1. 1 Consensus-Based Distributed Least-Mean Square Algorithm Using Wireless Ad Hoc Networks Gonzalo Mateos, Ioannis Schizas and Georgios B. Giannakis ECE Department, University of Minnesota Acknowledgment: ARL/CTA grant no. DAAD19-01-2-0011

2. 2 Motivation Estimation using ad hoc WSNs raises exciting challenges  Communication constraints  Limited power budget  Lack of hierarchy / decentralized processing Consensus Unique features  Environment is constantly changing (e.g., WSN topology)  Lack of statistical information at sensor-level Bottom line: algorithms are required to be  Resource efficient  Simple and flexible  Adaptive and robust to changes Single-hop communications

3. 3 Prior Works Single-shot distributed estimation algorithms  Consensus averaging [Xiao-Boyd ’ 05, Tsitsiklis-Bertsekas ’ 86, ’ 97]  Incremental strategies [Rabbat-Nowak etal ’ 05]  Deterministic and random parameter estimation [Schizas etal ’ 06] Consensus-based Kalman tracking using ad hoc WSNs  MSE optimal filtering and smoothing [Schizas etal ’ 07]  Suboptimal approaches [Olfati-Saber ’ 05], [Spanos etal ’ 05] Distributed adaptive estimation and filtering  LMS and RLS learning rules [Lopes-Sayed ’ 06 ’ 07]

4. 4 Problem Statement Ad hoc WSN with sensors  Single-hop communications only. Sensor ‘ s neighborhood  Connectivity information captured in  Zero-mean additive (e.g., Rx, quantization) noise Each sensor, at time instant  Acquires a regressor and scalar observation  Both zero-mean w.l.o.g and spatially uncorrelated Least-mean squares (LMS) estimation problem of interest

5. 5 Centralized Approaches If, jointly stationary Wiener solution If global (cross-) covariance matrices, available Steepest-descent converges avoiding matrix inversion If (cross-) covariance info. not available or time-varying Low complexity suggests (C-) LMS adaptation Goal: develop a distributed (D-) LMS algorithm for ad hoc WSNs

6. 6 A Useful Reformulation Introduce the bridge sensor subset  For all sensors, such that  For, there must such that Consider the convex, constrained optimization Proposition [Schizas etal ’ 06]: For satisfying 1)-2) and the WSN is connected, then

7. 7 Algorithm Construction Problem of interest Two key steps in deriving D-LMS  Resort to the alternating-direction method of multipliers Gain desired degree of parallelization  Apply stochastic approximation ideas Cope with unavailability of statistical information

8. 8 Derivation of Recursions Associated augmented Lagrangian Alternating-direction method of Lagrange multipliers Three-step iterative update process Multipliers Dual iteration Local estimates Minimize w.r.t. Bridge variables Minimize w.r.t. Step 1: Step 2: Step 3:

9. 9 Multiplier Updates Recall the constraints Use standard method of multipliers type of update Requires from the bridge neighborhood

10. 10 Local Estimate Updates Given by the local optimization  First order optimality condition  Proposed recursion inspired by Robbins-Monro algorithm  is the local prior error  is a constant step-size Requires  Already acquired bridge variables  Updated local multipliers

11. 11 Bridge Variable Updates Similarly, Requires  from the neighborhood  from the neighborhood in a startup phase

12. 12 D-LMS Recap and Operation In the presence of communication noise, for Simple, fully distributed, only single-hop exchanges needed Step 1: Step 2: Step 3: Sensor Rx from Tx to Bridge sensor Tx to Rx from Steps 1,2:Step 3:

13. 13 Further Insights Manipulating the recursions for and yields Introduce the instantaneous consensus error at sensor The update of becomes Superposition of two learning mechanisms  Purely local LMS-type of adaptation  PI consesus loop tracks the consensus set-point

14. 14 Network-wide information enters through the set-point Expect increased performance with Flexibility D-LMS Processor Local LMS Algorithm Sensor j PI Regulator To Consensus Loop

15. 15 Mean Analysis Independence setting signal assumptions for (As1) is a zero-mean white random vector, with spectral radius (As2) Observations obey a linear model where is a zero-mean white noise (As3) and are statistically independent Define and Goal : derive sufficient conditions under which

16. 16 Dynamics of the Mean Lemma: Under (As1)-(As3), consider the D-LMS algorithm initialized with. Then for, is given by the second-order recursion with and, where Equivalent first-order system by state concatenation

17. 17 First-Order Stability Result Proposition: Under (As1)-(As3), the D-LMS algorithm whose positive step-sizes and relevant parameters are chosen such that, achieves consensus in the mean sense i.e., Step-size selection based on local information only  Local regressor statistics  Bridge neighborhood size

18. 18 Simulations node WSN, Regressors: i.i.d. Observations: D-LMS:, True time-varying weight:

19. 19 Loop Tuning Adequately selecting actually does make a difference Compared figures of merit:  MSE (Learning curve):  MSD (Normalized estimation error):

20. 20 Concluding Summary Developed a distributed LMS algorithm for general ad hoc WSNs Intuitive sensor-level processing  Local LMS adaptation  Tunable PI loop driving local estimate to consensus Mean analysis under independence assumptions step-size selection rules based on local information Simulations validate mss convergence and tracking capabilities Ongoing research  Stability and performance analysis under general settings  Optimality: selection of bridge sensors,  D-RLS. Estimation/Learning performance Vs complexity tradeoff