用于射频指纹识别的改进多尺度残差网络

An Improved Multi-scale Residual Network for Radio Frequency Fingerprint Identification

射频指纹识别可以区分高度相似的无线通信设备,已被广泛用于频谱管理和物理层安全通信。然而,很多网络模型在低信噪比环境下表现出性能下降的情况。为了提高低信噪比环境下的识别精度,设计了改进多尺度残差网络模型。该模型首先提取出解调信号的同相(I)和正交(Q)特征作为神经网络的输入,然后改进基础残差块以增加网络的每一层感受野并在细粒度水平上学习更多的特征信息。最后,将多头自注意力机制引入残差块中,进一步增强特征提取能力。在公开数据集上的测试结果表明,该网络在信噪比为0~20 dB时的平均识别准确率为85.36%,表现出比1D-ResNet网络及其变体模型更好的性能,能够在低信噪比环境下更好地完成射频指纹识别。

Radio frequency fingerprint identification(RFFI)can distinguish highly similar wireless communication devices and has been widely used in spectrum management and physical layer secure communication.However,many network models exhibit performance degradation in low signal-to-noise ratio(SNR)environments.In order to improve the identification accuracy in low SNR environments,an multiscale residual network model is designed.The model first extracts in-phase(I)and orthogonal(Q)features of the demodulated signal as inputs to the neural network.Secondly,it improves the basic residual blocks to increase the receptive field of each layer of the network and learn more feature information at a fine-grained level.Finally,the multi-head self-attention mechanism is introduced into the residual block to further enhance the feature extraction ability.The test results on open dataset show that the average identification accuracy of the network is 85.36%when the SNR is from 0 dB to 20 dB,demonstrating better performance than the 1D-ResNet network and its variant models.It can better complete RFFI in low SNR environments.