I'm experimenting with running sub-patches at lower sample rates. In the screenshot, when I go below 1/4 sample rate (at 44.1kHz) I'm hearing some distortion. Doesn't sound like aliasing to me...

Seems odd bc an 880 Hz sample rate should theoretically be able to handle 440Hz without aliasing, right? Am I missing something?

yes but I think you have to use an antialiasing/interpolation filter of some kind afterwards
sampling causes a (positive and negative) spectrum to repeat at multiples of the sample rate, so whenever you sample you have to put a lowpass on it both before (to eliminate the errors in the actual capture, which is irrelevant here because it's a sinewave lower than nyquist) and after (to eliminate the repeated images created by the sampling).
normally this second filter would be the job of your soundcard/dac, but in your case you are outputting at the higher sample rate
so here I think you need an antialiasing lowpass for the nyquist of your downsampled rate (1/8th of your output rate) after the outlet
edit: you could also try just using [outlet~ lin] instead if you want something easy but imperfect (or [outlet~ pad] if you want it to be even more aliased). sample-and-hold and linear interpolation are actually types of crude lowpass (convolving w/ a rectangular or triangular window)

Indeed, stairs require lowered anti-aliasing filtering at the output.

But does always any DAC need filtering, even for sines below Niquist??

Hmmm here the stairs are in the human hearing-range....

@lacuna said:

Indeed, stairs require lowered anti-aliasing filtering at the output.

There used to be CD players boasting of "8x (or 16x) oversampling" DACs.

They would do exactly what you're seeing here: 1. Zero-order hold the 44.1 kHz samples up to 352.8 kHz or 705.6 kHz. 2. Then, the final filter could be a gentle rolloff at a much lower order, rather than a very tight, high-order 20k - 22.5k filter.

But does always any DAC need filtering, even for sines below Nyquist??

If there's a zero-order hold, then yes, you do need to filter away the sharp corners. The added samples extend the frequency range above the original Nyquist -- but the zero-order hold does not accurately reflect the original signal. The sharp corners introduce high frequency energy that couldn't possibly exist at the original sample rate. It is necessary to remove those.

Reconstructing the signal from the samples could be done in a way other than upsampling and filtering. Sinc interpolation convolves the samples against a band limited impulse (band limited impulse = (sin x) / x except y = 1 when x = 0), and this can reconstruct the original signal at any time resolution, without filtering. But it's expensive, compared to upsample-and-filter.

hjh

@bklindgren I'm not hearing any distortion or aliasing, and only hearing aliasing once the downsampling factor reaches 0.125. Maybe my headphones suck, I dunno. Others here have written about removing the aliasing, but I was assuming you wanted it. Here's a way to do it that doesn't restrict you to the frequencies available to you via downsampling:

Thanks all for the responses!

I think my misunderstanding was around the stair-stepping produced by the downsampling. I understand that the hardware dac needs a low pass filter to smooth out stair-stepping, but I forgot that I'd need to do the same for the artificially low sample rate situation I created.

The below image solves the problem (for 1/8 downsampling). An original (not downsampled) osc~ and the downsampled one sound nearly identical now

@seb-harmonik-ar thanks for responding so quickly. did not know about the interpolation option - very cool
@lacaca thank you for the helpful graphic
@ddw_music super clear thank you. Example 3.audio.examples/J07.oversampling demonstrates the oversampling approach really well I think
@jameslo thanks for sharing - personally I think the foldover sound can be really cool, but this was more of a theoretical PD question