|
Dear list, I've had a number of helpful replies on and off the list that have helped a good deal with clarifying things. The most significant point is the right method for establishing the level of noise. I was presuming that noise should be observable from the correlation between successively taken impulse responses. Though this seems plausible, it led me to the impression that
there was less noise from shortish sweeps. In fact, one can measure the noise floor from a single impulse response, perhaps most elegantly by using Schroeder's reverse integration method, for which there is a clear change in the function slope when the tail of the IR gives way to noise. Using this method, the benefit of longer sweeps becomes much more apparent. In a quiet office environment, the noise is about 50 dB down with a 20 s sweep, whereas it is only about 35 dB down for a 2 s sweep. So the noise rejection improves with sweep length, but is still pretty good even with a short one. Correlation also has the
problems that offsets in time can lead to a drop in value, which cannot easily be compensated (cross-correlation will not help unless the delay is an integer number of sampling periods),
and that one has to take more than one measurement to do the correlations. John.
|