5. Concluding Remarks
3.5 Simulation Parameters
No. of user 32
FFT size 2048
Total bandwidth 6 MHz Channel model SUI-1 and SUI-5
3.6.2 Effect of Multiuser Scheduling on the Fairness of Multi-carrier System
Figure 3.5 shows the fairness by using the IEEE 802.16a SUI-5 channel models in simulation. For the sake of fitting in with IEEE 802.16a OFDMA physical layer standard, 2048 FFT size used, we divide the total bandwidth into 2,4,8,16 and 32 subchannels. We still observe that when the number of subcarriers increases, the system fairness performance becomes better even in the SUI-5 channel model.
3.6.3 System Performances Comparison of Different Resource Allocation Techniques
Figures 3.6 and 3.7 compares the fairness and throughput performances of different resource management algorithms in SUI-1 and SUI-5 channel models, respectively. In SUI-1 channel model, the fairness performance can be maintained easily. However, SUI-5 channel suffer from more severe fading.
Figure 3.6 shows that dynamic power allocation and maximum C/I scheduling policies do not have obvious difference in fairness performance in SUI-1 channel mod-els. Because frequency and multiuser diversity exist in the multiuser multi-carrier environment, fairness performance is very good. However, the fairness performance
of the maximum C/I scheduling is worse than that of the power allocation scheme about 3.5%. Nevertheless, the value about 0.96 of the fairness index of the maximum C/I scheduling algorithm still means it is a fair resource allocation.
At the same time, we observe Fig. 3.7. We find that the throughput per-formances of the maximum C/I scheduling policy always better than that of the power allocation scheme whether in the SUI-1 channel model or in the SUI-5 channel model. In the SUI-1 channel model, the difference of the throughput performances of the maximum C/I scheduling algorithm and the power allocation scheme is very small. The maximum C/I performs better than the dynamic power allocation about 5%. On the other hand, the throughput performance of the maximum C/I schedul-ing policy is better than that of the power allocation algorithm about 13%. In short, we observe that the system performances of the maximum C/I scheduling algorithm and the power allocation policy are similar in the SUI-1 channel model. However, in the SUI-5 channel model, the maximum C/I scheduling enhance the system through-put about 13% more than the power allocation at the expense sacrificing 3.5% of the fairness performance. Therefore, we concludes that good fairness performance is easily achieved in the multiuser multi-carrier system even when the maximum C/I scheduling adopted.
3.6.4 System Performances Comparison of Different Scheduling Techniques
In addition to comparing the system performance of the power allocation and the maximum scheduling schemes, we compare the system performances of the maximum C/I and proportional scheduling algorithms in this subsection.
Figure 3.8 compares the fairness performance of the maximum C/I scheduling and the proportional fair scheduling in the IEEE 802.16 SUI-1 and SUI-5 channel
models. In the IEEE 802.16 SUI-1 channel model, the fairness performance can be easily maintained. However, because of more severe fading, it is more difficult to maintain the short-term fairness performance in the IEEE 802.16 SUI-5 channel than that in the IEEE 802.16 SUI 1 channel. The proportional fair scheduling takes the great part of frequency diversity and multiuser diversity when the channel variation is not severe, so it also performs well. Furthermore, from the figure, we see that in the IEEE 802.16 SUI-1 channel model, the difference of fairness performance between the maximum C/I and the proportional fair scheduling is insignificant. Even in the IEEE 802.16 SUI-5 channel model, although the fairness of the proportional fair scheduling scheme is still better than the maximum C/I scheduling scheme, the difference of the fairness index between the two scheduling algorithms is less than 3.5%.
Figure 3.9 shows that main advantage of using maximum C/I in a multiuser multi-carrier system. In Fig. 3.9, we compare the throughput performance of both scheduling schemes. In the SUI-1 channel, the throughput performances of the two algorithms are about the same. Interestingly, when consider the SUI-5 channel model with more severe fading, Fig. 3.9 indicates that maximum C/I can take advantage of severer fading and maximize the system throughput. Summarizing from Figs.
3.8 and 3.9, we find that the maximum C/I scheduling can improve the throughput performance by 20% over the proportional fair scheduling at the cost of degrading the fairness index by only 3.5%.
Consequently, the maximum C/I is sufficiently used in the OFDMA system.
We do not need other complicated resource allocation algorithms, such as proportional fair scheduling or power allocation method, to achieve good fairness performance at the expense of throughput. By adopting this simple maximum C/I scheduling schemes, we can obtain good fairness performance and the best throughput perfor-mance simultaneously. In SUI-5 channel model, the maximum C/I improves total system throughput about 13% compared to the power allocation without dynamic
subcarrier allocation. Moreover, the maximum C/I scheduling algorithm even in-crease more than 20% of system throughput than that using the proportional fair scheduling policy.
3.6.5 Discussions
In the scenario described above, we should decide to whom all subcarriers belong every transmission time interval (TTI). It is impractical to do this in such a short time. In fact, because IEEE 802.16a is a fixed wireless application, the channel does not change frequently. Hence, we do not need to schedule users every TTI.
Considering the coherence time of the system, the maximum Doppler frequency is 20Hz (SUI-5 channel), and then we will calculate the coherence time based on (3.20) [33]. Coherence time is the time duration over which two received signals have a strong potential for amplitude correlation. The Doppler spread and coherence time are inversely proportional to one another. The equation (3.20) is defined as the time over which the time correlation function is above 0.5. For example, when the maximum Doppler shift fd = 2Hz, and the coherence time Tc is about 90 ms. Therefore, the maximum C/I scheduling approach is practical in the system.
Tc = 9
16πfd (3.20)
3.7 Conclusions
In this chapter, we have demonstrated that the simple maximum carrier-to-interference scheduling scheme can be a fair scheduler in the OFDMA system, although it is viewed as an unfair scheduling scheme in the single carrier TDMA/CDMA systems. Using this simple maximum C/I scheduling algorithm in the OFDMA system can exploit
multiuser diversity and frequency diversity thoroughly, thereby achieving both high throughput and good fairness performances. Moreover, using this simple maximum C/I scheduling algorithm can combat the worse channel effect and observe the good fairness performance in a multiuser OFDM system.
5 10 15 20 25 30 0
0.1 0.3 0.5 0.6 0.7 0.8 0.9 0.95 0.98 0.9998
# of subchannels
Fairness index
8 users 16 users 24 users 32 users
Fig. 3.5: Fairness index with the number of subchannels varying in different numbers of users when the IEEE 802.16 channel models are used.
5 10 15 20 25 30 0.8
0.95 0.96 0.99 0.995 0.996 0.997 0.998 0.999 0.9995
time
fairness index
Max C/I in SUI−1 Max C/I in SUI−5 PA in SUI−1 PA in SUI−5
Fig. 3.6: Comparison of fairness performance of dynamic sbucarrier allocation and power allocation (1T T I = 2048/6M Hz = 341µs)
0 5 10 15 20 25 30 35 1
1.02 1.04 1.06 1.08 1.1 1.12 1.14 1.16 1.18
time (TTI)
Normalized System Throughput
Max C/I in SUI−1 Max C/I in SUI−5 PA in SUI−1 PA in SUI−5
Fig. 3.7: Comparison of throughput performance of dynamic sbucarrier allocation and power allocation (1T T I = 2048/6M Hz = 341µs)
5 10 15 20 25 30 0.8
0.95 0.96 0.99 0.995 0.996 0.997 0.998 0.999 0.9995 0.9999
time
fairness index
Max C/I in SUI−1 Max C/I in SUI−5 Prop fair in SUI−1 Prop fair in SUI−5
Fig. 3.8: Comparison of fairness performance of max C/I and proportional scheduling (1T T I = 2048/6M Hz = 341µs)
0 5 10 15 20 25 30 35 0.9
0.95 1 1.05 1.1 1.15 1.2 1.25 1.3 1.35 1.4
time (TTI)
Normalized System Throughput
Multicarrier Max C/I and Proportional Fair Scheduling Algorithms Simulation−−Throughput
Max C/I in SUI−1 Max C/I in SUI−5 Prop fair in SUI−1 Prop fair in SUI−5
Fig. 3.9: Comparison of throughput performance of max C/I and proportional scheduling (1T T I = 2048/6M Hz = 341µs)
Channel-aware Subcarrier Allocation and QoS Provisioning for OFDMA Systems
with Multi-type Traffic
The orthogonal frequency division multiple access (OFDMA) is becoming an impor-tant technique for the future wireless systems. Through parallel multi-carrier trans-missions, the inter-symbol interference (ISI) can be easily handled in transmitting high speed data. Furthermore, OFDMA systems bring a new dimension for allocat-ing radio resource - subcarrier. By exploitallocat-ing frequency diversity in the wide frequency spectrum, a suitable subcarrier allocation technique can further enhance throughput for the OFDMA system. This chapter addresses the issue of allocating subcarriers for providing both real-time and non-real-time traffic in the OFDMA system. We sug-gest a categorized subcarrier allocation (CSA) technique to improve throughput for non-real-time traffic, while satisfying the quality of service (QoS) requirement for the real-time method. In the proposed CSA technique, subcarriers are categorized into two groups based on their quality: good and fair. The real-time traffic will be assigned by the subcarrier with fair condition, while the non-real-time traffic will be assigned with good subcarriers. We find that such a subcarrier allocation method can apply the maximum carrier-to-interference (C/I) scheduling to maximize the throughput in good conditioned subcarriers, while the delay for the real-time traffic can be con-trolled by allocating enough fair-conditioned subcarrier through a queueing analytical method. Compared to other methods, such as dynamic subcarrier allocation (DSA)
and random subcarrier allocation (RSA), our results show that the CSA technique outperforms other methods in terms of throughput dropping probability and fairness performances.
4.1 Introduction
With the growing demand of high data rate communication, orthogonal frequency division multiple access (OFDMA) is becoming an important technology. OFDMA has been used in some broadband wireless systems, such as the IEEE 802.16a wire-less metropolitan area network (WMAN) [13] [20]. In single carrier systems, many scheduling schemes are discussed in [14–18,23]. Different from single carrier systems, the channel allocation in such multicarrier OFDMA systems has more dimensional consideration to support high-data-rate services. Besides, future wireless communi-cation networks are expected to support multi-type traffic, such as voice, video and data. Therefore, allocating radio resource to different types of services efficiently to meet quality of service (QoS) requirements of each service is an issue of concern. In [1], many conventional subcarrier allocation schemes, fixed subcarrier allocation (FSA), dynamic subcarrier allocation (DSA) and random subcarrier allocation (RSA), are listed to try to enhance the system performances of constant data rate services. Nev-ertheless, in single carrier systems, if the real-time user with higher priority enters the wireless networks, the non-real-time user will delay the transmission due to lower priority. However, in multicarrier systems, the real-time users can be served by the enough good subcarriers without delay and the non-real-time users use other good subcarriers to achieve the throughput requirements at the same time.
Good scheduling algorithms should have the following characteristics: (1) channel aware, (2) high throughput, (3) fair resource allocation and (4) achieving quality of service. There exist some scheduling algorithms discussed to assure QoS
requirements of different types of traffic in single carrier code division multiple ac-cess (CDMA) systems [2–6]. To provide both minimum service rate guarantees and dynamic channel bandwidth allocation to all users , generalized processor sharing (GPS) [7] [8] discipline is a scheduler candidate. In [2], the author employs fair queueing algorithm to minimize queueing delays in wireless networks. In [3] and [4], the author proposes a GPS based dynamic fair scheduling scheme, called code di-vision GPS (CDGPS) for wideband direct sequence code didi-vision multiple access (DS-CDMA) networks to support multi-type traffic. Furthermore, in [3], the author develops a credit-based CDGPS (C-CDGPS) to improve capacity by trading off short-term fairness. The CARR (channel-aware round robin) scheduler [5] utilizes channel information to increase system capacity and guarantees to allocate certain amount of time slots in an assignment round period in code division multiple access 2000 high data rate (CDMA2000 HDR) [9] or wideband code division multiple access high speed downlink packet access (WCDMA HSDPA) [10] systems. In [6], the idea of the FPLS (fair packet loss sharing) scheduling algorithm is to schedule the session of multimedia packets in the way that all the users share the packet loss fairly de-pending on their QoS requirements and to maximize the system capacity under the QoS constraints. However, in multicarrier systems, such as OFDM, if radio resource management makes use of the frequency diversity, the system performance can be improved. In [11], the author discusses the adaptive modulation and proposes dy-namic GPS (DGPS) scheduling for OFDM wireless communication systems, which exploits both multiuser diversity and frequency diversity. Yet, in [12], the proposed proportional rate adaptive optimization considers subcarrier and power allocation in the multiuser orthogonal frequency division multiplexing (MU-OFDM) system.
In this chapter, we develop a channel-aware and quality of service (QoS) provi-sioning scheduling subcarrier allocation algorithm, categorized subcarrier allocation (CSA), for the OFDMA systems. Frequency diversity inherently exists in OFDMA
systems, while multiuser diversity can be achieved by adopting scheduling algorithms.
Our proposed algorithm makes use of both diversity gains to support non-real-time service flows to achieve high throughput and considers the queueing analysis to al-locate the suitable amount of resource to the real-time service flows. Taking advan-tage of the specific characteristics of channels in OFDMA multicarrier environments, the proposed categorized subcarrier allocation (CSA) scheme can satisfy QoS delay constraint of real-time services and higher throughput requirements of non-real-time services at the same time. Moreover, the proposed subcarrier allocation algorithm can maintain good fairness performance in the multicarrier systems. As described above, we dynamically allocate subcarriers for of different types of service flows. This is a cross-layer design of radio resource management. Furthermore, it can be regarded as another form of water-pouring. We name it Service-oriented Water-pouring, which satisfy the QoS requirements of multi-type services, respectively. In addition, we manage the radio resource allocation from the viewpoint of users. In other words, it is the users that select the subcarriers that can assure their service-oriented QoS requirements.
In a multiuser wireless system, different users may have different channel re-sponses with respect to a time varying wireless channel. Thus, one user may view a channel as a bad channel, whereas the others may view it as a good channel. Con-sequently, for each channel, if the system can first pick a user with the best channel quality among a group of users and then deliver the service to this target user, the system capacity can be significantly improved. We call this capacity improvement as the multiuser diversity gain. However, in addition to multiuser diversity, we also make use of the correlation of subcarriers to efficiently allocate radio resource to real-time and non-real-time services, respectively.
The rest of this chapter is organized as follows. Section 4.2 introduces the quality of service (QoS) scheduling service specified in the IEEE 802.16 standard.
Section 4.3 describes the motivation. Section 4.4 formulates the problem. In Section 4.5, we explain our proposed a QoS provisioning subcarrier allocation method and describe the merit. Numerical results are given in Section 4.6. In Section 4.7, we give our concluding remarks.
4.2 IEEE 802.16 Scheduling Service
In the IEEE 802.16 specification for fixed broadband wireless access (BWA) sys-tems [34], scheduling services are designed to improve the efficiency of the poll/grant process. Owing to different quality of service (QoS) requirements of various service flows, the IEEE 802.16 standard defines four types of services, unsolicited grant ser-vice (UGS), real-time polling serivce (rtPS), non-real-time polling Serser-vice (nrtPS) and best effort (BE) service [34, 35]. We will illustrate each type of service later.
First, the UGS supports real-time service flows with fixed size data packets periodically, such as T1/E1 and Voice over IP (VoIP). The UGS type of service eliminates the delay of subscriber stations (SS) and assures that grants are available to meet the requirements of the service flows. In a word, UGS has the highest priority to access the network resource among the four types of services.
Second, the rtPS supports real-time service flows with variable size data pack-ets on periodic basis, such as MPEG video. This type of service requires more request overhead than UGS, but supports variable grant sizes to optimize the efficiency of data transmission.
Third, the nrtPS is designed to support non-real-time service flows which re-quire variable size Data Grant Burst Types regularly, such as high bandwidth FTP.
This service offers unicast polls on a regular basis which assures that the flow receivers request opportunities even during network congestion.
Finally, the BE service provides efficient service to best effort traffic. This
type of service has the lowest priority to access network but needs higher quality transmission. In other words, it is not tolerant of higher bit error rate.
4.3 Motivation - Channel Characteristics of OFDMA Systems
To develop an efficient subcarrier allocation scheduling algorithm, we have to com-prehend the characteristic of the communication channel. At first, we assume that in the multicarrier environments, each subcarrier channel response to each user is independent. Therefore, we can exploit the multiuser diversity and frequency di-versity to enhance the system throughput and maintain good fairness performance.
Nevertheless, in the practical OFDMA multicarrier environments, we observe that the subcarrier channel responses have some relationship among different users. We describe the characteristic by Fig. 4.1. In Fig. 4.1, each subcarrier is judged for 12000 times. Each user gives the good mark to the first 682 best subcarriers, medium mark to the following 682 subcarriers and bad mark to the first worst 684 subcarriers among the total 2048 subcarriers. Fig. 4.1 (A) shows that for each subcarrier, how many users think it is good for himself. Fig. 4.1 (B) points how many users think the subcarriers medium while Fig. 4.1 (C) represents the number of users who think subcarriers bad.
Here, we illustrates the subcarrier correlation for users by Fig. 4.2. We only care the medium subcarriers and good subcarriers because the bad subcarriers are scheduled lastly. In Fig. 4.2, we observe that the ratio of the number of good users to the number of medium plus good uers for each subcarrierfar far away 0.5 is in the majority, which means that for the same subcarrier, the most part of users regard it as the same rank, good, medium or bad. Of course there are exceptions. But the
exceptions is minor. From Fig. 4.2, we can find that the number of subcarriers whose ratio is less than 0.2 or greater than 0.8 is 82.6% of all subcarriers. The fact shows that the subcarriers are correlated among users. We will make use of the characteristic of the OFDMA channel to design our subcarrier allocation scheduling algorithm.
In the following, we see Fig. 4.3 and Fig. 4.4. The two figures show that the distribution different opinions to each subcarrier. Figure 4.3 is the summation of Fig.
4.1 (A), (B) and (C). We can find that for the most part of subcarriers, opinion of users on them are almost the same. We take some examples from Fig. 4.4 which is selected a section of 4.3 for the sake of clear observation. For the 2001st subcarrier, there are 68 users regarding it as a good subcarrier, while there are 11100 and 832 users seeing it as a medium and bad subcarrier, respectively. Taking another example, for the 2018th subcarrier, 11618 users regarding it as a good subcarrier, while 382 and 0 users seeing it as a medium and bad subcarrier. These two subcarriers have common consensus of users. However, there are still subcarriers with different opinions. For instance, 6296 users regard the 2008th subcarrier as a good subcarrier while 5699 users see it as a medium subcarrier and 5 users think it good. Nevertheless, this kind of subcarriers is minor that the fact can be observed by Fig. 4.2. Based on the characteristics of the channel, we will design a useful subcarrier allocation algorithm for OFDMA systems.
4.4 Problem Formulation
In the third-generation and beyond or the future communication systems, there will be a mixture of different traffic classes. Therefore, what is the suitable ratio of real-time service and non-real-real-time service resource allocation is a research topic. How to utilize the limited radio resource for various types of service flows with different QoS requirements is a very important issue.
0 200 400 600 800 1200 1400 1600 1800 2000 0
5000 10000 15000
(a)
0 200 400 600 800 1200 1400 1600 1800 2000
0 5000 10000 15000
(b)
0 200 400 600 800 1000 1200 1400 1600 1800 2000
0 5000 10000 15000
subcarrier index (c)
Fig. 4.1: (a) The number of users that judge the subcarrier is good for each subcarrier; (b) The number of users that judge the subcarrier is medium for each subcarrier; (c) The number of users that judge the subcarrier is bad for each subcarrier.