Measured performance of line in and line out of SGTL5000

[MarkT]

I took the Waveforms Audio example, set the frequencies to 1000Hz, wave type to sine, and maxed the levels with

  sgtl5000_1.lineInLevel(0);
  sgtl5000_1.lineOutLevel(13);

And looked at the line outputs with my QA403 audio analyzer:

Which is a bit shy of the -100dB THD and -85dB THD+N figures from the datasheet for 1Vrms from the DAC (though the outputs at maximum are a little higher than 1Vrms).

Next I took the PassThroughStereo example and maxed the levels in the same way, and thus tested the ADC as well, generating the signal on the QA403 at a level of 1Vrms (a.k.a. 0 dBV)

The THD against signal level plot is interesting:

The normal noise-limited slope down should bottom out once distortion dominates, but here there is a strange hump around -1 dBV - the dip at 0dBV might be a cancellation thing, but its still a little odd.

BTW the QA403 is good down to -110dB or better usually, so the distortion shown is all due to the SGTL5000.

The spreading of the 1kHz peak at the bottom is also definitely due to the SGTL5000, indicative of phase noise on its clock perhaps, or its internal jitter.

 

[TomChiron]

Is there a difference in the THD between sine and sine_hires?

 

[Nantonos]

Is there a difference in the THD between sine and sine_hires?

I would expect to see a big difference, if the sine wavelength is not an integer multiple of the exact sampling frequency. See

High Precision Sine Wave Synthesis Using Taylor Series

www.pjrc.com

 

[MarkT]

Good point. Will retry later.

 

[MarkT]

AudioSynthWaveformSineHires seems to make a very small difference of 0.5dB or so:

 

[PaulStoffregen]

Thanks for measuring the high-res sine wave! I've often idly wondered this. Turns out the linear interpolation on table lookup is pretty good.

Would be interesting to know if the excessive Taylor Series approach make a difference with a higher performance DAC.

 

[h4yn0nnym0u5e]

Are the line out caps part of the circuit you’re measuring? If so, could they be contributing to the distortion?

 

[Nantonos]

There is a little more spread at the base of the odd harmonics, although the FFT sizes are also different. I had expected more of a difference, and I was wrong.

 

[KaoH]

FFT on your graph shows bandwidth of 5Hz -> 80kHz. What is the bandwidth for your THD or THD+N measurement. Is it the same? I was looking SGTL5000 datasheet but they do not spec that for their measurement. They just say samplerate 48kHz. I could not find THD spec from there. Maybe on another document.

Quite a bump on 50 Hz :|

 

[MarkT]

Yes I think 80kHz for the measurement bandwidth - which may be a little unfair.

Thanks for measuring the high-res sine wave! I've often idly wondered this. Turns out the linear interpolation on table lookup is pretty good.

Well you could measure it directly by sending to USB and capturing the raw data without using any DAC for a definitive answer - the SGTL DAC is clearly dominating the difference between Hires and not.

 

[MarkT]

Are the line out caps part of the circuit you’re measuring? If so, could they be contributing to the distortion?

Yes, its an unmodified Audio adapter D that I've had for a while but only just soldered headers too and started using. Since the QA403 has high input impedance the caps distortion should be relatively low.

I hope they are C0G/NP0 type, because they need to be into a low impedance load as high-K dielectrics distort horrendously with current flowing through them.

And they have cross-over distortion as well as voltage-coefficient as I discovered - here's a residual plot of one:

 

[omjanger]

Happy with the QA analyser? Have wondered about ordering one..

 

[MarkT]

Yes it works well for me - and QA are fast to respond to questions/suggestions