Subspace-Based SNR Estimation for Underwater Acoustic Communications

Abstract

This paper proposes a signal-to-noise power ratio (SNR) estimation method for underwater acoustic environments that combines Hankel matrix transformation, singular value decomposition (SVD), and a Gaussian mixture model (GMM). Conventional approaches such as the second- and fourth-order moments, the time-domain power ratio, and GMM-based methods tend to lose accuracy under low-SNR conditions or in time-varying channels. The proposed method transforms the received signal into a Hankel matrix and applies SVD to decompose it into subspaces. Unlike conventional Hankel–SVD methods that truncate the rank solely by singular-value magnitude, the proposed framework constructs directional variation features from the left and right singular vectors and, through GMM-based probabilistic clustering, robustly classifies the signal and noise subspaces even when the number of sources is unknown. By estimating the power levels of the separated signal and noise subspaces, the proposed method achieves stable SNR estimation performance even in adverse environments. Computer simulations and real-sea experiments demonstrate that the proposed method provides superior SNR estimation accuracy and robustness compared with conventional methods.