Research On Distributed Estimation Over Adaptive Networks

Distributed networks are a collection of nodes that are distributed over a geographic areaand collect data from their environments. Distributed signal processing can deal withextraction of useful information from data collected at nodes. In same applications, eachnode in a network could collect noisy observations related to a certain parameter of interest.Distributed estimation is one of branches of distributed signal processing, which aims toestimate the unknown parameter of interest from the collected noisy observations. Indistributed estimation, the convergence rate, steady-state misalignment and robustness ofvarious algorithms differs from each other. To address the slow convergence rate of thedistributed least-mean-square algorithms for correlated input signals and the highcomputational cost of the distributed affine projection algorithms, in this paper twodistributed subband adaptive filtering algorithms are proposed, i.e., the incremental anddiffusion subband adaptive filtering algorithms. The distributed subband adaptive filteringalgorithms partition the signals of each node to reduce their correlation so that theirconvergence rate can be increased. The computational costs of the distributed subbandadaptive filtering algorithms are close to these of their corresponding distributed least-mean-square algorithms due to the decimation operation included in the filter banks usedfor subband partition. To solve the problem of weak robustness of the distributed least-mean-square algorithms and the distributed affine projection algorithms, this paper alsopresents two distributed affine projection sign algorithms, i.e., the incremental anddiffusion affine projection sign algorithms. The diffusion affine projection sign algorithmsare based on the L1-norm optimization criterion and therefore the incremental and diffusionaffine projection sign algorithms are robust against impulsive interference.