DITHERLESS SAMPLE AND BIT RATE CONVERSION SECRET

[dannywild]

Has anyone ever tried this technique to avoid the noise from dithering when changing sample rates and bit depth using a DAC and an ADC?

Let's say you recorded your tracks at 24bit 96k but you want it in 16bit 44.1k. Instead of dithering you do the following:

Go digital out from your computer or DAW into a DAC. From there go analog out of your DAC to the analog ins on a separte ADC. Finally go from the digital outs on the ADC at 16bit 44.1k to your CD burner.

Let me clarify the signal path:

24 bit 96khz digital file sent out of a digital output to:

Digital input of the D/A converter sent out of the analog outputs to:

Analog input of the A/D converter sent out of the digital outputs at 16bit 44.1khz to:

Digital input recorded bit for bit of a CD burner or computer ect.

The sound is playing into the analog ins on the second converter. You simply set that second converter to convert the incoming analog signal into digital at 16bit 44.1kHz.

Because you've already mixed and recorded at a high resolution, it is simply taking an analog signal of pristine quality and making it compatible for a standard commerical CD. Because the signal is now completely standard commerical CD compatible at 16bit 44.1Khz you can go digital out of that second converter via ADAT LightPipe, AES/EBU, S/Pdif, etc. into your external CD burner or other recording device and burn your master tracks at single speed for extremely high fidelity. Seems like it would work but I have not tried it. Has anyone done this? Can someone try it and post the results?

[heinz]

sure it would work, but not sure da-ad conversion would sound better than a fully digital conversion. Maybe if the gear is super high end.

why is dither bad again?

[dannywild]

Originally posted by heinz
sure it would work, but not sure da-ad conversion would sound better than a fully digital conversion. Maybe if the gear is super high end.

why is dither bad again?

Dither = More noise in the signal...no?

[chriscavell]

the way you propose performing bit rate conversion involves multiple steps of AD/DA conversion...each of which will add it's own noise to the chain, probably more than dithering will add. What you're doing also amounts pretty much to truncation...which can cause distortion in low level signals...such as fade outs. This is precisely one of the things dither prevents from occuring. It's a trade-off: a touch of uncorrelated noise (much lower than the actual noise floor usually and typically ignored by the ear/brain) is added in such a way that will prevent low level distortion and increase "perceived" dynamic range. Another advantage to the dither noise is due to the fact that it's uncorrelated...which means adding multiple dither stages together should only increase the overall noise by three dB, unlike many other forms of noise which would increase by 6 dB with each doubling.

Dither is generally used b/c it provides advantages that far outweigh the trade-offs in relation to other methods of conversion without dithering...and these advantages are not acheived in your "workaround".

SRC (sample rate conversion) is completely seperate from the process of dithering and the two are totally unrelated. Your proposed method could lead to "moire" effects in the audio...which if encountered would be far worse than any software algorithm I know of for performing SRC. Once again, you're also introducing more AD/DA stages than necessary...with added noise at each.

BUT,

try it out if you like...and if it sounds better to your ears...then by all means use this method. It is what the final result is to our ears that really matters in the end...just know that if the result is more pleasing, it has nothing to do with dithering or SRC and everything to do with the analog components of the converters and gear you intend to do this with.

[jbuntz]

Originally posted by dannywild
Dither = More noise in the signal...no?

If you can hear dither noise in a mix, you are a better man than I. If i'm not mistaken, dither noise RMS is a few dB over the -96 dB bottom limit when dithering to 16 bit and even lower on higher bitrates and theroretically be totally inaudible if under -96 dB being that the human hearing dynamic range is about 96 dB. To run out and back in to a converter means it has to run through two more low pass filters which it to blame for a lot of the digital "coldness."

[dannywild]

...just know that if the result is more pleasing, it has nothing to do with dithering or SRC and everything to do with the analog components of the converters and gear you intend to do this with.

I agree. You would definitely want to use very highend converters to do this.

[s0nguy]

Originally posted by jbuntz
If you can hear dither noise in a mix, you are a better man than I. If i'm not mistaken, dither noise RMS is a few dB over the -96 dB bottom limit when dithering to 16 bit and even lower on higher bitrates and theroretically be totally inaudible if under -96 dB being that the human hearing dynamic range is about 96 dB. To run out and back in to a converter means it has to run through two more low pass filters which it to blame for a lot of the digital "coldness."

With the Waves stuff I can hear noise shaping......very high piych wine...I se my noise shaping to "none" on the L2. Is that bad?

But hearing dither....hmm....dunno.

-S0nguy

[tele_player]

songuy: are you implying that noise shaping is something other that one of the forms of dither?

[Simon Lomax]

Originally posted by s0nguy
With the Waves stuff I can hear noise shaping......very high piych wine...I se my noise shaping to "none" on the L2. Is that bad?

But hearing dither....hmm....dunno.

-S0nguy

Me also, I can hear the difference with dither/noise shaping on an L2 or even just with pow-r in logic. UV22 sounds the most transparent to my ears.

Simon

[Bob Olhsson]

The A to D converter is still going to dither or truncate to 16 bits.

Dither is always going to be less audible than the modulation distortion from truncation and it will also accumulate less per generation than the correlated distortion from not dithering accumulates.

In other words, dither prevents considerably more distortion in successive bit reductions than it contributes as noise. This can be hard to hear because many systems don't truncate until you actually hit the final media being recorded to. This leaves you always monitoring at the full bit depth of the DAC without actually hearing the truncation until you play it back.

[gko]

burn your master tracks at single speed for extremely high fidelity

It is not adviceble to burn a CD at single speed. The chance to induce errors is very high. The optimum burn speed is 2 or 4.

some dither theory

Gertjan Koning

[Nathan Boddy]

I see the point you are making; however ADC's introduce extra dither noise in there signal chain and because convertors are often the weakest part of the chain, will degrade the quality of the signal far more than dithering.

Regards
Nathan

[StoneinaPond]

Originally posted by chriscavell
What you're doing also amounts pretty much to truncation...which can cause distortion in low level signals...such as fade outs.

Originally posted by Bob Olhsson
[B]The A to D converter is still going to dither or truncate to 16 bits.

I'm sorry, but this is not truncation. Truncation involves loosing 8-bits of information that is present in all of the 24-bits of a 24-bit word length.

Capturing audio via 16-bit converters simply spreads the audio information over 16-bits. Nothing is lost other than dynamic range. I do this all the time but for different reasons. I come out of a 24-bit converter into analog summing and capture that result to 16-bit DAT. The results are fine.

I know that then comparing that file capture to a BTD with dither is a stretch as I have just added additional processing. But it works well. I just wish CDs supported 24-bit data. Then we wouldn't have to worry about this.

[chriscavell]

I'm sorry, but this is not truncation. Truncation involves loosing 8-bits of information that is present in all of the 24-bits of a 24-bit word length.

Capturing audio via 16-bit converters simply spreads the audio information over 16-bits. Nothing is lost other than dynamic range. I do this all the time but for different reasons. I come out of a 24-bit converter into analog summing and capture that result to 16-bit DAT. The results are fine.

Please tell me you're joking...tutt

It sounds to me like you think that the 16 bit converters simply compress the 24 bit signal down into 16...which is far from correct. Please tell me this is not what you're trying to communicate.

I know that then comparing that file capture to a BTD with dither is a stretch as I have just added additional processing. But it works well. I just wish CDs supported 24-bit data. Then we wouldn't have to worry about this.

What you're really doing is just using an analog summing buss instead of a digital one and then relying on the dither at the input of the 16 bit converter to keep from enjoying least bit distortion. If the clocks of your D to A and your final 16 bit DAT are sync'ed, you probably won't run into moire effects...but your understanding of what is actually occuring seems to be incorrect, unless I'm misinterpreting your post. You really are truncating, it's just that prior to truncating the converter on the DAT deck is dithering it.

[jbuntz]

Originally posted by StoneinaPond
I'm sorry, but this is not truncation. Truncation involves loosing 8-bits of information that is present in all of the 24-bits of a 24-bit word length.

Capturing audio via 16-bit converters simply spreads the audio information over 16-bits. Nothing is lost other than dynamic range. I do this all the time but for different reasons. I come out of a 24-bit converter into analog summing and capture that result to 16-bit DAT. The results are fine.

I know that then comparing that file capture to a BTD with dither is a stretch as I have just added additional processing. But it works well. I just wish CDs supported 24-bit data. Then we wouldn't have to worry about this.

If you go from 24 to 16 bits, where do those other bits go again? In loss in bitrate is a form of truncation.

[StoneinaPond]

[QUOTE]Originally posted by chriscavell
[B]Please tell me you're joking...tutt

It sounds to me like you think that the 16 bit converters simply compress the 24 bit signal down into 16...which is far from correct. Please tell me this is not what you're trying to communicate.[QUOTE]

OK. We're talking about a 24-bit signal at a D/A. Prior to the conversion, <i>all</i> 24-bits contain information about the signal. Correct? So if you simply lop off 8-bits (admittedly low level information,) you are distorting the actual content of the data stream.

But once it is output in the analog domain, it is no longer 24-bits. It's analog. If you then recapture using 16-bit converters, you are spreading all the information over 16-bits. You loose nothing other than what 16-bit processing delivers in the first place - lower dynamics, slightly collapsed stereo field, etc. Something we have to live with in the CD Red Book world.

By the way, I did make casual comparisons a while back between straight out of 24-bit converter (analog) into DAT. I then compared that with a file that was BTD using Waves dither. I was hard pressed to hear the difference. Again, not scientific, but good enough for me to work the way I want to.

Compressing? What gave you that idea?

[chriscavell]

Compressing? What gave you that idea?

This is what...it seems that you think "all" the information is now contained within that 16 bits which once took 24 to contain. Yes, you lose dynamic range with 16 bits...but it's a tad more complicated than that. The analog dynamic range of the signal you're sending to the 16 bit converter may be more than what 16 bits can contain. In which case, without dither, you'd encounter least bit distortion...regardless of the fact that you went analog in between.

One bit of the 16 bit word contains the same amount of information as is contained in one bit of the 24 bit word. There is no getting around that. Information is lost in the process of going from 24 to 16.

This isn't sandwiches...it's audio. The process does not involve "spreading all the information" that was once contained in 24 bits down into 16. That's just not how it works....that's not how the bits work, not how the digital word works, not how the converters work, and not how the analog summing buss works, and not how any combination therein works.

Ask yourself, where does the information that was once contained in the lowest passages (the least bits) go exactly? Yep, that's right, out into the aether. It's lost, i.e. truncated.

[StoneinaPond]

The analog dynamic range of the signal you're sending to the 16 bit converter may be more than what 16 bits can contain.

What do you mean by that? Are you talking about distorting the analog front end. Are you talking about clipping the converters? Because properly recorded 16-bits will deliver as much dynamic range as it is capable of, no more, no less.

One bit of the 16-bit word contains the same amount of information as is contained in one bit of the 24-bit word.

Yes, a bit is a bit.

There is no getting around that. Information is lost in the process of going from 24 to 16.

Agreed, but that is true of digital bounce or recapture of audio. The difference being that truncation removes actual data that is embedded in the signal flow. Capturing at 16-bits means that the full data is captured with less precise resolution. But the whole signal is still present. It isn’t compressed, it just has less resolution.

Thinking while I’m writing, perhaps we have an issue with semantics.

To truncate is to shorten by cutting off. In computer terms, when information is truncated, it is ended abruptly at a certain spot. For example, if a program truncates a field containing the value of pi (3.14159265...) at four decimal places, the field would show 3.1415 as an answer. This is different from rounding, which would result in the answer of 3.1416.
For another example, consider a text entry field in an electronic form. If a program limits the size of the field to 255 characters, it may allow you to continue typing past the maximum number of characters. However, when the information is submitted or saved, the program truncates the data to the first 255 characters and any additional characters are disregarded.

When you bounce to disc, you are indeed truncating, but adding noise to help deal with the effect of that process. When you recapture audio, how can you truncate an analog signal?

This isn't sandwiches...it's audio. The process does not involve "spreading all the information" that was once contained in 24 bits down into 16. That's just not how it works....that's not how the bits work, not how the digital word works, not how the converters work, and not how the analog summing buss works, and not how any combination therein works.

Your mixing (to use another culinary pun) your data streams. Obviously analog summing does not involve bits.

Ask yourself, where does the information that was once contained in the lowest passages (the least bits) go exactly? Yep, that's right, out into the aether. It's lost, i.e. truncated.

When performing a digital transfer from 24-bits to 16-bits without any processing, yes indeed, information is lost. It is lost because you throw away 8-bits of information. Yes, yes and yes. But when you capture using 16-bit converters, you are sampling the entire analog signal with less overall resolution than you would with 24-bit converters.

Two different processes. Please try and be specific.

[chriscavell]

Capturing at 16-bits means that the full data is captured with less precise resolution. But the whole signal is still present.

Not the case with digital audio...not if the analog signal going into the 16 bit converters consists of information that would fill more than 96 dB's of dynamic range.

Okay, let's look at it this way (an extremely simplified way of looking at it, but an albeit accurate way of getting the idea across as it pertains to digital audio):

You perform the bounce to an "ideal" 24 bit converter as your final converter. Then perform the same bounce to an "ideal" 16 bit converter. Neither with any form of dithering. If we continue to look at it as an ideal process: The upper 16 bits of the 24 bit bounce should ideally be bit for bit identical to the 16 bit bounce. Where did the eight bits go? NONE of the information that would be contained in the least eight bits of the 24 bit bounce would be contained in the 16 bit bounce. It would be gone....in essence, truncated. This is how digital audio works as I understand it. Make the levels low enough in the bounce, and you'll hear least bit distortion in the 16 bit bounce where there isn't any in the 24 bit bounce...both coming from an analog source.

Maybe this would help: every bit of a word contains the same resolution, independent of whether or not it is a 16 or 24 bit word. In order for a capture of 16 bits to contain all the information that a 24 bit capture would contain, the individual bits of the 16 bit word would have to contain more info than the bits of the 24 bit word....that's obviously not how it works.

[StoneinaPond]

You perform the bounce to an "ideal" 24 bit converter as your final converter. Then perform the same bounce to an "ideal" 16 bit converter.

This statement makes no sense.

You don't bounce to a converter. You bounce to a hard drive as a file.

I ask you again, how can you truncate an analog signal?

(And I don't need some smart Alec to bring up filtering. That is not the issue.)

[chriscavell]

I'm using the term bounce to mean any dubbing of the downmix to any medium...as most of us would use it. Bounce to tape, bounce to disk...I usually follow the term "bounce" with the medium I'm dubbing to.

[StoneinaPond]

Originally posted by chriscavell
I'm using the term bounce to mean any dubbing of the downmix to any medium...as most of us would use it. Bounce to tape, bounce to disk...I usually follow the term "bounce" with the medium I'm dubbing to.

Regardless, you do not bounce to converters. You can bounce digitally (since this is what we are talking about) and then output that file through either a 24-bit converter or a 16-bit converter.

And if you do not dither the 24-bit file prior to outputing through a 16-bit converter you will indeed truncate the file with not very pleasant results.

But that is not the entire focus of the conversation. The original poster talked about outputing a 24-bit file through a 24-bit converter and recapturing through a 16-bit A/D. In this instance you are not truncating. You are recording with less resolution.

There is a profound difference between the two processes.

[Bob Olhsson]

There have not been 16 bit A to D converter chips made since shortly after the introduction of the CD. The ones from the early '90s were 18 bit and required you to use a dither box to get the best sound out of them!

[Teacher]

couldn't you just use a 24 bit converter and set it to 16?

[AlphaDingo]

.....this amp goes up to 11.

[StoneinaPond]

Originally posted by AlphaDingo
.....this amp goes up to 11.

Yes, but if you use an amp simulator plug-in that only goes to 8, you loose those 3-bits to the ether, never to return, except to be blended in with all the other poor truncated signals that are at this moment travelling at the speed of binary signals to the outer reaches of our solar system.

If one could capture the cake mix of truncated signals and spread it evenly around the baking dish, making sure that we bounce to the oven, ahh, then what beauty might we find there?

[Beech]

"Let's say you recorded your tracks at 24bit 96k but you want it in 16bit 44.1k. "

This was part of the original post but it seems to me that the thread has gone on to dealing with the issue of getting from 24 bits to 16. I see (and hear!) the Sample Rate Conversion from 96k to 44.1k as being a much bigger problem.

First let me mention a test I did the other day.

I had an stereo orchestal mix in a 24bit / 44.1k session. When I had it mastered I did two versions:

1. Bounce to disk in Pro Tools with the Waves L2 set to 16 bit , type 1 and Normal. PTs was set to 16bit in the BTD page.

2. Then I did an alternative version, this time with no dithering on the L2. PTs was set to 24bit in the BTD page

Then I imported both bounces into Wave Burner Pro. I put the dithered L2 version on one cd. Then I put the non dithered one on another cd, this time with POW-R (Waveburner's own dither)
on the second.

So, if you''re still following me, I now had two cds, one with Waves IDR dither, and the other with POW-R.

I then imported these two back into a 16bit Pro Tools session just the hear the difference, if any.

I decided that I prefered the Waves IDR but there really was not much to choose between them.

Overall, I have to say no matter which way I do it, the loss in quality is mimimal and I purposely used some orchestral music figuring that this would be most critical.

BUT, BUT, BUT. Going from 24/96 to 16/44.1 is an entirely different matter. In this situation the quality lose is bad! This leads me to believe that the software (PTs) SRC is not to be taken seriously, at least not for making cds.

So, getting back to the original poster's dilemma, I would be of the view that going through a good D/A from the 24/96 recording and into an A/D and onto the cd is a good option. Also, it would be a good time to do any analog processing necessary.

Now if the Rosetta 200 does all it's supposed to do when they finally release it, I'm going to be very interested. It's the first device that is an A/D, D/A and has SRC built in, at least for less then $2000.

[AlphaDingo]

Yes, but if you use an amp simulator plug-in that only goes to 8, you loose those 3-bits to the ether, never to return, except to be blended in with all the other poor truncated signals that are at this moment travelling at the speed of binary signals to the outer reaches of our solar system.

....this amp goes up to 11.

[StoneinaPond]

Originally posted by AlphaDingo
....this amp goes up to 11.

YOU NEED TO TURN IT DOWN!!!

Or else just go and sell shoes. You can do that, can't you?

[StoneinaPond]

Originally posted by Beech
"Let's say you recorded your tracks at 24bit 96k but you want it in 16bit 44.1k. "

This was part of the original post but it seems to me that the thread has gone on to dealing with the issue of getting from 24 bits to 16. I see (and hear!) the Sample Rate Conversion from 96k to 44.1k as being a much bigger problem.

First let me mention a test I did the other day.

I had an stereo orchestal mix in a 24bit / 44.1k session. When I had it mastered I did two versions:

1. Bounce to disk in Pro Tools with the Waves L2 set to 16 bit , type 1 and Normal. PTs was set to 16bit in the BTD page.

2. Then I did an alternative version, this time with no dithering on the L2. PTs was set to 24bit in the BTD page

Then I imported both bounces into Wave Burner Pro. I put the dithered L2 version on one cd. Then I put the non dithered one on another cd, this time with POW-R (Waveburner's own dither)
on the second.

So, if you''re still following me, I now had two cds, one with Waves IDR dither, and the other with POW-R.

I then imported these two back into a 16bit Pro Tools session just the hear the difference, if any.

I decided that I prefered the Waves IDR but there really was not much to choose between them.

Overall, I have to say no matter which way I do it, the loss in quality is mimimal and I purposely used some orchestral music figuring that this would be most critical.

BUT, BUT, BUT. Going from 24/96 to 16/44.1 is an entirely different matter. In this situation the quality lose is bad! This leads me to believe that the software (PTs) SRC is not to be taken seriously, at least not for making cds.

So, getting back to the original poster's dilemma, I would be of the view that going through a good D/A from the 24/96 recording and into an A/D and onto the cd is a good option. Also, it would be a good time to do any analog processing necessary.

Now if the Rosetta 200 does all it's supposed to do when they finally release it, I'm going to be very interested. It's the first device that is an A/D, D/A and has SRC built in, at least for less then $2000.

Beech,

Did you do the SRC and dithering at the same time? And in what way was the result bad?

The Rosetta 200 should be a dream.

And going out of a good D/A with analog processing into a good A/D is how many mastering engineers do it.

By the way, I prefer POW-r to IDR, but that is just a personal opinion. In fact my favourite was Apogee's UV22 TDM plug-in that did not make it to OS X.

Peace,

Yorik