基于CNN的PCMA信号调制识别方法

A modulation identification approach for PCMA signals based on convolutional neural networks

已有的针对成对载波多址(PCMA)信号调制识别的方法,其主要思想是利用高阶统计量构造识别特征量对调制类型进行识别。由于高阶统计量的估计需要较多的符号数才能达到较高的精度,故在符号数较少的情况下,这些方法的性能较差。基于深度学习的思想和卷积神经网络(CNN)的特点,提出一种新的PCMA信号调制类型识别方法,仅需要较少的符号数就能够有效识别PCMA信号调制类型。卷积神经网络的应用使得该方法对频偏和相偏具有很强的鲁棒性,且在不同的信噪比下都能够保持良好的识别性能。仿真实验结果表明,利用1000个符号,在信噪比为10dB时,新方法的调制类型正确识别率达到90%,明显优于其它方法,但是其算法复杂度较高。

The basic idea of the existing methods for modulation identification of PCMA signals is to utilize the high-order statistics of the signals to construct the corresponding features for modulation recognition. However, sufficient transmitted symbols are needed to achieve high estimation accuracy for high-order statistics. Therefore, the existing methods possess poor identification performance without sufficient transmitted symbols. Based on the idea of deep learning and the property of CNN, a novel approach for modulation recognition of PCMA signals is proposed, which can effectively identify the modulation type of the PCMA signals and achieve high identification accuracy with less transmitted symbols. Moreover, the application of CNN makes the proposed approach much robust to frequency offset and phase offset, and the approach has good identification performance under different SNR. The simulation results show that the correct modulation recognition rate of the proposed approach is more than 90% with 10dB SNR and 1000 transmitted symbols, which obviously outperforms the existing methods. However, the time consumption of the proposed approach is higher than the existing methods.