A Source Number Enumeration Method at Low SNR Based on Ensemble Learning

Abstract

Source number estimation is one of the important research directions in array signal processing. To solve the difficulty of estimating the number of signal sources under a low signal-to-noise ratio (SNR), a source number enumeration method based on ensemble learning is proposed. This method first preprocesses the signal data. The specific process is to decompose the original signal into several intrinsic mode functions (IMF) by using Complementary Ensemble Empirical Mode Decomposition (CEEMD), and then construct a covariance matrix and perform eigenvalue decomposition to obtain samples. Finally, the source number enumeration model based on ensemble learning is used to predict the number of sources. This model is divided into two layers. First, the primary learner is trained with the dataset, and then the prediction result on the primary learner is used as the input of the secondary learner for training, and then the prediction result is obtained. Computer theoretical signals and real measured signals are used to verify the proposed source number enumeration method, respectively. Experiments show that this method has better performance than other methods at low SNR, and it is more suitable for real environment. Keywords—Source number estimation; Array signal processing; SNR; IMF; CEEMD; Ensemble learning.