Detecting the Number of Sources Using Modified EGM

In the array signal process field, to estimate the direction of arrival (DOA) of sources, it is a key problem to estimate the number of signal source correctly. Luo Jingqing proposed a set of eigenvalue gradient methods (EGMs), which showed better performances of number-detection than traditional AIC and MDL method. However, EGMs doesn´t take the false-alarm probability (PFA) into account, which causes high PFA while no source exists. In this paper, after some analysis to the average eigenvalue-gradient distribution of the pure noise auto-correlation matrix, a modification is introduced to improve the PFA performance of EGMs. Computer simulation results shows that the modification is efficient, and bring only tiny loss of number-detection probability in very low SNR