From: <dan...@br...> - 2004-06-21 09:35:45
|
Hi Scott > A little while ago Daniel talked about replacing the FHT with another scheme to generate the impulse response, which appears to have been done. Can you confirm that it is as fast as the FHT? I don't know if it's as fast as FHT, it's certainly fast enough for desktop computers. I have no idea how they do it, but fftw doesn't seem to be that much slower for n^2-1 fft lengths than n^2 lengths. >Does the correlation still create a perfect unit response as it should with the MLS? I'm not sure what you mean with a perfect unit response. But I'm sure the new scheme is much more accurate than the FHT. I have compared the results from the sweep mode with the MLS mode and they are the same. The problem with FHT MLS is that you assume that the soundcard doesn't degrade the signal at all. It's in fact even more sensitive to this than you might think. Even with an almost perfectly flat freqency response you will still not get good results with the FHT. I think the FHT should be viewed as a last resort, only to be used when speed is top priority and not accuracy. Although some of the FHT problems could be solved with calibration I still don't think you will get as good results as when you correlate in the frequency domain. I'm not sure that the FHT errors are static enough to be effectivly removed with calibration. > I must get started on a test/demo app. I have almost finished a waveform viewer/library that can be used from other appplications, so you might want to wait for that. I will make it public as soon as I figure out where to put it. > One final thing. If we have upgraded the project name to level 1 then I think we need to upgrade the major revision number to 1 as well. Perhaps 1.0. I agree on this, although I don't think we should go 1.0 until the library is stable. /Daniel |