Regularization Techniques for Extracting OSNR From Low Resolution WDM Channel Monitors

Channel monitors that are based on scanning the wavelength division multiplexing (WDM) spectrum with a tunable filter must have sufficient resolution to resolve the ASE floor between channels to enable the OSNRs to be determined. If the bandwidth of the scanning filter is too broad then adjacent channels in the measurement become smeared together thus obscuring this level of ASE. However, in practice it is desirable to use as broad a filter as possible since this will increase mechanical robustness and reduce cost. In this paper, we describe and experimentally demonstrate a signal processing technique that can successfully extract the OSNR out of such smeared measured spectra. Such inverse problems are notoriously ill-posed. To deal with this our technique employs two stages of Tikhonov regularization and uses the known spectral response of the scanning filter together with the knowledge of the expected spectral structure of a single channel. As with all ill-posed inverse problems it can be expected that the extracted OSNR would be extremely sensitive to measurement errors and inaccuracies in the known filter and expected channel shapes. We investigate the sensitivities in OSNR to such errors and our results show the technique to be fairly robust, yielding uncertainties in OSNR below 1 dB over a wide range of realistic system noise levels and spectral variations.