Dilated Convolution Based CSI Feedback Compression for Massive MIMO Systems

Although the frequency-division duplex (FDD) massive multiple-input multiple-output (MIMO) system can offer high spectral and energy efficiency, it requires to feedback downlink channel state information (CSI) from users base station (BS), in order fulfill precoding design at BS. However, large dimension of CSI matrices MIMO makes very challenging, is urgent compress CSI. To this end, paper proposes a novel dilated convolution based network, namely D ilated xmlns:xlink="http://www.w3.org/1999/xlink">C hannel xmlns:xlink="http://www.w3.org/1999/xlink">R econstruction xmlns:xlink="http://www.w3.org/1999/xlink">Net work (DCRNet). Specifically, convolutions are used enhance receptive field (RecF) proposed DCRNet without increasing size. Moreover, advanced encoder decoder blocks designed improve reconstruction performance reduce computational complexity as well. Numerical results presented show superiority over conventional networks. In particular, compared state-of-the-arts (SOTA) networks, achieve almost same while floating point operations (FLOPs) by about 30%.