Sample Rate Conversion

[philbagg]

Hello all

Inspired by jedblue's interesting thread on 24 bit/16 bit, I'm interested to know
what the reasons are for recording at let's say 48kHz, or even 88.2kHz, mixing
and mastering at those SR's and then converting back down to 44.1kHz for the
final product. I presume it's for similar points as mentioned in the bit depth
thread, but I'm not entirely sure

I'm told it sounds better than something that started at 44.1kHz, but the
question I ask is.... why.... (how philosophical )

Cheers,
Phil

 

[LeeRosario]

Hello all

Inspired by jedblue's interesting thread on 24 bit/16 bit, I'm interested to know
what the reasons are for recording at let's say 48kHz, or even 88.2kHz, mixing
and mastering at those SR's and then converting back down to 44.1kHz for the
final product. I presume it's for similar points as mentioned in the bit depth
thread, but I'm not entirely sure

I'm told it sounds better than something that started at 44.1kHz, but the
question I ask is.... why.... (how philosophical )

Cheers,
Phil

Well it's been discussed at lengths, but I enjoy brushing up on anything audio so....


The orginal intent for sample rates actually goes back to something called the "Nyquist Theory". Believe it or not, sample rates where actually figured out long before digital existed. It was developed as a means to transmit electrical signals over a hard line (telephone cables).

The theory stated that to sample any audio source effectively, you must capture at least double it's highest frequency. Ergo, that's why 44.1 is such a reoccurring thing.

Someone figured that the highest frequency you need for most things in audio production shouldn't go past around 22,000 hertz. Not to mention at the highest end of our possible hearing range, which is 20khz *normally*.

You factor that maybe things like cymbals or any high frequency content would benefit most from higher sample rates.


So the original goal had nothing to do with digital processing period. Most people will argue that you can't tell the difference between 44.1 and 192khz. That it depends on the quality of other things. You get into A/D converters and anti aliasing...just ridiculous amounts of arguments. Also, the standard by which PCM wave files are created. The whole concept is actually kind of simple.

The common practice *scientifically* speaking is that you stay at your target sample rate as long as you can. Right up until the end.

 

[noisewreck]

In jedblue's post I linked to an earlier post that I had made where I was arguing that having a project run at higher sample-rates WHEN GENERATING SOUNDS INSIDE THE DAW, by the means of plugin/software synths and samplers can produce audible differences, depending on how well those synths and samplers were doing bandlimiting themselves. This is independent of the AD/DA conversion process.

However, when recording EXTERNAL SOUND SOURCES, there is far more to take into account in the AD conversion process than just sample-rates. There are way to many variables involved, not the least of which is the quality of the converter itself and how it handles higher sampling frequencies, as clock jitter for example becomes more of an issue at higher sample-rates that might adversely affect the integrity of the audio signal coming into the converter.

 

[philbagg]

Cheers for the responses guys.

I've read about the Nyquist theorem and sample rates and understand how
they work. I'm just curious if there's more accuracy/precision aspects of
recording at higher sample rates and then converting down at a final stage,
and how?

Cheers,
Phil.

 

[LeeRosario]

I'm just curious if there's more accuracy/precision aspects of
recording at higher sample rates and then converting down at a final stage,
and how?

Cheers,
Phil.

that's exactly where it gets dicey. Some people hear a difference, some people don't. In fact, some of the best engineers in the world where locked off in a room, doing all the double blind stuff, best listening environment...you name it...

Came back with a half/half. The one half felt a difference, the other half didn't. There was no general accuracy as to anyone who could honestly hear a difference past 50% of the time.

It was rumored that a huge knife fight/blood bath ensued right after in the parking lot.


In all seriousness, it's said that plug ins perform better at higher sample rates, at the cost of higher CPU loads. And still, I hate talking in neutral tongues because I am one that feels a difference. The anti-aliasing filter (which has a noticeable effect on what you hear) is pushed off to a much higher end of the spectrum in higher sample rates...but again, that's a function of the hardware you own.


That's honestly the state we are currently in. You're gonna get half and half responses on this. Half will tell you it matters, the other half will tell you it dosn't.

I do it because I can afford to, but it's not a hard rule and shouldn't be treated as such.

 

[SouthSIDE Glen]

I'm just curious if there's more accuracy/precision aspects of recording at higher sample rates and then converting down at a final stage.

Technically, the answer is "no", there is no more precision at audible frequencies based strictly upon sample rates higher than 44.1. Because what Nyquist actually shows is that a capture sample rate of double the frequency is all that's required for lossless (i.e. complete and 100% accurate) reproduction of a limited bandwidth analog signal.

What complicates that answer in the real world is that it is difficult to physically implement - design and build - converter circuitry that will be true to the math without injecting it's own artifacts or distortions. The upshot is that there are some converters that may sound better at a higher sample rate than 44.1, and at the same time other converters that may not. This is not because of the actual sample rate itself, but rather because of idiosyncrasies in the purity of operation of various converters. It's like having one auto engine that delivers maximum power at 3000 RPM and another whose power band is closer to 4500 RPM. So Joe Mixer may get better results at 44.1 than Bob Mixer gets at 88.2, or vice versa.

Either way, it's really an issue that's generally considered pretty far down the list of worries and priorities when talking about actual conversion. Any difference that sample rate will make will be of a smaller order of magnitude than most problems we have to deal with in the studio.

One exception to this is if you're going to be working with audio for video. A standard in pro video is to use 48k sample rates, so if you're working on audio for video that will wind up in 48k, if your choice is between that and 44.1km you might as well work in 48k from the get-go.

G.

 

[Bristol Posse]

I always enoy these beyond the range of human hearing nyquist discussions especially since most mics don't have a frequency response of above 20 - 30 khz so the whole recording the harmonics you can't hear at 96 or even 192 thing is kind of a moot point too because the mics can't "hear" them either and even $2k (each) monitors won't produce frequencies above 30khz, so even if the invisible/inaudible harmonics were recorded they wouldn't be reproduced but I'm sure we'll hear a lot about it in future replies.

 

[philbagg]

...it's said that plug ins perform better at higher sample rates, at the cost of higher CPU loads... That's honestly the state we are currently in. You're gonna get half and half responses on this. Half will tell you it matters, the other half will tell you it dosn't.

I do it because I can afford to, but it's not a hard rule and shouldn't be treated as such.

I'm gathering from this, that as an PT LE user myself (limited CPU), I should
go with 44.1kHz/24 bit to save some processing power? Given that as you
said, its at the cost of something that half of the best ears in the world can
barely hear themselves

However I've taken this point on board too;

One exception to this is if you're going to be working with audio for video. A standard in pro video is to use 48k sample rates, so if you're working on audio for video that will wind up in 48k, if your choice is between that and 44.1km you might as well work in 48k from the get-go.

44.1 for music, 48 for post? Or I have to walk 44.1km if I don't want to use
48k SR? Sorry for being so eagle-eyed

It was rumored that a huge knife fight/blood bath ensued right after in the parking lot.

Damn sound engineers up
to no good once again

Cheers,
Phil

 

[noisewreck]

44.1 for music, 48 for post? Or I have to walk 44.1km if I don't want to use
48k SR? Sorry for being so eagle-eyed

44.1kHz if the end result is an audio CD.
48kHz if the end result is a video DVD.
96kHz if the end result is DVD-A (audio-only DVD).

 

[SouthSIDE Glen]

Or I have to walk 44.1km if I don't want to use
48k SR?

"Everywhere is within walking distance if you have the time." - Steven Wright



G.

 

[pezking]

I always enoy these beyond the range of human hearing nyquist discussions especially since most mics don't have a frequency response of above 20 - 30 khz so the whole recording the harmonics you can't hear at 96 or even 192 thing is kind of a moot point too because the mics can't "hear" them either and even $2k (each) monitors won't produce frequencies above 30khz, so even if the invisible/inaudible harmonics were recorded they wouldn't be reproduced but I'm sure we'll hear a lot about it in future replies.

Well I think you're missing the point.

The idea of recording at higher sample rates is to give more "room" for the necessary low-pass anti-aliasing filter to operate. Steep filters (i.e. those with a greater dB per octave slope) affect the phase of the audio signal far more than a shallow filter. Furthermore, if you have a higher sample-rate, and therefore a higher nyquist frequency, you can place the start point of that filter outside of the audible range - further reducing the impact of the phase effects of the filter.

 

[noisewreck]

Why is it that I have a certain sense of Déjà vu?

https://homerecording.com/bbs/showthread.php?t=288096

 

[SouthSIDE Glen]

The idea of recording at higher sample rates is to give more "room" for the necessary low-pass anti-aliasing filter to operate.

Which is why 44.1k was selected as the standard sample rate instead of 40k, to give the filter some 2k of room above 20k to do its thang.

G.

 

[Bristol Posse]

Well I think you're missing the point.

The idea of recording at higher sample rates is to give more "room" for the necessary low-pass anti-aliasing filter to operate. Steep filters (i.e. those with a greater dB per octave slope) affect the phase of the audio signal far more than a shallow filter. Furthermore, if you have a higher sample-rate, and therefore a higher nyquist frequency, you can place the start point of that filter outside of the audible range - further reducing the impact of the phase effects of the filter.

I don't think I'm missing the point. I've seen about a million of thee threads all over the net and people invariable claim the higher rates record the harmonics you can't hear and will make your recordings more natural

I must admit I wasn't aware the developers had gotten so sloppy that you potentially need 400% over and above the audible spectrum (when you go to 192 anyway) in which to hide al the artifacts that are being generated by their plugs and processors

 

[LeeRosario]

I must admit I wasn't aware the developers had gotten so sloppy that you potentially need 400% over and above the audible spectrum (when you go to 192 anyway) in which to hide al the artifacts that are being generated by their plugs and processors

well I mean it's just expensive (but it's getting better) to make it right. Like the Digidesign 002 rack vs the MBox. Especially in the "Do it yourself" era we live in now. As a former electronics lab technician, I've seen first hand what happens when corners have to be cut. People want cheap and great and expect it to work like the industrial stuff. Something has to give. In fact there's a theory in product management that I use for alot of things:

Every product involves time, money, and quality. The rule is, you can only have any two of the three which will directly affect the remainder. So for example, if you want it done fast (time) and cheap (money), then quality is gonna suffer. Or if you say you want a good product (quality) for cheap (money), then time is affected because time costs money and most people don' work for free (time). Just a fun fact.

You can only cram so much into a budget box before it's too expensive to be worth anything. Of course, if you can afford something like a Radar Nyquist V, then you're all set

 

[pezking]

Which is why 44.1k was selected as the standard sample rate instead of 40k, to give the filter some 2k of room above 20k to do its thang.

G.

Precisely. I should have added that to what I said!

I must admit I wasn't aware the developers had gotten so sloppy that you potentially need 400% over and above the audible spectrum (when you go to 192 anyway) in which to hide al the artifacts that are being generated by their plugs and processors

Who said anything about being sloppy? We're talking about the theory of higher sample rates and anti-alias filters. Unless I misunderstand you?

Another point I missed out from my previous post was that higher sample rates allow creative use of noise-shaping. With a sample rate which allows higher than audible frequencies to be recorded, it becomes possible to shape the bulk of the noise into the super sonic range. This allows quantization error, and therefore quantization distortion to become near inaudible.