Machine Learning for Spectrum Information and Routing in Multihop Green Cognitive Radio Networks

Research works on cognitive radio networks (CRNs) together with energy harvesting (EH) promise to address the spectrum scarcity and limited battery power problems on the wireless communication nodes. While radio frequency (RF) signal of primary user (PU) can be used in EH, its absence (non-transmission state) offers an unused spectrum for the availability of the secondary user (SU) data transmission. Spectrum sensing (SS) process that detects the presence or the absence of PU signal, occupies a considerable slot in time-frame and energy consumption in wireless nodes, hence, both SU throughput and battery power get reduced. To address the problems, this work explores a support vector machine (SVM) based PU activity (transmit/non-transmit mode) prediction without accomplishing any SS task in the system operation. Thereafter, a Deep Q-networks (DQN) based energy and spectrum efficient routing strategy is suggested to maximize the CRNs throughput. The proposed joint EH and routing scheme improve the sum throughput, spectrum and energy efficiencies by
$\sim 33.48\%$
,
$\sim 34.24\%$
, and
$\sim 30.76\%$
, respectively over the conventional dedicated SS based approach.