Distributed estimation over adaptive networks.

摘要

Recently a lot of interest has been shown in parameter estimation using ad hoc wireless sensor networks. An ad hoc wireless sensor network is devoid of any centralized fusion center and thus, has a distributed structure. Several algorithms have been proposed in the literature in order to exploit this distributed structure in order to improve estimation. One algorithm that was practically sound as well as fully distributed was called Diffusion LMS (DLMS) algorithm. In this work, variations to the DLMS algorithm are incorporated.;The first algorithm improves the DLMS algorithm by varying the step-size of the algorithm and eventually the Variable Step-Size DLMS (VSSDLMS) algorithm is setup. Well known VSSLMS algorithms are compared, then the most suitable algorithm identified to provide the best trade-off between performance and complexity is chosen.;Next, an algorithm is derived using the constraint that the noise variance is known. This algorithm is akin to the VSSDLMS algorithm but is computationally more complex. Convergence and steady-state analyses are carried out in detail for both algorithms. The effect of mismatch in noise variance estimate is studied for the constraint based algorithm. Extensive simulations are carried out to assess the performance of the proposed algorithms. Simulation results are found to corroborate the theoretical findings.;Finally a new scenario is investigated. All the algorithms existing in literature assume knowledge of regressor data. However, this information is not always available. This work studies blind algorithms for adaptive networks. Inspired by second order statistics based blind estimation methods, two algorithms are first converted into recursive block blind algorithms. Then these algorithms are applied to the adaptive network scenario using the diffusion scheme. Simulation results are carried out to assess the performance of the algorithms under different scenarios.;Keywords: Diffusion least mean square algorithm, Variable step-size least mean square algorithm, Noise constrained least mean square algorithm, Blind estimation algorithm, Distributed network.