The effects of fundamental and ancillary algorithm differences on the performance of three noniterative factor analysis spectral resolution algorithms on noisy and overlapped bilinear matrix-formatted spectral data are evaluated and compared. The evaluation consists of the analysis of simulated fluorescence excitation-emission matrices in which the spectral overlap, noise type, and level were systematically varied. The results indicate that the conventions used to exclude low-intensity, high-noise rows and columns from consideration as component spectra estimates and to choose the first estimates of the component spectra have significant impact on resolution algorithm performance. The results of the application of the algorithms to ideal data are nearly identical; however, there are several distinctions in the performance of the algorithms on noisy data. Verifiable estimates of the component spectra were resolved from data matrices degraded by white and Poisson noise that have signal-to-noise (S/N) ratios above 10 by all three algorithms regardless of the noise level and the degree of spectral overlap. The impact of pink noise was uniformly deleterious at S/N below 15.

PDF Article

Cited By

You do not have subscription access to this journal. Cited by links are available to subscribers only. You may subscribe either as an OSA member, or as an authorized user of your institution.

Contact your librarian or system administrator
Login to access OSA Member Subscription