Dear noise fans, I've released a bugfix (msnoise 1.5.1) today to correct: * a nasty issue when computing autocorrelations with multiple filters configured, resulting in a damaged spectrum (see https://github.com/ROBelgium/MSNoise/issues/105) * reverted the usage of next_fast_len in favor of the previous nextpox2 for the MWCS FFT computations. "next_fast_len" is an optimised value for the FFT library (faster computation), but usually results in a smaller number than nextpow2. Then, as currently the smoothing of the spectrum (and cross-spectrum) is done in number of samples (not number of samples/Hz) AND because of the small number of points in the FFT, the resulting MWCS values were highly noisy. The smoother provides some stability and for sure, somehow enhances the mean coherence measured. From tests I've done, the final results of the analysis (at the compute_dtt step) is identical, but the noise (scatter) level was much higher. This reverts the behaviour to what was obtained for the original version. * a small bug that prevented the "-i" parameter to be passed to msnoise stack. to update: $ pip uninstall msnoise $ pip install msnoise and you're done! All you need to do is then to reset the DTT jobs and rerun compute_mwcs & compute_dtt. ---- MSNoise 1.6, the next release, will most probably appear around April 2018. It will include : * rewritten job logic (new job types, 1 for each step) * rewritten compute_cc that is muuuuuuuuch faster for most jobs * support for other spectral whitening methods (brutal "all to 1.0", division by PSD) * support for nomalization by the mean 3C PSD energy, to allow post-processing analysis (particle motion, or rotation to RR and TT after computing cross-components) * rewritten stack, compute_mwcs and compute_dtt to make them parallelisable * a tutorial on how to use msnoise on a 2000 core HPC using PBS Best regards from Brussels ! Cheers Thomas