Distributed Compressed Estimation Based on Compressive Sensing for Wireless Sensor Networks

Transcription

1 Distributed Compressed Estimation Based on Compressive Sensing for Wireless Sensor Networks Joint work with Songcen Xu and Vincent Poor Rodrigo C. de Lamare CETUC, PUC-Rio, Brazil Communications Research Group, Department of Electronics, University of York, U.K.

2 Outline Review of Compressive Sensing Introduction System model Proposed distributed compressed estimation (DCE) scheme Measurement matrix optimization Proposed adaptive algorithm Simulations Conclusions

3 Review of Compressive Sensing Compressive sensing has gained substantial interest amongst researchers and was applied to areas such as image processing, radar, sonarand wireless communications. Compressive sensing theory states that an S-sparse signal of length M can be recovered with high probability from O (S log2m) measurements. Mathematically, the vector with length D that carries sufficient information about can be obtained by the linear model =, where is the D x M measurement matrix. A reconstruction algorithm is then applied to obtain an estimate of. Prior work on distributed compressive sensing is limited and has not considered compressed transmit strategies and estimation techniques. G. Quer, R. Masiero, G. Pillonetto, M. Rossi, and M. Zorzi, Sensing, compression, and recovery for WSNs: Sparse signal modeling and monitoring framework, IEEE Transactions on Wireless Communications, October 2012

4 Introduction (1/2) Distributed signal processing algorithms are of great importance for statistical inference in wireless networks. Such techniques enable a network to perform statistical inference from data collected from nodes that are geographically distributed. Inthiscontext, foreachnodeasetofneighbournodescollectandprocess their local information, and transmit their estimates to a specific node. Then, each specific node combines the collected information together with its local estimate to generate improved estimates. C. G. Lopes and A. H. Sayed, Diffusion least-mean squares over adaptive networks: Formulation and performance analysis, IEEE Transactions on Signal Processing, vol. 56, no. 7, pp , July 2008.

5 Introduction (2/2) Key problems: In many scenarios, the unknown parameter vector to be estimated can be very large, sparse and contain only a few nonzero coefficients. Most distributed algorithms to date do not exploit the structure of the unknown parameter vector and process the full-dimensional data. Our contributions: Inspired by Compressive Sensing, we present a scheme that incorporates compression and decompression modules into the distributed processing. We present a design procedure and an adaptive algorithm to optimize the measurement matrices, resulting in improved performance.

6 System model (1/2) We consider a wireless network with N nodes: We assume that the network is partially connected. Communication protocols: incremental, consensus and diffusion. Main tasks: To collect and process data locally at the node and then transmit estimates. To perform distributed estimation by exchanging information about local estimates. Application scenario: Wireless sensor networks Wireless Network with N nodes k S. Chouvardas, K. Slavakis, Y. Kopsinis, and S. Theodoridis, A sparsity promoting adaptive algorithm for distributed learning, IEEE Transactions on Signal Processing, vol. 60, no. 10, pp , Oct 2012.

7 System model (2/2) System model: Problem: min k Adapt-then-combine (ATC) diffusion LMS algorithm: Wireless Network with N nodes F. S. Cattivelliand A. H. Sayed, Diffusion lmsstrategies for distributed estimation, IEEE Transactions on Signal Processing, vol. 58, pp , March 2010.

8 Proposed DCE scheme (1/3) Measurement: where and is the D x 1 input signal vector. Adaptation: where Information exchange: Combination: with D parameters Decompression: Computation of with the OMP algorithm.

9 Proposed DCE scheme (2/3) Each node observes the M x 1 vector ( ) The D M measurement matrix obtains the D x 1 compressed version (i), which is the input to the compression module. The estimation of is performed in the compressed domain. A decompression module employs a measurement matrix and a reconstruction algorithm to compute an estimate of at each node.

10 Proposed DCE scheme (3/3) Measurement, compression and adaptation Transmission Combination Decompression: reconstruction of estimates

11 Proposed DCE scheme (4/4) Computational complexity: Proposed DCE scheme: O (NDI + ND 3 ) Distributed NLMS algorithm: O(NMI) F. S. Cattivelliand A. H. Sayed, Diffusion lmsstrategies for distributed estimation, IEEE Transactions on Signal Processing, vol. 58, pp , March 2010 Distributed sparsity-aware NLMS algorithm: O(3NMI) P. Di Lorenzo, S. Barbarossa, and A.H. Sayed, Distributed spectrum estimation for small cell networks based on sparse diffusion adaptation, IEEE Signal Processing Letters, vol. 20, no. 12, pp , Dec Distributed compressive sensing: O(NMI +ND 3 I) C. Wei, M. Rodrigues, and I. J. Wassell, Distributed compressive sensing reconstruction via common support discovery, in IEEE International Conference on Communications, Kyoto, Japan, June 2011

12 Measurement matrix optimization (1/3) Design of the measurement matrix: MSE metric Adaptive and can track changes in the environment Distributed and robust against link failures Minimization of MSE cost function: where = ( ) (i) is the estimate by sensor k and the measured signal is = ( ) (i)+ = ( ) (i)+.

13 Measurement matrix optimization (2/3) Expanding the cost function and taking the first 3 terms, we obtain Exploiting the fact that the random variable ( )is statistically independent from the other terms and has zero mean, we have

14 Measurement matrix optimization (3/3) We compute the gradient of the terms of the cost function and equate them to a zero matrix: where and The resulting system of equations is given by Note that the above expression cannot be solved in closed-form because is unknown. Solution: adaptive algorithms

15 Proposed adaptive algorithm (1/2) Development of proposed adaptive algorithm: We employ instantaneous values for the statistical quantities: Steepest descent rule with instantaneous values: where is the step size and is the M x 1 unknown parameter vector.

16 Proposed adaptive algorithm (2/2) The D x 1 parameter vector is used in the reconstruction of the M x 1 parameter vector : where f OMP {. } denotes the orthogonal matching pursuit (OMP) algorithm. Replacing (i) with ( ), we obtain the proposed measurement matrix optimization algorithm: Computational complexity: O(NDI + ND 3 I)

17 Simulations (1/5) The proposed DCE scheme and algorithms are assessed in a wireless sensor network application. We consider a network with N = 20 nodes, parameter vectors with M = 50 coefficients, D=5, and sparsity of S = 3 non-zero coefficients. The input signal is generated as and where =0.5 is a correlation coefficient. The compressed input signal is obtained by The measurement matrix is an i.i.d. Gaussian random matrix. The noise is modelled as complex Gaussian noise with variance

18 Simulations (2/5) The proposed DCE scheme and algorithms are compared with: Distributed NLMS algorithm: O(NMI) F. S. Cattivelliand A. H. Sayed, Diffusion lmsstrategies for distributed estimation, IEEE Transactions on Signal Processing, vol. 58, pp , March 2010 Distributed sparsity-aware NLMS algorithm: O(3NMI) P. Di Lorenzo, S. Barbarossa, and A.H. Sayed, Distributed spectrum estimation for small cell networks based on sparse diffusion adaptation, IEEE Signal Processing Letters, vol. 20, no. 12, pp , Dec Distributed compressive sensing: O(NMI +ND 3 I) C. Wei, M. Rodrigues, and I. J. Wassell, Distributed compressive sensing reconstruction via common support discovery, in IEEE International Conference on Communications, Kyoto, Japan, June 2011

19 Simulations (3/5) MSE performance x time: The DCE scheme has a significantly faster convergence and a better MSE performance than other algorithms. This is because of the reduced dimension and the fact that compressive sensing is implemented in the estimation layer.

20 Simulations (4/5) MSE performance x time: The DCE scheme can achieve an even faster convergence when compared with the DCE scheme without the measurement matrix optimization and other algorithms.

21 Simulations (5/5) MSE performance x D: We compare the DCE scheme with the distributed NLMS algorithm with different levels of resolution in bits per coefficient, reduced dimension D and sparsity level S. With the increase of the sparsity level S, the MSE performance degrades. The MSE performance improves with the increase of the number of bits.

22 Conclusions We have proposed a novel DCE scheme and algorithms for sparse signals and systems based on CS techniques. In the DCE scheme, the estimation procedure is performed in a compressed dimension which reduces the required bandwidth. We have also proposed a measurement matrix optimization strategy and an algorithm to design the measurement matrix online. The results for a WSN application show that the DCE scheme outperforms prior work in terms of convergence rate, bandwidth and performance.

23 Thanks! For further details please contact me at The paper outlining the mains ideas here appeared in IEEE Signal Processing Letters: Xu, S.; de Lamare, R.C.; Poor, H.V., "Distributed Compressed Estimation Based on Compressive Sensing,"IEEESignal Processing Letters,vol.22, no.9, pp.1311,1315, Sept. 2015

24 References (1/4) [1] C. G. Lopes and A. H. Sayed, Diffusion least-mean squares over adaptive networks: Formulation and performance analysis, IEEE Transactions on Signal Processing, vol. 56, no. 7, pp , July [2] S. Xu and R. C. de Lamare, Distributed conjugate gradient strategies for distributed estimation over sensor networks, in Proc. Sensor Signal Processing for Defence 2012, London, UK, [3] S. Xu, R. C. de Lamare, and H. V. Poor, Adaptive link selection strategies for distributed estimation in diffusion wireless networks, in Proc. IEEE International Conference on Acoustics, Speech, and Signal Processing, Vancouver, Canada [4] F. S. Cattivelliand A. H. Sayed, Diffusion lmsstrategies for distributed estimation, IEEE Transactions on Signal Processing, vol. 58, pp , March [5] Y. Chen, Y. Gu, and A.O. Hero, Sparse LMS for system identification, in Proc. IEEE ICASSP, pp , Taipei, Taiwan, May [6] P. Di Lorenzo, S. Barbarossa, and A.H. Sayed, Distributed spectrum estimation for small cell networks based on sparse diffusion adaptation, IEEE Signal Processing Letters, vol. 20, no. 12, pp , Dec 2013.

25 References (2/4) [7] D. Angelosante, J.A Bazerque, and G.B. Giannakis, Online adaptive estimation of sparse signals: Where rls meets the l1 norm, IEEE Transactions on Signal Processing, vol. 58, no. 7, pp , July [8] S. Chouvardas, K. Slavakis, Y. Kopsinis, and S. Theodoridis, A sparsity promoting adaptive algorithm for distributed learning, IEEE Transactions on Signal Processing, vol. 60, no. 10, pp , Oct [9] G. Mateos, J. A. Bazerque, and G. B. Giannakis, Distributed sparse linear regression, IEEE Transactions on Signal Processing, vol. 58, no. 10, pp , Oct [10] P. Di Lorenzo and A. H. Sayed, Sparse distributed learning based on diffusion adaptation, IEEE Transactions on Signal Processing, vol. 61, no. 6, pp , March [11] Z. Liu, Y. Liu, and C. Li, Distributed sparse recursive least-squares over networks, IEEE Transactions on Signal Processing, vol. 62, no. 6, pp , March [12] M. O. Sayin and S. S. Kozat, Compressive diffusion strategies over distributed networks for reduced communication load, IEEE Transactions on Signal Processing, vol. 62, no. 20, pp , Oct [13] D. L. Donoho, Compressed sensing, IEEE Transactions on Information Theory, vol. 52, no. 4, pp , April [14] E.J. Candes and M.B. Wakin, An introduction to compressive sampling, IEEE Signal. Proc. Mag.,vol.25,no.2,pp.21 30, March2008.

26 References (3/4) [15] J. Romberg, Imaging via compressive sampling, IEEE Signal Processing Magazine, vol. 25, no.2,pp.14 20, March2008. [16] W. Bajwa, J. Haupt, A. Sayeed, and R. Nowak, Compressive wireless sensing, in Proc. IEEE Inform. Process. Sens. Netw., Nashville, TN, April ] Y. Yao, A. P. Petropulu, and H. V. Poor, MIMO radar using compressive sampling, IEEE Journal on Selected Topics in Signal Processing, vol. 4, no. 1, pp , Feb [18] C. Wei, M. Rodrigues, and I. J. Wassell, Distributed compressive sensing reconstruction via common support discovery, in IEEE International Conference on Communications, Kyoto, Japan, June [19] D. Baron, M. F. Duarte, M. B. Wakin, S. Sarvotham, and R. G. Baraniuk, Distributed compressive sensing, in ECE Department Tech. Report TREE-0612, Rice University, Houston, TX, November [20] G. Quer, R. Masiero, G. Pillonetto, M. Rossi, and M. Zorzi, Sensing, compression, and recovery for WSNs: Sparse signal modelingand monitoring framework, IEEE Transactions on Wireless Communications, vol. 11, no. 10, pp , October [21] J. A. Troppand A. C. Gilbert, Signal recovery from random measurements via orthogonal matching pursuit, IEEE Transactions on Information Theory, vol. 53, no. 12, pp , December 2007.

27 References (4/4) [22] Y. C. Pati, R. Rezaiifar, and P. S. Krishnaprasad, Orthogonal matching pursuit: recursive function approximation with applications to wavelet decomposition, in Proc. 27th Annu. Asilomar Conf. Signals, Systems, and Computers, pp vol.1, Pacic Grove, CA, Nov, [23] G. Davis, S. Mallat, and M. Avellaneda, Adaptive greedy approximations, Constructive Approximation, vol. 13, no. 1, pp , [24] C. G. Lopes and A. H. Sayed, Incremental adaptive strategies over distributed networks, IEEE Transactions on Signal Processing, vol. 48, no. 8, pp , Aug2007. [25] L. Xie, D.-H. Choi, S. Kar, and H. V. Poor, Fully distributed state estimation for wide-area monitoring systems, IEEE Transactions on Smart Grid, vol. 3, no. 3, pp , September [26] A. Bertrand and M. Moonen, Distributed adaptive node specific signal estimation in fully connected sensor networks part II: Simultaneous and asynchronous node updating, IEEE Transactions on Signal Processing, vol. 58, no. 10, pp , [27] Y. Yao, A. P. Petropulu, and H. V. Poor, Measurement matrix design for compressive sensing based MIMO radar, IEEE Transactions on Signal Processing, vol. 59, no. 11, pp , November [28] S. Haykin, Adaptive Filter Theory, Prentice Hall, Upper Saddle River, NJ, USA, third edition, 1996