- Aug 18, 2017
But an average *is* a low pass filter. Look up any text on DSP: https://www.researchgate.net/publication/299485551_Average_Filtering_Theory_Design_and_Implementation#:~:text=A special implementation of a,that carrying high frequency distortion.
"A special implementation of a low pass algorithm is the averaging filter. It calculates the output sample using the average from a finite number of input samples. The averaging filter is used in situations where is necessary to smooth data that carrying high frequency distortion. "
Basically, an average is a crappy low pass filter in that you have no control over its cut off frequency or its strength. It gets used because it is so fast and easy to compute.
Let's test that. In my measurements I am showing the left and right response. People often average those two and build a filter based on that. Problem with that scheme is that the filter is no longer match either channel now.
Harman did a study of their Room EQ where they tested optimizing for one location versus a few locations around the seating position. The one optimized for one location won in listening test versus weighted averaged of multiple locations.
In my amplifier SINAD rating bar graphs I compute the geometric mean. That way, you have a real value of an amplifier that has as many products below it, than it has above it. Make an average and it doesn't stand for any and can be easily skewed by a single very high or very low value. This is actually a problem when I was testing headphones with B&K 5128. It is common to get a single outlier and with it, generate really bad "average." It is best to use GeoMean or discard that sample.
Anyway, we really don't need any of these arguments. Basic intuition shows that the two graphs that were show are essentially the same and could not come from two different designs:
If I breath on my fixture I can get one or the other. No way any kind of crude signal processing like averaging bail you out of the inaccuracy of the measurements. And the small sample size to boot.
Inaccuracy of measurement due to random variables can certainly be improved by averaging. No need to argue here, but, again, S/N of any signal with noise can be improved by averaging and has nothing to do with headphone measurements but with noise statistics.