基于轻量化卷积神经网络的水声通信前导信号检测方法

The detection of the preamble signal is the key link of underwater acoustic communication. Only when the detection is successful can the receiver be woken up for subsequent communication data processing to complete the communication. Currently, the commonly used normalized matching filter detection algorithm is easy to realized and has good anti-noise performance, but it cannot effectively combat the multi-path effect, and the detection performance will be significantly reduced under the condition of complex channel structure. Convolutional neural network(CNN), which has achieved excellent results in image classification field in recent years, is applied to the field of underwater acoustic communication preamble signal detection. It can still achieve high-performance detection under the condition of complex channel structure. However, the detection algorithm based on classical CNN has a large amount of computation and parameters. It is difficult to meet the requirements of timeliness of underwater acoustic communication and low energy consumption of underwater communication equipment. Therefore, in this paper, a compact neural network was designed based on Lenet-5 for the detection of underwater acoustic communication preamble signal by using the depth-separable convolution and global average pooling technologies, considering the specific characteristics of the problem of underwater acoustic communication preamble signal detection. The filter pruning technology based on inter-channel independence and post-training quantization technology were used to further compress the trained compact network, and finally a lightweight neural network was obtained for the preamble signal detection of underwater acoustic communication. The Qiandao Lake experimental results show that, the detection performance of the lightweight neural network was not much lower than that of the classical CNN, it can effectively combat the complex channel environment, and the required parameters and calculation amount were significantly reduced compared with the classical CNN, it can meet the requirements of timeliness of underwater acoustic communication and low energy consumption of underwater communication equipment well.