Stochastic analysis of the diffusion least mean square and normalized least mean square algorithms for cyclostationary white Gaussian and non‐ <scp>Gaussian</scp> inputs

The diffusion least mean square (DLMS) and the normalized (DNLMS) algorithms are analyzed for a network having fusion center. This structure reduces dimensionality of resulting stochastic models while preserving important properties. analysis is done in system identification framework cyclostationary white nodal inputs. parameters vary according to random walk model. cyclostationarity modeled by periodic time variations input powers. holds all types distributions except with infinite variance. derived consist simple scalar recursions. These recursions facilitate understanding mean-square dependence upon 1) weighting coefficients, 2) kurtosis cyclostationarities, 3) noise powers, 4) unknown parameter increments. Optimization node coefficients studied. Also investigated stability two coefficients. Significant differences found between behaviors DLMS DNLMS non-Gaussian Simulations provide strong support theory.