Some time ago I found out that my trusty old SB PCI128 card had a feedback loop hardware bug and I upgraded to a PCI512. When I did that my measurements went downhill rapidly. My assumption was that the card was lousy but over time and much research I have realized that much of the problem actually related to an improvement in the card - increasing the internal sampling rate from 44.1KHz to 48KHz!

This started with some perceptive questions on the forum that got me thinking about why there was so much noise in received MLS waveforms. When testing MLS waves I found that the noise level actually went up as sample size increased - and it should go down. I fruitlessly worked with this off and on until recently when I changed the synchronization scheme (and tried DirectOutput). Directoutput failed but the new synchronization is working so I decided to try averaging multiple MLS tests to reduce the noise level. When I did this it became clear that there was a discrepency between output sample rate and input sample rate - small but discernible (1/4000). Looking at the MLS inputs it became clear that resampling was occuring (something that I had trivially seen with DirectSound). Now, the Windows sound spec only allows inquiring about 11050, 22100, and 44.1KHz sampling rates but there is one sentence that says "other rates can be supported". I tried, therefore, using 48KHz with the SB PCI512 and lo-and-behold the resampling stopped and I was getting sample accurate data.

As a result, 0.85 now has the new synchronization and it includes support for 12000, 24000, and 48KHz sampling rates (if the sound card can use them) as well as the usual 11025, 22050, and 44.1KHz rates. If you have a newer sound card and it supports the new rates you SHOULD use them.

Here are two comparison charts:

This is a recording done at a 64K sample size and 44.1KHz sample rate

This was done 10 seconds later at a 64K sample size and a 48KHz sample rate. You decide.

Speaker Workshop is currently