基于联合卷积与记忆神经网络的信号自动调制识别

AUTOMATIC MODULATION RECOGNITION FOR SIGNALS BASED ON CONVOLUTIONAL AND MEMORY NEURAL NETWORK

为了构建具有自我学习、自主决策能力的自动调制识别神经网络,将卷积神经网络与长短期记忆层结合设计一种并行双路神经网络来自动提取信号数据的调制信息,用于完成多种复杂无线电信号调制识别任务,其在24种信号的无线电信号数据集上识别率最高可达到97.54%。无线电信号IQ两路信号兼具时域信息和空间关系,还设计了增加IQ分离数据作为辅助通道的多通道联合卷积与长短期记忆神经网络。该方法在训练参数量相近的情况下加快了收敛速度,在调制识别准确度上也有相应提高。

In order to build automatic modulation recognition neural networks with self-learning and autonomous decision-making ability,we combined convolutional neural network with long and short-term memory layer,and designed parallel neural networks,so as to automatically extract the modulation information of the signal data.It was used for complex signal modulation recognition task,and the recognition rate could reach 97.54%on the 24 kind of radio signal dataset.The two dimensions of signal IQ data had both time domain information and spatial relationship,so we designed a multi-channel network which combined convolution and long and short-term memory by adding IQ separate data as the auxiliary channel.This method accelerates the convergence speed and improves the accuracy of modulation recognition accordingly in the case that the training parameters are similar.