复杂噪声场下对角减载技术的原理及应用

Principle and application of diagonal reducing method in the complex noise fields

实际水下噪声场是非常复杂的,它具有一定的相关性,且各阵元接收到的噪声的功率不相等,因此归一化的噪声协方差矩阵不是单位矩阵,会使得一些阵列信号处理算法的性能下降.针对这个问题,本文充分分析了复杂噪声场的物理特性,建立了噪声协方差矩阵的物理模型,提出了一种复杂噪声场下的协方差矩阵对角减载技术.首先将数据协方差矩阵减去一个减载系数矩阵,在使得波束输出信噪比达到最大的约束条件下,获得了减载系数矩阵的理论表达式和近似表达式.然后利用最小二乘方法,估计减载系数矩阵,并且理论分析了噪声场的相关性及输入信噪比对估计误差的影响.仿真实验和湖试数据处理结果表明,在复杂噪声场条件下,该算法提高了输出信噪比,改善了阵列信号处理算法的性能,并且该算法计算复杂度低,可以实时处理.

Acoustic environment has low signal-to-noise ratio （SNR）; hence, array signal processing is always used for reducing noise and enhancing signal. Because the delay-and-sum beam forming method is robust, so it is almost widely used, but the array gain is limited by the array aperture. The actual underwater ambient noise is complex, which includes uncorrelated noise and correlated noise. The noise powers of array elements are unequal to each other. The noise covariance matrix is not a scaled identity matrix. Consequently, the performance of array signal processing method decreases obviously. Aiming at these two problems, a diagonal reducing method of the covariance matrix in the complex noise field is proposed. Firstly, a reducing matrix, which is defined as a diagonal matrix with unequal diagonal elements, is subtracted from the covariance matrix so as to reduce the noise, and a new matrix is obtained. Secondly, the delay- and-sum beamforming is done by using the new matrix to obtain the beaming output. The analytic solution and approximate solution of reducing matrix are obtained under the constraint condition that the output SNR attains its maximum. Thirdly, the estimation of the reducing matrix is determined by minimizing the function that is defined as the error between the covariance matrix and the estimated covariance matrix. This minimization problem is accomplished in an iterative method. Fourthly, if the noise is uniform white noise or the nonuniform white noise, this proposed method performs well. While, under the complex noise field the performance of the proposed method may be deteriorated. So the effects of the correlation of the noise field and the input SNR on the estimated error are analyzed. In fact, the weaker the correlation is, or the larger the input SNR is, the smaller the estimated error is. Lastly, the simulation experiment and the lake trial are implemented. The simulation results show that the diagonal reducing method of the covariance matrix reduces some ambient noises, the noise output power decreases, the output SNR increases, and the proposed method improves the performance of array signal processing. The experimental results show that the output SNR of the target by using the proposed method is increased by about 14 dB. The diagonal reducing method of covariance matrix has definite value for engineering application, and is computationally attractive.