In this paper, we presented an improved approach to design a broadband beamforming in time domain and its architecture on FPGA. The FIR digital filter with rigid linear phase was obtained by linearly constrained LMS algorithm. An iterative algorithm was used to enhance the performance of the FIR digital filter through adaptively optimizing cost function according to amplitude error function. Simulational and experimental results based on FPGA for a 10-element uniform linear array showed that the adaptive algorithm proposed in this paper was effective, can satisfied the real time requirement of underwater high speed signal processing. Moreover, this beamforming method was valuable in practical engineering in sonar systems, and can be applied to arbitrary array.