[Paper Review] Fast-MUSIC for Automotive Massive-MIMO Radar

Bin Li, Shuseng Wang+3 more|arXiv (Cornell University)|Nov 18, 2019

Radar Systems and Signal Processing46 references4 citations

TL;DR

This paper proposes fast-MUSIC, a computationally efficient variant of the MUSIC algorithm for automotive massive-MIMO radar, using randomized low-rank approximation to accelerate subspace computation by orders of magnitude while preserving high-resolution angle-of-arrival (AoA) estimation accuracy. The method achieves near-exact pseudo-spectrum performance compared to conventional MUSIC, enabling real-time high-resolution sensing in mobile and automotive applications.

ABSTRACT

Massive multiple-input multiple-output (MIMO) radar, assisted by millimeter-wave band virtual MIMO techniques, provides great promises to the high-resolution automotive sensing and target detection in unmanned ground/aerial vehicles (UGA/UAV). As one long-standing challenging problem, however existing subspace methods may suffer from either the low resolution/accuracy or the high time complexity. In this study, we propose two computational efficient methods to accomplish the high-resolution estimation of angle of arrival (AoA) information. By leveraging randomized low-rank approximation, our fast-MUSIC approaches, relying on random sampling and projection techniques, would speed up the subspace computation by orders of magnitude. At the same time, we establish the theoretical bounds of our proposed approaches, which ensure the accuracy of approximated pseudo-spectrum. As shown, in the case of high signal-to-noise ratio, the pseudo-spectrum acquired by our fast-MUSIC is highly precise, when compared to the exact MUSIC. Comprehensive numerical study demonstrates that our new methods are tremendously faster than MUSIC, while the AoA estimation accuracy are almost as good as MUSIC. As such, our fast-MUSIC enables the high-resolution yet real-time sensing with massive MIMO radar, which has great potential in the emerging mobile computing and automotive applications.

Motivation & Objective

• Address the high computational complexity of traditional subspace methods like MUSIC in massive-MIMO radar systems.
• Enable real-time, high-resolution angle-of-arrival (AoA) estimation for automotive and unmanned vehicle applications.
• Develop a computationally efficient alternative to MUSIC that maintains high estimation accuracy under practical signal-to-noise ratio (SNR) conditions.
• Theoretical analysis is conducted to bound the error introduced by low-rank approximation, ensuring reliability of the pseudo-spectrum.

Proposed method

• Leverages randomized low-rank approximation to accelerate the computation of the signal subspace in the MUSIC algorithm.
• Employs random sampling and projection techniques to reduce the dimensionality of the covariance matrix while preserving essential spectral information.
• Constructs an approximate signal subspace using a smaller, randomly projected matrix, significantly reducing computational cost.
• Applies the MUSIC algorithm on the approximated subspace to estimate the pseudo-spectrum for AoA estimation.
• Derives theoretical bounds on the approximation error to guarantee the accuracy of the resulting pseudo-spectrum.
• The method maintains the structure of MUSIC but replaces the full SVD computation with a faster randomized SVD on a reduced matrix.

Experimental results

Research questions

• RQ1 Can randomized low-rank approximation be effectively applied to accelerate the MUSIC algorithm in massive-MIMO radar without sacrificing AoA estimation accuracy?
• RQ2 What are the theoretical bounds on the error introduced by the low-rank approximation in the context of MUSIC-based AoA estimation?
• RQ3 How does the performance of fast-MUSIC compare to conventional MUSIC in terms of resolution and computational efficiency under high SNR conditions?
• RQ4 To what extent can the computational complexity of MUSIC be reduced while maintaining near-optimal pseudo-spectrum accuracy?

Key findings

• The proposed fast-MUSIC methods achieve speedups of multiple orders of magnitude in subspace computation compared to conventional MUSIC.
• Under high signal-to-noise ratio (SNR), the pseudo-spectrum of fast-MUSIC closely matches that of exact MUSIC, indicating near-identical AoA estimation accuracy.
• Theoretical bounds confirm that the approximation error remains small, ensuring reliable and accurate pseudo-spectrum estimation.
• Comprehensive numerical studies demonstrate that fast-MUSIC maintains almost identical AoA estimation performance to MUSIC while being significantly faster.
• The method enables real-time, high-resolution sensing in massive-MIMO radar systems, making it suitable for dynamic automotive and unmanned vehicle applications.