Coherent multi-beam ray-based blind deconvolution with curvature compensation for channel impulse response estimation

This paper presents a method to mitigate the degradation of ray-based blind deconvolution for channel impulse response (CIR) estimation in the presence of near-field propagation and array deformation. The method uses sparse Bayesian learning to locate dispersed beam peaks and coherently combines the corresponding per-beam CIRs to sharpen the target wavefront. It then estimates inter-element delays by extracting the dominant arrival time in the combined CIR at each sensor and uses them to construct a curvature-compensated steering vector that removes phase errors and recovers the multipath structure. Monte Carlo simulations and at-sea data validate the approach.