Channel Polarization Based Block Compressive Sensing SoftCast System

Abstract

The existing image or video transmission system of joint blocked compressive sensing and SoftCast (BCS-SoftCast) has an advantage in energy or bandwidth, but the peak signal to noise ratio (PSNR) is still poor in the case of low signal to noise ratio (SNR). We find that the indexes of the non-zero data, which correspond to the columns of the measurement matrix, have an impact on system performance. In this paper, by using the idea of channel polarization, we propose two column selection methods. For the first method, namely polarization column selection (PCS), the columns of measurement matrix are selected based on the column reliability and non-repetition. In order to further improve the performance in low SNR, the second method, namely hybrid column selection (HCS), partially modifies the first method. That is to say, some low reliability columns selected in the first method are replaced by the columns with the high weight. The simulation results show that both methods can improve the PSNR performance of BCS-SoftCast. Especially, in the low-SNR region, the PSNR performance with the HCS construction dramatically outperforms that with the conventional method.