Low complexity MUSIC algorithm for FMCW radar systems

Abstract

Frequency Modulated Continuous Wave (FMCW) radars have the advantages of lower cost and complexity than pulse radars. In particular, since it is possible to detect the distance using Fast Fourier Transform (FFT), it has advantages of the speed and easiness of implementation. However, when a number of detected targets are adjacent to each other, there is a disadvantage in that the multiple targets can not be distinguished through the FFT method. In order to solve this problem of the FFT method, the super resolution algorithms such as Multiple Signal Classification (MUSIC) and Estimation of Signal Parameters via Rotational Invariance Technique (ESPRIT) have been proposed. The MUSIC algorithm has a very high resolution compared to FFT by using the correlation matrix of the received signal and by separating the subspaces of the signal and the noise. However, when the size of the correlation matrix is large, there is a problem that the computational complexity significantly increases. In this paper, we propose an MUSIC algorithm with very low complexity by adaptively controlling the size of correlation matrix according to conditions in FMCW radar systems. First, we use FFT in order to estimate the approximate frequency of the beat signal obtained from the FMCW transmit and receive signals. Secondly, we generate the correlation matrix for only one period of the sinusoid signal which has the obtained frequency by FFT. By doing so, the complexity becomes significantly reduced by applying the MUSIC algorithm compared to the conventional MUSIC algorithm. In order to evaluate the computational complexity reduction of the proposed algorithm, we compared the number of multiplications used in eigen value decomposition (EVD) of the correlation matrix which is the main operation in the MUSIC algorithm in the case of the two targets. As a result, the proposed algorithm achieved from 10 times to 1000 times reductions in computational complexity compared with the conventional MUSIC algorithm while achieving similar performance.