First some background. I am using a Turtle Beach Santa Cruz soundcard on WindowsME. So even though it has WDM drivers installed, it will always use an api like directsound or mme or waveout which in turn will always go via the kmixer. I wanted some way to bypass kmixer (ofcourse without upgrading to win2k/xp rightaway). So I installed ASIO4ALL which translates ASIO calls from the application to WDM-KS calls. Basically it will bypass the kmixer. (but keep in mind though it is not really ASIO, it is finally Kernel Streaming provided by WDM). So here we go..

Test Setup -> recording via RMAA loop (line-in to line-out) and playing a 44.1Khz/16bit test signal(wav) from Foobar.

First the THD and IMD using Directsound.




THD and IMD using ASIO4ALL (ASIO->WDM/KS).


As you can see the distortion reduces a lot. Whether the difference is audible or not I dont know. Maybe on a hi-end or better system it might be. This means that the soundcard's resampler isn't too bad also.

Another thing I want to show you guys is what happens when you use a good resmapler - like fb2k's SSRC.
Here's the THD using Directsound with SSRC ON. As you can see even though its going thru kmixer its not too bad. (probably because the kmixer and the soundcard do not attempt to resample the already resampled stream)

Maybe a dumb question:

Is THD the right metric for this test? Won't that miss random noise, like from quantization? Or am I misunderstanding what THD means in this case?

Also, I don't think either is really better. At least according to this test, all of the outputs produce distortion levels that are below the -96dB, so unless you're using heavy noise shaping and dither, you'll never actually be able to hear any of that.

Also2, I'm guessing that zero dB spike is an artifact? Reason I ask is I've been noticing something like that on my DAC during testing, though mine was around 11kHz. Its always there, no matter what FFT length I use. Its also less then .25Hz wide on my system (highest resolution FFT I tried).

THD and IMD were the only things severely affected so I posted only that. Noise, DR, Crosstalk, etc.. were ok.


No. Its the fundamental.
Likely the spike at 11Khz is also a fundamental trying to test the guts of the filter. What/how do you use to test your DAC ?

As mike already said, you will not be able to hear it. Even in a super quiet room (almost never the case), you will not be able to hear lower than 70dB dynamic range without damaging your ears.

Try it yourself: set the volume knob of your amp so that the music is at a loudness which you can barely stand(i.e. you will never hear music louder than that). Then decrease the digital preamp downwards...... i bet you wont hear anything anymore at about -55 to -60db.

- Lyx

Doesn't kmixer always dither the audio?

What do you mean by "Its the fundamental"? This is a frequency sweep at constant power right? It doesn't really seem possible that there could be 100dB spike in the middle unless I'm misunderstanding what you're testing.

I wrote a LabVIEW program to test mine. Basically just step a sin wave through the spectrum, and record the result at each frequency. I also tried white noise and using an FFT to recover the spectrum. Both seemed to give about the same result.

It is not a frquency sweep. As the graphs say, this is just the Total Harmonic Distortion + Noise measured at 1KHz.

The spikes beyond the fundamental are the actual distortion components.

@ -117dB they are down at the levels of probably not trusting the accuracy of the measurements.

-120dB = 0.0001% distortion.

-117db is where the noise floor is. The harmonic components are at about -90 to -100db, which is also quite far, but certainly not where you can't trust the measurements.
Anyway, the fact that the measurement is "clean" with SSRC enabled (and the fact that the soundcard by itself has excellent noise and thd results) indicates that the slightest harmonic components that are on the charts are real and not untrustworthy.

I didn't realize that was the test tone. Thanks for explaining.