, the DRA must wait for an ACK/NACK message from the MS before selecting the HARQ process again.

Mobile Netw Appl (2008) 13:385–397

391

mobile with the best channel quality indicator (CQI) for transmission, according to the following rule: k ðnÞ ¼ arg max Ri ðnÞ i2f1;...;K g

n ¼ 0; 1; 2; . . .

ð10Þ

The CI scheduler is very effective in resources allocation in terms of capacity maximization, but is totally unfair and completely blind towards user QoS requirements. 4.3.2 Proportional Fair (PF) The PF scheduler was proposed in [23] as a means to combat the inefficiencies of the CI scheduler. At the beginning of each radio frame n the scheduler selects for transmission the mobile with the highest ratio of estimated maximum data rate Ri(n) to current average throughput Ti(n), according to the following rule:

Figure 4 WiMAX resource allocation map (RAM)

k ðnÞ ¼ arg

4.3.1 Max C/I (CI) In order to address the throughput maximization potential in the proposed DRA design, we have used the Max C/I scheduler [22]. The CI scheduler opportunistically assigns resources to the user with the highest channel gain by scheduling, at the beginning of each radio frame n, the

Phase D1 Process ACK/NACK messages received in TTI n-1

Phase A1 Compute Resource Allocation Map TTI n

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

No transmission

No transmission

Phase B1 Brodcast Resource Allocation Map TTI n

Phase C1 Transmit Packets according to Resource Allocation Map TTI n Phase D1 Process ACK/NACK messages received i TTI n TTI n+1

ð11Þ

4.3.3 Round Robin (RR)

RA Info

TTI n

; n ¼ 0; 1; 2

If the radio channels to the users are symmetrical and i.i.d., the PF scheduler is fair in terms of resource allocation and effective in terms of capacity maximization.

4.3 Scheduler

TTI n-1

max Ri ðnÞ i2f1;...;K g Tn ðnÞ

Phase A1 Compute Resource Allocation Map TTI n+1 Base Station HARQ channel

Figure 5 WiMAX DRA cycle

RA Data ACK/NACK No transmission

No transmission

Phase D2 Send an ACK/NACK message if MS has received a packet in TTI n-1

TTI n-1

Phase A2 Receive Resource Allocation Map TTI n Phase B2 Receive data (packets) if MS is in the Resource Allocation Map TTI n Phase C2 Determine if re-transmission is required TTI n Phase D2 Send an ACK/NACK message if MS has received a packet during TTI n

Mobile Station HARQ channel

TTI n

392

Mobile Netw Appl (2008) 13:385–397

priority to each user in a sequential way without taking into consideration any user QoS requirements.

5 Resource map definition A schematic description of the Resource Allocation (RA) map for the DL sub-frame of the WiMAX TDD frame is summarized in Fig. 4. The Resource Allocation Space is a matrix of subchannels per OFDM symbols. The minimum granularity of resource allocation is the Radio Access Unit (RAU). The RAU constitutes one FEC (Forward Error Correction) block and is physically mapped into four consecutive slots, i.e. one sub-channel and four OFDM symbols, for the DL FUSC channelization mode. In the Resource Allocation Space each burst is composed of a group of continuous RAUs. Each RAU is referenced by a pair of indices RAU(i, j), where i and j are respectively the sub-channel and the FEC block indices. Each RAU stores information regarding the assignment of the MS and MCS scheme used. Subsequent to scheduling and mapping of the IP packets to the available RAUs, the BS broadcasts the map with the resource allocations (Resource Allocation MAP—RAM) to the network via DL-MAP control fields in the DL subframe. Each MS utilizes this signaling information to anticipate packet transmission and to perform the demodulation and decoding of the transmitted data using the appropriate MCS scheme. The Resource Allocation Map is updated every frame by the BS. In the beginning of each TDD frame the DRA assigns free slots from the resource map to packets stored in the user’s queues, according to the individual priorities assigned to each packet by the scheduler. Whenever a MS is selected by the DRA for DL transmission over the PHY interface, the DRA withdraws the selected packet from the respective queue. The amount of slots requested for the transmission of the packet depends on the size (on bits) of each individual slot. The size is computed by the Link Adaptation Module. The whole process elapses along a cycle equivalent to four frames and is illustrated in Fig. 5. & &

&

Phase A1: The BS computes the Resource Allocation Map to be used for data transmission during the Phase C1 that follows. Phase B1: The BS broadcasts the Resource Allocation Map determined during Phase A1, so that all users in the cell have the information required to demodulate the packets they will receive during Phase C1. Phase C1: The BS sends data according to the Resource Allocation Map determined in Phase A1.

&

Phase D1: The BS receives ACK messages from users which have received their packets successfully during Phase C1, and likewise receives NACK messages from users which have received their packets un-successfully during Phase C1. The MS receives and processes each packet mapped onto a set of slots. If the received packet is only another version of a first packet transmission the MS performs Chase Combining on the different replicas of the same packet. After decoding the received packet the MS sends an ACK/NACK message according to the quality of the received packet. This is to inform the BS about the future status of the HARQ transmission buffer. In order to achieve continuous data transmission four different HARQ processes are considered (from 0 to 3) per user. Each HARQ channel is mapped onto one of a total of four TDD frames and each HARQ process of each MS handles only one packet mapped onto one or more slots of the resource map. 5.1 Resource allocation procedures In what follows, the proposed Resource Allocation Map computation is described for the HARQ process 0, but it is equally applicable for all other HARQ processes. Here, we describe the resource allocation procedures to encompass the methodologies for packet scheduling, resource map allocation and link adaptation. The proposed resource allocation procedures are sub-divided in four steps: 5.1.1 Determining mobile stations lists Two lists of MSs are computed: the initial transmission and retransmission lists. The first transmission list is filled with any MS with its HARQ process number 0 inactive and with packets waiting in the user queue for transmission. The retransmission list is filled in with any MS with its HARQ process number 0 active and waiting for a retransmission opportunity. It is assumed that the BS hosts a waiting queue for each MS. 5.1.2 Scheduling re-transmissions For each RAU, the BS initially checks the re-transmission list. If the slot is not allocated to a MS of the re-transmission list, then it is temporarily not allocated. The current total data transmission power Pmax is Pbudget computed. It is the sum of the transmit power allocated to each slot. The remaining transmission power is updated according to the following expression: Pbudget ¼ Pmax  Pdata , where Pmax is the maximum BS transmission power available for data. The retransmissions are allocated the same slots, MCS scheme and transmit power as for the first transmission.

Mobile Netw Appl (2008) 13:385–397

393

Table 1 WiMAX system level simulation parameters Symbol

Quantity

Value

fc Bs Pmax No NFm CT dis Gbs

Carrier frequency System bandwidth BS max. tx. power Thermal noise density MS noise figure Cell type Inter-site distance BS antenna gain Shadowing standard deviation Shadowing de-correlation length Path loss in dB for urban environment

2.5 GHz 10 MHz 43 dBm −174 dBm/Hz 7 dB Tri-sectored 2.7 km 15 dBi 8 dB 20 m 128.1+ 37.6log10(dm)

σs

dsh PL

tion of bursts, the MCS scheme used in the transmission and respective transmission power assigned to each slot.

6 Simulation results 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, the users are uniformly distributed across the cell and the path loss values, user positions and random shadowing fading are drawn and held constant throughout the whole run, whilst fast fading is simulated on a TTI basis. User positions and new values for the shadowing and path loss are re-computed at the beginning of each new run.

5.1.3 Scheduling new transmissions and link adaptation 6.1 Simulation assumptions The first transmission list is split into two sub-lists: one list of high priority MSs and the other is the low priority list. &

&

High Priority List—MSs with packets waiting in the queue for a period of time longer than a specific threshold are placed into the high-priority list. The delay threshold is service specific: for real-time services such as voice services or near-real time video, the delay threshold is zero; for non-real time services such as the web traffic, the delay threshold can be larger. Low Priority List—the remaining users are put into the low priority list.

Both lists are then sorted in descending order of the priority as defined in each one of the three schedulers proposed for analysis of the DRA. The BS selects the mobiles with highest priority in each list, where mobiles from the high priority list have preference. Once a mobile is assigned to the RAU, the link adaptation mechanism selects the appropriate MCS scheme according to the predicted SINR (SINRpred), based on the OFDM symbol transmitted in the frame’s preamble and reported from each mobile. The MCS allocation criteria is given by Eq. (8). The traffic model considered in the simulations is the full queue. In full queue, all users are assumed to have an infinite amount of bits waiting in their dedicated queues located at the BS. In addition, only one user is allowed to transmit at each transmission time interval (TTI) which, in the WiMAX system, is equal to one TDD frame. This means that all resources available in the frame are assigned to a single user at each TTI.

Users are uniformly drawn along the three sectors of the central BS of a cellular network composed of three tiers of BSs. Neighboring cells are assumed as transmitting with full power at full load, i.e. they contribute with inter-cell interference. A SISO transmission scheme with a simplified Maximum Ratio Combiner (MRC) is used in the receiver at each MS. Each slot transmits with the same power resulting from the maximum transmit power at the BS for data transmission uniformly distributed along the whole set of slots in the frame. It is also assumed channel quality is estimated with no errors and there is no error in the feedback of the CQI report by each MS. CQI is reported with a delay equal to two frames. Simulations were conducted assuming an urban deployment model, i.e., assuming tri-sectored cells and mobile speeds of 3 km/h. Each simulation run lasted for 100,000 TTIs, with 50 snapshots (runs) in total. In each TTI we prioritized the users according to the proposed DRA architecture considered. Within each run 12 users are dropped per sector within the central cell and with full queue traffic model. Table 1 provides the key simulation parameters for the WiMAX system level simulations. The following evaluation metrics were used for performance evaluation: &

Geometry (G factor) defined as: GðMS Þ ¼

1 antGðCell 0 ; MS Þx PLðCell 0 ;MS ÞxSH ðCell 0 ;MS Þ N P k¼1

1 antGðCell k ; MS Þx PLðCell k ;MS ÞxSH ðCell k ;MS Þ

ð12Þ 5.1.4 Interaction with the physical layer Finally the Resource Allocation Map is broadcasted by the BS, providing the respective MSs, with data mapped in the resources map, with control information regarding the alloca-

Where Cell0 and Cellk is the respective serving sector and interfering sector of MS; antG(Cell0, MS) is the antenna gain between the serving sector and the MS; PL(Cell0, MS) and SH(Cell0, MS) are, respectively, the path-loss and shadowing

394

Mobile Netw Appl (2008) 13:385–397 4

5

CDF of User Average Throughput

User Number of Transmissions vs. G

x 10

100%

FER = 0 FER>0

4.5

% of users with avg Throughput < X

4

User Number of Transmissions

MAX C/I RR PF

90%

3.5 3 2.5 2 1.5 1

80% 70% 60% 50% 40% 3%

0.5

20%

0 -5

0

5 10 User Geometry in dB

15

20

0

20

40

60

80

100

120 140 160 X in Mbps

180

200 220

240

Figure 6 No. of user transmissions vs. G factor

Figure 8 CDF of user average throughput

loss in linear scale between the BS of the serving sector and the MS respectively and N is the number of interfering sectors.

distinguished in terms of FER. Users with bad channel conditions have a Frame Error Rate metric greater than zero (FER>0) and alternatively users with good channel conditions have FER=0. It can be seen that the number of transmissions increases with the geometric factor since the scheduler is clearly providing preferential treatment to users closer to the BS. Furthermore, the transmission samples are highly populated between 7–13 dB which indicates the benefits of the proposed architecture in terms of throughput achievements and the quality of the signal transmitted. It is shown that users with lower channel quality are also provided service, since users experiencing several retransmissions and consequently delayed are given higher priority than first time transmissions. The beneficial effects of the proposed DRA architecture can also be visualized from Fig. 7, which shows the average user service throughput vs. the geometric factor as a function of FER. It can be seen that users closer to the cell boundary can also be serviced thus enhancing the quality of service delivered by the proposed DRA architecture for the WiMAX system. It can be also seen that the number of transmissions increases with the geometric factor, since the scheduler is clearly providing preferential treatment to users closer to the BS. Furthermore, the benefits of the proposed architecture can be seen here. The inherent ability of the DRA architecture to deliver better QoS due to the “retransmission priority tables” means that users further away from the BS will have increased opportunities to deliver their packets error-free in comparison to the classical CI scheduling approach.

&

Service cell throughput is used to study the network throughput performance and is measured as: Rservice ¼

b k  xT

ð13Þ

Where b is the total number of correctly received data bits by all MSs in the simulated system over the whole simulated time, k is the number of cells/sectors in the simulation and T is the total simulated time. 6.2 DRA performance Figure 6 shows the number of scheduled transmission attempts vs. the geometry factor for the CI scheduler, where users are 8

2.5

User Service Throughput vs. G

x 10

FER=0 FER>0

User Service Throughput in bps

2

1.5

1

0.5

0 -5

6.3 DRA schedulers usage

0

5 10 Geometry in dB

15

Figure 7 Average user service throughput vs. G factor

20

The proposed DRA architecture was evaluated for the three different types of schedulers: Round Robin (RR), Proportional Fairness (PF) and Max C/I (CI). As can be seen from Figs. 8, 9 and 10, and as expected, the schedulers

Mobile Netw Appl (2008) 13:385–397

395

CDF Average Packet Delay Per User (ms) 2.5

1

x 10

8

User OTA Throughput vs. distance MAX C/I RR PF

2

0.8

User OTA Throughput

Prob. Average Packet Delay < Abcissa

0.9

0.7 0.6 0.5 0.4

1.5

1

0.3

0.5

0.2 CI PF RR

0.1 0

0

50

100

0

0

150

200

400

600

800

1000

1200

1400

1600

1800

Distance in meters

Average Packet Delay Per User (ms)

Figure 9 CDF of packet delays

Figure 11 User average throughput vs. distance

performance, in terms of system throughput, packet delay and spectral efficiency, are ordered in the following manner: RR
delays than PF is the fact that this scheduler is efficient on resources allocation. A user selected with high channel quality translates to a much less robust MCS scheme and, hence, a more spectral efficient transmission, resulting in a higher transmission capacity and, consequently, smaller packet transmission delay. Furthermore the CI scheduler consistently selects users closer to the BS, thus resulting in very high channel quality reports. The probability of the Link Adaptation mechanism in selecting robust MCS schemes is almost negligible, as shown in Fig. 10. The continuous selection of higher order MCS schemes results in higher spectrum efficiencies. Figure 11 plots the average user OTA throughput vs. the distance from the BS. It can be seen that users closer to the cell edge have reduced transmission opportunities.

6.3.1 Max C/I It can be observed from Figs. 8 and 10 that the CI scheduler is unfair in the assignment of transmission opportunities for mobiles closer to the cell edge, resulting in a large number of dropped packets due to the expiration of the maximum allowable delay. Moreover, Fig. 8 shows a high throughput is achieved at the expense of fairness as 40% of the users in the cell do not transmit. Figure 9 shows the CDF of packet delays for the three schedulers. The reason for the CI to have smaller packet

Modulation and Coding Scheme Utilization 0.5

0.4

DL Average Service Throughput

MAX C/I RR PF

18 16

0.35

14

0.3

12

0.25

10

Mbps

% of Blocks sent with MCS in x axis

0.45

0.2

8

0.15

6

0.1

4 2

0.05

0

0 QPSK 1/2 QPSK 2/3 QPSK 5/6 16QAM 1/2 16QAM 2/3 16QAM 3/4 16QAM 5/6

Modulation and Coding Scheme

Figure 10 Modulation and coding scheme distribution

CI

PF DL/UL ratio = 1.0

Figure 12 Average user service throughput vs. G factor

RR

396

6.3.2 Proportional fair As expected, it can be noticed that in terms of the average throughput per user, the performance of the PF scheduler is between the CI and RR schedulers, as can be observed from Fig. 8. It can also be noted that users closer to the cell edge are not allowed to transmit because of their poor channel quality. The comparative analysis of Figs. 8, 9 and 10 clearly highlight the trade-off between fairness and spectral efficiency that can be achieved by the PF scheduler.

Mobile Netw Appl (2008) 13:385–397

that uses both CSI and packet retransmission status for assigning user priority. The proposed DRA also implements a practical architecture that can be implemented in a real WiMAX network. The DRA architecture has been specifically designed and tested for a WiMAX 802.16e system. The simulation results show that the DRA scheme has the capacity to increase the delivered QoS within WiMAX whilst utilizing spectrum efficiently and moreover has inherent flexibility to regulate the degree of QoS provisioning, and support heterogeneous traffic.

6.3.3 Round Robin References For RR, scheduling performance is contrary to that of the CI scheduler. It is completely fair and disregards the SINR at the user location, which results in a less efficient scheduler in terms of maximization of system capacity. In Fig. 8, it can be seen that almost all users in the cell are selected for transmission, but 80% of the users have an average throughput lower than 20 Mbps. Moreover, it can be seen from Figs. 10 and 11 that the inherent fairness attribute of the RR leads to users closer to the cell edge being selected for transmission. However, the poor channel conditions results in the lowest order MCS selection leading to a low user throughput. In contrast with the CI scheduler, the users at the cell edge have fewer transmission opportunities leading to the lowest average user throughput. Figure 12 shows the achieved average throughput for each one of the three schedulers. As the UL sub-frame resource allocation and scheduling is not implemented, the DL/UL ratio is equal to 1.

7 Conclusions This article proposes a DRA architecture framework for the Mobile WiMAX system. The major functional blocks of the proposed DRA are the Scheduler, Link Adaptation, Resources Allocation and Hybrid Automated Repeat Request modules. A prioritization scheme was proposed that assigns priority according to the status of the Arq Process and uses cross-layer information to ensure costeffective transmission by assigning the most appropriate transmission format according to the CSI and required QoS profile. The proposed DRA architecture was evaluated in terms of user throughput and packet delay for well-known scheduling policies that includes the Proportional Fair scheduler, Round Robin and Max C/I schedulers. The proposed DRA architecture has been designed to provide a degree of trade-off between maximizing system capacity and user QoS, thanks to the inherent prioritization scheme

1. ITU-R (2003) Vision, framework and overall objectives of the future development of IMT-2000 and systems beyond IMT-2000. draft new recommendation ITU-R M. [imt-vis] [doc. 8/110], Feb 2. Cimini LJ (1985) Analysis and simulation of a digital mobile channel using orthogonal frequency division multiplexing. IEEE Trans Comm COM 33(7):665–675 June doi:10.1109/TCOM.1985.1096357 3. Kwon T, Lee H, Choi S, Kim J, Cho D (2005) Design and implementation of a simulator based on a cross-layer protocol between MAC and PHY layers in a WiBro compatible IEEE 802.16e OFDMA System. IEEE Comm Mag :136–146, Dec doi:10.1109/MCOM.2005.1561931 4. Cimini LJ (2006) A tutorial on cross-layer optimization in wireless networks. IEEE J Sel Areas Comm 24(8):1452–1463 Aug doi:10.1109/JSAC.2006.879351 5. WiMAX forum, http://www.wimaxforum.org 6. IEEE Std 802.16-2004 (2004) IEEE standard for local and metropolitan area networks—Part 16: air interface for fixed broadband wireless access systems. Oct 7. IEEE802.16e (2005) [IEEE Std 802.16e, “Part 16: Air Interface for Fixed and Mobile Broadband Wireless Access Systems”, Dec 8. Yaghoobi H (2004) Scalable OFDMA physical layer in IEEE802.16. Intel Technol J 8(03), Aug 9. Mobile WiMAX—Part I: a technical overview and performance evaluation, Aug. 2006, WiMAX Forum 10. Balachandran K et al (2007) Design and analysis of an IEEE 802.16e-Based OFDMA communication system. Bell Labs Tech J 11(4):53–73 doi:10.1002/bltj.20196 11. Kwon T, Lee H, Choi S, Kim J, Cho D (2005) Design and implementation of a simulator based on a cross-layer protocol between MAC and PHY layers in a WiBro compatible IEEE 802.16e OFDMA System. IEEE Commun Mag :136–146, Dec doi:10.1109/MCOM.2005.1561931 12. Wang F, Ghosh A, Sankaran C, Fleming P (2006) WiMAX overview and system performance. IEEE VTC-2006, pp 1–5, Fall 13. Wang F, Ghosh A, Sankaran C, Benes S (2007) WiMAX system performance with multiple transmit and multiple receive antennas. IEEE VTC-2007 :2807–281, Spring 14. Hoymann C (2005) Analysis and performance evaluation of the OFDM-based metropolitan area network IEEE 802.16. Comput Networks 49(3):341–363 Oct 15. Liu P, Berry R, Honig ML (2003) Delay-sensitive packet scheduling in wireless networks. IEEE WCNC 3:1627–1632 Mar 16. Song G, Li Y (2003) Adaptive resource allocation based on utility optimization in OFDMA. Proc. IEEE GLOBECOM 2:586–590 17. ITU-R M.1225, 1997

Mobile Netw Appl (2008) 13:385–397 18. Li Y, Huang X (2000) The generation of independent Rayleigh Faders. IEEE ICC 1:41–45 Jun 19. Xiandong et al (2003) A two-dimensional channel simulation model for shadowing processes. IEEE Trans Veh Technol 52(6), Nov 20. Fan Wang et al (2005) 3GPP TR IEEE802.16e System performance analysis and simulation results. Proc of PMIRC, Sept

397 21. TR 25.892 (2004) Feasibility study for OFDM for UTRAN enhancement. v1.1.0, Mar 22. Viswanath P, Tse D (2002) Opportunistic beamforming using dumb antennas. IEEE Trans Inform Theor 48(6), June 23. Jalali A et al (2000) Data throughput of CDMA-HDR—a high efficiency-high data rate personal communication wireless system. IEEE VTC 3:1854–1857 May