A Study of Coordinate Attention Mechanism Scheme Under Digital Signal Modulation Recognition
To solve the problem that it is difficult for neural network to extract spatial features of digital signals under low signal-to-noise ratio,a digital signal recognition scheme based on coordinate attention mechanism is proposed.Eight kinds of digital signals are orthogonally modulated,pre-encoded according to their amplitude and phase information sequences,and the key features of amplitude and phase of the digital signals are extracted and analyzed under different training steps,and the appropriate hyperparameters of the neural network are selected to make the network to reach the fitting surface.The coordinate attention mechanism encodes the digital signal features into two one-dimensional features to capture the remote dependence of amplitude and phase along the longitudinal and transverse directions,respectively;and the generated digital signal features are encoded into a pair of direction-aware and position-sensitive weight coefficients for the recalibration of the digital signal features.Simulation results show that:under eight digital signals,when the recognition rate of the modulation method is higher than 95％,the signal-to-noise gain of the coordinate attention mechanism in the Convolutional Neural Network(CNN)is about 4 dB,and the signal-to-noise gain of the coordinate attention mechanism in the residual neural network is about 8 dB.The coordinate attention mechanism achieves a higher recognition rate as well as a better signal-to-noise gain,and is more suitable for digital signal demodulation applications,compared with the channel attention mechanism and the spatial attention mechanism.
digital signalmodulation recognitioncoordinate attention mechanismweight coefficient
国家科技期刊平台
NSTL
万方数据
