Stochastic resonance with reinforcement learning for underwater acoustic communication signal

In the underwater acoustic communication (UAC), the receive signal is submerged in the heavy noise which make it hard to be detected. Stochastic resonance (SR) utilizes noise instead of eliminating it to improve the signal to noise ratio (SNR) and has been an attractive topic in the field of weak signal detection. However, as a nonlinear system, the SR requires sophisticated system design and critical parameter choice to meet its oscillatory condition so as to keep the balance among signal, noise and the nonlinear system. To solve this problem, the parameters that influence SR system have been analyzed in this paper and an adaptive method called as Reinforcement Learning & Genetic Algorithm (RLGA) for adjusting the parameters of the SR system has been proposed. By combining the reinforcement learning (RL) with genetic algorithm (GA), the method ameliorates the local search ability and accelerates the convergence speed of the traditional GA, which make it more suitable for the signal detection in UAC. Numerical simulations and outfields experiments shown that the proposed RLGA_SR can increase the output signal to noise ratio (OSNR) evidently and has robust in different kinds of marine noise and UAC channels. (C) 2020 Elsevier Ltd. All rights reserved.