By Mohammed E Eltayeb Electrical Engineering Department King

By Mohammed E. Eltayeb Electrical Engineering Department King Fahd University of Petroleum and Minerals (KFUPM) E-mail: melgaily@kfupm. edu. sa Thesis Advisor : Dr. Yahya S. AL-Harthi (1 of 70)

Presentation Outline • • Opportunistic Communication. Problem Statement. Literature Survey. Thesis Contributions. OS in Single Carrier Systems. OS in Multi-Carrier Systems. Summary. Future Research. 2

Opportunistic Communication • Unlike wired channels, wireless channels are random in nature and unpredictable. • The wireless channel is characterized by the variations of the channel coefficients for each user over time and frequency. • These variations occur due to the reflection, diffraction and scattering of EMW as they propagate which results in multiple versions of the transmitted signal (multipath). fading channel AWGN channel Tx Rx 3

Opportunistic Communication (cont’d) • A way to combat this fading is by using any of the classical diversity techniques which provides multiple versions of the transmitted signal at the receiver. • Multiuser diversity is a way to exploit the fading characteristics of the channel by selecting one user from an array of connected users [6]. • Multiuser diversity uses fading to maximize the overall system capacity. fading channel AWGN channel Tx Rx 4

Opportunistic Communication (cont’d) • As the number of users increases, multiuser diversity gain also increases. Figure 2. 1: Time varying channel of two users undergoing Rayleigh fading. 5

Opportunistic Communication cont’d • Multiuser diversity gain increases with the randomness of the channel. Figure 2. 2: Multiuser diversity gain for Rayligh and Ricain fading channels with Rician factor = 5 and average SNR = 0 d. B. 6

Opportunistic Communication (cont’d) • Schedulers that exploit multiuser diversity are known as opportunistic schedulers as they take the channel conditions into consideration prior any scheduling decision. U 1 s nt ra Gra uest s Req G eq ue st s R U 4 Requests U 3 Requests Grants U 2 BS 7

Opportunistic Communication • Opportunistic schedulers can be classified as fair, semi-fair and greedy. • Opportunistic Round Robin schedulers grant the channel resource to the user with the best channel conditions, but they make sure that all users get an equal share of the resources. • Proportional Fair Schedulers maximizes the system spectral efficiency with a fairness constraint • Greedy algorithms are rate optimal and always schedule the user with the best channel conditions. 8

Opportunistic Communication Figure 2. 3: Average spectral efficiency for MCS, PFA, ORR, RR. Figure 2. 4: Normalized feedback load for MCS, PFA, ORR, RR. 9

Opportunistic Communication 10

Opportunistic Communication • A good scheduling algorithm should seek these goals – – Efficiently utilization resources Provide fairness Meet Qo. S guarantees Minimize scheduling delay and thus feedback load. • In this thesis, we focus on these goals (except for fairness). • We concentrate on greedy algorithms mainly for two reasons. – Greedy algorithms are rate optimal and thus act as an upper bound in which our algorithms should not deviate from. – Greedy algorithms impose full feedback overhead, as such, any feedback reduction scheme imposed here can be easily implemented on other less greedy schemes. 11

Issues in Scheduling • A good scheduling algorithm should seek these goals – Efficiently utilization resources – Provide fairness – Meet Qo. S guarantees – Minimize scheduling delay and thus feedback load. 12

Problem Statement • Is it possible to reduce the feedback overhead (load and rate) and guard time requirements imposed by opportunistic schedulers without any significant degradation in spectral efficiency or Qo. S? 13

Literature Survey • A thorough discussion on the topic of feedback reduction has been conducted in the literature. • Relevant literature can be roughly grouped into the following categories: – Schemes that allow feedback from a group of users only based on an SNR threshold. – Schemes that employ compression of feedback information. – Schemes that feedback quantized values of the SNR instead of the analogue values. – Schemes that use a combination of one or more combination of the above methods. 14

Literature Survey • Gesbert et al. , proposed scheduling algorithms that permitted users which had channel qualities above a predetermined threshold to feedback while other users remain silent [9][10][11]. • In [11], Gesbert and Alouni stated that if no user is found with channel quality that exceeded the threshold, a random user is selected. • The work was extended Hassel et al. where full feedback was requested when no user is found. • Hassel et al. further extended the work by using multiple thresholds to reduce the feedback load. • As seen, these schemes reduced the feedback load with the expense of either MUD gain loss or delay. 15

Literature Survey • Floren et. al. focused on feedback quantization to reduce the feedback load. • Floren et. al. showed that only a few quantization levels were required to capture most of the MUD gain. • Sanayei and Nosratina extended showed that only 1 bit of feedback information was able to achieve the optimal capacity growth. • Xue and Kaiser considered imperfect feedback channel and allowed all users which were above a threshold to feedback and then randomly choose a user. • Harthi et. al. considered a probing system with discrete rates. The first user with quantized SNR above threshold was given the channel resource. This reduced the feedback load in high SNR regime with no loss in spectral efficiency. This work was extended to multicarrier systems in [16]. 16

Literature Survey • Other work concentrated on feedback reduction for multi-carrier systems [17]-[20] and [28]-[35]. • Majority of the work focuses on the following – Feedback of either the best or a group of subcarriers instead of all sub-carriers. – Feedback of information from the best users only. – Feedback of the indices of the strongest clusters only. – Feedback a code that gives an estimate of the current SNR as compared to the SNR of the previous carrier (delta-modulation based). – Feedback compression by exploiting the correlation between neighboring carriers. 17

Thesis Contributions • For Single-Carrier Systems – We introduced an algorithm that reduced the feedback load, rate and guard time. – We derived closed form expressions for the feedback load. • For Multi-Carrier Systems – We presented two algorithms that reduced the feedback load and average guard time requirements. – We derived closed form expressions for the average spectral efficiency for both algorithms. – We derived closed form expressions for the feedback load for both algorithms. 18

19

System Model • In this section we assume the following – A probing based system with a single antenna and K i. i. d. users employing discrete rates. – The channel is assumed to be experiencing flat Rayleigh fading. – Scheduling is performed at the downlink with perfect feedback channel. – The system can detect collisions. 20

Algorithm Description • In this section we detailed description of our proposed 1 -bit feedback scheme and compare it with the optimal and DSMUDiv schemes [8]. – The scheduler (base station) sequentially arranges all the users and assigns each user a unique ID. – The base station broadcasts a query message (first stage) to all users with the highest threshold level, allowing all users that are above or equal to the threshold to feedback a 1 in one mini slot with probability p= 1. – If no user feeds back, the threshold is sequentially lowered until at least one user feeds back or the lowest threshold level has been reached. – If one or more users feedback a 1, then the scheduler knows that at least one user has an SNR lying within the broadcasted threshold level and goes into the second stage (search mode). 21

Algorithm Description (cont’d) – The scheduler then randomly probes two users and allows them to contend for a new mini slot with probability p = 1. – Assuming that the probing request is heard by all users, the user with the higher index number will feedback a 1 (lower will feedback -1) if it has an instantaneous SNR above or equal to the threshold. If the SNR is below the threshold, the user will remain silent. – If one of the users feeds back successfully, then the user is identified and given the channel resource. – If a collision occurs, then the channel resource is given randomly to any of the two users. – If none of the two users respond, then another set of two users are allowed to contend for another minislot. 22

Algorithm Description (Optimal Scheme) γ(4) q(2) q(4) q(3) User 2 DATA γ(3) 19 23

Algorithm Description (DSMUDiv Scheme [8]) γ(4) q(2) q(4) User 2 DATA 24

1 -Bit Binary Feedback Algorithm Description Stage 1 2 1 1 User 3 DATA γ(4) γ(3) Stage 2 25

Performance Analysis • The average spectral efficiency is the average transmitted data rate per unit bandwidth (bits/sec/Hz). where, is the cumulative distribution function (CDF). • The average feedback load (AFL) is defined as the average number of consumed mini-slots until a user is scheduled. The feedback load in this case consists of two terms: 26

Performance Analysis (cont’d) where, and given that and 27

Performance Analysis (cont’d) • Considering feedback traffic degradation, we define the average system capacity as (bits/channel use): • Considering the guard time effect, we define the average system throughput as the amount of bits transmitted per unit time (bits/sec/Hz) • The guard time duration is the time duration of the consumed minislots until a scheduling decision has been made. It is expressed as: 28

System Parameters 29

Numerical Results 30

Numerical Results (cont’d) 31

Numerical Results (cont’d) 32

Numerical Results (cont’d) 33

Conclusion • As the number of users increase, feedback load and guard time requirements increase. • Transmission of data over short time slots to a large number of users degrades the system performance. • We introduced a scheduling scheme that: – Reduced feedback rate due to one bit feedback. – Reduced feedback load with no loss in average spectral efficiency when compared to the optimal scheme. – Improved system throughput due to reduced scheduling delay. – Improved system capacity due to reduced feedback load and rate. 34

35

System Model • Multi-carrier systems employing opportunistic scheduling impose heavy feedback overhead. • In this section, we introduce two algorithms that reduce the feedback load and guard time requirements and then we compare our results with the optimal, DSMUDiv and DSMUDiv-EEA schemes. 36

System Model (cont’d) 37

Algorithm 1 Description (Contribution 2) 38

Algorithm Description - (DSMUDiv-EEA Scheme [16]) γ(4) q(2) q(3) q(1) User 2 DATA q(2) q(4) User 3 DATA q(1) User 1 DATA γ(3) 19 39

MC-ALGO 1 (Contribution 2) γ(1) γ(2) γ(3) γ(4) q(2) User 1 DATA q(2) q(4) User 3 DATA q(1) User 2 DATA γ(3) 19 40

Algorithm Description (cont’d) 41

Algorithm 2 Description (Contribution 3) 42

Performance Analysis 43

Performance Analysis (cont’d) 44

Performance Analysis (cont’d) 45

Performance Analysis - Sub-state steady-state probability for Algorithm 1 • In order to derive the average feedback load and average spectral efficiency, we need to derive the steady-state probability of being in substate w(n, k). • The probability of being in substate w(n, k) (steady-state probability) is : 46

Performance Analysis - Sub-state steady-state probability for Algorithm 1 (cont’d) 47

Performance Analysis - Sub-state steady-state probability for Algorithm 2 The probability of being in substate w(n, k) (steady-state probability) is : 48

Performance Analysis - Sub-state steady-state probability for Algorithm 2 (cont’d) 49

Performance Analysis (Feedback Load) • The average feedback load (AFL) is defined as the average number of probes sent until the subchannel is assigned. The average feedback load conditioned on k and n is [8]: 50

Performance Analysis (Spectral Efficiency – Algorithm 1) • The average spectral efficiency (ASE) is defined as the average transmitted data rate per unit bandwidth in bits/sec/Hz for specified power and target error performance. The average spectral efficiency conditioned on k and n is: 51

Performance Analysis (Spectral Efficiency – Algorithm 1) • Average over all possible number of users in all the possible states, the average spectral efficiency at the lth scheduling process is: 52

Performance Analysis (Spectral Efficiency – Algorithm 2) • Average over all possible number of users in all the possible states, the average spectral efficiency at the lth scheduling process is: 53

Performance Analysis (Probability of Access) • At each subchannel, at most one user can be scheduled, therefore, the probability of access per subchannel given that k users are competing for the subchannel assignment is [8]: and thus the probability of access is 54

Performance Analysis (Scheduling Delay) • The subchannel scheduling time delay is the time needed to schedule a subchannel and it is a function of the number of probes, which is the feedback load. • The average time delay to schedule the lth subchannel is: • The average guard time is: 55

Performance Analysis (System Throughput) • The system throughput is defined as the amount of bits transmitted per unit time. • The average system throughput (ASTH) is derived by taking into account the effect of the guard time duration. • The normalized average system throughput is: 56

System Parameters 57

Numerical Results 58

Numerical Results (cont’d) 59

Numerical Results (cont’d) 60

Numerical Results (cont’d) 61

Numerical Results (cont’d) 62

Numerical Results (cont’d) 63

Conclusion • Two scheduling algorithms that reduced the feedback load and guard time in a polling-based multi-carrier system were introduced. This was achieved by gradually varying the probing thresholds and removing users from the scheduling process once they have been granted a channel resource. • The effect of scheduling delay on the overall system performance was analyzed and it was showed that the guard time has great impact on the ASTH when the AP transmits over short time slots. • Our numerical results showed that the proposed algorithms further reduced the feedback load (especially in low to mid SNR regions) with slight penalty loss in average spectral efficiency. 64

Thesis Summary and Future Work • We showed that to optimally schedule users, full feedback was required from all users. However, this resulted in high guard time, scheduling delay and feedback overhead. • We proposed a scheduling algorithm that dramatically reduced the feedback load and rate as compared to the optimal (full feedback) scheme with no performance degradation. • We also showed that only one bit of feedback information can be used to schedule two users instead of one user. • We introduced another two scheduling algorithms that reduced the feedback load in a multi-carrier system with a slight penalty loss in capacity. • The proposed algorithms reduced the transmission latency and increased the overall system throughput notably when the system employs short time slot durations in low SNR regimes. 65

Thesis Summary and Future Work • Future research – Future research work could consider feedback reduction schemes for multi-user multicarrier multiantenna systems which are gaining much attention recently due to their high data-rates and diversity gains. – Schemes for mobile users could be considered as more feedback information would be required to track the user’s channel. – Feedback reductions algorithms could exploit subchannel correlation to reduce the feedback load. – Feedback quantization optimization and determine the possibility of scheduling more users using only 1 bit feedback information. 66

Publications • M. E. Eltayeb and Y. S. Al-Harthi, "Multiuser Diversity with Binary Feedback”, submitted to Wireless Personal communications. • M. E. Eltayeb and Y. S. Al-Harthi, "Opportunistic Multiuser Scheduling Algorithm For Multi-carrier Wireless Data Systems”, submitted to IET. 67

Bibliography [1] D. Tse and P. Viswanath, “Fundamentals of Wireless Communication”, Cambridge University Press, September 2004. [2] T. Rappaport, “Wireless communications: principles and applications”, Pearson education incorpoaration, Singapore, 2 nd edition, 2002. [3] S. Jha and M. Hassan, “Engineering Internet Qo. S”, Artech House, 2002. [4] B. Sklar, “Rayleigh Fading Channels in Mobile Digital Communication Systems Part I: Characterization”, IEEE Signal Processing Lett. , IEEE Communications Magazine, July 1997. [5] B. Sklar, “Rayleigh Fading Channels in Mobile Digital Communication Systems Part II: Characterization”, IEEE Signal Processing Lett. , IEEE Communications Magazine, July 1997. [6] P. Viswanath, D. Tse, and R. Laroia, “Opportunistic beamforming using dumb antennas”, IEEE Transactions on Information Theory, vol. 48, pp. 1277 - 1294, June 2002. [7] R. Knopp and P. A. Humblet, “Information capacity and power control in single cell multiuser communications”, in Proc. IEEE International Communication Conference (ICC’ 95), Seattle, WA, June 1995. [8] Y. Al-Harthi, A. Tewfik, and M. Alouini, “Multiuser Diversity with Quantized Feedback”, IEEE Transactions on Wireless Communications, vol. 6, no. 1, January 2007. [9] V. Hassel, M. -S Alouini, G. Øien and D. Gesbert, “Rate-Optimal Multiuser Scheduling with Reduced Feedback Load and Analysis of Delay Effects”, EURASIP Journal on Wireless Communications and Networking, vol. 2006, no. 26424 pp. 7, 2006. [10] V. Hassel, M. Alouni, D. Gesbert, G. Øien, “ Exploiting multi user diversity using multiple feedback thresholds”, IEEE Proceedings on Wireless Communications, vol. 9, no. 5, June 2005. [11] D. Gesbert and M. Slim-Alouini, “How much feedback is multi-user diversity really worth? ”, in Proc. IEEE International Communication Conference (ICC’ 04), Paris, France, June 2004, pp. 234 -238. 68

Bibliography [12] S. Catreux, V. Erceg, D. Gesbert and R. Heath, “Adaptive modulation and MIMO coding for broadband wireless datanetworks”, IEEE Communication Magazine, vol. 40 no. 6 pp. 108 -115, June 2002. [13] F. Flor´en, O. Edfors, and B. -A. Molin, “The effect of feedback quantization on the throughput of a multiuser diversity scheme”, in Proc. IEEE GLOBECOM, vol. 1, (San Francisco, CA), pp. 497501, Dec. 2003. [14] S. Sanayei and A. Nosratinia, “Exploiting multiuser diversity with 1 -bit feedback”, in Proc. IEEE Wireless Communications and Networking Conference (WCNC’ 05) , New Orleans, LA, March 2005, pp. 978 -983. [15] Y. Xue and T. Kaiser, “Exploiting multiuser diversity with imperfect one-bit channel state feedback”, in IEEE Transactions on Vehicular Technology , vol. 56, no. 1, pp. 183 -193, January 2007. [16] Y. Al-Harthi, A. Tewfik, and. M. S. Alouini, “Multiuser diversity-enhanced equal access with quantized feedback in multicarrier OFDM system”, in Proc. IEEE Vehicular Technology Conference (VTC-Fall’ 05), Dallas, Tx, Septemper 2005, pp. 568 -572. [17] Y. Xue, T. Kaiser and A. Gershman, “Channel-aware aloha-based OFDM subcarrier assignment in single-cell wireless communications”, IEEE Transactions on Communications, vol. 55, no. 5, May 2007, pp. 953 -962. [18] Y. -J. Choi and S. Bahk, “Selective channel feedback mechanisms for wireless multichannel scheduling”, in Proc. IEEE Wo. WMo. M 2006, Niagara-Falls, NY, USA, June 26 -29, 2006. [19] Y. -J. Choi, S. Bahk and M. -S. Alouini, “Switched-based reduced feedback OFDM multi-user opportunistic scheduling”, in Proc. IEEE 16 th International Symposium on Personal, Indoor and. Mobile Radio Communications (PIMRC 05), September 2005. [20] Z. -H. Han and Y. -H. Lee, “Opportunistic scheduling with partial channel information in OFDMA/FDD systems”, IEEE Vehicular Technology Conference (VTC), vol. 1, pp. 511514, September 2004. [21] T. Tang and R. W. Heath Jr. , “Opportunistic feedback for downlink multiuser diversity”, IEEE Communications Letters, vol. 9, no. 10, pp. 948 -950, October 2005. [22] X. Qin and R. Berry, “Opportunistic splitting algorithms for wireless networks with heterogeneous users”, in Proc. of the 38 th Conference on Information Sciences and Systems (CISS ’ 04), Princeton, NJ, USA, March 2004. [23] H. Koubaa, V. Hassel, and G. E. Øien, “Multiuser diversity gain enhancement by guard time reduction”, in IEEE Workshop on Signal Processing Advances in Wireless Communications (SPAWC), New York, NY, USA, June 2005. [24] H. Koubaa, V. Hassel, and G. E. Øien, “Contention-less feedback for multiuser diversity scheduling”, in Proc. IEEE VTC Fall 2005, Dallas, Texas, USA, September 2005. [25] C. -S. Hwang and J. M. Cioffi, “Achieving multi-user diversity gain using user identity feedback”, IEEE Transactions on Wireless Communications, vol. 7, no. 8, pp. 2911 -2916, August 2008. 69

Bibliography [26] Y. Park, D. Park and D. Love, “On scheduling for multiple-antenna wireless networks using contention-based feedback”, IEEE Transactions on Communications, vol. 55, no. 6, pp. 1174 -1190, June 2007. [27] S. Patil and G. Veciana, “Reducing feedback for opportunistic scheduling in wireless systems”, IEEE Transactions on Wireless Communications, vol. 6, no. 12, pp. 4227 -4232, December 2007. [28] M. Nicolaou, A. Doufexi and S. Armour, “A Selective Cluster Index Scheduling Method in OFDMA”, IEEE Vehicular Technology Conference (VTC), pp. 1 -5, September 2008. [29] P. Svedman, S. Wilson, L. Cimini Jr. and B. Ottersten, “A simplified opportunistic feedback and scheduling scheme for OFDM”, IEEE Vehicular Technology Conference (VTC), vol. 4, pp. 1878 - 1882, May 2004. [30] K. Bai and J. Zhang, “Opportunistic multichannel Aloha for clustered OFDM wireless networks”, IEEE Vehicular Technology Conference (VTC), vol. 55, no. 3, pp. 848 -855, May 2006. [31] Y. -J. Choi, J. Kim and S. Bahk, “Qo. S-aware Selective Feedback and Optimal Channel Allocation in Multiple Shared Channel Enviroments”, IEEE Transactions on Wireless Communications, vol. 5, no. 11, pp. 3278 -3286, November 2006. [32] M. -G. Cho, W. Seo and D. Hong, “A joint feedback reduction scheme using delta modulation for dynamic channel allocation in OFDMA systems”, IEEE Transactions on Wireless Communications, vol. 6, no. 1, pp. 46 - 49, January 2007. [33] T. -S. Kang and H. -M. Kim, “Opportunistic Feedback Assisted Scheduling and Resource Allocation in OFDMA Systems”, 10 th IEEE Singapore International Conference on Communication systems, ICCS 2006. [34] J. Chen, R. Berry and M. Honig, “Limited feedback schemes for downlink OFDMA based on sub-channel groups”, IEEE journal on Selected Areas in Communications, vol. 26, pp. 1451 -1461, October 2008. [35] T. Eriksson and T. Ottosson, “Compression of Feedback in Adaptive OFDM Based Systems using Scheduling”, IEEE Communications Letters, vol. 11, pp. 859 -861, November 2007. [36] T. Eriksson and T. Ottosson, “Compression of feedback for adaptive transmission and scheduling”, IEEE Proceedings, vol. 95, no. 12 pp. 2314 -2321, December 2007. [37] R. Jain, D. Chiu, and W. Hawe, “A quantitative measure of fairness and discrimination for resource allocation in shared computer systems”, DEC Research Report TR-301, Digital Equipment Corporation, Maynard, MA, USA, September 1984. [38] V. Hassel, M. Alouni, M. Hanssen, G. Øien, “Spectral Efficiency and Fairness for Opportunistic Round Robin Scheduling”, in Proc. IEEE International Conference on Communications, (ICC 06), Istanbul, Turkey, June 2006. [39] J. Holtzman, “Asymptotic analysis of proporional fair algorithm”, in Proc. IEEE Symp. on Personal, Indoor and Mobile Radio Communications (PIMRC), vol. 2, San Diego, CA, pp. F-33 -F-37, September. 2001. [40] M. Corson, R. Laroia, V. Park and G. Tsirtsis, “A new paradigm for IP-based cellular networks”, IT Professional, pp. 20 -29, November/December 2001. [41] IEEE Standards Department, “IEEE Std. 802. 11. paert 11: Wireless Lan medium access control(MAC) and physical layer (PHY) specifications”, Technical Report, IEEE, NJ, September 1999. [42] V. Hassel, “Design Issuses and Performance Analysis for Opportunistic Scheduling Algorithms in Wireless Networks”, Ph. D. Thesis, Norwegian University of Science and Technology, January 2007. 70