Modulation recognition method of UAV signals based on MobileNetv2 neural network

DOI:	10.11809/bqzbgcxb2023.03.030
Keywords:	UAV signal; modulation recognition; MobileNetv2 lightweight neural network; short time Fourier transform (STFT); energy threshold denoising
Abstract:	UAV image transmission signals have the problems like a low recognition rate of the existing modulation recognition algorithm under a low SNR, and high storage cost and complex calculation of the traditional deep network model. Besides, it is also difficult for the signals to be applied to 6G intelligent edge devices with limited storage space. In this view, this paper proposes a modulation recognition method of UAV image transmission signals based on time frequency analysis and MobileNetv2 lightweight neural network model. The one dimensional time domain signal is converted into a two dimensional time frequency image by the short time Fourier transform (STFT), and the time frequency image features are denoised and normalized through the energy threshold denoising method. Finally, MobileNetv2 lightweight neural network is used to identify the signal characteristics. The experiment uses six common single carrier digital communication signals and one multi carrier OFDM modulation signals, and is carried out in the additive white Gaussian noise (AWGN) channel environment. The experimental results show that, compared with the image features without denoising, the proposed method improves the recognition rate by about 6% when SNR=-12 dB, and achieves 93.33% for UAV image transmission signals in seven different modulation modes in an SNR=-12~0 dB Gaussian white noise environment. In addition, about 313M times of calculation are required to complete one identification. The number of model parameters is about 3.5M, and the model size is about 13M.Compared with other modulation recognition methods, the proposed method not only has higher recognition accuracy and better stability, but also significantly reduces the cost of network model storage and computation, so it is easy to be applied to mobile devices and embedded devices with limited storage resources.
Published:	2023-03-28
PDF HTML