Abstract: For local ethernet in embedded machine vision system, low transmission efficiency and pseudo retransmission will happen due to the fact that RTO lags behind RTT when transmission congestion occurs in the transmission of a large amount of image data. So a method of ethernet congestion prevention based on modified LwIP is proposed in the multiple embedded vision system. Firstly, the ethernet communication platform of embedded machine vision system which contains Cortex-A8, μC/OS-III and LwIP must be built. Then, it introduces NewReno algorithm in the TCP congestion control mechanism of LwIP to overcome the influence of transmission efficiency caused by many times of rapid recovery and slow start when multiple packets lose in a send window. Finally, it introduces the normalized least mean square error algorithm to reduce false retransmission in the TCP data transmission when the RTO valuations are imprecise. Experimental results show that the method can more effectively tackle the problem of packets loss and reduce the pseudo data retransmission. What’s more, it has also increased the network throughput in some degree.

Key words: LwIP, congestion control mechanism, timeout retransmission?mechanism, embedded?machine vision system

摘要： 针对嵌入式机器视觉系统局域以太网中，在大量图像数据传输过程中出现拥塞时，存在发送效率低和RTO滞后RTT导致伪重传问题，为此提出一种基于改进型LwIP的嵌入式视觉系统以太网防拥塞方法。Cortex-A8、μC/OS-III和LwIP嵌入式机器视觉系统以太网通信平台，确保网络任务的实时响应;引入NewReno算法改进LwIP中TCP数据包的拥塞控制机制，克服了一个发送窗口多个数据包丢失引起多次快速恢复或慢启动对发送效率的影响；引入归一化最小均方误差算法改善在RTT变化较大时RTO估值不精确的情况，减少TCP数据传输中的伪重传。实验结果表明，该方法能够更加有效地处理数据包丢失问题，有效减少了数据伪重传，并且数据吞入量提高了约12.3%。

关键词: LwIP, 拥塞控制机制, 超时重传机制, 嵌入式机器视觉系统