Improved Recognition of Sub-Nyquist Sampled Cyclic Features using Graph Filter-bank De-noiser

In virtual antenna array based cognitive femtocell system, most of the research works have utilized conventional Nyquist rate sampling at the receiver side to recover the desired signal-of-interest (SOI). In this work, modulated wideband converter (MWC) based sub-Nyquist sampling technique is used to capture multi-band filtered OFDM signal by the femto-cell receiver. The compressive cyclostationary features of the received signal are processed by a graph filterbank and a deep auto-encoder respectively. Computer simulation results show that the cyclic feature classification performance of the proposed method is better than that of only deep auto-encoder, under frequency selective Rayleigh fading channel and additive white Gaussian noise (AWGN).