[Paper Review] A Simple DFT-aided Spatial Basis Expansion Model and Channel Estimation Strategy for TDD/FDD Massive MIMO Systems

Hongxiang Xie, Feifei Gao+2 more|arXiv (Cornell University)|Nov 16, 2015

Advanced MIMO Systems Optimization4 citations

TL;DR

This paper proposes a DFT-aided Spatial Basis Expansion Model (SBEM) for TDD/FDD massive MIMO systems that reduces training overhead and feedback by exploiting angle reciprocity and ULA characteristics. By representing channels with few DFT-based spatial basis vectors, it enables low-complexity, low-pilot-overhead uplink/downlink estimation and mitigates pilot contamination, while a greedy user scheduling scheme enhances spectral efficiency with minimal CSI feedback.

ABSTRACT

This paper proposes a new transmission strategy for the multiuser massive multiple-input multiple-output (MIMO) systems, including uplink/downlink channel estimation and user scheduling for data transmission. A discrete Fourier transform (DFT) aided spatial basis expansion model (SBEM) is first introduced to represent the uplink/downlink channels with much few parameter dimensions by exploiting angle reciprocity and the physical characteristics of the uniform linear array (ULA). With SBEM, both uplink and downlink channel estimation of multiuser can be carried out with very few amount of training resources, which significantly reduces the training overhead and feedback cost. Meanwhile, the pilot contamination problem in the uplink raining is immediately relieved by exploiting the spatial information of users. To enhance the spectral efficiency and to fully utilize the spatial resources, we also design a greedy user scheduling scheme during the data transmission period. Compared to existing low-rank models, the newly proposed SBEM offers an alternative for channel acquisition without need of channel statistics for both TDD and FDD systems based on the angle reciprocity. Moreover, the proposed method can be efficiently deployed by the fast Fourier transform (FFT). Various numerical results are provided to corroborate the proposed studies.

Motivation & Objective

• Address the high training overhead and pilot contamination in massive MIMO systems, especially in FDD mode where channel reciprocity does not hold.
• Reduce feedback and training resource requirements in multiuser massive MIMO by exploiting spatial sparsity and physical array characteristics.
• Develop a low-complexity, statistics-free channel estimation method applicable to both TDD and FDD systems using DFT-based basis vectors.
• Improve spectral efficiency during data transmission through a greedy user scheduling scheme based on spatial orthogonality.
• Enable efficient deployment via fast Fourier transform (FFT) without requiring channel covariance matrices or complex compressive sensing.

Proposed method

• Introduce a DFT-aided Spatial Basis Expansion Model (SBEM) that represents channel vectors using a small set of DFT beams, reducing parameter dimensionality.
• Leverage angle reciprocity in TDD and physical ULA structure to enable uplink spatial signatures to be used for downlink training in FDD systems.
• Use the DFT of steering vectors to generate fixed, orthogonal spatial basis vectors that represent dominant user directions.
• Apply least-squares (LS) estimation with the SBEM basis to estimate channels using minimal pilot overhead (e.g., L=32 instead of M=128).
• Design a greedy user scheduling algorithm that selects users with orthogonal spatial signatures to transmit simultaneously, maximizing spectral efficiency.
• Deploy the method efficiently using FFT for basis vector computation and channel estimation, avoiding high-complexity EVD or CS reconstruction.

Experimental results

Research questions

• RQ1 Can a DFT-based spatial basis model reduce training overhead in massive MIMO systems without relying on channel statistics?
• RQ2 How does the proposed SBEM mitigate pilot contamination in uplink training while maintaining estimation accuracy?
• RQ3 To what extent can the SBEM be applied to both TDD and FDD systems using angle reciprocity?
• RQ4 How does the spectral efficiency of the proposed method compare to conventional LS and covariance-based schemes like JSDM under mobility and low SNR?
• RQ5 Can the FFT-based implementation of SBEM achieve low-complexity channel estimation and feedback reduction in practice?

Key findings

• The proposed SBEM achieves significantly higher average spectral efficiency (AASR) than conventional LS when training overhead is high or SNR is low, with T=64 and SNR=10 dB, AASR improves by over 30%.
• At low SNR and short coherence time, SBEM outperforms LS due to reduced noise amplification from fewer training pilots.
• The BER performance of SBEM is comparable to JSDM at low SNR, with only ~0.5 dB gap, despite not requiring channel covariance matrices.
• Under user mobility, SBEM maintains stable performance as statistical angular spread increases (up to 20°), while JSDM degrades significantly due to inaccurate covariance estimates.
• The method reduces training overhead from M=128 to L=32 pilots for 128-antenna systems, cutting training overhead by ~75%.
• The use of fixed DFT basis vectors enables efficient FFT-based implementation, avoiding the need for EVD or nonlinear CS reconstruction.