Cloud Radios with Limited Feedback Dr. Kiran Kuchi Indian Institute of Technology, Hyderabad Dr. Kiran Kuchi (IIT Hyderabad) Cloud Radios with Limited Feedback 1 / 18
Overview 1 Introduction 2 Downlink Channel Capacity 3 Capacity with Partial CSI 4 Clustering 5 Conclusion Dr. Kiran Kuchi (IIT Hyderabad) Cloud Radios with Limited Feedback 2 / 18
Cloud Networks 3GPP LTE introduced COMP 1 Base stations are connected to central processing unit All base band signal processing done at cloud Functions of base station-carrier power amplification, transmission-rrh Cloud facilitates traffic load balancing through joint scheduling among RRH 1 3GPP TR 36.819 v11.0.0, Coordinated Multi-Point Operation for LTE, 3GPP TSG RAN WG1, Sept. 2011. Dr. Kiran Kuchi (IIT Hyderabad) Cloud Radios with Limited Feedback 3 / 18
Motivation Interference a bottleneck in performance Mitigation of interference through coordinated transmission Costa s Dirty paper coding 2 (DPC) to make interference free networks 3 Suboptimal methods 4 with reduced complexity precoders are available like ZF-DPC, THP 2 M. H M Costa, Writing on dirty paper, IEEE Transactions on Information Theory, May 1983. 3 Lozano, A, Heath, R.W, Andrews, J.G., Fundamental Limits of Cooperation,IEEE Transactions on Information Theory, Sept. 2013 4 G. Caire and S. Shamai.On the achievable throughput of a multiantenna gaussian broadcast channel. 2003 Dr. Kiran Kuchi (IIT Hyderabad) Cloud Radios with Limited Feedback 4 / 18
Main Results Stochastic geometric frame work 5,6 for analysing cloud networks Characterisation of received SINR of a typical cloud user Analytical bounds on the rate performance of a cloud network without clustering ZF-DPC for full channel knowledge ZF-DPC for limited channel knowledge Bounds on the rate performance of a cloud network with clustering ZF-DPC for full channel knowledge ZF-DPC for limited channel knowledge 5 Andrews, J.G., Baccelli, F., and Ganti, R.K., A tractable approach to coverage and rate in cellular networks, IEEE Trans. on Commun., November 2011. 6 Stoyan. D, Kendall, W.S, and Mecke, J., Stochastic Geometry and its Applications. Wiley series in probability and mathematical statistics, second edition, 1995. Dr. Kiran Kuchi (IIT Hyderabad) Cloud Radios with Limited Feedback 5 / 18
Received Signal Vector Assuming k-pairs of BS-UEs in one random instance of the network. Received signal vector in downlink 7, where, Y = [y 1 y 2... y k ] T, H = Y = H X + N h 11 r α/2 11 h 12 r α/2 12... h 1k r α/2 1k h 21 r α/2 21 h 22 r α/2 22... h 2k r α/2 2k. h k1 r α/2 k1..., h k2 r α/2 k2... h kk r α/2 kk where, X = [x 1 x 2... x k ] T is the transmitted symbol vector with E[ x i 2 ] = P T and N = [n 1 n 2... n k ] T, with, n i CN (0, σi 2 ) is AWGN. 7 Downlink is vector Gaussian broadcast channel (BC) and uplink, a vector Gaussian multiple access channel (MAC). Dr. Kiran Kuchi (IIT Hyderabad) Cloud Radios with Limited Feedback 6 / 18
W = Q H can cancel interference from BSs with indices j > i. The remaining interference terms with indices j < i are taken care of by applying DPC successively. Dr. Uplink-Downlink Kiran Kuchi (IIT Hyderabad) duality can Cloud be Radios applied. with Limited Feedback 7 / 18 Zero-forcing Dirty Paper Coding Channel matrix known at cloud, H = LQ. Precode signal with W = Q H. Received signal vector, Y = HWX + N Y = HQ H X + N = LQQ H X + N y 1 l 11 0... 0 y 2. = l 12 l 22... 0.... y k l 1k l 2k... l kk x 1 x 2. x k + n 1 n 2. n k For i th UE, y i = l ii x i + j<i l ij xj + n i
Capacity of ZF-DPC and Theoretical Bounds The post processing SINR for the UE i is given by SINR i = l ii 2 P T σ 2 i The achievable rate for UE i is given by ( C i = log 1 + l ii 2 ) P T σi 2 bps/hz Dr. Kiran Kuchi (IIT Hyderabad) Cloud Radios with Limited Feedback 8 / 18
Capacity with Partial CSI As the size of the cloud (k) increases, the amount of CSI required for precoding increases as O(k 2 ). For instance, in a cloud with k = 2, the number of individual channel states required is 4, whereas this number increases to 900 for a cloud with 30 BSs. Obtaining such high CSI in a cellular network is practically not feasible. ZF-DPC with Partial CSI Channel matrix with known channel states, H p = L p Q p. Precoding matrix, W = Q H p Received signal vector, Y = LQQ H p X + N QQ H p I, leaves residual signals in post processing. Dr. Kiran Kuchi (IIT Hyderabad) Cloud Radios with Limited Feedback 9 / 18
Comparison of Conventional and Cloud Networks CDF 1 0.5 0 Conv. Cell. L = 2 L = 4 L = 6 L = 20 L = 50 ZF-DPC 0 5 10 15 Rate in b/s/hz Figure: Rate performance comparison of conventional and cloud networks. Here, D = 200m, P T = 4.5741W and SNR = 15 db at a distance 200m. Dr. Kiran Kuchi (IIT Hyderabad) Cloud Radios with Limited Feedback 10 / 18
Rate Profiles Table: Rate profile comparison for cloud network with partial CSI and D = 200m. % Cell. L = 2 L = 4 L = 6 DPC 5% 0.0763 0.1934 0.4247 0.6658 5.8298 10 % 0.1635 0.3657 0.7513 1.1026 6.8065 50 % 1.2332 2.0752 3.1329 3.8244 10.2199 80 % 3.4520 4.8626 6.0563 6.9433 13.3781 Mean 2.1671 3.0108 3.9216 4.6015 10.7759 Dr. Kiran Kuchi (IIT Hyderabad) Cloud Radios with Limited Feedback 11 / 18
Clustering In real life, the size of the cloud is limited the propagation delay in optical fiber communication other implementation constraints Cluster: Groups of BSs are connected to a central processor Cloud processing is applied in each cluster independently Cluster edge users interference will be high Dr. Kiran Kuchi (IIT Hyderabad) Cloud Radios with Limited Feedback 12 / 18
Received Signal Vector δ=no.of BSs in the cluster =No.of BSs in the whole area of consideration y 1, y 2,..., y δ be the received signals, y 1 l 11 0... 0 x 1 y 2 l 21 l 22... 0 x 2. y δ =..... l δ1 l δ2... l δδ x δ g 1δ+1 g 1δ+2... g 1 g 2δ+1 g 2δ+2... g 2 +.... g δ+1 g δ+2... g where g ij = h ij r α/2 ij The SINR of a typical user of the cluster, x δ+1 x δ+2. x SINR i = P T l ii 2 /(σ 2 + I i (Φ)) + n 1 n 2. n δ, (1) where I i (Φ) = b Φ P T h i,b 2 r α i,b and Φ = Φ \ {b 1, b 2,..., b δ } Dr. Kiran Kuchi (IIT Hyderabad) Cloud Radios with Limited Feedback 13 / 18
Rate Profile for Clustering 1 0.8 CDF, FC (t) 0.6 0.4 0.2 0 Conv. Cell. Full L = 2 L = 4 L = 6 0 2 4 6 8 10 12 Rate, t, in b/s/hz Dr. Kiran Kuchi (IIT Hyderabad) Cloud Radios with Limited Feedback 14 / 18
Rate Profiles Table: Rate profile comparison for clustered cloud network with partial CSI and SNR = 15 db at a distance 200m. % Cell. L = 2 L = 4 L = 6 Full 5 % 0.0763 0.1566 0.2740 0.3367 0.4113 10 % 0.1635 0.3222 0.5165 0.6298 0.7768 50 % 1.2332 1.9110 2.6508 3.1821 3.8840 80 % 3.4520 4.5474 5.5583 6.1874 7.1035 Mean 2.1671 2.8224 3.5259 3.9504 4.5748 Dr. Kiran Kuchi (IIT Hyderabad) Cloud Radios with Limited Feedback 15 / 18
Conclusion Cooperation can provide significant interference cancellation gain Practically feasible feedback of order 4-6 channels is enough to produce remarkable improvement Notable improvement in cell edge rate can be seen only with a PCSI of 2 with clustering 20 BSs Clustering of 20 BSs can improve the edge users rate to 4.5 times and mean rate to twice with knowledge of 6 nearest BSs of the cluster Pilot design to be studied Dr. Kiran Kuchi (IIT Hyderabad) Cloud Radios with Limited Feedback 16 / 18
References E.Telatar, Capacity of multi-antenna gaussian channels ATT-Bell Labs Internal Tech. Memo, 1995, pp. 585595. A. Gorokhov Coordinated joint transmission in WWAN, R. Irmer, H. Droste, P. Marsch, M. Grieger, G. Fettweis, S. Brueck, H.-P. Mayer, L. Thiele, and V. Jungnickel,, Coordinated multipoint: Concepts, performance, and field trial results, Feb. 2011 M. K. Karakayali, G. J. Foschini, and R. A. Valenzuela, Network coordination for spectrally efficient communications in cellular systems IEEE Trans. on Communications, Aug. 2006. A. Lozano, R. W. Heath, and J. G. Andrews, Fundamental limits of cooperation. Dr. Kiran Kuchi (IIT Hyderabad) Cloud Radios with Limited Feedback 17 / 18
Thank you! Dr. Kiran Kuchi (IIT Hyderabad) Cloud Radios with Limited Feedback 18 / 18