SPARSH BANDIT ALGORITHMS FOR NON-CONTIGUOUS CHANNEL SELECTION FOR ALOT NETWORKS

Abstract

Wireless communication networks require smart technique to discover resources in restricted shared non-contiguous spectrum in order to make large-scale AIoT a reality. Wideband spectrum analyzer, based on sub-Nyquist sampling and used in Artificial Intelligence of Things (AIoT) gateway, solves this problem. Because the nature of the channels available to us is noncontiguous, so understanding of their occupancy is required. The multi play multi armed bandit (MP-MAB) algorithm is used to model problem of selection of subset. In this project, we show the ability of learning of such a task using several machine learning algorithms, with a subset having K channels within it, that leads to no reconstruction failure. Here we observe the comparison among five algorithms which are as follows:  K subset learning with UCB  K subset learning with SUCB having fixed sparsity  K subset learning with SUCB having variable sparsity  K subset learning with Thompson  K subset learning with Sparse Thompson K subset learning with SUCB having variable sparsity and K subset learning with Sparse Thompson are the main contribution of my research. Key-words: Multi-armed bandit, Upper Confidence Bound, Sparse UCB, non-contiguous wideband spectrum analyzer, Thompson sampling, sub-Nyquist sampling.