Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

End-to-End Deep Learning for TDD MIMO Systems in the 6G Upper Midbands

Juseong Park, Foad Sohrabi, Amitava Ghosh, Jeffrey G. Andrews

TL;DR

This work designs end-to-end deep learning solutions for TDD SU-MIMO and MU-MIMO in 6G upper midbands, addressing UL pilot overhead by introducing channel-adaptive analog pilots and DNN-based joint CSI acquisition and DL precoding. A theory-guided MU-MIMO precoding module leveraging UL-DL duality provides a structured inductive bias, with complexity reductions achieved via matrix inversion tricks. Evaluations on realistic Nokia mid-band and QuaDRiGa mid-band datasets show substantial sum-rate gains at moderate/high DL SNR and under constrained UL pilots, surpassing traditional LMMSE-based baselines. The approach demonstrates practical potential for efficient precoding in 6G FR3 channels, enabling scalable end-to-end optimization for large BS arrays and moderate UE configurations.

Abstract

This paper proposes and analyzes novel deep learning methods for downlink (DL) single-user multiple-input multiple-output (SU-MIMO) and multi-user MIMO (MU-MIMO) systems operating in time division duplex (TDD) mode. A motivating application is the 6G upper midbands (7-24 GHz), where the base station (BS) antenna arrays are large, user equipment (UE) array sizes are moderate, and theoretically optimal approaches are practically infeasible for several reasons. To deal with uplink (UL) pilot overhead and low signal power issues, we introduce the channel-adaptive pilot, as part of an analog channel state information feedback mechanism. Deep neural network (DNN)-generated pilots are used to linearly transform the UL channel matrix into lower-dimensional latent vectors. Meanwhile, the BS employs a second DNN that processes the received UL pilots to directly generate near-optimal DL precoders. The training is end-to-end which exploits synergies between the two DNNs. For MU-MIMO precoding, we propose a DNN structure inspired by theoretically optimum linear precoding. The proposed methods are evaluated against genie-aided upper bounds and conventional approaches, using realistic upper midband datasets. Numerical results demonstrate the potential of our approach to achieve significantly increased sum-rate, particularly at moderate to high signal-to-noise ratio (SNR) and when UL pilot overhead is constrained.

End-to-End Deep Learning for TDD MIMO Systems in the 6G Upper Midbands

TL;DR

This work designs end-to-end deep learning solutions for TDD SU-MIMO and MU-MIMO in 6G upper midbands, addressing UL pilot overhead by introducing channel-adaptive analog pilots and DNN-based joint CSI acquisition and DL precoding. A theory-guided MU-MIMO precoding module leveraging UL-DL duality provides a structured inductive bias, with complexity reductions achieved via matrix inversion tricks. Evaluations on realistic Nokia mid-band and QuaDRiGa mid-band datasets show substantial sum-rate gains at moderate/high DL SNR and under constrained UL pilots, surpassing traditional LMMSE-based baselines. The approach demonstrates practical potential for efficient precoding in 6G FR3 channels, enabling scalable end-to-end optimization for large BS arrays and moderate UE configurations.

Abstract

This paper proposes and analyzes novel deep learning methods for downlink (DL) single-user multiple-input multiple-output (SU-MIMO) and multi-user MIMO (MU-MIMO) systems operating in time division duplex (TDD) mode. A motivating application is the 6G upper midbands (7-24 GHz), where the base station (BS) antenna arrays are large, user equipment (UE) array sizes are moderate, and theoretically optimal approaches are practically infeasible for several reasons. To deal with uplink (UL) pilot overhead and low signal power issues, we introduce the channel-adaptive pilot, as part of an analog channel state information feedback mechanism. Deep neural network (DNN)-generated pilots are used to linearly transform the UL channel matrix into lower-dimensional latent vectors. Meanwhile, the BS employs a second DNN that processes the received UL pilots to directly generate near-optimal DL precoders. The training is end-to-end which exploits synergies between the two DNNs. For MU-MIMO precoding, we propose a DNN structure inspired by theoretically optimum linear precoding. The proposed methods are evaluated against genie-aided upper bounds and conventional approaches, using realistic upper midband datasets. Numerical results demonstrate the potential of our approach to achieve significantly increased sum-rate, particularly at moderate to high signal-to-noise ratio (SNR) and when UL pilot overhead is constrained.
Paper Structure (32 sections, 2 theorems, 31 equations, 11 figures, 1 table)

This paper contains 32 sections, 2 theorems, 31 equations, 11 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Work
  3. Contributions
  4. Efficient analog CSI feedback mechanism
  5. Joint CSI acquisition and precoding modules tailored for TDD systems
  6. A theory-guided MU-MIMO precoding module
  7. Application to 6G mid-band channels with two distinct datasets
  8. System Model
  9. DL Data Transmission and UL Pilots Models
  10. Problem Formulation
  11. Analog CSI Feedback for Joint CSI Acquisition and Precoding in SU-MIMO
  12. Channel-Adaptive Pilot
  13. Joint CSI Acquisition and Precoding
  14. Training Strategy
  15. MU-MIMO Systems
  16. ...and 17 more sections

Key Result

Proposition 1

For a TDD-based noiseless UL analog feedback system, there exists a matrix $\mathbf{P}$ such that the optimal DL SU-MIMO precoder can be derived from the feedback $\mathbf{Y} = \mathbf{H} \mathbf{P}$, provided that $N_p \geq N_s$.

Figures (11)

  • Figure 1: The proposed deep neural network structure for UL analog channel training and DL precoding in MU-MIMO.
  • Figure 2: The proposed MU-MIMO scheme for joint CSI acquisition and precoding.
  • Figure 3: DL SU-MIMO channel capacity performance versus DL SNR for $N_p=\{1, 2\}$ and $N_s=2$ with fixed UL $\text{SNR}=10\, \text{dB}$.
  • Figure 4: DL SU-MIMO channel capacity performance versus UL SNR for $N_p=1, 2$ and $N_s=2$ with fixed DL $\text{SNR}=20\, \text{dB}$ using Nokia mid-band dataset.
  • Figure 5: DL MU-MIMO sum-rate performance versus DL SNR for $K=4$, $N_s=2$, and $N_p=1$ with fixed UL $\text{SNR}=10\, \text{dB}$.
  • ...and 6 more figures

Theorems & Definitions (5)

  • Proposition 1
  • proof
  • Remark 1
  • Proposition 2
  • Remark 2