Underwater Acoustic Channel Replay Based on Singular Value Decomposition and Autoregressive Model

The underwater acoustic channel replay method effectively captures the effects of marine environments, enabling the performance evaluation of underwater acoustic communication technologies and algorithms. To avoid the complexity of describing correlated scattering paths, the time-varying transfer function is used for replay and efficiently decomposed into principal and random components using singular value decomposition. This decomposition requires a single estimation of the decomposition order, determined by the minimum description length criterion. By regenerating random components with different sets of singular values, various time-varying underwater acoustic channels are obtained. Subsequently, autoregressive (AR) models are established for the principal and random components, yielding model parameters such as coefficients, poles, and complex Gaussian noise terms. Using the inverse process of the AR model with interpolated parameters, a time-varying underwater acoustic channel can be replayed. Compared with the traditional time-domain interpolation of time-varying impulse responses, the proposed method enables the replay of time-varying multipath cluster delays and generates replayed channels with better temporal correlation that more closely align with the original measured channels. The effectiveness of the proposed replay method is validated by comparing channel properties and communication simulation results between the original channel and replayed channels generated via different methods. Specifically, the time-varying channel properties derived using the local scattering function demonstrate that the channel generated using the proposed method exhibits time–frequency variations and statistical distributions that closely match those of the original channel.