The estimation of wideband coherent sources number with different noises is a challenging problem in array signal processing. Aiming at the characteristics of wideband signals, a novel method is proposed based on two-element array, which employs virtual array extension technique to virtualize the different frequencies of incident signals into varying interval virtual elements. Additionally, the correlation among incident sources can be weakened by spatial smoothing technique. However, the noise eigenvalues of the virtual array covariance matrix is seriously divergent due to the non-strictly flat noise power spectrum, which neutralizes classical information methods. In this paper, a new loading value based on diagonal loading is proposed, which can effectively smooth the noise eigenvalues. Furthermore, the ratio criterion is applied to improve the information methods in this paper. Simulations demonstrate this improved method can achieve better performance than conventional Gerschgorin disk estimator with white and colored noises.
