An Improved Sliding Window BATS Code
Sachini Jayasooriya, Jinhong Yuan, Yixuan Xie · 2021
Batched sparse (BATS) codes, a class of random linear network coding, are proposed to transmit a collection of packets through erasure networks. BATS codes generalize the fountain codes and preserve properties such as ratelessness and low encoding/decoding complexity. In BATS codes, the destination node starts decoding information packets only after receiving a sufficient number of coded packets. This induces a latency. The larger the size of information block is, the longer the latency. In this paper, we apply a sliding window framework over BATS codes, that divides the information block into smaller sub-blocks, thus helping to reduce the latency. In the proposed sliding window BATS coding scheme, multiple degree distributions are designed based on the overlapping pattern of input packets in a window, and coded packets are generated accordingly. Based on asymptotic performance analysis, we formulate a heuristic optimization problem to jointly optimize the degree distribution and the window selection probability for each sub-window. Simulation results show that the proposed sliding window BATS code outperforms the standard BATS codes and existing sliding window BATS codes in terms of overhead and latency.