Finding Fountain Codes for Real-Time Data by Fixed Point Method

摘要

We study fountain coding-like transfer method over a packet erasure channel for data with real-time requirements, such as streaming video or voice. The proposed method uses a sliding window, which defines the current section of non-expired data. The coding is systematic, with new blocks entering the window being initially sent intact, followed by a possible correction packet. Two different strategies for correction packet sending are studied. In both methods correction packets are generated in fountain-coding like fashion using bitwise XOR-operation for addition of source blocks. The correction packets are constructed so as to maximize the probability of immediately decoding useful data at the receiver's end, based on the sender's belief on the receiver's state. After sending the correction packet this belief is updated. Fixed point iteration is used to find the stationary solution in long streams. The two methods differ in the details how the window is moved and how exactly the correction packets are generated. We study the performance of proposed methods and give theoretical and simulation results for both of the used approaches.