Strange FFT / FFT not stable

Using windows version of the GUI for the QA40x , version 1.197 formal release. Notice at some input levels a strange FFT behavior which I do not seem to be able to solve. It is 3 FFT runs ok, then an FFT run it is strange.

Lowering the input signal slightly solves it.
Changing type of FFT window does not solve it.

Doing a THD=f(Pout) curve in the QA403 audio analyzer tool also is impacted by this effect.

Doing THD=f(Pout) using a Python script seems to give ok results.

Using QA403, USB voltage reported is 4.883V.

The oscilloscope signal of QA403 app looks clean.
Using a real scope (Picoscope), the FFT is also ok-esh, except the FFT on my scope does not go to such a low noise floor (only about -75dBV).

To be complete, I am testing a boosted class-D amp running with fpwm=384kHz and a booster inside the chip boosting the supply in a signal tracking mode. Amp is close to clipping but not clipping yet.

Any idea / suggestions?

Hi @JP-Huijser,

I see you have averages turned off. Sometimes, if you are averaging and you encounter a glitch in the signal, the glitchy spectrum will overwhelm the averages until the glitch clears. That is, if you doing 10 averages you get a glitch in your signal, it will take 10 more acqs to clear the glitch.

But that doesn’t look to be what you are encountering.

The next time you catch a glitch like this, can you switch to time domain and look at the signal? You can often see what is causing the disruption in the time domain. You can use the mouse to zoom in on the time domain signal and look to see if any tips are maybe cut off or don’t look quite right (soft clipping).

But your second second harmonic is in roughly the same place in the good and bad plot. So, it’s probably not a clipping issue. The rise in the noise floor suggests a discontinuity someplace. A discontinuity will have a lot of energy across the entire spectrum and will raise the noise floor considerably.

It could come from a bad connection or a dropped USB packet. The time domain should help diagnose. Please share whatever you find out if you can, as I think this topic is broadly interesting.

Hi Matt,

Thanks for your reply and advices.
Think I am close at the root cause. It is not a bad USB connection, but my impression is that the issue is to be found in I2S signal disturbance to the amplifier (DUT) due ground spikes caused by PWM switching of the amplifier with fast PWM slopes (<10nsec rise/fall time).

Maybe first some extra info:

  1. For measuring the output signal, I am using an active filter box, which I designed myself
    This box is also a 5th order 43kHz low pass filter and BTL to SE transfer.
  2. alternative I also have a 1:1 audio transformer box and a simple 2nd order low pass filter to achieve the same

With both filter methods and using a scope to look to the differential signal I noticed something strange. Triggering on the differential filtered output signal of the amplifier but on a level larger then the sine wave signal I could capture a ringing which occurs every time 1 out of 20 to 50 periods.

My impression is that this ringing is caused by a TDM/I2S signal error to the amplifier.

For your info, I am validating an amplifier chip which we developed in my company.
Looking to the build-in debug tools of this chip, I notice also a I2S error flag being triggered and as such I am for 99.9% sure that the problem is located at the input of the amplifier chip.

For creating the I2S signal, I use an ADC with level-shifters (3.2V to 1.8V/1.2V domain which is used for the amplifier). Reason to use the ADC is the fact I cannot find easily the matching connector for the to the front of the QA403 to get the I2S signals out (and back in) in the market.
I expect the cause to be that I pick up some disturbance at this interface when using the QA403.
When testing same chip in the lab at my company using APX driving directly 1.8V I2S signals the signal and FFT of APX software looks clean

Will dig a little bit further and will keep you updated.

Hi @JP-Huijser, I think it’s a generic 1.27mm pitch connector. The drawings on the page linked below have the part numbers (assuming I understood your statement correctly).

If the I2S error flag is getting raised, you might see an instant muting of the output in your DUT? And that discontinuity could give the rise to the noise floor. In any case, sounds like you have everything under control…but these types of problems can be difficult to sort out for sure!