Instrumental noise covariance matrix, whose diagonal square root is currently named radiometric noise, is one of the most important characteristics of hyper-spectral infrared sounders. It is used in spectral data inversion and atmosphere parameters estimation as a part of observation-error covariance matrix. In this paper new results of radiometric noise covariance matrix characterization in measurements of the infrared Fourier transform spectrometer IKFS-2 onboard the Meteor-M No. 2 meteorological satellite are presented. The main factors leading to the inter-channel noise correlation are considered. The results of the IKFS-2 noise covariance matrix estimation are compared obtained by various methods on the base on (1) reference calibration sources measurements, (2) calibrated atmospheric spectra and (3) imaginary part of the calibrated atmospheric spectra. The results presented are in agree with calculations and can be used in the IKFS-2 spectral data assimilation in numerical weather forecast models of Roshydromet.