I am a little worried with my QA401 noise floor when I do mesurment from an class D amplifier. I don’t use the QA401 so often and last time it what 2 or 3 years ago for an home made class D amplifier.
Back then, I had a noise floor of around -110 dBV. I told myself I’d do a better job with the design next time.
Today, I measured an assembled TPA3255 260-2-29A module from 3E Audio, and to my surprise, I’m getting a noise floor of -115 dBV. Amirm from Audio Science Review tested this module (paired with a Mean Well power supply) and achieved much better results with his lab-grade equipment:
Noise floor around -140 dBV (with a 1 kHz peak at 0 dBV), THD+N of 0.0008%, and SINAD of 101 dB.
Me : Noise floor around -115 dBV (with a 1 kHz peak at 0 dBV), THD+N of 0.007%, and SINAD of 83 dB (with Hypex A180 power supply).
When I do the calibration test everything is OK. IN+ and IN- shorted to GND I have a noise floor of -145dbV (FlatTop and no attenuator). But it increase around DC (-110dBV).
When I link OUT+ to IN+ and IN- shorted to GND I have a noise floor of -155dbV (FlatTop) and a few harmonics
Any idea why I have this spike near DC and why the noise floor is so high?
Were you plotting spectrum or noise spectral density? Only the latter has valid noise floor values since it compensates for FFT bin size and window function equivalent noise bandwidth. I think the best resource about this is here: https://holometer.fnal.gov/GH_FFT.pdf
In particular which flattop window? Do you know its properties? What FFT size and sampling rate? They all come into the formula for meusuring noise from an FFT spectrum.
-140 dBV is 100 nV. Without a bandwidth its not to meaningful. I think it translates into 100 nV. Thats the noise of a 100 Ohm resistor in 20K bandwidth. way less than an amps output noise.
Understanding noise is way more complex in practice. QuantAsylum has good tools for that. First- measuring noise density with the QA401 you need to use the rectangular window to get the correct numbers. The QA403 corrects for the window function. Its expresses in volts per root Hertz, 1 hertz bandwidth. This makes it more interchangable with other measurements.
Second, you want to know the “input referrred noise” which is the effective noise at the input, which is amplified by the gain of the amplifier. Then you want to know the noise of the source.. Get your input refered noise 6 dB below the source noise and further reduction will make no difference.
But notice too that in both of these you can see the RMS noise is the same. And that is what is so useful about the RMS is that works no matter your bin size. Whereas the “noise floor” of the FFT depends on the size of the FFT. And as noted above, what you really want is noise density. That removes the FFT size from the equation. But when displaying noise density, then your peaks are wrong. So, it’s really a specialized mode.
The bump at he left end is DC bleeding into the spectrum. As your FFT gets larger, that bump will push left.
Thanks everyone. Sorry for the late reply; I was on holiday.
Ok, I understand that the displayed FFT plot depends on the FFT resolution. But is that resolution linked to the SINAD and THD results?
Because I find it strange that everyone (the chip manufacturer’s datasheet, the amp manufacturer, and reviewers) reports distortion in the 0.000x% range, whereas I’m getting 0.00x%.
Measuring a Class D amplifier is not so easy. It requires a lowpass filter with a steep roll off just above 20kHz. This is needed because of some high frequency components which fold back in the audio band if not filtered with a good quality LPF.
There is a very good article from Matt about measuring Class D amps in the blog section of the QuantAsylum website.
Concerning the results of your measurements you should be careful what the chip and the module manufacturers are publishing. This is often very optimized to certain conditions. In the end whether it is 0.00x% or 0.000x% THD resp. SINAD does not really matter, both is pretty good.
