| Paper ID | F-3-3.6 |
| Paper Title |
Enhanced Channel Tracking in THz Beamspace Massive MIMO: A Deep CNN Approach |
| Authors |
Navjot Kaur, Seyyed Saleh Hosseini, Benoit Champagne, McGill University, Canada |
| Session |
F-3-3: Signal Processing Systems for AI |
| Time | Thursday, 10 December, 17:30 - 19:30 |
| Presentation Time: | Thursday, 10 December, 18:45 - 19:00 Check your Time Zone |
|
All times are in New Zealand Time (UTC +13) |
| Topic |
Signal Processing Systems: Design and Implementation (SPS): |
| Abstract |
In this paper, we propose a novel model-driven deep learning approach to improve the performance of the channel tracking process in terahertz (THz) massive MIMO (m-MIMO) systems. Specifically, a recently introduced a priori aided (PA) channel tracking scheme which exploits the kinematics of the mobile users, is first used to obtain preliminary estimates of the THz m-MIMO channel. Then, a deep convolutional neural network (DCNN) based on the visual geometry group network architecture (VGGNet) is employed to refine these estimates, where the DCNN is trained offline to learn strong features of the non-linear map between PA-based channel estimates and the true channels. Simulation results demonstrate that the proposed DCNN-based approach significantly outperforms its traditional counterpart in terms of normalized mean square error. The resulting gains in accuracy can be traded to reduce the pilot overhead or required signal-to-noise ratio (SNR). |