| Summary: | 碩士 === 國立交通大學 === 電子研究所 === 107 === Deployments of video streaming applications in
wireless mesh networks increase in recent years. However, the packet loss is inevitable in wireless channels. TCP applies retransmissions once a packet is lost, which makes TCP not suitable for multimedia communication. Unlike TCP, UDP does not require retransmission. It retains a high/constant transmission rate in a channel of high loss rate; however, the transmission quality is not guaranteed. The fountain code is a forward error correction method which guarantees to recover data as long as a sufficient amount of data is received.
In this thesis, we implement a fountain code scheme on top of UDP for streaming applications. Due to the low throughput of a software implementation, we implement a
fountain code hardware decoder in 65nm process. This decoder achieves 488.2MB/s under 500 Mhz. Under this fountain code scheme, a throughput higher than that of TCP in a high packet loss rate channel can be achieved. Since a fountain code decoder starts the decoding process only when enough codewords are received. This may cause video delay. We proposed a fountain code scheme compressed sensing algorithm that can start the decoding process even when there are not enough received codewords. In this way, it can be regarded as a scalable video coding scheme.
|
|---|
