what sample rate do you record at?

[SouthSIDE Glen]

At best, a DAC attempts to reduce stairstepping by running at a faster rate and stepping a fraction of the distance several times,

THERE IS NO SUCH THING AS "STAIRSTEPPING". It's just not done that way. Any article that tries explaining analog reconstruction in terms of stairstepping is written by someone who just has no clue as to what the Nyquist theorem is actually all about.

If one tried reconstructing analog voltages from digital data via that method, even a 96kHz sample rate would be woefully inadequate to reproduce the majority of the audio spectrum. Here's how that would breakdown:

At a 44kHz sample rate, here's the number of samples taken for a complete wave cycle at given audio frequencies

20kHz - 2
10kHz - 4
5kHz - 8
2kHz - 22
1kHz - 44
400Hz - 110
200Hz - 220
60Hz - 733
20Hz - 2200

If done that way, even with several times oversampling, the CD standard of 44.1kHz would be woefully inadequate for re-creating anything other than the bass frequencies of the audio spectrum.

DACs do NOT connect the dots or smooth out staircases created by intrinsic digital values. Instead, those individual values are converted to analog values, and those values are then run through a cascade of bandpass filters to seperate them into their constituant analog wavelets. Then they are run through some Nyquist-related mathematical functions that reconstruct the original analog waveform. Using this proven method, even only two samples can be reconstructed into a full wavecycle at 20kHz, 4 samples at 10kHz, etc.

Read that Lavry article that Farview referenced a couple of posts ago, and pay attention to the medthods used for creating the waveforms - namely the summing of constituant wavelets - to get a flavor for how Nyquist and DAC actually works.

As far as all those different converter types mentioned in WikiWorld, I believe those are referring to the conversion of the original digital value into the raw analog voltage *before* Nyquist waveform reconstruction takes place.

G.

 

[Farview]

In theory, yes, you can reconstruct the original waveform that way. In practice, however, DACs don't work that way. At best, a DAC attempts to reduce stairstepping by running at a faster rate and stepping a fraction of the distance several times, but you can't even guarantee that. A more typical DAC will jump at the sample rate from one value to the next, then will smooth off the resulting 44.1 kHz (or whatever) stairstep with a low pass filter. There's no trigonometric anything going on in a DAC.

http://en.wikipedia.org/wiki/Digital-to-analog_converter

That is the biggest problem with Wikipedia, anyone can change it.

There is no stairstepping. There never was.

Read Dan Lavry's article. Dan Lavry designs some of the finest converters in the world, he is intimatley familiar with nyquist, reconstruction, etc... Wiki is just plain wrong.

 

[pipelineaudio]

I agree theoretically that if your converter sounds "better" at a higher rate than 44.1khz, your converters' filters/filter math, is not right

That said I dont personally own multichannel converters that sound or test as well at 44.1khz as they do at 48khz.

My friend just scored a 2 channel Lavry off ebay, so I'll test those but so far for me, TASCAM DA-x8's, RME ADI DS's, Apogee AD8k's, Behringer AD8k's, and Presonus Firepods all test better at 48khz than 44.1khz.

Looking on a scope, it doesnt even seem like the frequency extension up top is even an issue, by the time they differ its WAY up in useless land...the problem I see is the foldback from a lot of aliasing at 44.1khz

I'll stick with 48k 24bit for now...and since my mastering engineer seems to use analog stuff, sample rate conversion is ZERO problem

 

[dgatwood]

DACs do NOT connect the dots or smooth out staircases created by intrinsic digital values. Instead, those individual values are converted to analog values, and those values are then run through a cascade of bandpass filters to seperate them into their constituant analog wavelets. Then they are run through some Nyquist-related mathematical functions that reconstruct the original analog waveform. Using this proven method, even only two samples can be reconstructed into a full wavecycle at 20kHz, 4 samples at 10kHz, etc.

Read that Lavry article that Farview referenced a couple of posts ago, and pay attention to the medthods used for creating the waveforms - namely the summing of constituant wavelets - to get a flavor for how Nyquist and DAC actually works.

As far as all those different converter types mentioned in WikiWorld, I believe those are referring to the conversion of the original digital value into the raw analog voltage *before* Nyquist waveform reconstruction takes place.

There are several different types of DACs, and Wikipedia describes the more common varieties. I can't imagine what wavelets could possibly have to do with DA conversion. Wavelets are a means of decomposing an already digital signal into multiple digital signals for the purpose of compression or spectral analysis.

BTW, Lavry appears to agree with me that 48 kHz is less than ideal. From page 27 of that article: "Sampling audio signals at 192KHz is about 3 times faster than the optimal rate." Notice that number was "3", not "4". That means that he thinks the ideal sampling rate is about 64 kHz. He and I are in complete agreement in that regard.

Also, I don't see anything in that article about DA design. It's all about AD design. The word "wavelet" does not appear anywhere in the article. Are you sure you're talking about the same article?

 

[SouthSIDE Glen]

There are several different types of DACs, and Wikipedia describes the more common varieties. I can't imagine what wavelets could possibly have to do with DA conversion. Wavelets are a means of decomposing an already digital signal into multiple digital signals for the purpose of compression or spectral analysis.

I believe what Wikiwackywoowoo is talking mostly about is the conversion of a digital value to a discrete analog value, not the reconstruction of a complete wave from a series of samples. In other words, how it decides that 11010011011100001010 equals a specific analog voltage. As I understand it, it's in the type and accuracy of that conversion stage where converters differ most. But it's the reconstruction of a continuous analog wave from those discrete sample values where information theory and the Nyquist-related theories and equations (such as the ones described in the Lavery paper) take over, and where - as I undertand it - all converters do more or less the same thing.

And once again, I ask you, how does one use a simple filter to smooth out the stairsteps in the 10-20kHz range when at a sample rate of 44kHz there's only somewhere between 2 and 4 samples taken per full wavecycle? It's virtually mathematically impossible. Not to mention a complete disregard of the whole point of the Nyquist theorem to begin with.

That means that he thinks the ideal sampling rate is about 64 kHz. He and I are in complete agreement in that regard.

Probably more like 66.15kHz, which would conviently be halfway between 44.1 and 88.2, but that's splitting hairs; you're right, he is on record as advocating the mid-60s as what he referrs to as the "sweet spot."

And yeah, I more or less agree with that as well. The idea there is that one is moving potential ailiasing problems out of the way as well as covering the controversial area in the 20+kHz range that some argue add "air" to a quality recording.

(Of course if this were to help at all, it would be most likely on a Telarc recording of Joseph Silverstein and the Boston Philharmonic; applying that to an indie release of stacked distorted wankers most likely wouldn't make an iota of difference. )

The problem is 66kHz is not an option offered by the manufacturers, and by the time you get up to 88 or 96kHz one is not only using up unnecessary resources, but - as Lavery and other Big Boy engineers discusses in detail in other papers - one is actually getting a diminishment of returns because of other info theory conditions that start to take effect at that speed.

And to avoid misundertsanding, this is all talking strictly about sample rate theory itself, and not inconsistancies in physical converter design where some makes/models sound better at some speeds and others sound better at others...not because of the sample rate itself, but because of differences in quality of design.

Also, I don't see anything in that article about DA design. It's all about AD design. The word "wavelet" does not appear anywhere in the article. Are you sure you're talking about the same article?

Maybe we're not . Because as I read it, practically the entire article is talking about the reconstruction of an analog waveform from the digital values using sophisticated sin integrations to build the full wave from it's component wavelets. It's easy to get lost in all the math, but when you get to the English language descriptions of the effect of those equaitons, along with the accompanying graphs and charts (did you look at it in PDF and not just a Google HTML translation?), he's clearly discussing building analog waves from discrete samples.

I think (I hope) we can agree that this is a very difficult subject. Not the least of which reasons is because it deals in wholely in information theory, which is just a half step away from quantum theory in lack of intuitiveness. (There's a good reason for that; the roots of much information theory are buried right in the middle of and strongly related to quantum theory.)

I admit I'm pushing the envelope of my knowledge when I get into DA and Nyquist - there will be some order of operation details and definitely some math which I could easily get wrong or misunderstand. I don't claim to follow all of Lavry's (or Nyquist's) equations in comprensive detail, for example, but I have a pretty good grasp on the overlying concepts.

One of the main concepts is that "stairstepping" is a nice graphical way to explain the basics of digital sampling of an analog signal, but it is very misleading or at least incomplete - to the point of downright wrong in some aspects - as a way to describe the conversion of digital back into analog. The electronic math used - regardless of converter design type - is not meant to connect the dots or smooth the slope between discreet analog values; not in the way inferred by such starstep graphics. There just simply are not enough dots to connect with any accuracy whatsoever, even at frequencies below 10kHz.

Rather, the conversion is based in the heavy-duty and hard to comprehend math of information theory as discovered by Nyquist and his contemporaries; where the analog values are desconstructed into their fundamental frequency components, which in turn are then used to "organically" (for lack of a better word) re-build the whole analog waveform from it's parts.

The magic of this, and one of the great results of Nyquist's discovery, is the economy of this method. It does indeed allow one to rebuild an entire wave form in it's entirity from only two discrete samples. It doesn't seem possible, but it's proven with every CD we put in our player.

G.

 

[dgatwood]

One of the main concepts is that "stairstepping" is a nice graphical way to explain the basics of digital sampling of an analog signal, but it is very misleading or at least incomplete - to the point of downright wrong in some aspects - as a way to describe the conversion of digital back into analog. The electronic math used - regardless of converter design type - is not meant to connect the dots or smooth the slope between discreet analog values; not in the way inferred by such starstep graphics. There just simply are not enough dots to connect with any accuracy whatsoever, even at frequencies below 10kHz.

Sure, there are. That's the crux of what Nyquist guarantees: that any artifacting produced by sampling will necessarily be at a characteristic frequency above the Nyquist point. I'll explain.

If you have four points representing a wave (10 kHz signal at 40-ish kHz sampling rate) and you generate a stair-stepped signal with a 48 kHz sampling rate, the step size is producing a square wave approximation of the sound, but the sharp points of that square wave are at a frequency of precisely 48 kHz, which means they're well outside the human hearing range.

Thus, while you may not be able to reproduce a perfect sine wave at 10 kHz, it is not necessary to do so because any artifacts caused by the lower sampling rate are just adding overtones that are similarly outside the human hearing range. You can replace that sine wave with even a raw sawtooth wave at 10 kHz and most people can't hear a difference because the lowest frequency harmonic added is at 20 kHz.

Since the signal produced by a simple DAC as described in Wikipedia would more closely resemble a square wave, and since you only have odd harmonics, the lowest possible harmonic is the next odd harmonic, which is three times the frequency (first harmonic is the fundamental, second is twice the frequency/an octave up, next harmonic is three times the frequency/an octave and a fifth up), or about 30 kHz, well outside the human hearing range.

As mentioned earlier, however, because it is just an approximation of a square wave, rather than precisely being a square wave, the frequency of the distortion is actually at 48 kHz.

As a result of the frequency of the distortion being outside the range of human hearing, to get as accurate a reproduction as the human ear can perceive, it is necessary only to add a low-pass brickwall filter at 22 kHz or so---maybe at 24 kHz just to avoid having to build such a steep analog filter.

In fact, if all you care about is human hearing, you don't technically even have to have a low-pass filter. The stairsteps will be well outside of the range of human hearing, as mentioned earlier. That said, everyone always puts one in anyway to avoid the risk of blowing out tweeters with a high intensity signal at 48 kHz.

And indeed, that's all a typical reconstruction filter is. It is a simple low pass filter set just above the range of human hearing. When people say that there is no stairstepping in the output of a DAC, it is not because they're doing some complex math on the output. The output is already analog, so there's not any math or manipulation that can practically be done. All that can be done once you're in the analog domain is to remove the frequencies outside the human hearing range that cause the stairstepping.

Even at half the sample rate, you are getting two points per cycle, and because each point holds at that value until the start of the next sample, you never get a sawtooth wave. It is always a square wave. Thus, even if you generate a 24 kHz test tone at 48 kHz sampling rate, you could reproduce it so accurately that even a dog couldn't hear that it wasn't a sine wave. You would have one sample high and one sample low, so it would be a perect 24 kHz square wave, the first harmonic of which is at 4 * 24 kHz = 96 kHz.

Of course, that's with only two samples per cycle. With a 10 kHz tone, you have almost five whole samples. That's huge by comparison!

BTW, I may well be wrong about the potential for amplitude loss near half the sampling frequency during capture. Oversampling may completely compensate for that. It would depend largely on how the ADC does its oversampling as to whether it uses a sum, a weighted average, some sort of max/min finding algorithm, etc. Any of those techniques will result in some type of error, but the error will be in different domains.

When reducing the oversampled signal to the desired sampling rate, a max/min finding algorithm would result in equal amplitude but potentially subtle errors in the frequency and phase distortion. I think it is likely that this would not be audible at the frequencies we're talking about here, but I wouldn't swear to it.

By contrast, a weighted average or unweighted average would probably be the most audibly pleasing if that phase distortion turns out to be audible. However, it would result in HF roll-off as you approach the sampling rate.

I'm not sure which technique modern oversampling ADCs use in practice.

 

[antichef]

I love how this started out as a lightweight casual discussion dominated by apparent newbs like me, and quickly degenerated into a well considered and sophisticated dialog, with enough great external references to keep me busy for a week. I'll be much better for reading through a third or fourth time, and carefully considering the cited material. Thanks for all this - keeps me coming back to this board.

I use 44.1k because Garage Band forces me to.

 

[danny.guitar]

Anyone using a sample rate less than 384KHz just doesn't know what they're doing...

 

[SouthSIDE Glen]

Sidenote comment -

This is the first time I've ever seen results like this - Don't know if it's a "home-rec" thing or what.

Every single poll I've seen of full-time industry professionals in recent years puts around 70-80% of them squarely at the target rate (44.1kHz).

In this poll, almost half (more than half? Wasn't paying *that* much attention) record higher. Never saw that coming...

And the *third* of them recording at 48kHz?!? What's that all about?

I think that's a result of the design of the poll as much as it is anything else, John. For exmple, I couldn't even vote because none of the options applied to me; I record at 44.1kHz, but not for any of the options given. I suspect that applies to most people who have been at this for a while.

Most of the options given don't make sense, and many valid options (e.g. "It depends upon the converter I'm using", "I record at 48k for video post", "I use 44.1k because that's more than good enough for 20-20k rock n' roll", "Other", etc.)

Polls on this board simply have to be taken with a large grain of salt for that very reason; they are written in ways that unintentionally skew the results. At the very least, every poll should have at least "Other" or "None of the above" as an option in order to allow full voting.

G.

 

[danny.guitar]

Digital tubes

 

[TerraMortim]

I always track in 24bit 96khz
Mix in 24bit 96khz
Master in 24bit 96khz
Enjoy a much better sounding CD in 16bit 44.1khz and realize it was worth all the trouble and extra money to build a machine actually capable of doing 24/96

yup... most people can't hear the difference...some people can. It's very worth the CPU power and dealing with know it all dickwads who want to tell us what we hear doesn't exist. We are crazy, maybe?

 

[TerraMortim]

I always track in 24bit 96khz
Mix in 24bit 96khz
Master in 24bit 96khz
Enjoy a much better sounding CD in 16bit 44.1khz and realize it was worth all the trouble and extra money to build a machine actually capable of doing 24/96

oh yeah, thanks a shitload for the quotes! Those will be helpful when dealing with endless hordes of former analogue engineers who tell me I'm insane for recording/mixing and mastering at 24 96

 

[TerraMortim]

Look at the poll again and add up the 44.1 people. 44.1k seems to win.

homerecording.com - - as much as I think your work is great, and you obviously know how to record a fucking song (brothers of true metal....are proud and standing tall! bahahaha manowar...you are god for recording them hahaha..seriously..what did you think about their latest wacken open air fest dvd with the full orchestra and choir from eastern europe...wow), I rest my case. That's like when people who try to use the "most americans believe 911 was a conspiracy (even though it was)...come on folks..that doesn't prove anything. please don't use that tactic.. nooo! *slaps forhead*

 

[TerraMortim]

So the verdict is some go for 44.1/48, some go for 96. Therefore i am going to do half my stuff in 48 and half in 96.

Seriously though, would it be better to look at it subjectively? Depending on the power of your machine of course, if you have a tune that doesn't use too many tracks or VST's then perhaps 96 is in order. If you are working on something fairly intensive then perhaps 48 is in order. Of course that doesn't work to well in practice. I start tunes with a small amount of tracks that soon turns into a large amount, by which point its too late. Been using 96 but thinking of going down to 48 just for the extra performance boost. But when I look at the graphs etc, it makes me want to stay at 96.

So, I'm going to build another music PC. I will use them both to record at the same time. One at 96, one at 48. That way I don't have the dilemma.

yep! It doesn't matter that much, most people don't hear the difference, and to those who do, the difference drives us nuts... but really, no one will care when they listen to the song whether it was recorded on the sexiest gear or some portadat with a $20 radioshack mic in an airplane hanger. The song is king, and that will never change, but after all us audio nerds need something to validate ourselves, don't we?

 

[TerraMortim]

Probably because someone is spreading the BS that higher sample rates are superior . Full time guys have been doing it enough and have commited enough mistakes to know otherwise. There are quite a few people that could be or ARE professionals here, but there are also a lot of people that use cracked software and ask questions like "what mic for best quality for rap" ?
so is it really surprising? It falls in the hands of the Learn-ed ones to educate the newer folks and offset the loads of BS that come through.

I do a LOT of DVD-Audio Classical/World Music/Persian/Acoustic productions and sometimes the people that pay me want high res for archiving and possible conversion to DSD. The Market I cater to is Picky classical listeners and snooty Audiophiles, so things are a little different. MY standard operating procedure is 24/44.1.

The thing is dude,

the arguement that if you're a successful producer or engineer you must have golden ears is the real BS. There are some world class guys that don't have as hearing as beginners and vice versa, that has more to do with medical reasons than your success in your job. Anyway, it's actually not that important, as good music is good music...do what works for you and what makes it work in your favor. I can head the difference, and so I prefer to record, mix and master at 24, 96... some people don't and it doesn't matter really. It's all art we're recording and the highest fi only matters to us audio geeks and 35 year old virgins who spend $20,000 on home stereos. We are the minority, and our customers are the majority. In short do what you feel like and just make some good music . . . .

 

[TerraMortim]

I think that's a result of the design of the poll as much as it is anything else, John. For exmple, I couldn't even vote because none of the options applied to me; I record at 44.1kHz, but not for any of the options given. I suspect that applies to most people who have been at this for a while.

Most of the options given don't make sense, and many valid options (e.g. "It depends upon the converter I'm using", "I record at 48k for video post", "I use 44.1k because that's more than good enough for 20-20k rock n' roll", "Other", etc.)

Polls on this board simply have to be taken with a large grain of salt for that very reason; they are written in ways that unintentionally skew the results. At the very least, every poll should have at least "Other" or "None of the above" as an option in order to allow full voting.

G.

Why are you sometimes so well thought out and rational and sometimes so far from it? Anyway...it is nice to see you thinking beyond the usual standards. Totally man. Do think outside the box, I like seeing that!

 

[TerraMortim]

I love how this started out as a lightweight casual discussion dominated by apparent newbs like me, and quickly degenerated into a well considered and sophisticated dialog, with enough great external references to keep me busy for a week. I'll be much better for reading through a third or fourth time, and carefully considering the cited material. Thanks for all this - keeps me coming back to this board.

I use 44.1k because Garage Band forces me to.

We are so important, and our craft is what everyone cares about! Nobody listens to music for any reason other than what we do to record it hehehe. I wish I could destroy the entire music recording industry in a second. People get so heated over which ultimately doesn't matter to anyone but us audio obsessed folks. Some amazing producers and engineers use 44.1 16bit, and some people hear the difference use more... with me, it just works for my ears, because there is a difference to me and the way I like to mutilate audio into "mixes". For The next guy, maybe it's not important...just the same as some people like film in the visual world, and some people love digital... it's all subjective, and only important to us in our little audio club, just how you don't watch a movie and say holy fuck I don't want to see that piece of crap, they used digital instead of film to record the visuals...nope, the story is king...and a great movie stands as a great movie, even if it was recorded with an old 8mm camera bought at a thrift store, and the sound was scratched with a tattoo gun into the inside of a sewage treatment plant holding tank by a schizophrenic elephant on tranquilzers. I'm part of all this, but I DO realize how silly it all is when you really think about it.

 

[SouthSIDE Glen]

Why are you sometimes so well thought out and rational and sometimes so far from it?

Well, that's a funny thing about people; they tend to see one as "well thought out and rational" when they agree with one's position, and "far from it" when they disagree with it. .

G.

 

[superspit]

Well, that's a funny thing about people; they tend to see one as "well thought out and rational" when they agree with one's position, and "far from it" when they disagree with it. .

G.

hehehehe!.....this is good!!
I want more!!....more.....more!!

 

[Farview]

homerecording.com - - as much as I think your work is great, and you obviously know how to record a fucking song (brothers of true metal....are proud and standing tall! bahahaha manowar...you are god for recording them hahaha..seriously..what did you think about their latest wacken open air fest dvd with the full orchestra and choir from eastern europe...wow), I rest my case. That's like when people who try to use the "most americans believe 911 was a conspiracy (even though it was)...come on folks..that doesn't prove anything. please don't use that tactic.. nooo! *slaps forhead*

that was a response to someone pointing out that it looked like the poll was against 44.1k. I was pointing out that you have to add all the different 44.1k questions to get the real percentage. It wasn't an endorsement, it was an observation.

The Earthshaker festival was cool. I just got done mixing David Shankle's new album. He (his band, DSG) will be playing the Magic Circle Festival on July 7 in Germany. That band is going to kill.