Steady-State Analysis of Diffusion LMS Adaptive Networks With Noisy Links

Combination weights for diffusion strategies with imperfect information exchange

“…Observe that in (7)-(10), we are including several sources of information exchange noise. In comparison, references [10], [12] only considered the noise source v (7) and one set of scaling coefficients {a 1,lk }; the other coefficients were set to c lk = a 2,lk = 0 for l = k and c kk = a 2,kk = 1. In other words, these references only considered (7) and the traditional CTA strategy without exchange of the data {d l (i), u l,i }.…”

“…Studies on the effect of link noise on performance appear in [7]- [9] for consensus-type strategies and in [10]- [12] for diffusion-type strategies; the latter references simply extend the analysis of [1], [2] in a manner similar to what was done before in [13] while studying the performance of stand-alone filters under non-stationary conditions. In this paper, our objective is to go beyond these earlier works and to study the effect of link noise on network adaptation and learning by taking into account several additional sources of imperfection, and by considering more general algorithmic structures.…”

Steady-state analysis of the deficient length incremental LMS adaptive networks with noisy links

“…The effect of noisy links on the performance of incremental and diffusion adaptive networks are studied in [11,12]. The importance of such study stems from this fact that the performance of distributed adaptive estimation algorithm can drastically be deteriorated in the presence of noisy links [11,12].…”

“…In all of the previous works [3][4][5][6][7][8][9][10][11][12], it is assumed that the length of the adaptive filter in each node is equal to that of the unknown parameter. Actually, the length of the unknown parameter similar to its coefficients is unknown.…”

Diffusion LMS strategies for parameter estimation over fading wireless channels

“…Previous works on distributed estimation and adaptation over networks examined the effect of noisy communication links on the performance of the diffusion learning schemes [4]- [6]. The main conclusion is that performance degradation occurs unless the combination weights are adjusted accordingly in order to counter the effect of noise over the links.…”