[Paper Review] Compressive sensing based differential channel feedback for massive MIMO

Wenqian Shen, Linglong Dai+3 more|arXiv (Cornell University)|Jul 16, 2015

Advanced MIMO Systems Optimization7 references3 citations

TL;DR

This paper proposes a compressive sensing (CS)-based differential channel feedback scheme for massive MIMO systems that exploits temporal correlation to generate a differential channel impulse response (CIR) with stronger sparsity than the original CIR. By compressing this differential CIR with a lower compression ratio, the scheme reduces feedback overhead by approximately 20% compared to direct CS-based feedback while maintaining comparable normalized mean square error (NMSE) performance.

ABSTRACT

Massive multiple-input multiple-output (MIMO) is becoming a key technology for future 5G wireless communications. Channel feedback for massive MIMO is challenging due to the substantially increased dimension of MIMO channel matrix. In this letter, we propose a compressive sensing (CS) based differential channel feedback scheme to reduce the feedback overhead. Specifically, the temporal correlation of time-varying channels is exploited to generate the differential channel impulse response (CIR) between two CIRs in neighboring time slots, which enjoys a much stronger sparsity than the original sparse CIRs. Thus, the base station can recover the differential CIR from the highly compressed differential CIR under the framework of CS theory. Simulations show that the proposed scheme reduces the feedback overhead by about 20\% compared with the direct CS-based scheme.

Motivation & Objective

• Address the challenge of high feedback overhead in massive MIMO systems due to the large dimension of the MIMO channel matrix.
• Overcome the limitations of conventional codebook-based and direct CS-based feedback schemes in high-dimensional, time-varying channels.
• Leverage temporal correlation in time-varying MIMO channels to enhance sparsity and reduce feedback requirements.
• Develop a differential feedback framework that enables reliable channel state information at the transmitter (CSIT) with reduced overhead.

Proposed method

• Model the time-varying MIMO channel as a sparse CIR with support and amplitude vectors that evolve over time.
• Generate the differential CIR between two consecutive time slots by differencing the CIRs, which exhibits significantly stronger sparsity than the original CIR.
• Apply compressive sensing (CS) to the differential CIR using a random Gaussian sensing matrix, enabling recovery from compressed measurements.
• Use the subspace pursuit (SP) algorithm for robust and low-complexity signal recovery at the base station.
• Implement periodic reinitialization every P time slots to prevent error propagation in CIR recovery.
• Characterize temporal correlation via a first-order Markov process for support evolution and an autoregressive model for amplitude evolution.

Experimental results

Research questions

• RQ1 Can temporal correlation in time-varying MIMO channels be exploited to generate a more compressible differential CIR than the original CIR?
• RQ2 Does compressing the differential CIR instead of the original CIR lead to a significant reduction in feedback overhead for massive MIMO systems?
• RQ3 How does the proposed differential CS feedback scheme compare in NMSE performance and feedback overhead to direct CS-based feedback under the same compression ratio?
• RQ4 What is the impact of error propagation on CIR recovery, and how can it be mitigated in a differential feedback framework?

Key findings

• The differential CIR generated between adjacent time slots exhibits significantly stronger sparsity than the original CIR, enabling more efficient compression.
• The proposed scheme reduces feedback overhead by approximately 20% compared to direct CS-based feedback while achieving similar NMSE performance.
• At a compression ratio of 25%, the proposed scheme achieves NMSE performance comparable to direct CS-based feedback at 45% compression ratio.
• In scenarios with insufficient feedback overhead, the direct CS-based scheme fails to maintain acceptable NMSE, while the proposed scheme remains effective.
• The scheme achieves a 4 dB SNR gain over the direct CS-based scheme at a 45% compression ratio, demonstrating superior robustness.
• Periodic reinitialization every P time slots effectively prevents error propagation and maintains long-term feedback accuracy.