48kHz vs 44.1kHz

grimepoch · Post by **grimepoch** » Sat Jan 21, 2006 6:02 am

Okay, so I've done enough research and experimenting to understand the difference between working at 16-bit, or working at 24-bit and dithering down to 16-bit. Even with my experiments, the difference was pretty clear. I hadn't realized that you could gain so much more in the sound by dithering, but when I saw photo examples of a 2-bit image dithered and not dithered, it became clear to me what was going on. I must say, I find it very fascinating to say the least how that works.

My next question is on sample rate. Everything I've found on dithering speaks of the dynamic range (value) issue, but I've not found anything relating to the conversion between 48kHz and 44.1kHz.

Theoretically, to me it seems like it wouldn't be a problem, because it's not the same thing we are doing with dithering. Instead, it would be like taking the 24-bit value and just dividing the size to get to 16-bits. (Which in essense would not introduce new noise, but it would greatly reduce your dynamic range and loose information as it quantizes to the smaller values.

So, for 48Khz, you can just interpolate between the data (picking a common multiple) and re-pick the points.

So my question comes from, is the sonic quality of recording at 48kHz and reducing to 44.1kHz with everything else held the same the best thing to do? To me, it would seem you'd have more detail in the higher frequencies, and this would affect the interpolated data as you mixed things, but I've not been able to find a definite answer.

The reason I've been thinking about this, the mastering and dithering guide from the makers of Ozone 3 don't discuss this. In fact, some of their more powerful dithering shapes only work it said at 44.1kHz because of the noise shaping.

Thoughts? Experience?

(I'd even try 96kHz, but I'm hardware limited on my ADA-8000 which only can do up to 48kHz/24 bit.

richardein · Post by **richardein** » Sat Jan 21, 2006 7:05 am

Rick, A stupid question, but if your final goal is a 16 bit 44.1 master, why not do that and not go through the conversion step?

grimepoch · Post by **grimepoch** » Sat Jan 21, 2006 7:19 am

Originally, that is what I thought. But after experimentation and a lot of reading, I discovered that if you start at a higher bit rate, mix, then convert down with dithering, you get a better sounding result. You aren't just scaling down when you convert, you truncate the 8 least significant bits. But before you do this, you add a little noise to the mix. The attempt is to preserve some of the dynamic range of the 24-bit signal beyond the 16-bit quality you'd get if you just converted by scaling the dynamic range.

As for the 44.1/48 question, I believe that working at the higher rate is still going to be better. People get hung up on 'what the ear can hear' debate, but it goes much deeper than that. That extra information you are recording makes a difference in the digital domain. Once you've quantized values, you loose the data inbetween. Those inbetween values in the analog domain can make a difference, especially to causal signals, you loose some of the definition.

But, like I said, try this stuff out. I wasn't buying it completely until I did some experimenting in my studio. I mean, someone tells you adding noise to a mix can make it better? (Specific noise that is), and now, I agree with them.

If you go to the Izotope website, look in their knowledge base. They wrote two great papers; one on mastering and one on dithering. They are both related to Ozone 3, but even they say, there is information for people who just want to know something behind the science of it.

Aramis · Post by **Aramis** » Sat Jan 21, 2006 9:12 am

grimepoch wrote:Okay, so I've done enough research and experimenting to understand the difference between working at 16-bit, or working at 24-bit and dithering down to 16-bit. Even with my experiments, the difference was pretty clear. I hadn't realized that you could gain so much more in the sound by dithering, but when I saw photo examples of a 2-bit image dithered and not dithered, it became clear to me what was going on. I must say, I find it very fascinating to say the least how that works.

My next question is on sample rate. Everything I've found on dithering speaks of the dynamic range (value) issue, but I've not found anything relating to the conversion between 48kHz and 44.1kHz.

Theoretically, to me it seems like it wouldn't be a problem, because it's not the same thing we are doing with dithering. Instead, it would be like taking the 24-bit value and just dividing the size to get to 16-bits. (Which in essense would not introduce new noise, but it would greatly reduce your dynamic range and loose information as it quantizes to the smaller values.

So, for 48Khz, you can just interpolate between the data (picking a common multiple) and re-pick the points.

So my question comes from, is the sonic quality of recording at 48kHz and reducing to 44.1kHz with everything else held the same the best thing to do? To me, it would seem you'd have more detail in the higher frequencies, and this would affect the interpolated data as you mixed things, but I've not been able to find a definite answer.

The reason I've been thinking about this, the mastering and dithering guide from the makers of Ozone 3 don't discuss this. In fact, some of their more powerful dithering shapes only work it said at 44.1kHz because of the noise shaping.

Thoughts? Experience?

(I'd even try 96kHz, but I'm hardware limited on my ADA-8000 which only can do up to 48kHz/24 bit.

Hello !!! Read this !

http://www.audioease.com/Pages/BarbaBat ... CTest.html

Aramis

carrythebanner · Post by **carrythebanner** » Sat Jan 21, 2006 9:25 am

The benefits of working at 24 bit all the way until the final stage and only then dithering down to 16 bit have been covered ad nauseum, so I won't touch that.

Regarding sample rate: While I can understand comparing 44.1/48 to 88.2/96 in terms of extended high frequency response, there's really not much difference between 44.1 and 48 in that regard.

Consider the highest frequency you can record at each sample rate (i.e., the Nyquist frequency): 22.05 kHz at 44.1, and 24 kHz at 48. In terms of musical intervals, the difference between 22.05 and 24 kHz is only around 1 semi-tone, so there's not much to be gained. Also, as various tests and white papers show, sample rate conversion is approached differently in different apps, so in the end you may actually lose a bit of high end working at 48 & converting down to 44.1 when compared with just working at 44.1.

Considering that, I work at 24 bit / 44.1, and convert down to 16 / 44.1 at the final stage.

Splinter · Post by **Splinter** » Sat Jan 21, 2006 10:09 am

carrythebanner wrote:The benefits of working at 24 bit all the way until the final stage and only then dithering down to 16 bit have been covered ad nauseum, so I won't touch that.

Regarding sample rate: While I can understand comparing 44.1/48 to 88.2/96 in terms of extended high frequency response, there's really not much difference between 44.1 and 48 in that regard.

Consider the highest frequency you can record at each sample rate (i.e., the Nyquist frequency): 22.05 kHz at 44.1, and 24 kHz at 48. In terms of musical intervals, the difference between 22.05 and 24 kHz is only around 1 semi-tone, so there's not much to be gained. Also, as various tests and white papers show, sample rate conversion is approached differently in different apps, so in the end you may actually lose a bit of high end working at 48 & converting down to 44.1 when compared with just working at 44.1.

Considering that, I work at 24 bit / 44.1, and convert down to 16 / 44.1 at the final stage.

I concur 100%. SRC has gotten much better, but it's just not worth it. Get a really good ADC, record at 24/44.1 and don't look back.

pcm · Post by **pcm** » Sat Jan 21, 2006 10:48 am

Splinter wrote:
carrythebanner wrote:The benefits of working at 24 bit all the way until the final stage and only then dithering down to 16 bit have been covered ad nauseum, so I won't touch that.

Regarding sample rate: While I can understand comparing 44.1/48 to 88.2/96 in terms of extended high frequency response, there's really not much difference between 44.1 and 48 in that regard.

Consider the highest frequency you can record at each sample rate (i.e., the Nyquist frequency): 22.05 kHz at 44.1, and 24 kHz at 48. In terms of musical intervals, the difference between 22.05 and 24 kHz is only around 1 semi-tone, so there's not much to be gained. Also, as various tests and white papers show, sample rate conversion is approached differently in different apps, so in the end you may actually lose a bit of high end working at 48 & converting down to 44.1 when compared with just working at 44.1.

Considering that, I work at 24 bit / 44.1, and convert down to 16 / 44.1 at the final stage.
I concur 100%. SRC has gotten much better, but it's just not worth it. Get a really good ADC, record at 24/44.1 and don't look back.

After working at 48k since the late eighties, I switched over to 44.l1 a few months ago. Stuff sounds much better now. But that's because the convertors are better, and that was always the problem to begin with.

Shooshie · Post by **Shooshie** » Sat Jan 21, 2006 10:54 am

Aramis wrote:Hello !!! Read this !

http://www.audioease.com/Pages/BarbaBat ... CTest.html

Aramis

Ok, to sum up that page, at least regarding MOTU's DP, it comes down to this: DP begins filtering early so that anti-aliasing occurs quickly, and aliasing is eliminated in time, but at a loss of high frequencies. Using 48K at least raises the bar to the threshold of hearing, and using 3rd party conversion on your final 2-tracks, or converting out-of-the-box after that, as in sending your 48k tracks to a mastering facility, would give you results that are right up there with anyone else's.

Logic's aliasing problem, on the other hand, apparently is just screwed, or was when that was written.

Am I reading that right?

Shooshie

Aramis · Post by **Aramis** » Sat Jan 21, 2006 1:02 pm

Shooshie wrote:
Aramis wrote:Hello !!! Read this !

http://www.audioease.com/Pages/BarbaBat ... CTest.html

Aramis
Ok, to sum up that page, at least regarding MOTU's DP, it comes down to this: DP begins filtering early so that anti-aliasing occurs quickly, and aliasing is eliminated in time, but at a loss of high frequencies. Using 48K at least raises the bar to the threshold of hearing, and using 3rd party conversion on your final 2-tracks, or converting out-of-the-box after that, as in sending your 48k tracks to a mastering facility, would give you results that are right up there with anyone else's.

Logic's aliasing problem, on the other hand, apparently is just screwed, or was when that was written.

Am I reading that right?

Shooshie

See !!!! my opinion ....on this !!!!!
After I saw that page and that I digested it .....
My opinion became more .....If you need a final tune at 44 , record it at this sample rate .....If you need 48 or 96 record it on such ....

Of course it depend on what system you own ....

Aramis

grimepoch · Post by **grimepoch** » Sun Jan 22, 2006 11:17 am

So this brings up some interesting questions. First, I listened to all those samples. My ears couldn't tell too much difference in most of them except for all the aliasing I heard. I must not be sensitive enough or my monitors aren't. Even though I've recently calibrated them with quite a flat response.

I'd love to see some of these papers though. Sure I agree that 48Khz might not be that much better to risk the interpolation, but even at 96kHz, I would think the extra data would be beneficial in curves between the points that are not linear. Does that extra 3.9Khz resolution help with non-linear curves? It would have to at 96kHz. To me it wouldn't be completely a means of nyquist limit at all, it would have to do with things like phase relationships between edges of say higher frequency data. Maybe I am getting too mathematical about all this.

That said, if someone has any more references, I'd love to read them. I think my only choice at this point is to record at 48Khz and 44.1kHz and see what I hear out of the mix

Thanks for all the information though!

mhschmieder · Post by **mhschmieder** » Mon Jan 23, 2006 3:57 pm

There's actually a few more issues besides those of pure audio quality in the original A/D conversion. But it also all depends on context.

As far as raw quality of different sample rates is concerned, I did verify a couple of years back that instruments such as celtic harp, which have complex decay patterns, sound far more realistic when converted at 96 kHz. I have heard this is also true for nylon guitar, and probably many other acoustic instruments. So I would say working (initially) in the high-resolution rates would be dictated by the instrumentation, style of music, etc.

The down-sampling is said to be more efficient and error-free in general, when using a divisor of two, but I also suspect that this was more true a few years ago than it is today (as with many of the "rules of thumb" about digital recording, many of them based on early limitations).

My own dilemma has also been between 44.1k and 48k, but not because of any incremental quality difference but due more to practical matters. I use TC Electronic outboard gear at mixing time for effects processing, and some modules work ONLY at 48 kHz (via S/PDIF). In looking through all my specs, 48 kHz seems to be the MOST COMMON sample rate when working with heterogeneous equipment and multiple transfer protocols; even though 44.1 kHz is the final target for Red Book audio CD's.

As a result, I think it makes the most sense to record at 48 kHz (this also helps if preparing tracks for dual use in soundtracks). Possibly mixed with 96 kHz for critical tracks like nylon guitar, but I haven't had the time yet to learn how to set up DP for allowing multiple sample rates in a single project. Maybe it doesn't, in which case one still probably benefits (for reasons discussed by an earlier poster) from RECORDING at 96 kHz and then downsampling to 48 kHz and disabling the 96 kHz track (or whatever it takes for DP to consider the project to be "sane" and mix-ready).

Our local pro audio store tells me that one has to go to the high end for 96 kHz to make much of a difference when recording signal directly into the preamps of an audio I/O converter; as opposed to my experiment of coming in at line level with prerecorded archival material from tape.

qo · Post by qo » Mon Jan 23, 2006 5:09 pm

I record at 88.2 and end up with 44.1 without using digital SRC. Just don't use SRC to get from 88.2 to 44.1 and you don't have to worry about all that digital stuff

It becomes a non-issue.

mhschmieder · Post by **mhschmieder** » Tue Jan 24, 2006 12:27 am

Just when I thought I had this stuff figured out, I see some on-line surveys that have made me rethink the entire approach to sample rate choice:

http://www.audioease.com/Pages/BarbaBat ... CTest.html
http://www.bias-inc.com/products/peakPr ... ePaper.pdf
http://src.infinitewave.ca/

OK, so I guess I need to stop using Digital Performer's convenient export function to Sample Rate Convert a 48k or 96k mix, always mix and export at the recorded sample rate, and use a quality outboard program (probably a stereo audio editor such as BarbaBatch or Peak), to do the SRC.

Looking at the results from Logic sure makes me glad that I chose DP over that program!

qo · Post by qo » Tue Jan 24, 2006 12:34 am

mhschmieder wrote: http://www.bias-inc.com/products/peakPr ... ePaper.pdf

Yep. I'd posted the Bias paper in a different thread some time back. Just avoid SRC completely and you don't have to deal with this:

DP-->DA-->analog_mixer-->AD-->other_daw

DA is set to e.g. 48, 88.2, 96, whatever

AD is set to 44.1

No SRC. Just say no to SRC

mhschmieder · Post by **mhschmieder** » Tue Jan 24, 2006 12:54 am

That is very enlightening, and thanks for reminding me of who posted the earlier reference paper weblink. I'm still pretty green at this digital game, after years of analog work, and often take other people's advice a worth more than my own intuition as a result.

Now I don't need to assume that going back to the analog realm after mixing, is going to degrade the quality of the result, as SRC does so much damage (though it is improving, and it seems that the 64-bit version of iZotope, not yet released, is the best yet, looking at the neutral report vs. the ones provided by BarbaBatch and Peak).

Of course, if I send something to a mastering lab, I assume that will happen already, but they have the megabucks for the best outboard gear ($7000 compressors, etc.). But if I'm doing something all on my own, I had assumed that I should do everything possible to stay in the digital domain.

And the spanner in the works was the discrepancy between different outboard gear from TC Electronic and others, in terms of their supported sample rates. That's what got me onto this recently revived thread.

I think you have me convinced that it's cheaper and better to avoid SRC, as a non-pro home studio, and look into what I should use to receive the 44.1k stereo buss back from the final mix processing from DP.

I need a stereo audio editor anyway for a more convenient way of archiving some of my tapes, unless I rebuy the Alesis MasterLink (which has WAY more capacity than when I sold my rev 1 unit a few years back, after frustration at how much time it spent in the repair shop).

Looks like I should start searching for threads on stereo audio editors in this forum, or maybe switch over to MusicPlayerNetwork for that task.

MOTUnation.com

48kHz vs 44.1kHz

48kHz vs 44.1kHz

Re: 48kHz vs 44.1kHz

Re: 48kHz vs 44.1kHz

Re: 48kHz vs 44.1kHz

Hmmmm....