Fountain Code-Based Multi-Path Reliable Transmission Scheme with Rnn-Assisted Predictive Feedback

Abstract

With the rapid advancement of Internet technology, the demand for network performance continues to grow. Due to the problems of low fault tolerance and limited available bandwidth of traditional single-path transmission methods, multipath transmission has received widespread attention. However, existing multipath transmission schemes may encounter problems such as packet disorder and mutual interference between paths when facing complex network environments, thus reducing network transmission efficiency. To cope with these problems, this paper proposes a multipath reliable transmission scheme (FC-MPRT) based on fountain codes and recurrent neural networks (RNN). The scheme fully uses the randomness and reliability of fountain codes to effectively solve the buffer-blocking problem caused by frequent packet loss retransmissions and the need for packets to arrive in order. In addition, a supervised prediction scheme based on recurrent neural networks (RNN) is designed at the data receiving end to solve the possible resource wastage problem during encoding and decoding. The transmission efficiency is effectively improved by predicting the arrival of packets. We evaluate the performance of FC-MPRT in different network scenarios on NS-3 and Mininet platforms. Simulation results show that FC-MPRT improves by 66.1% and 47.6% in throughput and latency, respectively, compared with existing multipath transmission schemes.

Keywords: Multipath, Raptor Codes, Machine Learning