Deep Learning-based Implicit CSI Feedback in Massive MIMO

TL;DR

A DL-based implicit feedback architecture to inherit the low-overhead characteristic is proposed, which uses neural networks (NNs) to replace the precoding matrix indicator (PMI) encoding and decoding modules and can achieve a more refined mapping between the precode matrix and the PMI compared with codebooks.

Abstract

Massive multiple-input multiple-output can obtain more performance gain by exploiting the downlink channel state information (CSI) at the base station (BS). Therefore, studying CSI feedback with limited communication resources in frequency-division duplexing systems is of great importance. Recently, deep learning (DL)-based CSI feedback has shown considerable potential. However, the existing DL-based explicit feedback schemes are difficult to deploy because current fifth-generation mobile communication protocols and systems are designed based on an implicit feedback mechanism. In this paper, we propose a DL-based implicit feedback architecture to inherit the low-overhead characteristic, which uses neural networks (NNs) to replace the precoding matrix indicator (PMI) encoding and decoding modules. By using environment information, the NNs can achieve a more refined mapping between the precoding matrix and the PMI compared with codebooks. The correlation between subbands is also used to further improve the feedback performance. Simulation results show that, for a single resource block (RB), the proposed architecture can save 25.0% and 40.0% of overhead compared with Type I codebook under two antenna configurations, respectively. For a wideband system with 52 RBs, overhead can be saved by 30.7% and 48.0% compared with Type II codebook when ignoring and considering extracting subband correlation, respectively.

Figures (6)

• Figure 1: (a) Example of a cross-polarized single-panel antenna array with $(N_1,N_2)=(2,2)$; (b) 2D oversampled DFT beams with $(O_1,O_2)=(4,4)$.
• Figure 2: Architecture of ImCsiNet-s for single RB. NNs are used to replace the PMI encoding module at the UE and the PMI decoding module at the BS. The input of the encoder is the eigenvector $\mathbf{v}$ extracted from the full channel matrix.
• Figure 3: (a) Architecture of bi-ImCsiNet, which uses bi-LSTMs instead of FC layers to extract the subband correlation; (b) The structure of a layer in the bi-LSTM network in the bi-ImCsiNet encoder.
• Figure 4: PSE distribution of eigenvectors for $N_{t}=8$ and $N_{t}=32$. PSE is a metric for the compressibility of eigenvectors.
• Figure 5: Recovery performance using different numbers of quantization bits, $B$, under a fixed number of feedback bits. The UMi(32T4R) dataset is used for simulation. Four fixed feedback bits are set, which are 156, 208, 312 and 624, respectively.