Publication

Abstract : Due to the data rate demand, reliability, and flexibility of 5G cellular systems, quest for new modulation schemes has gained importance. The high mobility use case family is one of the many new use case families for 5G and is one that presents technological difficulties. In the future, it is anticipated that there will be a demand in growth for mobile services in vehicles, high-speed trains as well as aircraft. Orthogonal Time Frequency Space (OTFS) is a recently proposed modulation scheme that is very well-suitable for high-speed scenarios where the Doppler shifts are quite high. To utilize spatial multiplexing, diversity gain, and downlink precoding computations, downlink channel responses need to be fed back to the base station in frequency division duplexing (FDD) operated systems. Acquiring accurate channel state information (CSI) in high mobility scenarios is extremely challenging. Conventional algorithms use compressed sensing but, the process of reconstruction is very slow. Hence, recent techniques use deep learning to compress the CSI with low dimensionality and the original channel matrix is recovered at the base station. This paper proposes a novel deep learning network named Doubly Selective Channel Feedback Network (DSC-FeedNet) which follows the concept of an autoencoder and inception block for reliable reconstruction of channel information. Simulation results show the superior performance of DSCFeedNet compared to the existing fast iterative shrinkage thresholding algorithm (FISTA) network and CSiNet. The proposed network provides an improvement of 11.5 dB, 14.7 dB, 19.55 dB, and 20.54 dB in NMSE over CSiNet for the reconstruction of the original channel matrix for the compression ratios 1/4, 1/8, 1/16, and 1/32 respectively.