Since interests in multimedia services have been growing with the proliferation of wireless networks, there is a need to provide an efficient loss-differential scheme for the wired-cum-wireless networks. This is important because with the aid of the differential scheme, the congestion control protocol can tell the packet loss due to the error-prone wireless link or the network congestion. Then, the congestion control protocol will avoid reducing the sending rate blindly in the present of the packet loss. In this work, we propose an end-to-end loss-differential approach, called Trend-and-Loss-Density-based (TD) scheme. Since the congestion losses are highly co-related to each other, the TD scheme considers (a) the trend to indicate where the packet loss happens, and (b) the loss density to examine how often the packet loss occurs. Specifically, we use the autocorrelation function to measure the dense degree of a packet loss in a given sequence. Moreover, the threshold of each trend in the TD scheme is decided according to the characteristics of the wireless channel so it is easily adapted to different wireless network situations. We evaluate the TD scheme via extensive simulations in ns2. The results show that compared with other differential schemes, the TD scheme can substantially improve wireless and overall misclassification rates while maintaining a comparable throughput in all experiments.