Research on the Modulation Recognition Method OFDM Signal Based on Short-time Fourier Transform and Convolutional Neural Network

In order to improve the modulation recognition accuracy of the Orthogonal Frequency Division Multiplexing (OFDM) signal under low signal-to-noise ratio, an OFDM signal modulation recognition method based on short-time Fourier transform and convolutional neural network is proposed. Using the cyclic stationarity of the communication signal with the symbol interval as the cycle, the method firstly performs a short-time Fourier transform with a window period, and converts the received time domain signal into time-frequency image; Secondly, the time-frequency image as a shallow feature is sent to the VGG-16 convolutional neural network model for training, and the features of the OFDM signal and several single-carrier communication signals are automatically extracted in the two dimensions of time and frequency; Then, the Softmax function is selected in the output layer of the network to map the output of the neuron to the probability space to calculate the probability of the category of the signal, and completes the modulation classification of the OFDM signal and other signals. Finally, the modulation recognition process of OFDM signal and several single carrier signals with equal symbol interval are simulated, and the results show that the proposed method can realize the modulation recognition of OFDM signal. When the signal-to-noise ratio is -8dB, the modulation recognition accuracy of OFDM signal can reach 98.3%. Compared with other OFDM signal modulation recognition algorithms, and the influence of different signal-to-noise ratios, different convolution kernel sizes, different activation functions and different optimization algorithms on the system's modulation recognition performance are analyzed.