The Signal Number Estimation In Array Signal Processing

The estimation of signal number in noise background is one of the key problems in array signal processing. It is widely used in radar, communication, biomedical, seismic signals and the field of electronic countermeasures and so on. The accurate number of signals is a prerequisite for many super-resolution array signal processing algorithms. If the estimated signal number is different with the actual number of signals, many super-resolution estimation algorithms will suffer from a severe performance degradation. Therefore, the signal number estimation is one of the basic tasks in array signal processing, as well as in radar, communications and other fields. Most algorithms are based on the presupposition that the background noise is stationary white Gaussian. But in practical applications, the noise is often non-stationary, non-Gaussian or even colored. Therefore, it is of great significance to study the effective signal number estimation algorithm.In this paper, we focus on the estimation of signal number in different noise environments. Some new algorithms are developed after the analysis of existing algorithms. The main contents of this paper are as follows:1. The fundamental algorithms for signal number estimation are systematically addressed. With the analysis of the principle of these algorithms, we compare their performances in different noise environments and verify their effectiveness through experiments. The weakness of traditional signal number estimation methods will be discussed in practical environment, and then we get the reasons for that.2. Through a large number of samples and repeating experiments, we sum up the distribution of noise eigenvalues with limited snapshots and then get the method to calculate the eigenvalues or estimate the eigenvalues. With the result of eigenvalues, we get a new method to improve the performance of signal estimation methods on snapshots limited circumstances, and verify its effectiveness through experiments.3. The signal estimation algorithm in non-stationary noise is discussed in detail. First, the distribution of the noise eigenvalues is estimated with finite snapshots. Second, a method to eliminate the effect of non-uniform noise is proposed by using a transformation matrix, and then followed by re-sampling method to improve the performance of the estimates. Finally, a novel method with low complexity is proposed to improve the estimation performance by using a diagonal matrix of special structure. Numerical experiments verify the effectiveness of the proposed method.