why does this square wave sound cleaner when downsampled?

I was experimenting with square waves in CEP2 and what I found out is that the square wave that I generated (3khz) sounded better on my cd player when it was first generated in 96khz then downsampled to 44.1khz, same goes for 48khz.

So 96khz was first, 48khz second and 44.1khz last and terrible may I add.
Why is this so?

Does this all relate to why we need to record in 96khz and higher.

Is downsampling/dithering rounding the corners of the waveform?

Does this all relate to why we need to record in 96khz and higher.

Don't include me in that "we" - I don't think it's at all necessary to record at such high sampling rates.

Graeme Posted:
--------------------------------------------------------------------------------
 Don't include me in that "we" - I don't think it's at all necessary to record at such high sampling rates.

Same here.... FWIW, I don't bother with anything over 44,100.  If you accomplish the task of tracking a good sounding, clean signal with good kit, then (personally) I can't see any benefit.

 

Er...what qualifies as "better" when listening to something as artificial as a square wave?  I suspect the previous guess about the corners being rounded in the downsampling could be what you're hearing...but in fact you're getting a less accurate wave out of it!

Bob

Mmm... dude... dig the sound of that square wav....

I don't know about you but I can't even stand to listen to a square wav these days unless it's been downsampled from 96khz.

Of course square wavs really sound the best on vinyl...

Chris

Speaking of Vinyl. I need a record player. I have a ton of 7 inches and LPs but no player.  

Are we saying that we're "hip", Karl?

 

To all the nay-sayers, did you actually check this out?  Teflon appears to be right, a 3khz square definately sounds better when downsampled from 96khz!


Don't believe me, here is the 96khz downsampled:


and here is the straight to 44khz one:



While in sample view, the 44khz version appears more sample-accurate, but the 96khz version has the frequency spectrum I think is correct (odd-n harmonics, with none of the strange background harmonics).  

It seems that even a 300hz square has a messed up spectrum in 44khz, as I thought this may be a problem of there only being a handful of samples per cycle.  Also I can't seem to produce a reliable 1khz square in 96khz, it has clicks throughout

What's going on here?

About the corner rounding...  downsampling does more than that, it mutilates the whole waveform - but rather than smoothly rounding it (which would admittedly sound more pleasing) it makes the waveform follow the 'wibbly' path that the line joining the dots wants to follow one a sample-accurate wave - if you follow my poor explanation.

Back at home (work is a DAW-Free zone) and did a little investigation myself.

Have a gander at this one (300 Hz):


The green plot is the straight 44KHz.  Red is the downsampled version.  Increasing the FFT size changes the downsampled plot some.

A Square wave is the sum of the fundamental sine + each subsequent harmonic.  It is interesting that downsampled file has only the odd-numbered harmonics.  Also the -84db minimum on the 44 KHz version compared to the -150 dB or so on the downsampled file.

You can't overlay freq plots of 44KHz and 96 KHz plots accurately (bug or future enhancement request?) but the unaltered 96 KHz freq plot is only slightly different than the downsampled one shown here, and also includes only the odd-numbered harmonics.

Hmmmm.

I just got here and the topic is interesting... but remember that a perfect square-wave has only odd harmonics, and a ramp or sawtooth has both odd and even harmonics.    The better you represent a square-wave the fewer even harmonics will be present.

Mac: The perfect square-wave you generated has odd harmonics up to infinity.  The squiggly lines are what the square-wave looks like without it's harmonics above 96kHz. or 48kHz. or 24kHz. - say for 96kHz. - all harmonics above 48kHz. will be gone, so the straight lines become squiggly, reveiling the summed sine-wave nature of wave-forms.    

The displays you posted show a slower slew rate or transition slope for the 44.1kHz. example that smears the screen, true, but this again simply  represents a square-wave with all harmonics above 22.050kHz. removed.  We could do the same excersise at 192kHz. and at 384kHz. and see the very same thing - it's a matter of scale.

Teflon: What does a better square-wave sound like.  What was better about it for you when you compared the two?

... sounded better on my cd player ...

which could also be saying something about the converters on your CD player.  No question that the downsampled waveform is distorted - but to be honest, I couldn't hear a difference in headphones.

remember that a perfect square-wave has only odd harmonics

So it does ... still makes me go Hmmm, just looking the other direction.  Why are they there in the 44KHz signal?  The samples don't show any lack of squareness even when zoomed WAYYY up.

I'm not getting any even harmonics on my generated square-waves which is as expected, even when i convert a 96kHz. to a 48kHz. or start with 48kHz. and convert up to 96kHz.  Maybe you have a bug Zemlin - try a different frequency maybe.