Channel Modeling for Underwater Acoustic Communication Systems Using Liquid State Machines with Reservoir Computing

In recent years, Underwater Acoustic Communication (UAC) system is defined as Acoustic Signals (AS) by which information between underwater devices exchanged. UAC environment faces several challenges including dispersion and multipath distortion due to nature of channel modeling for reservoir computing. Hence, this research proposes Liquid State Machine-based Reservoir Computing (LSM-RC) for channel modeling with fixed input data called a “liquid” reservoir for processing input signals. Initially, the data is collected from Watkins Marine Mammal Sound Database (WMMSD) to analyze human activities. After that, preprocessing is performed using Quadrature Phase Shift Keying (QPSK) modulation to generate continuous signals through amplifier circuit. A pretrained model, Liquid State Machine (LSM) is used for modulation recognition of input signal to produce output signal. For prediction, RC is used to memorize and produce network modulation features which are capable of forecasting large-scale systems in UAC. Finally, LSM-RC predicts UAC signals with enhanced robustness and adaptability in underwater environments. The proposed LSM-RC achieved superior performance in terms of Mean Absolute Error (MAE) (0.0270) and Mean Absolute Percentage Error (MAPE) (0.4320) when compared to existing Orthogonal Time-Frequency Space (OTFS).