Abstract: A blind deconvolution method using multiple beam outputs produced by Sparse Bayesian Learning(SBL) is proposed to estimate the Channel Impulse Response(CIR) of underwater acoustic multipath channel.The method firstly uses a Vertical Linear Array(VLA) and the multi-frequency SBL algorithm to obtain the multi-frequency signal in the complex domain at each vertical arriving angle;and then,it uses the phases picked out from the signal to matched filter the received signals at each hydrophone in the VLA,so as to obtain the CIR corresponding to a vertical arriving angle,which in fact corresponds to a propagation path in the waveguide;finally,it coherently adds the CIRs at all paths to get the final CIR result.Numerical simulations and the sea trial data show that the proposed method has several advantages when compared with the existing multi-path blind deconvolution algorithm based on alternate projection:(1) it does not need to accurately estimate the number of multiple paths in the waveguide;(2) the estimated angle of the multi-path signals is more accurate and the estimated signal phase is more reliable;(3) the CIR between the vessel and the VLA is effectively obtained,and the average time error of the CIR of the weak paths is reduced from 4.7 ms to 1.0 ms,so the time accuracy of the CIR of the weak path is significantly improved.As a result,it can be concluded that the proposed blind deconvolution method using SBL can effectively improve the performance of the blind estimation of CIR in underwater acoustic multipath channel.