Adoption of Auto-Encoders to Estimate CSI in MASSIVE MIMO Networks
Keywords:
Massive Multiple-Input Multiple-Output (MIMO), Channel State Information (CSI), Deep Learning based Auto-Encoder (DLAE) Model, COST 2100 dataset.Abstract
Massive Multiple-Input Multiple-Output (MIMO) heavily relies upon channel feedback mechanism to complete precoding operations. Thus, Channel State Information (CSI) estimation is massively responsible for high-performance gain. Thus, in this article, deep learning-based CSI feedback system is presented to ensure high compression efficiency at the encoder side and better reconstruction quality at the decoder side. Further, auto-encoders are adopted to separate energy coefficients from downlink CSI matrices so that objections functions are formulated. The effect of Complex Normalized Conjugates (CNC) is observed on CSI matrices in a massive MIMO system. An architecture of CNC supported CSI matrix system is presented. The dataset utilized by the proposed DLAE model is COST 2100 and this database contains a large amount of data to train the network efficiently. Simulation results are obtained in terms of Normalized Mean Square Error (NMSE) and Correlation efficiency against compression ratio. Further, obtained performance results are compared against varied traditional channel estimation techniques and a substantial gain is observed in terms of spectral efficiency enhancement of the CSI feedback system.
Downloads
- Requires Subscription PDF