数据加窗对MUSIC方法的分辨率和抗噪性能的改进

基于矩阵特征分解的频率及功率谱估计方法中的MUSIC(Multiple Signal C1assification)方法有很高的分辨率，但它的缺陷在于抗噪声性能还有待提高，而且由于是对有限个观测数据进行处理，导致了误差的存在。在大量反复实验的基础上，结合几种常用窗函数的特点，提出了一种改进方法，这种方法先对数据进行加窗，进一步提高了MUSIC方法频率分辨率和抗噪声性能，同时提高了频率估计的准确性。

Effectively Dealing with Short Data Length and Low SNR to Improve the Quality of MUSIC Spectral Estimation

The effectiveness of MUSIC (Multiple Signal Classification) spectral estimation deteriorates when the data length is short and/or the SNR is low. We aim to improve MUSIC spectral estimation with weighted data against deterioration under short data length and/or low SNR. In the full paper, we explain in some detail how to improve MUSIC spectral estimation with weighted data; here we give only a briefing. We use five typical window functions (Bartlett, Hanning, Hamming, Blackman, Kaiser) to truncate the data through weighting. Then, we use these weighted data to improve the resolution and lower the sidelobe. We give many simulation results based on sine waves with white noise signals for three cases: (1) SNR=37.9 dB, data length N=64, f_1=0.10, f_2=0.14, f_3=0.20; (2) SNR=37.9 dB, N=64, f_1=0.10, f_2=0.12, f_3=0.20; (3) SNR=0 dB, N=64, f_1=0.10, f_2=0.14. For case (1), MUSIC spectral estimation can still resolve the three frequencies, but the sidelobe is relatively high. For case (2), although case (2) differs from case(1) only in that f_2=0.12 instead of 0.14, MUSIC spectral estimation can no longer resolve the two frequencies f_1 and f_2, but weighted-data MUSIC spectral estimation can. For case (3), because SNR is very low, only 0 dB, weighted-data MUSIC spectral estimation gives much better resolution of the two frequencies f_1 and f_2 than MUSIC spectral estimation and the sidelobe given by weighted-data MUSIC spectral estimation is much lower than that given by MUSIC spectral estimation.