AbstractA lost packet recovery method is proposed by using (n, k, m) convolutional codes. The amount of computation in encoding and decoding as well as the lost packet recovery capability are evaluated. In the proposed method, a packet with a length of q bits is considered as a symbol on GF(2q) for application of the convolutional codes. It is recognized that on the Internet the packet loss position can be identified. Then, decoding is applied to a lossy communications channel on the packet level. First, the procedures for coding and decoding are presented. Subsequently, the packet loss recovery characteristic in the proposed method is analyzed. The condition for recovery of all lost packets is discussed. In the proposed method, linear simultaneous equations can be developed from the generating matrix and the packet loss positions. If these equations have a unique solution, all packet losses can be recovered. Also, by means of the lost packet recovery simulation, the recovery capability and computational complexity of the proposed method are evaluated. Under the model in which packets are independently lost, the method is compared with the packet recovery method using (n, k) Reed–Solomon codes with identical n and k. Simulation shows that packet recovery by (n, k, m) convolutional codes is effective in the area where the packet loss rate is rather low. © 2005 Wiley Periodicals, Inc. Electron Comm Jpn Pt 3, 88(7): 1–13, 2005; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/ecjc.20155