基于FLOCC-ESPRIT的极化阵列参数估计方法

FLOCC-ESPRIT based polarization array parameter estimation method

针对基于分数低阶矩类阵列波达方向(DOA)估计方法仅适用于独立同分布(i.i.d)SαS背景噪声的缺点,提出了一种线性极化阵列DOA和极化参数联合估计的分数低阶循环相关(FLOCC)极化参数联合估计(ESPRIT)算法。该方法首先利用入射信号的循环平稳特性,采用分数低阶循环相关函数抑制α和高斯混合噪声及与信号循环频率相异的任何循环平稳干扰信号。在此基础上,利用阵列信号参数与噪声子空间的正交性,采用ESPRIT算法直接求取了信号的DOA和极化参数。该方法对于α和高斯混合噪声及与信号循环频率相异的任何循环平稳干扰信号具有很强的抑制能力。即使对于空域内靠得非常近的信源,该方法也可利用极化信息的差异进行区分。实验结果表明,在α和高斯混合信噪比(SNR)为0 d B,信干比(SIR)为3 d B时,其DOA和极化参数估计的均方根误差分别为0.23°和0.54°;并且在实测数据环境下,当SNR为10 d B时,本文算法仍然有效。

Aiming at the deficiency that fractional lower order moment-based array Direction Of Arrival （DOA） estimation methods are only suitable for the independent identically distributed （i. i. d） SaS background noise, a fractional lower order cyclic correlation （FLOCC）-based estimation of signal parameters via rotational Invariance techniques （ESPRIT） algorithm for DOA and polarization parameter joint estimation of linear polarization array is proposed in this paper. Firstly, based on the cyclic stationary characteristic of the incident signal, the fractional lower order cyclic correlation function is used to suppress the mixed noise of a and Gaussian noise and any cycle stable interference signal with the cycle frequency different from the signal frequency. On this bass, with the orthogonality of the array signal parameters and noise subspace, the ESPRIT algorithm is used to obtain the DOA and polarization parameters of the signal directly. The proposed method has strong suppression ability for the mixed noise of a and Gaussian noise and any cycle stable interference signal with the cycle frequency different from the signal frequency. The proposed method can distinguish the sources very close in space domain with the polarization information difference. The simulation and experiment results show that while the mixed signal to noise ratio （SNR） of a and Gaussian noise is 0 dB and the signal to interference ratio （SIR） is 3 dB, the root mean square errors of DOA and polarization parameter estimation are 0.23 degree and 0.54 degree, respectively. And in the actually measured data environment, while SNR is 10 dB, the proposed algorithm is still effective.