The measuring equipment used on this website?

[oratory1990]

The frequency is 5th harmonic that is 5khz for 1khz fundamental. If the sampler doesn't see the peak, it doesn't measure the 5th harmonic.
It occurs 4 times in a cycle but not in full cycle, the FFT sees half cycle after the beginning and half before the end.

Are you familiar with the Nyquist theorem? If you sample at 40 kHz, you see *every* wave below 20 kHz.

 

[Hayk]

It must be a problem with the software that needs very long period of accumulated samples to end of seeing the 5th harmonic that my software in use doesn't.
That is, using short period, you can show lower distortion than it should as is the case in #13.

 

[Hayk]

This is an example of TDA7293 at 1w 8ohm measured by QuantAsylum and the residual of a similar amplifier.
One can see clearly the crossover distortion that should exhibit 5th harmonic, where is it?

 

[NTK]

One can see clearly the crossover distortion that should exhibit 5th harmonic, where is it?

Are you sure you can just eyeball that oscilloscope trace and give a decent estimate of what HD5 would be?

 

[Hayk]

This is a class AB amp and the measurement is that of classA. The fifth harmonic should be the dominant of the THD.
The DS of TDA7293 gives about 0.006% THD for // and 0.01 for single. The measurements are for // at 1w 8ohm.

 

[NTK]

Do you have any reference or evidence to show that the 5th harmonic should be dominant?
Even for a waveform like the one below, the magnitudes of the harmonics do not increase with order.

Anyway, the 5th harmonic of a 1 kHz signal is 5 kHz. Unless you can prove Nyquist, Shannon, et al. were wrong (and be Nobel prize worthy world famous for the effort), when the sampling rate is far above 10 kHz, the ability to get an accurate magnitude of the 5th (5 kHz) harmonic is not in doubt.

 

[Hayk]

You can see yourself that the measurement shows 0.0006% where the DS shows 0.006%. 90% of the THD is missing.
I'll post tomorrow a simulation of crossover harmonics and show you the difference between classAB and classA harmonics.
For Nyquist-Shannon, as I said, you need long period of accumulated data to see the fifth harmonic. The problem is with software, time to time it will sample the spike but will it consider it as noise or harmonic is the trouble. The above analyzer doesn't see as harmonics.

 

[staticV3]

as I said, you need long period of accumulated data to see the fifth harmonic

44.1kHz sample rate is entirely sufficient to fully capture the 5th harmonic of a 1kHz fundamental, including all effects from crossover distortion.

Increasing the sample rate will not magically enable you to measure the fundamental's 5fth harmonic at 5kHz more accurately.

It will only allow you to analyze higher frequencies, at the cost of increased acquisition time if you want to keep the same noise floor.

The low frequency data will not change.

 

[danadam]

For Nyquist-Shannon, as I said, you need long period of accumulated data to see the fifth harmonic.

It only needs to be long enough for the noise floor to be lower than the harmonic level. After that, increasing the period further will actually slightly lower the level of the harmonic because the bin-width gets smaller and less of the noise is counted as the harmonic. Here's the signal with 2nd harmonic at -105 dBFS and FFT 2K, 32K and 512K, 9 averages for each. The whole signal is 47s long. For FFT 32K I took the first 3s and for FFT 2K I took the first 0.19s.

The 5th harmonic in your spectrum seem to be at -113 dB. It won't be much different with different lengths of FFT.

 

[Hayk]

I made a simulation of an amp feeding a sampler at 44khz.

Adjusted to generate crossover 5th harmonic

Then the fft of the sampler with 100 samples is this

Adjusted to be classA without extra 5th harmonic

And the fft now becomes

Notice how the 5khz that disappeared with classA is so tiny when it should be the dominant.

 

[Geert]

Notice how the 5khz that disappeared with classA is so tiny when it should be the dominant.

Difficult to gauge since the FFT’s are missing their vertical axis labels. I also don't see the 2nd harmonic on these FFT’s (which should be at -99dB if I’m not mistaken, with the 5th at -87dB). But maybe my understanding of your experiment is wrong.

 

[Hayk]

I made a simpler simulation. I created pulses at zero crossing and added to the sine wave.

The fft of this signal is

The FFT of the sampled 44khz over 50 cycles is

Where are the harmonics?

 

[danadam]

Where is anti-alias filter in those schematics? Does A1 element do it internally?

 

[staticV3]

Where are the harmonics?

All that happened is the noise floor was raised. The harmonics stayed where they where.

There seems to be a fundamental misunderstanding.

Increasing the sample rate while keeping the acquisition time constant will decrease the measurement precision as FFT bins get wider and noise floor creeps up.

If you want to improve the measurement accuracy of the relevant measured harmonics, you need to either: A: keep the sample rate but increase FFT size at the cost of increase acquisition time or B: reduce the sample rate.

Increasing the sample rate is never the answer.

 

[Hayk]

No filter.

 

[Hayk]

All that happened is the noise floor was raised. The harmonics stayed where they where.

Are you serious?

 

[staticV3]

Are you serious?

100%

 

[voodooless]

I think this is just a limitation of the way LTSpice does FFT, it cannot do averaging by default... There seem to be some ways around this, none of them are super trivial, but also not very hard.

 

[RayDunzl]

For Nyquist-Shannon, as I said, you need long period of accumulated data to see the fifth harmonic.

Harmonics of a 10Hz square wave, measured in-room with a UMIK-1

Noise along the bottom, around 1,150 harmonic spikes on out to the end of the line.

A couple of USB spikes (8 and 16kHz) for whatever reason.

I don't think it took very long.

 

[danadam]

The FFT of the sampled 44khz over 50 cycles is

To me this doesn't look like a spectrum of a signal sampled at 44.1 kHz because it shows content beyond 22.05 kHz.

No filter.

Well, it's not a proper sampling then. Here I generated a signal similar to yours:

It is at 705.6 kHz sampling rate. Let's pretend it is our input analog signal. Its spectrum goes far beyond 22.05 kHz:

If we sample it without anti-aliasing filter (in this example it just means picking every 16th sample), then obviously all that stuff will alias back to the 0-22k band. But if we sample it properly, i.e., with anti-alias filter, then all those lower harmonics will stay clearly visible: