Research on Influence of Source Number Estimation on Application of Blind Source Separation Algorithms

Source number estimation is the premise for applying blind source separation algorithms, and it's a technical difficulty. The influence of source number estimation on application of blind source separation algorithms is researched in present paper. Results show that, in the condition of over-determined and positive-determined, if the estimated source number is more than practical source number, then the separated signals would be more than practical source signals, usually all of the source signals can be separated, and the part of separated signals more than practical source signals would be the practical source signals' copy. When the estimated source number is less than that of practical source signals, the separated source signals would be part of the practical source signals. In the condition of under-determined, when estimated source number is more than practical source number, all columns of the practical mixing matrix can be estimated, and the other estimated columns are the practical columns' copy or redundancy, all of the practical source signals can be separated, and the other part are the practical source signals' copy or redundancy. If the estimated source number is less than that of practical source signals, part columns of the practical mixed matrix can be estimated, and part of the practical source signals can be recovered.