Abstract:

Compressed spectrum sensing can obtain broadband frequency hopping(FH) communications signals at a rate much lower than that by the Nyquist sampling,but under-sampling signal reconstruction,as a key component of compressed spectrum sensing,will more directly determine the receiving performance of FH communication.Aiming at the problems of low under-sampling reconstruction accuracy,high computational complexity,and long iterative reconstruction time in compressed FH communications signal sensing,this paper proposes a deep learning reconstruction method for the under-sampling FH communication signal.In the proposed method,deep learning is introduced into the reconstruction of the wideband sparse under-sampling signal,a suitable input layer network structure is designed for under-sampling samples,and then a generative reconstruction network is constructed to replace sparse optimization.Finally,under-sampling signal reconstruction without iteration is realized.The influence of parameters such as under-sampling structure configuration,network model setting,and signal-to-noise ratio on signal reconstruction performance is simulated.Simulation results show that compared with the classical sparsity adaptive matching pursuit(SAMP),orthogonal matching pursuit(OMP) signal reconstruction algorithms,and the under-sampling signal reconstruction method based on the convolutional neural network,the proposed method has a better performance in reconstruction error and reconstruction time.The proposed method can reconstruct the under-sampling FH communications signal accurately,efficiently,and in real time,and can be one of the effective ways to solve the bottleneck of receiving and processing signals of wideband FH communications.

Key words: compressed sensing, frequency hopping communications signal, deep learning, signal reconstruction