I think you should elucidate a little on exactly what you understand to be the correlation between bits and sample rate and "resolution." The fact that you used that R word has me a bit concerned as I find it undescriptive and generally indicative of a lack of understanding of the actual correlations.

If you tell us what you know we can probably fill in enough gaps that you can discover the answers for yourself?

Nika

If you are going to record a project to output only for CD - only use 44.1k. If for DVD, it is ok to use 48k if that is your final output.
Only use Sample rates that will be exact when dividing by 2 for example if your output is 44.1k & you have the option to do 88.2 - you're good - other wise all of your high end frequencies will get chopped off and you will lose quality in the long run.... it's too much to type on explaining that. :?

That is the rule of thumb for Sampling rates.

For bit depth, use as high as you can get. There is a huge difference between 24 to 16 bit.
If you don't have a clean signal recording in, it won't really matter what bit rate you use cause 'bad sound in, bad sound out'. However, if you have a very clean input signal, the sound quality is an immense difference.

Also, if you are a cubase user, you get the option of '32 bit' recording. Now most people scoff at this because they have no idea what it is. But when you bench mark test vocals or instruments its amazing.

32 bit recording is not actually 'analog to digital' conversion but a software 'analog compression' if you will. It 'will not' clip when recording. the only way you can clip is by doing so 'before' you get into cubase. All sounds are very meaty and full when recording in 32 bit mode.

At the end of the day, just use the sample rate your going to mixdown to (44.1k for CD, or higher if for DVD) and use the highest bit rate possible if hard disk space is not an issue.
200gb hard drives are around $100 on pricewatch these days and you can get external cases very inexpensively - its much easier to just use a hard drive for backup files instead of DVD data or tape drive backups too.

Hope that helps a bit.

gn wrote: If you are going to record a project to output only for CD - only use 44.1k.

I'm not so sure.

Why?

1. Can't go against science (Nyquist/Shannon/others/math). I guess we agree. The rules, same for mathematician and for converter builders.

vs

2. There are many reports (from those who have knowledge, experience and best toys) for better sound on higher SR (48/96/192/384 kHz) even after SRC to 44.1 for CD.

Possible reasons (some of):

a) Converters now have better clocks, filters, electronic design, parts, algorithms…whatever

b) Some filters working better in 96 and 192 mode than for 44.1 or 48

c) Frequencies above 20 kHz (we can't hear) in some way affect frequencies below 20kHz (we can hear)

d) Theory and real world could be different. There always exist some gap which science can’t explain. Jet.

e) Subjective errors

f) Marketing tricks, fashion, in

I was digging around and found something very interesting. Big guys are talking about the same problem. Great discussion.

(dead link removed)

"The Old Yellow Board"

Topic: “Recording at 192 is a waste”

If they're wrong, I'm wrong too. Don't shoot me.

If you are recording for a project ending up on cd there is NO reason whatsoever to choose another sample rate then 44.1

The resampling process will mess up your sound.

All of the information above the nyquist freq. (half the sample rate) will be lost. ALL of it. I can't stress this enough.

I’ve asked this question to a real expert on the topic, one of the most famous guys in the UK who does all the mix to everybody there, he said that raising the bit rate results always in a better sounding material at the end. For instance: 24 bits is better than 16 bits, and 32 better than 24 bits. But, that sample rate is just a commercial issue to sell soundcards cos, despite the fact that it’s true that you get a more compact wave file by the end, it is not worth the trouble cos the difference is imperceptible.

So he said: To record at 96kHz it’s not worth the trouble, it’s better to record always at 44.1kHz, even if you go 32 bits recording.

Honestly, I would just pay attention to what this guy has said, cos it’s a 50 years old man that has being in business since 15 years old and knows all about analog and all about digital, it’s a real genius.

Hmm, I'm a little confused on this.

Wouldn't it make sense to always go to the highest recording rate you can use? 44.1kHz means 44,100 steps a second.

So wouldn't going to 88.2 mean you're tracking twice the amount of information (ie smaller incrimental steps will pick up greater information)?

I'm a total noob to this stuff, so if logic doesn't apply, please have me shot. But this is making me think about digital photography: Even if you know you only want a 640x480 image, you're still going to want to capture it at the highest resolution your camera can do, right? And then resample down....

Or is the issue that resampling will do more harm than good?

Dave.

Recording and digital photography are a bit different.

I would say that 88.2 is the only thing to use if you are going to use a higher sample rate. but yes, anything else, even the 88.2 is going to do more harm than good.

As stated, even if you use 88.2, its not going to give nearly as noticable a quality as the bit rate will do.

Easiest solution - use higher bit recording and save the time and energy for creativity and not sample rate conversions.

dczoner wrote: I'm a total noob to this stuff, so if logic doesn't apply, please have me shot.

David,

Logic does apply, but you need all of the parts of the equation before you can apply the logic.

I have a piece of graph paper in front of me and I am drawing a circle on the paper somewhere. Let's say I give you three coordinates of this circle and I tell you that each of these three points is on the circle somehow. You create a circle that passes through those three points on your own graph paper. How accurately will the circle on your graph paper look like the one I have on my paper?

Let's apply logic now. Will more coordinates help make your circle more accurate?

Sample rates and the Nyquist theorem, as counterintuitive as it may seem, require the same logic.

Nika

Nika wrote:
I have a piece of graph paper in front of me and I am drawing a circle on the paper somewhere. Let's say I give you three coordinates of this circle and I tell you that each of these three points is on the circle somehow. You create a circle that passes through those three points on your own graph paper. How accurately will the circle on your graph paper look like the one I have on my paper?

Let's apply logic now. Will more coordinates help make your circle more accurate?

Sample rates and the Nyquist theorem, as counterintuitive as it may seem, require the same logic.

Nika

Three points, as long as they're not on the same line, uniquely define a circle. The intersection of the three perpendicular bisectors of the sides of that triangle from each opposite vertex intersect at the center of the circle.

Can someone give a more detailed mathematical answer as to why sampling at 96KHz will give you a lower-quality product when it is eventually dumped down to 44.1 for a CD than either 44.1 or 88.2?

I don't see the point in increasing the word length from 16 to 24 bit and increasing the dynamic range when it's just going to be squashed in mastering anyway.

The results from sample rate conversions are not as obvious as the sound difference from bit length.

The importance of higher bit length is this - all individual tracks are going to be at a higher quality resolution. If you have a song with 24 to 32 to 64 tracks or higher and each track is at 24 bits, then your entire project's instruments are going to sound much clearer and crisp. So when you mix down - you'll send a 24 bit 44.1k file to the mastering plant - then they use their high quality converters and dither down to 16 bit.

The result? all of your individual tracks have higher quality resolution & depth when listening to the mix than they would if they were all in 16 bit. If I'm not mistaking, bit rate also increases headroom giving more signal to add to each track (32 bit floating cannot clip)

You'll find 16 bit depth and clarity is not as clear as 24 & 32 bit resolutions - the difference is very obvious when you have many tracks together.

the only way to do this is to get in your studio and bench mark it for yourself - this is the only thing that will give you the answers for your particular needs.

experiment w/ bit rates 1st - they are easier to physically hear a difference in any environment.
with sample rates, as stated before - the results can be so miniscule that its not worth all the sample rate conversions, but you'll need to listen on a very high end set of monitors.

Also, mastering does not squash bits, it squashes sound. Also, its not mastering that squashes but compressors & limiters...
Bits are what help with the headroom to have more material to work with....

charles77 wrote: Can someone give a more detailed mathematical answer as to why sampling at 96KHz will give you a lower-quality product when it is eventually dumped down to 44.1 for a CD than either 44.1 or 88.2?

I don't see the point in increasing the word length from 16 to 24 bit and increasing the dynamic range when it's just going to be squashed in mastering anyway.

Hi charles,
As far as I know, the theory for sampling frequencies is the math during converting. if you record at 44.1, nothing needs to be screwed with, therefore no information is lost or summed. if you record at 88.2, every 2 samples need to be summed into one sample, which is relatively simple math. however if you record at 96, every 2.17687.... samples need to be summed into 1 sample. but since, as far as i know, a conversion algorithm can't take 0.17687.... of a sample, it simply sums two samples here and 3 samples there to have the 44.1KHz end result. don't quote me on this I'm just pulling it out of my poop-chute. but if this is true, it would really defeat the whole purpose of recording at higher sample rates... I also believe that this "better" sound people hear when recording at higher sample rates and converting down is subjective and should be called "different" because they may just be hearing the artifacts from the converting process (because of the nyquest theorem they can't possibly be hearing those higher harmonics that "affect the lower frequencies")

another thing to take into account is that unless you are recording with microphones, preamps, etc. that are capable of reproducing frequencies above 20KHz, why bother? as far as I know, there aren't many companies who even consider higher frequencies than 20KHz when designing their products.

also with bit depth, the main reason for recording at 24 bits is that every 6dB loss in signal means you are using 1 less bit. effectively raising the ugly digital noise floor. therefore, when tracking instruments at 24 bits it gives you a full 48dB more dynamic range. thats 48dB higher above the noise floor, so you don't need to worry as much about "slamming the daw" with signal and risking clipping. you can give yourself alot of room to play and still be able to bring things up without raising the noise floor substantially.

there is probably alot in this post that scientists would scoff at, but it makes sense to me 8-)

Now,
this is a subject that has been discussed quite a lot on the forums. And as always, there are a lot of misunderstandings and outright incorrect things said. One example might be from above:

>>Only use Sample rates that will be exact when dividing by 2 for example if your output is 44.1k & you have the option to do 88.2 - you're good - other wise all of your high end frequencies will get chopped off and you will lose quality in the long run

That statement is outright wrong which is proven by extensive experience.

You might try to attack the problem from a theoretical point of view, making complicated mathematical formulas and so on. In my mind, avoid that totally. One reason is that most formulas are gross simplifications when you compare them to the real world. (PLEASE! Theory is an important tool, but when theory meets reality, you often have to improve the theory).

Instead attack the questions from a practical point of view. What difference does it make with my actual equipment? Does it sound better or worse? It can easily be shown that there is a difference sometimes, and sometimes you cannot hear it. The difference is generally difficult to explain theoretically, instead it is dependant on the weaknesses and errors in the real world.

One example might be sample rate conversion. There are several very bad SRC implementations out there that does destroy sound. So if you start with a good recording at, say, 96 kHz and convert it to 44.1 it will sound bad. This is not because EVERY CONCEIVABLE SRC algorithm is bad. Instead it is this exact algorithm that is the problem.

What algorithm does your DAW have? How can you verify that it sounds good? Well, on first question, I have so far not talked to many persons who actually know but many are guessing. And the answer to the second question, use your ears!

So, dont think too much. Listen to practical experiences and tips from knowledgeable engineers in order to get started. Then use your own ears and listen carefully. You might end up recording at 8kHz 8 bit for what I know if that is what you like most.

Be aware that most soundcards sounds better at one of the available frequencys, and that it differs between different cards. (This is from my own very limited experience, I cannot really say it is an absolute rule.)

Most engineers nowaday try to go for at least 20bit recording, often selecting 24. The reasons may be different, but for me it is a question of beeing lazy. Using 24 bits I can aim for -10 dB at recording and avoid clipping. Same thing using 16 bit and I get too much noise, so I would have to be very much more careful with levels. End product of a well-done 16bit recording is as good as for the 24bit.

When it comes to sample rate, I run at 44.1kHz because my main target is CD. I try higher now and then but with my current equipment (Motu 828mkII) and sources (classical symphony orchestra) there is not a difference in my ears that is worth the extra risk. (Risk since my PC is heavier loaded, and that increases the risk for it stopping).

Anyway, use your ears. Don´t use too much theory as in the end most of it is about the imperfections in your hardware and software.

Gunnar.

Thank you Gunnar!

Oh for the love of God. I'm seeing so much BS on this topic. I'm really surprised that Nika hasn't gone Ape-sh*t here!

The algorithm for SRC can be as complex as you want it to be. It can handle floating point and intergers so the stuff after the decimal can certainly be computed!

As for -- anything recorded above 44.1 will simply be cut off and therefore ruin the sound -- WHAT?!? :?

It's called -- IT IS FILTERED -- not cut off! This filtering is a very accurate process. One steeped in numbers not Voodoo. It's a rather simple process which has NO effect on the sound. Think about it - there are only 2 samples required to any frequency. Those samples must divide the frequency into 2 parts and thus must be twice the stated frequency. If you record 22,000 Hz, you need a sample rate of 44,000 (actually 44,001 but that's more theoretical "voodoo" that I won't get into.)

So, what happens if you record a 20Hz computer generated wave at 44,1000 kHz sampling. Is the sound reproduced at ANY higher quality than if you used 45 Hz sampling rate?? NO!

Now, there are overtones to be dealt with in the real world - so, if a pipe organ produced that same 20 Hz tone, there would be thousands of sympathetic frequencies and therefore the higher frequency sample rate will reproduce this tone more accurately.

Where higher sampling is important is in a few areas.

1. Archival. This is kind of a fail safe. In case we ever do move to a different format than standard PCM and it involves higher frequency content than 20kHz, it doesn't hurt to actually have it on tape.
2. Effects - particularly EQ. Equalization (as well as other effects) produces out of band distortion and phase shifting. Much of the stuff that happens out of band is actually quite noticable. Keeping the SR high until the last conversion ensures that all of these changes stay as you've heard them and don't get altered dramatically at one of the Sample Rate Conversions somewhere down the line.

As for Bit Depth - sure there is a lower DIGITAL noise floor, yet ironically, it's almost unimportant. Very few people in this BBS have a room to record in that would challenge the digital noise floor of a 16 bit recording.

Where it does make a difference is simple. You do not have to keep your needles at peak the whole time to capture the sound without adding significant noise.

In other words - your signal to noise ratio on BOTH a 16 or 24 bit recording will be exactly the same. However, as you increase the 24 bit recording, you will not be bringing up a lot of digital noise with the signal. Now, imagine that you process the signal numerous times raising the amplitude a little each time - compressing your dynamic range. Slowly but surely (actually rather quickly) the digital noise floor will start creeping in quite noticably. The lower the Digital Noise Floor to start with (ie 24 bit) the more processing and amplitude changes you can make without introducing a serious level of digital noise.

FWIW, I record the majority of the symphony work that I do in 44.1kHz if for nothing else to preserve HD space. I don't hesitate to bump it up to 96k, 192k, or even DSD if I think it's a project where someone might claim to hear a difference. (Especially if that someone is me. As I've stated time and time again, personally, I believe I can hear a significant difference between standard SR and high SR recordings. I know Nika isn't a fan of this concept and certainly isn't a fan of DSD, but the fact is I'm not alone in my ability/belief to hear a difference.)

But by all means - stop spouting crap from someone who obviously is giving more opinion about digital than fact. I mean, com'on - you can't sit here and try to give informative details about digital when all you can claim as a source is some old cat who claims to know a crap load about digital.

J.

Cucco wrote: But by all means - stop spouting crap from someone who obviously is giving more opinion about digital than fact. I mean, com'on - you can't sit here and try to give informative details about digital when all you can claim as a source is some old cat who claims to know a crap load about digital.

J.

Fair enough J,
Like i said, i was just thinking out loud, i don't pretend to know what i'm talking about :?. i dunno about others... didn't mean to push any buttons :)

interested to hear opinions on this though:

Amo Audio wrote: another thing to take into account is that unless you are recording with microphones, preamps, etc. that are capable of reproducing frequencies above 20KHz, why bother? as far as I know, there aren't many companies who even consider higher frequencies than 20KHz when designing their products.

anything valid here or was I way off the mark?

Why not try it out for yourself?

Personally I think it may make a difference. As a general statement I would expect the difference to lay mainly in how the software effects inside the DAW will handle the sound. As for the recording part, I would expect it to differe between different types of cards. I believe AD-converters to be designed doing a lot of compromises, and these will influence the sound. More or less at different frequencies.

So, stop worrying, try it on your equipment and stay happy.

Gunnar

Hey AmoAudio,

There's a little truth here, but it's a little off the mark.

Mainly, audio equipment is more than capable of recording and playing back beyond 20kHz. However, most manufacturers don't spec past that point. Meaning, until recently, you would never have seen a speaker or amplifier or mic for that matter give you it's measurement tolerances beyond 20kHz.

About 7 years ago, home audio manufacturers (led by Harmon Int'l) started this big push for wide frequency range amplifiers and preamplifiers (or receivers too). Microphones, by their very nature will record beyond 20 kHz in most situations. (Though many mics naturally roll off beyond 15kHz or so.)

Most mic preamps are designed to work to far beyond 20kHz.

Of course, the big question is, why? Considering that most people truly can't hear above 20 kHz. Again, it's a matter of perception. I've spoken before about the experiments concerning high-frequency and directional cues. Though we may not receive the information of ultra-sonic frequencies the same way we do with other frequencies, evidence is certainly there to suggest that we do process the information.

J.

ghellquist wrote:
So, stop worrying, try it on your equipment and stay happy.

i dunno about others, but i'm not loosing any hair over this :). I just like to hear others opinions and learn stuff.

thanks for the info guys. I'd also like to hear thoughts on this:

Amo Audio wrote: I also believe that this "better" sound people hear when recording at higher sample rates and converting down is subjective and should be called "different" because they may just be hearing the artifacts from the converting process (because of the nyquest theorem they can't possibly be hearing those higher harmonics that "affect the lower frequencies")

Cheers

Cucco wrote:

Of course, the big question is, why? Considering that most people truly can't hear above 20 kHz. Again, it's a matter of perception. I've spoken before about the experiments concerning high-frequency and directional cues. Though we may not receive the information of ultra-sonic frequencies the same way we do with other frequencies, evidence is certainly there to suggest that we do process the information.

J.

Imagine if every piece of gear in your signal chain had a -3 dB point at 20 Hz and 20 KHz. As you pass through each subsequent piece it filters the spectrum from the previous one. And so on. By the time you get to the end, you don't have flat amplitude response in the audible region and the phase (time) response is distorted, as well (a 'linear', vice 'non-linear' distortion).

Therefore, one good argument for wide bandwidth electronics is to ensure that the net response of the system is distortionless in amplitude and phase. You want the transducers (mics and speakers) to provide the ultimate bandwidth limiting, not the electronics. IMO