Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Belief Information based Deep Channel Estimation for Massive MIMO Systems

Jialong Xu, Liu Liu, Xin Wang, Lan Chen

TL;DR

This work tackles the spectral efficiency loss caused by pilot overhead in downlink massive MIMO-OFDM channel estimation. It introduces a Belief Information Module (BIM) that injects per-antenna belief, derived from LS-based reliability, into DL-based estimators to exploit spatial correlation across antennas. The BIM is lightweight and plug-and-play, and it integrates with backbones such as ReEsNet, Channelformer, and CMixer to produce BIM-enhanced variants (BReEsNet, BChannelformer, BCMixer). Experimental results in a 5G urban macro setting show NMSE gains of about 1–2 dB and the ability to reduce pilot overhead by roughly 1/3 to 1/2 at low EbNo, while maintaining manageable storage and computational costs.

Abstract

In the next generation wireless communication system, transmission rates should continue to rise to support emerging scenarios, e.g., the immersive communications. From the perspective of communication system evolution, multiple-input multiple-output (MIMO) technology remains pivotal for enhancing transmission rates. However, current MIMO systems rely on inserting pilot signals to achieve accurate channel estimation. As the increase of transmit stream, the pilots consume a significant portion of transmission resources, severely reducing the spectral efficiency. In this correspondence, we propose a belief information based mechanism. By introducing a plug-and-play belief information module, existing single-antenna channel estimation networks could be seamlessly adapted to multi-antenna channel estimation and fully exploit the spatial correlation among multiple antennas. Experimental results demonstrate that the proposed method can either improve 1 ~ 2 dB channel estimation performance or reduce 1/3 ~ 1/2 pilot overhead, particularly in bad channel conditions.

Belief Information based Deep Channel Estimation for Massive MIMO Systems

TL;DR

This work tackles the spectral efficiency loss caused by pilot overhead in downlink massive MIMO-OFDM channel estimation. It introduces a Belief Information Module (BIM) that injects per-antenna belief, derived from LS-based reliability, into DL-based estimators to exploit spatial correlation across antennas. The BIM is lightweight and plug-and-play, and it integrates with backbones such as ReEsNet, Channelformer, and CMixer to produce BIM-enhanced variants (BReEsNet, BChannelformer, BCMixer). Experimental results in a 5G urban macro setting show NMSE gains of about 1–2 dB and the ability to reduce pilot overhead by roughly 1/3 to 1/2 at low EbNo, while maintaining manageable storage and computational costs.

Abstract

In the next generation wireless communication system, transmission rates should continue to rise to support emerging scenarios, e.g., the immersive communications. From the perspective of communication system evolution, multiple-input multiple-output (MIMO) technology remains pivotal for enhancing transmission rates. However, current MIMO systems rely on inserting pilot signals to achieve accurate channel estimation. As the increase of transmit stream, the pilots consume a significant portion of transmission resources, severely reducing the spectral efficiency. In this correspondence, we propose a belief information based mechanism. By introducing a plug-and-play belief information module, existing single-antenna channel estimation networks could be seamlessly adapted to multi-antenna channel estimation and fully exploit the spatial correlation among multiple antennas. Experimental results demonstrate that the proposed method can either improve 1 ~ 2 dB channel estimation performance or reduce 1/3 ~ 1/2 pilot overhead, particularly in bad channel conditions.
Paper Structure (7 sections, 9 equations, 4 figures, 3 tables)

This paper contains 7 sections, 9 equations, 4 figures, 3 tables.

Table of Contents

  1. Introduction
  2. System Model
  3. Belief Information Based Deep Channel Estimation
  4. Belief Information
  5. Belief Information Module
  6. Experimental Results
  7. Conclusion

Figures (4)

  • Figure 1: The downlink transmission of the massive MIMO-OFDM communication system.
  • Figure 2: The structure of Belief Information Module and structures of BIM enhanced convolution layer, transformer layer, and MLP-Mixer layer.
  • Figure 3: Performance of BIM enhanced channel estimation networks and basic channel estimation networks.
  • Figure 4: Performance of BIM enhanced denoising sub-networks and BIM enhanced expansion sub-networks.