Detection of DSSS Signals Based on Autocorrelation Compressed Sampling and Deep Learning

This paper proposes a method that combines compressed sampling and deep learning to reduce computational complexity while ensuring the performance of detecting direct sequence spread spectrum (DSSS) signals. We consider two scenarios: one where the prior condition of the spreading code sequence period is known, and the other where this prior condition is unknown. By exploiting signal sparsity in the correlation domain, we use random measurements to reduce dimensionality before feeding data into a convolutional neural network (CNN) for detection. Extensive simulation experiments validate the feasibility of this method for direct detection of DSSS signals under low signal-to-noise ratio (SNR) conditions. The approach accomplishes signal detection tasks without signal reconstruction.