The diffusion least mean square algorithm with variable q-gradient

• Published in: Digital signal processing Vol. 127; p. 103531
• Main Authors: Cai, Peng, Wang, Shiyuan, Qian, Junhui, Zhang, Tao, Huang, Gangyi
• Format: Journal Article
• Language: English
• Published: Elsevier Inc 01.07.2022
• Subjects: Diffusion adaptive filter; Diffusion LMS; Distributed estimation; Optimization; q-gradient; Distributed estimation; Diffusion adaptive filter; q-gradient; Optimization; Diffusion LMS
• ISSN: 1051-2004; 1095-4333
• DOI: 10.1016/j.dsp.2022.103531

•We propose a novel v-q-DLMS based on the q-gradient method.•The analysis for v-q-DLMS is conducted in the mean and mean square sense.•The variable q-version ARV is proposed.•The superiorities of the proposed algorithm are validated by simulations. To avoid the local minima in optimization for distributed adaptive filters, the q-gradient diffusion least mean square (q-DLMS) algorithm uses the q-gradient vector to estimate parameters of interest in distributed networks. However, the q value in q-DLMS is fixed and required to be determined beforehand. To this end, a novel variable q-DLMS (v-q-DLMS) algorithm is proposed in this paper to improve the performance of q-DLMS and avoid the selection issue of q, simultaneously. The theoretical results of the proposed v-q-DLMS algorithm regarding accuracy and convergence are provided for performance analysis. In addition, the variable q-version of combination rule is derived by minimizing mean square derivation. Simulation results on distributed networks validate the correctness of obtained theoretical results and illustrate the superiorities of the proposed v-q-DLMS algorithm from the aspects of accuracy, convergence rate, and robustness.