DRL-based Optimization of Fountain Codes with Intermediate Feedback in Buffer-limited Scenarios

Rateless codes, also known as fountain codes, are very suitable for communication in unknown and complex channel environments. However, the overhead and complexity of rateless codes will increase sharply when the receiver's buffer is limited. In this paper, we first present a transmission process for Luby Transform (LT) code with intermediate feedback. Subsequently, we propose a buffer-limited degree distribution optimization method based on deep reinforcement learning (DRL). The proposed method is applicable both with and without intermediate feedback. Furthermore, we analyze and verify the relationship between buffer capacity and the optimal feedback point under the condition of single intermediate feedback. Simulation results show that under the condition of limited buffer, the proposed method outperforms conventional schemes in terms of intermediate recovery rate, bit error rate and overhead performance.