Decentralized Resource Allocation and Effective CSI Signaling in Dense TDD Networks 1 Decentralized Resource Allocation and Effective CSI Signaling in Dense TDD Networks Antti Tölli with Praneeth Jayasinghe, Ganesh Venkatraman, Jarkko Kaleva, Markku Juntti, Matti Latva-aho and Le-Nam Tran, e-mail: atolli@ee.oulu.fi Centre for Wireless Communications, University of Oulu, Finland 2016 IEEE Communication Theory Workshop, Nafplio, Greece 16 May, 2016 G. Venkatraman, A. Tölli, L-N. Tran & M. Juntti, Traffic Aware Resource Allocation Schemes for Multi-Cell MIMO-OFDM Systems, IEEE Transactions on Signal Processing, vol. 64, no. 11, pp. 2730 2745, June 2016. P. Jayasinghe, A. Tölli & M. Latva-aho, Bi-directional Signaling Strategies for Dynamic TDD Networks in Proc. IEEE SPAWC 2015, Stockholm, Sweden, July, 2015 G. Venkatraman, A. Tölli, M. Juntti & L-N. Tran Queue Aware Precoder Design via OTA Training, in Proc. 2016 IEEE 17th International Workshop on Signal Processing Advances in Wireless Communications (SPAWC), Edinburgh, UK, July 3 6, 2016
Decentralized Resource Allocation and Effective CSI Signaling in Dense TDD Networks 2 Heterogeneous Network Setting Heterogeneous network composed of Large macro cells with (massive) MIMO antenna arrays, Small cells and relays with (distributed) MIMO arrays, and D2D communication with base station coordination Backhaul / control Data
Decentralized Resource Allocation and Effective CSI Signaling in Dense TDD Networks 3 frequency UL and DL control channels Dynamic TDD UL or DL data channels time Figure: Flexible TDD frame structure 1 Significant load variation between adjacent cells Flexible UL/DL allocation provides large potential gains in spectral efficiency 2 More challenging interference management 1 Nokia Networks, 5G radio access system design aspects, Nokia white paper, Aug. 2015. Available: http://networks.nokia.com/file/37611/5g-radio-access 2 3GPP TSG RAN WG1, Study on scenarios and requirements for next generation access technologies TR 38.913, 3rd Generation Partnership Project 3GPP, www.3gpp.org, 2016
Decentralized Resource Allocation and Effective CSI Signaling in Dense TDD Networks 4 Dynamic TDD Figure: UL-DL/DL-UL interference in Dynamic TDD Additional UL-to-DL and DL-to-UL interference associated with the dynamic TDD Interference mitigated by coordinated beamforming. More measurements and info exchange also at the terminal side Similar interference scenarios in underlay D2D transmission 3 3 A. Tölli, J. Kaleva & P. Komulainen, Mode Selection and Transceiver Design for Rate Maximization in Underlay D2D MIMO Systems, in Proc. IEEE ICC 2015, London, UK, June, 2015
Q 1 Q 2 Q 3 Q 4 Q 5 Q 6 U 6 Decentralized Resource Allocation and Effective CSI Signaling in Dense TDD Networks 5 System Model & Problem Formulation U 4 U 3 OFDM system with N sub-channels and N B BSs, N T TX antennas per BS U 1 U 2 U 5 K users each with N R antennas Desired signal Interference signal Goal: minimize the number of packets in BS queues via joint TX/RX design and resource allocation over spatial and frequency resources
Decentralized Resource Allocation and Effective CSI Signaling in Dense TDD Networks 6 Queueing Model Each user is associated with backlogged packets of size Q k packets. Queued packets Q k of each user follows dynamic equation at the ith instant as [ + Q k (i + 1) = Q k (i) t k (i)] + λk (i) (1) where t k = N n=1 L l=1 t l,k,n denotes the total number of transmitted packets corresponding to user k λ k represents the fresh arrivals of user k at BS b k
Decentralized Resource Allocation and Effective CSI Signaling in Dense TDD Networks 7 JSFRA Formulation 4 The optimization objective of joint space-frequency resource allocation (JSFRA) to design transmit precoders is minimize t l,k,n N L q a k Q k t l,k,n k U n=1 l=1 (2) where a k are arbitrary weights used to control the priorities Exponent q = 1, 2,..., plays different role based on the value it assumes Inherent maximum rate constraint: N L n=1 l=1 t l,k,n Q k Special cases (when Q k > N L n=1 l=1 t l,k,n k): q = 1: Sum rate maximization q = 2: Queue-Weighted Sum Rate Maximization (Q-WSRM) 4 G. Venkatraman, A. Tölli, L-N. Tran & M. Juntti, Traffic Aware Resource Allocation Schemes for Multi-Cell MIMO-OFDM Systems, IEEE Transactions on Signal Processing, vol. 64, no. 11, pp. 2730 2745, June 2016.
Decentralized Resource Allocation and Effective CSI Signaling in Dense TDD Networks 8 JSFRA Formulation (MSE Reformulation) The queue minimization problem can be solved by utilizing the relation between the MSE and the SINR as ɛ l,k,n = (1 + γ l,k,n ) 1 (3) Equivalence is valid only when the receivers are designed with the mean squared error (MSE) objective, i.e., using MMSE receivers t l,k,n = log 2 (ɛ l,k,n ) (4a) ɛ l,k,n = E [ (d l,k,n ˆd l,k,n ) 2] = 1 w H l,k,n H bk,k,nm l,k,n 2 + w l,k,n H H b i,k,nm j,i,n 2 + Ǹ 0 (4b) (j,i) (l,k)
Decentralized Resource Allocation and Effective CSI Signaling in Dense TDD Networks 9 JSFRA Formulation (MSE Reformulation) Queue minimization via MSE reformulation minimize t l,k,n,m l,k,n, ṽ q (5a) ɛ l,k,n,w l,k,n subject to t l,k,n log 2 (ɛ l,k,n ) l, k, n (5b) ɛ l,k,n 1 wl,k,n H H b k,k,nm l,k,n 2 + w H l,k,n H bi,k,nm j,i,n 2 + Ǹ 0 l, k, n (j,i) (l,k) N L tr (m l,k,n m H l,k,n ) Pmax b. (5d) n=1 k U b l=1 1 q where ṽ k ak (Q k N L n=1 l=1 t l,k,n) The nonconvex (difference of convex) rate constraints are approximated via successive convex approximation (SCA) method Receive beamformers are designed by the MMSE receivers using the converged TX precoders (5c)
Decentralized Resource Allocation and Effective CSI Signaling in Dense TDD Networks 10 Dynamic Traffic Scenario - Centralized Performance Total backlogged bits after each tx slot, Σ k (Q k (i) - t k (i)) + 140 120 100 80 60 40 20 JSFRA with q= JSFRA with q=2 JSFRA with q=1 Q-WSRM Q-WSRME Sum arrivals Σ k λ k (i) 0 0 50 100 150 200 250 Time Slots Figure: Queue dynamics for {N, N B, K, N T, N R, A k } = {4, 2, 12, 4, 1, 6} [G. Venkatraman, A. Tölli, L-N. Tran & M. Juntti, Traffic Aware Resource Allocation Schemes for Multi-Cell MIMO-OFDM Systems, IEEE Transactions on Signal Processing, vol. 64, no. 11, pp. 2730 2745, June 2016.]
Decentralized Resource Allocation and Effective CSI Signaling in Dense TDD Networks 11 Distributed Methods Q 1 Q 2 Q 3 Q 4 Q 5 Q 6 U 4 U 3 U 1 U 2 U 6 U 5 Desired signal Overhead of the centralized design is large as the network size grows Distributed approaches based on primal decomposition or ADMM can be used to reduce the signaling Interference signal Precoder design by solving the KKT expressions of the JSFRA problem (5) via MSE reformulation Practical approach to design precoders with minimal backhaul usage
Decentralized Resource Allocation and Effective CSI Signaling in Dense TDD Networks 12 KKT Expressions for (5) m (i) l,k,n = ( w (i) l,k,n = ( L ) 1α α (i 1) y,x,n H H b k,x,n w(i 1) y,x,n wy,x,n H (i 1) (i 1) H bk,x,n + δ b I NT x U y=1 L ) H bx,k,nm (i) y,x,nmy,x,nh H (i) 1 H b + N x,k,n 0I NR Hbk,k,n m (i) l,k,n x U y=1 ɛ (i) 1 l,k,n = H (i) w l,k,n H b k,k,nm (i) l,k,n t (i) l,k,n = log 2 (ɛ(i 1) [ ( σ (i) l,k,n = ak q Q log(2) k α (i) l,k,n = α(i 1) l,k,n + ρ(i) l,k,n ) ( ɛ (i) N l,k,n ɛ(i 1) l,k,n log(2) ɛ (i 1) l,k,n L l,k,n HH b k,k,n w(i 1) l,k,n 2 + w H (i) l,k,n H b y,k,nm x,y,n (i) 2 + w l,k,n 2 N 0 (x,y) (l,k) ) t (i) l,k,n n=1 l=1 ( σ (i) l,k,n α (i 1) ɛ (i) l,k,n l,k,n ) (q 1) ] + )
Decentralized Resource Allocation and Effective CSI Signaling in Dense TDD Networks 13 Decentralized Precoder Design - Strategy A Over-the air (OTA) based iterative algorithm 5 with Bi-directional training (BiT) 6 m (i) l,k,n = ( w (i) l,k,n = ( L ) 1α α (i 1) y,x,n H H b k,x,n w(i 1) y,x,n wy,x,n H (i 1) (i 1) H bk,x,n + δ b I NT x U y=1 L ) H bx,k,nm (i) y,x,nmy,x,nh H (i) 1 H b + N x,k,n 0I NR Hbk,k,n m (i) l,k,n x U y=1 l,k,n HH b k,k,n w(i 1) l,k,n Transmit precoders m l,k,n depend on H H b k,x,n w y,x,n, i.e., effective uplink channel Receive beamformers w l,k,n depend on H bx,k,nm y,x,n, i.e., effective downlink channel Can be measured locally at each node in TDD using precoded pilots 5 P. Komulainen, A. Tölli & M. Juntti, Effective CSI Signaling and Decentralized Beam Coordination in TDD Multi-Cell MIMO Systems, IEEE Transactions on Signal Processing, vol. 61, no. 9, pp. 2204 2218, May 2013 6 Changxin Shi; Berry, R.A.; Honig, M.L., Bi-Directional Training for Adaptive Beamforming and Power Control in Interference Networks, IEEE Transactions on Signal Processing, vol.62, no.3, pp.607 618, Feb.1, 2014
Decentralized Resource Allocation and Effective CSI Signaling in Dense TDD Networks 14 Signaling Requirement for OTA based Updates 7 8 Bi-directional training phase m k α k w k α k w k Strategy A Data m k α k w k Q k Strategy B Data Forward pilots Backward training pilots Figure: TDD frame structure with bidirectional signaling 7 P. Komulainen, A. Tölli & M. Juntti, Effective CSI Signaling and Decentralized Beam Coordination in TDD Multi-Cell MIMO Systems, IEEE Transactions on Signal Processing, vol. 61, no. 9, pp. 2204 2218, May 2013 8 P. Jayasinghe, A. Tölli & M. Latva-aho, Bi-directional Signaling Strategies for Dynamic TDD Networks in Proc. IEEE SPAWC 2015, Stockholm, Sweden, July, 2015
Decentralized Resource Allocation and Effective CSI Signaling in Dense TDD Networks 15 Assumptions and Evaluation Model Every BS and user terminal uses orthogonal pilots in UL and DL over-the-air (OTA) signaling For simplicity, pilot transmissions used to convey the equivalent channel information in one BiT iteration - consume η resource share. 9 Under this assumption, the effective rate by considering the signaling overhead is given as t l,k,n = (1 I max η) t l,k,n (6) Total number of backlogged packets is evaluated as - χ = K k=1 [Q k t k ] + In all simulations, we consider η = 1% 9 In practice, the performance depends on the amount of available pilots and the size of coherence block.
avg. backlogged pkts for all users Decentralized Resource Allocation and Effective CSI Signaling in Dense TDD Networks 16 Average Backlogged Packets - Distributed Design 5500 5000 4500 4000 3500 3000 Centralized Design Uncoordinated Design Strategy A (OTA-3 with Mem) Strategy A (OTA-5 with Mem) Strategy B (OTA-3) Strategy B (OTA-5) 2500 2000 1500 1000 500 0 1 2 3 4 5 6 7 avg. arrival pkts per user Figure: Average backlogged packets for {N, N B, K, N T, N R } = {3, 2, 12, 4, 2} evaluated over 250 slots with f d T s 0.1 [G. Venkatraman, A. Tölli, M. Juntti & L-N. Tran Queue Aware Precoder Design via OTA Training, in Proc. 2016 IEEE 17th International Workshop on Signal Processing Advances in Wireless Communications (SPAWC), Edinburgh, UK, July 3 6, 2016]
Decentralized Resource Allocation and Effective CSI Signaling in Dense TDD Networks 17 OTA Signalling in Dynamic TDD 10 DL channel BS1 UL-DL interference Downlink Cells DL-DL interference BS3 BS2 Figure: Interference at DL terminal Uplink Cell DL cell forward phase Users measure DL cell BS pilots and UL cell user pilots DL cell backward phase BSs measure UL cell BS pilots and DL cell user pilots 10 P. Jayasinghe, A. Tölli & M. Latva-aho, Bi-directional Signaling Strategies for Dynamic TDD Networks in Proc. IEEE SPAWC 2015, Stockholm, Sweden, July, 2015
Decentralized Resource Allocation and Effective CSI Signaling in Dense TDD Networks 18 OTA Signalling in Dynamic TDD Uplink Cells UL channel BS1 UL-UL interference DL-UL interference BS3 Figure: Interference at UL BS BS2 Downlink Cell UL cell forward phase BSs measure UL cell user pilots and DL cell BS pilots UL cell backward phase Users measure DL cell user pilots and UL cell BS pilots
Decentralized Resource Allocation and Effective CSI Signaling in Dense TDD Networks 19 Bi-directional Signalling: Simulation Setup 4-antenna BSs, 4 2-antenna UEs per BS
Decentralized Resource Allocation and Effective CSI Signaling in Dense TDD Networks 20 Comparison of Different Signalling Strategies - Sum Rate Actual Rate at SNR = 20 db 28 26 24 22 20 18 16 14 12 Strategy A Strategy B Strategy C Uncoordinated method 10 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 Overhead Figure: Actual sum rate at SNR = 20dB vs overhead for different bi-directional signaling strategies, {α, β, γ} = {0, 3, 6}dB. [P. Jayasinghe, A. Tölli & M. Latva-aho, Bi-directional Signaling Strategies for Dynamic TDD Networks in Proc. IEEE SPAWC 2015, Stockholm, Sweden, July, 2015]
Decentralized Resource Allocation and Effective CSI Signaling in Dense TDD Networks 21 Conclusions Cross layer design of transmit and receive beamformers based on the number of residual packets was studied An iterative solution is found by solving a series of convex subproblems A practical approach via iterative computation of KKT expressions Extensions of the proposed work in time-correlated fading scenario with limited information exchange Iterative OTA signalling methods can be used in Dynamic TDD and/or underlay D2D to handle the interference due to cross-user channels Future/current work: pilot allocation/decontamination, dynamic cell mode (UL/DL) selection