48kHz vs. 44.1kHz Sample Rate @ 24 Bits

[snow lizard]

"44.1 sample rate of CDs can still leave deficiencies in the higher frequencies approaching 20k, because of the filters needed to reject the content above the Nyquist limit from being audible and that these problems are gone above around 60k. "

This is what I was talking about in the difference between 44.1 and 48. The filtering is slightly different, and to say that this tiny difference is not noticeable is like saying that something with a top end response of 10k is no different than one with a response of 12.5k... That can make a BIG difference, but again: The gear used and the operator of that gear makes a much bigger difference. I have done stuff at 44.1 that sounds much better than certain projects I have heard that were done at 96k or 48k or 88.2 or whatever.

Anti aliasing filters that are nice sound better than crap filters, the rest is data... (gross oversimplification).

I agree, and there's no dispute to Bob's Mods's original claim. I'm just concerned that mixing in the box and doing an SRC might not work as well as leaving things at the target rate, which seems to be a popular idea around here.


sl

 

[mshilarious]

since Nyquist theory suggests any number of sine waves can generate/represent a complex signel, then one would summize more points to create more sine waves would result in more accuracy rather than less... I beleive Dan L's point in his article is that the ADC/DAC are constrained in their ability to convert to the desired word size and at around 60K (at the time of his article) you hit the limit of orderly data vs. conversion handling. with the presumption that 96K was probably as far as people should go.

No, he picked 60 because that provided more than ample room for the anti-aliasing filter to have no attenuation of at audible frequencies.

I'm gonna run a test of high frequencies at 44.1 vs 48, in and out of the same converter, and see if I can graph a difference. If I fail, that is not necessarily conclusive, could be quality of the converter's analog circuitry, resolution of the analysis tool, etc. But if it's there . . . well then. An important issue is whether the converter's anti-aliasing filter handles the two sample rates differently.

According to the spec sheets for my converters, it does, so we'll see. Although, based on the specs, it will be very hard to find the attenuation, because at 44.1 kHz, by 19.845 kHz, the filter has no effect. And these aren't new converters! (Crystal 4220 chip) But I'll be running D/A/D, which should double the attenuation.

Then maybe I'll try a 48 to 44.1 SRC vs. a 44.1 original. Stay tuned . . .

 

[mshilarious]

I agree, and there's no dispute to Bob's Mods's original claim.

I will dispute the claim, to the extent that there is a 9% increase in accuracy at audible frequencies. Ignoring the effects of anti-aliasing filters, there is no difference in audible accuracy at all. Disputing that requires a refutation of Nyquist, which I haven't seen here. So instead we should be looking at high frequencies and devising tests to capture any difference.

As for his listening test, I have no opinion, because I wasn't there, nor do I know his test methodology. As for the claims of a more tape-like sound, I don't know what that means. Tape is reknown for its natural compression effect, but I am not aware of any compression effect from higher digital sample rates.

I asked on another similar thread for a little 10 second wave file so I could hear the difference too, but in that case, the sample files were two different performances with a large difference in level. Maybe I'll create sample files from the demonstration program on my old Yamaha keyboard (familiar to all who entered the original Rumble)

 

[Farview]

To basically do an accurate conversion following this line of reasoning, the SRC would have to bring 48k up 160 times to bring it down to 44.1 evenly. Converters do exist that can handle this kind of SRC, but I'm not sure that my Delta 44 is up to this task. 88.2 to 44.1 does seem like less of a challenge.


sl

I'm talking about sample rate conversion, I don't think the Delta does on-the-fly SRC.

 

[Bob's Mods]

What I did was create a short diddy and record it at both 44.1 and 48, mixed them the same and listened to the playback of each A/B. If the filter is active in the 44.1 diddy, then that could account for increased sense of high end causeing the 48k diddy to appear warmer as the filter is further out on the frequency spectrum. There's only caps in those filters and if they are of the cheap variety, they will most certainly cause audio problems.

Bob the Mod Guy

 

[mshilarious]

There's only caps in those filters and if they are of the cheap variety, they will most certainly cause audio problems.

No, the filtration occurs after the A/D conversion at the master clock rate, which is something north of 2 mHz or even higher. The filtration is accomplished with a digital algorithm.

 

[Farview]

What I did was create a short diddy and record it at both 44.1 and 48, mixed them the same and listened to the playback of each A/B. If the filter is active in the 44.1 diddy, then that could account for increased sense of high end causeing the 48k diddy to appear warmer as the filter is further out on the frequency spectrum. There's only caps in those filters and if they are of the cheap variety, they will most certainly cause audio problems.

Bob the Mod Guy

It would make more sense if the filter was farther than 1 musical step up the spectrum. I'm not arguing what you are hearing, I just think that it has to be something else causing the difference.

 

[snow lizard]

I'm talking about sample rate conversion,

Right. The thread on Dan Lavry's forum that Mshilarious posted goes into depth about it. The concept that I'm wondering about is that if a conversion uses an interger ratio (like 2:1, available if you're doing a conversion from 88.2 to 44.1) versus a non-interger ratio (48 to 44.1 for example) how clean would the results be in comparison? There seems to be speculation that if you're not using a high grade conversion device for a non-interger conversion then you get phase distortions that tromp on the imaging. The math is much easier at interger rates, unless you're using something like a Weiss SRC which converts everything to interger rates. (48 to 44.1 becomes 160:147 - not all SRC's do this)

In the case of 88.2 to 44.1, the space in the time domain between samples is exactly half at the lower rate, so there is an analogy to simply throwing out every other sample. The ones that remain at 44.1 should have been in the same spot at 88.2 - you've got exactly half the samples. The ratio of 2:1 would be easier to deal with for something like a computer processor.

I don't think the Delta does on-the-fly SRC.

It's got a series of crystal oscillators that are synced to a master clock. The oscillators are divided by a number to generate the sample rate frequency. Usually the software will set the A/D to whatever rate you tell it to, but it's possible for the monitor section of the card to run at a different rate.


sl

 

[Bob's Mods]

What I was trying to learn is if there is anyone else that can hear a difference between 44.1 and 48?

As far as the conversion process itself is concerned (from 48 to 44.1), the math may be more complex but your processor is up to the task. Sure some rounding off may occur but it should be inaudible. When I downsample I've largely found the sonic character has been maintained. Its one thing to copy a tape recorded mix into a 44.1/16 bit format. Alot of vinyl has been put on CD and it sounds just as it did on vinyl, even better. When your mixing in box however, I believe it advantageous to have more more bits for more precise math. To me, its more than just a bandwidth issue, its that more bits provide for better math for mixing. Hence tracks should have better definition. Recording at 96k I got great ITB mixing results but 96k is to much of a pain. The downsample for 96 to 44.1 for the most part maintained the character of the original. I've used the free Voxengo convertor and Cool Edit Pro to downsample and they all worked fine. I believe the losses of downconversion may be overstated.
Its been proven that working in 24 bit is superior to 16 bit. The same is true for sample rate. More sample data points, 96k and 48k, provide for more precise math for those plugs and track mixes. At 48 its abit subtle but there.

Bob the Mod Guy

 

[Bob's Mods]

No, the filtration occurs after the A/D conversion at the master clock rate, which is something north of 2 mHz or even higher. The filtration is accomplished with a digital algorithm.

Thanks for pointing that out msh..., I did not realize that was the case. At 44.1 I think that roll off begins at 20k? Thats this much higher than I can hear. The acuraccy of that 20k data would be something in the order of 2.2 samples. Right on the edge of accuracy. At 96k that 20kHz tone is sampled 4.8x. This would provide for more accurate high frequency data which should translate to more precise math for math intensive operations like mixing and plugs. Digital mixing is about math, analog mixing is about summing voltages. Two very different worlds. I'm just not so sure I'm hearing any filter action at 44.1, I suspect its more that this standard rate is too close to the Nyquist value than anything else. Its probably fine for one track but when we factor in the math processes that take place in mixing, just meeting Nyquist may not be good enough any more.

 

[kq300325]

since Nyquist theory suggests any number of sine waves can generate/represent a complex signel, then one would summize more points to create more sine waves would result in more accuracy rather than less...

Isn't this Fourier Analysis rather than Nyquist?

 

[easychair]

No, the filtration occurs after the A/D conversion at the master clock rate, which is something north of 2 mHz or even higher. The filtration is accomplished with a digital algorithm.

My understanding is that filtering needs to happen before conversion, to prevent sampling of frequencies above the Nyquist limit, which causes aliasing in the converted signal. And that the distortion caused by aliasing, like digital clipping, cannot be removed afterwards. Thus the name "anti-aliasing filters".

 

[Bob's Mods]

My understanding is that filtering needs to happen before conversion, to prevent sampling of frequencies above the Nyquist limit, which causes aliasing in the converted signal. And that the distortion caused by aliasing, like digital clipping, cannot be removed afterwards. Thus the name "anti-aliasing filters".

I think this is probably true. It makes sense that you want to roll off any frequency effects that exceed the sample rate before the convertor to prevent aliasing problems. This is done in an analog world so caps would be present. These would cause phase shift and some hysterisis issues in the higher end of the spectrum. This stuff gets pushed up the frequency spectrum to an inaudible range as the sample range goes up. These effects could well be what I'm hearing in the 44.1 range making the mix sound slightly brighter. Just my guess here though.

Bob the Mod Guy

 

[mshilarious]

My understanding is that filtering needs to happen before conversion, to prevent sampling of frequencies above the Nyquist limit, which causes aliasing in the converted signal. And that the distortion caused by aliasing, like digital clipping, cannot be removed afterwards. Thus the name "anti-aliasing filters".

Yes, except the actual sample rate is in the mHz range, so the analog filter to eliminate that will have no effect on audio. The decimation process in the converter downsamples that rate to the data rate--again filtration is necessary to avoid aliasing, but this time is done digitally.

 

[mshilarious]

What I was trying to learn is if there is anyone else that can hear a difference between 44.1 and 48?

I'm trying to learn that too! It is theoretically possible, and I'm planning on doing my tests tonight.

 

[Blue Bear Sound]

All the in-depth technical analysis and speculation is all fine and good, but the bottom-line is that if someone can't get good sound at 24/44.1, then their problem has absolutely nothing to do with the sample rate or digital recording in general!

 

[gullfo]

I don't think the point is so much that the A/D/A's are constrained at 60k, and he's not mentioning word size.

part of the issue is the ADC - how large a word can you reliably create at a given sample rate.... I think the concern was that at 192K you begin to have errors in the conversion and so larger word sizes don't necessarily help - your better off with a lower sample rate and word size if they're more accurately representing the audio.

 

[Bob's Mods]

All the in-depth technical analysis and speculation is all fine and good, but the bottom-line is that if someone can't get good sound at 24/44.1, then their problem has absolutely nothing to do with the sample rate or digital recording in general!

I believe I'm getting decent results at 24/44.1. I've also got choices of 48 and 96 up the food chain and I'm curious as to what that will get me. I found 96 to be a more of pain but did give good mix definition. 48 seems to help with the mixing abit without the issues of tracking at 96.

 

[Bob's Mods]

Follow Up

As a follow up on this issue I raised, I've recently did another demo at 48kHz and found it sounded exactly like the demo in 44.1kHz. I suspect my earlier attempt may have been skewed slightly by small changes between the two earlier demos recorded at the differing sample rates.

It is very easy to see the upper end roll off in both the 44.1 and 48kHz demos using Voxengo's Span. Is seems the upper end roll off for both sample rates begins at approx 17.5 kHz! This is no difference in roll off between the two at all. It seems the 48kHz sample buys you a bit more in band accuracy but that just doesn't seem to translate into any meaningful sonic improvement.

Now when I view a tune I recorded at 96kHz, there was no roll off whatsoever. None. There was this nice slope with data all the way out to 30kHz (the upper limit of Span). So clearly, there is harmonic stuff going on in the upper (inaudible range) thats being chopped out at 44.1 and 48kHz.

Does anyone know, in those tape based studios, when those began to roll off the high end?

 

[mshilarious]

As a follow up on this issue I raised, I've recently did another demo at 48kHz and found it sounded exactly like the demo in 44.1kHz. I suspect my earlier attempt may have been skewed slightly by small changes between the two earlier demos recorded at the differing sample rates.

It is very easy to see the upper end roll off in both the 44.1 and 48kHz demos using Voxengo's Span. Is seems the upper end roll off for both sample rates begins at approx 17.5 kHz! This is no difference in roll off between the two at all. It seems the 48kHz sample buys you a bit more in band accuracy but that just doesn't seem to translate into any meaningful sonic improvement.

Thanks for the bump, I'm still planning on testing my converters too.

Your result is inconclusive without knowing if the converter's filter behavior is different between the two rates. See if you can find the spec sheet for the converter chip.

Mine is supposed to change the filter rolloff frequency, so we'll see.