What's the secret of high frequency harmonics?

[ChrisHooker]

Edit: Are you sure the oscillator in your first pic wasn't set at a slightly lower frequency, like 10,300 by accident?
If you're certain it was set to 11,00 on the nose...

I will say that your first screenshot does seem to show a frequency modulation, which completes a cycle every 23 samples, or 6 repetitions of the waveform. You can see the waveform repeat shape almost identically. Perhaps instead of an actual modulation, it could be a result of the osc freq not being an exact multiple of the Sample Rate (or really vice-versa), but the visual "modulation" for this case of 11K osc should be at about 110 samples shouldn't it?

44,100 samples per sec /11,000 cycles per sec = 4.0090909090909090909090909090909 samples per cycle
(instead of an exact 4 samples per cycle, it's over by 0.00909 samples per cycle)
A sampling shortage of 0.00909 samples per cycle x 110 samples = 1 full cycle
(puts the moment of sampling back at the beginning of the cycle).

Incidentally, this could be the rate of sampling error that the second screenshot is beginning to show, so I do find the appearance of this faster frequency modulation on your first pic interesting... perhaps a result of the osc generator itself.... but when I connect Martin's Osc, it does not show this, so I don't know how you got that result, unless it was set at a diff freq, or was modulated.

A close osc freq that should show exactly the same sample level each cycle would be 11.025K. I believe this freq would be better to use as a test for the accuracy of FS's high frequency generation.

However, The oscillator that I was trying to get you AWAY from (for the purposes of avoiding aliasing in your listening test) was Martin’s. The one I had connected in my schematic that I wanted you to use generates from stock oscillator prims from FlowStone.
Here is a screenshot of what I get at 11K:

Stock Osc 11K.png (127.84 KiB) Viewed 7773 times

You can this has the slower expected sampling error, just like the wavetable-generated screenshot you posted.

For reference, and proof that Flowstone can be just as accurate on producing high freqs as anything else, here's a screenshot of 11,025Hz (an exact multiple of the SR). Note that it shows NO difference cycle-to-cycle (pretty accurate):

Stock Osc 11025 Hz.png (122.86 KiB) Viewed 7773 times

[martinvicanek]

@Roxy, I haven't quite understood your goal in this thread. Just wanted to clarify that my polyBLEP oscillators are not a good reference at 10 kHz - they were designed for musical purposes, where the fundamental rarely goes beyond 2 kHz.

An "ideal" digital square wave would have all harmonics in full strength up to Nyquist (and none above). Such an oscillator can be devised, and I think FS wavetable osc's come close. However, at 44 kHz sample rate anything but the fundamental of a square wave is beyond Nyquist, hence cannot be represented. Any attempt to do so would result in aliasing.

Don't be fooled by the visual representation of waveforms at such high frequencies. The straight line segments between samples are (hopefully) not what you actually hear! Your DAC certainly does a better job at interpolating.

In the digital domain all you have is samples - there is nothing in between, no staircase, no straight line, just the amplidude values at discrete points in time given by the sample rate used. When we represent waveforms we usually draw a curve connecting the samples to guide the eye, which has often led to misconceptions. The space between the samples is only filled once we convert to the analog domain.

[Spogg]

I could discuss this further but I don't want to attract any 'experts' and have an instant replay of last week.

I'm concerned about this part of your post Roxy. The message I get from this is the implication that you have previously attracted comments from self-proclaimed experts who are not capable. I've not seen any evidence of this, quite the opposite in fact.

Everyone here who offers assistance has always done so with the best intentions, even if they are ultimately proved to be less than ideal (which has happened to me on several occasions!).

Personally I love debate but not when it gets personal. Ad hominem statements never move things forward!

Cheers

Spogg