Kernel adaptive filter-based channel prediction for adaptive underwater acoustic OFDMA system

Abstract

This paper presents a dynamic adaptive forgetting sparse kernel recursive least squares (DAFS-KRLS) model for predicting time-varying underwater acoustic (UWA) channels and applies it to an adaptive orthogonal frequency-division multiple access (OFDMA) system. By introducing an offline-online joint training mechanism, the DAFS-KRLS model adapts to the time-varying nature of UWA channels, thereby improving the real-time performance and stability of channel prediction. Simulation and sea trial data are used to validate the DAFS-KRLS model, with comparisons to traditional recursive least squares (RLS), approximate linear dependency based KRLS (ALD-KRLS), and convolutional neural networks (CNN) combined with long short-term memory (LSTM) models (CNN-LSTM). Experimental results show that the DAFS-KRLS model achieves robust performance even with a smaller data volume, outperforming other models in accuracy and stability.