Low-Overhead Channel Estimation via 3D Extrapolation for TDD mmWave Massive MIMO Systems Under High-Mobility Scenarios

TL;DR

This work tackles pilot overhead in TDD mmWave massive MIMO under high mobility by introducing a three-domain (3D) channel extrapolation framework. It combines a knowledge-and-data driven uplink spatial-frequency extrapolation network (KDD-SFCEN) with a slot-level temporal extrapolation network (TUDCEN) to recover full CSI across space, frequency, and time from sparse pilots. The KDD-SFCEN integrates a knowledge-driven coarse estimator with ASEEM-based extrapolation, while the TUDCEN uses uplink-downlink calibration and a generative Transformer to autoregressively predict future downlink channels. Results show up to 4x reductions in spatial-frequency pilot overhead and another 4x reduction in temporal pilot overhead, yielding substantial gains in spectral efficiency in high-mobility scenarios. The framework demonstrates strong performance across varying compression ratios, SNRs, and velocities, offering a practical path to scalable, low-overhead CSI acquisition in future mmWave systems.

Abstract

In time division duplexing (TDD) millimeter wave (mmWave) massive multiple-input multiple-output (MIMO) systems, downlink channel state information (CSI) can be obtained from uplink channel estimation thanks to channel reciprocity. However, under high-mobility scenarios, frequent uplink channel estimation is needed due to channel aging. Additionally, large amounts of antennas and subcarriers result in high-dimensional CSI matrices, aggravating pilot training overhead. To address this, we propose a three-domain (3D) channel extrapolation framework across spatial, frequency, and temporal domains. First, considering the effectiveness of traditional knowledge-driven channel estimation methods and the marginal effects of pilots in the spatial and frequency domains, a knowledge-and-data driven spatial-frequency channel extrapolation network (KDD-SFCEN) is proposed for uplink channel estimation via joint spatial-frequency channel extrapolation to reduce spatial-frequency domain pilot overhead. Then, leveraging channel reciprocity and temporal dependencies, we propose a temporal uplink-downlink channel extrapolation network (TUDCEN) powered by generative artificial intelligence for slot-level channel extrapolation, aiming to reduce the tremendous temporal domain pilot overhead caused by high mobility. Numerical results demonstrate the superiority of the proposed framework in significantly reducing the pilot training overhead by 16 times and improving the system's spectral efficiency under high-mobility scenarios compared with state-of-the-art channel estimation/extrapolation methods.

Figures (11)

• Figure 1: A TDD massive MIMO system with the hybrid precoding architecture at the BS. (a) The TDD massive MIMO system; (b) The hybrid precoding architecture at the BS; (c) The 5G frame structure; (d) An exampled TDD slot pattern with a general pilot training scheme; (e) An exampled TDD slot pattern with the SLCE-aided pilot training scheme; (f) An exampled uplink SRS pattern.
• Figure 2: The overall framework of the proposed spatial, frequency, and temporal channel extrapolation method. The proposed method first realizes uplink channel estimation with the KDD-SFCEN to reduce the spatial-frequency domain pilot training overhead $C_{sl}$, and then conducts accurate slot-level channel extrapolation with the TUDCEN to reduce the times of uplink channel estimations $\frac{T}{T_p}$, thereby systematically reducing the pilot training overhead $C_o$.
• Figure 3: The architecture of the proposed ASEEM.
• Figure 4: The architecture of the (inverse) spatial-frequency sampling embedding layers and the generative Transformer.
• Figure 5: The NMSE performance of uplink channel estimation versus the frequency compression ratio $R_f$ at $R_s=1$, $v=60$ km/h, and $\gamma_{SNR}=20$ dB.