The traditional TCP assumes that all packet losses result from the congestion even in networks with high bit error rates (BER), and this causes a significant performance degradation. In this paper, we propose a novel end-to-end congestion control mechanism called TCP VenoDB, which joins the relative delay and bandwidth estimation together as a judgment condition to distinguish the network states, and refines the traditional fast retransmit and fast recovery phase with partial ACK mechanism. Simulation results by NS-2 show that TCP VenoDB provides more significant performance improvement in throughput, bandwidth utilization and fairness than TCP Veno, TCP Westwood and TCP NewReno in networks with high BER, and has minimal influence of reverse link congestion. Furthermore, TCP VenoDB is friendly towards TCP NewReno for practical purpose.