Help: Dithering in Logic 44.1 at 24 bit?

[JAY55]

Hi, I tried searching w/out success. I'm tracking in Logic with 24bit recording enabled at 44.1k. When bouncing down the mix to burn on a CD, do I need to use the Dithering feature? Please help, thanks.

[Luna Sound]

Yes!

[Ashermusic]

Originally Posted by Luna Sound

Yes!

Luke, as a Mastering Engineer, perhaps you could expand on why you say that and how you decide which dithering algo to use.

[Luna Sound]

Originally Posted by Ashermusic

Luke, as a Mastering Engineer, perhaps you could expand on why you say that and how you decide which dithering algo to use.

Okay, the final output of a mix in Logic is 24bit. CDs are 16bit. If you bounce the mix straight to a CD without dithering then the last 8 bits will just get thrown away. Using the dither will help to minimise the effect of the losses. As for which dither to use you can always try bouncing the mix with each of the 3 options available and see which one you like/ which one suits the style of music.

[Ashermusic]

Originally Posted by Luna Sound

Okay, the final output of a mix in Logic is 24bit. CDs are 16bit. If you bounce the mix straight to a CD without dithering then the last 8 bits will just get thrown away. Using the dither will help to minimise the effect of the losses. As for which dither to use you can always try bouncing the mix with each of the 3 options available and see which one you like/ which one suits the style of music.

Any thoughts on the various Pwr- options etc.?

[Lagerfeldt]

Originally Posted by JAY55

Hi, I tried searching w/out success. I'm tracking in Logic with 24bit recording enabled at 44.1k. When bouncing down the mix to burn on a CD, do I need to use the Dithering feature? Please help, thanks.

Bounce to a 44.1 kHz 16 bits WAV or AIFF using dither or noise shaping.

You can choose between POW-r#1 (colored dither), POW-r#2 (noise shaping) and POW-r#3 (noise shaping). There is no TPDF (flat dither) option available. Turning dither off will truncate your signal, simply discarding the last bits. Logic does self-dither in other cases but that's not relevant here.

Which POW-r algorithm is best for the job depends on the material. I find #3 works best on most material though I sometimes use #1 for heavily compressed songs. I never use #2 as it's not as well implemented for some reason, though it can sound great in the Weiss hardware.

Use interleaved stereo (not split files) or you will burn two mono files instead of one stereo file.

Turn off normalizing, as you've already mastered your file and do not want to change the level or ceiling.

Bounce in realtime so you can hear what's going on while processing, naturally this is required if you're using the I/O plug-in with outboard.

Once you've bounced your 16 bit dithered file do not alter the file in any way , i.e. no gain change, no processing, no fading. Any alteration willl destroy the dither. You can trim (cut) the start or end point of the file, but that's it.

If you experience buffer residue when bouncing (a small click of sound just before the beginning of the song) set the locators one beat or bar earlier, and cut this extra space off after bouncing by using the trim function in the sample editor.

[Ashermusic]

Originally Posted by Lagerfeldt

Bounce to a 44.1 kHz 16 bits WAV or AIFF using dither or noise shaping.

You can choose between POW-r#1 (colored dither), POW-r#2 (noise shaping) and POW-r#3 (noise shaping). There is no TPDF (flat dither) option available. Turning dither off will truncate your signal, simply discarding the last bits. Logic does self-dither in other cases but that's not relevant here.

Which POW-r algorithm is best for the job depends on the material. I find #3 works best on most material though I sometimes use #1 for heavily compressed songs. I never use #2 as it's not as well implemented for some reason, though it can sound great in the Weiss hardware.

Use interleaved stereo (not split files) or you will burn two mono files instead of one stereo file.

Turn off normalizing, as you've already mastered your file and do not want to change the level or ceiling.

Bounce in realtime so you can hear what's going on while processing, naturally this is required if you're using the I/O plug-in with outboard.

Once you've bounced your 16 bit dithered file do not alter the file in any way , i.e. no gain change, no processing, no fading. Any alteration willl destroy the dither. You can trim (cut) the start or end point of the file, but that's it.

If you experience buffer residue when bouncing (a small click of sound just before the beginning of the song) set the locators one beat or bar earlier, and cut this extra space off after bouncing by using the trim function in the sample editor.

Lagerfeldt, this is perhaps an embarrassing admission,, but when I burn my 24 bit files in Waveburner with POW-r #1, #3, and no dithering and then bring them all into Logic, I hear very little difference.

I guess my ears are old and I am not a mastering engineer so I just leave it on POW-r #1.

[Lagerfeldt]

Originally Posted by Ashermusic

Lagerfeldt, this is perhaps an embarrassing admission,, but when I burn my 24 bit files in Waveburner with POW-r #1, #3, and no dithering and then bring them all into Logic, I hear very little difference.

I guess my ears are old and I am not a mastering engineer so I just leave it on POW-r #1.

Did you mean that you import 24 bit files into WB and dither down to 16 bit (or truncate) when bouncing, then burn and import into Logic? You said "burn" 24 bit files.

I can understand if the difference for you is not big with #1/2/3 but the difference is clearly noticeable when leaving the dither off (truncating). Especially on reverb tails, fades, low level passages, etc. Even with less dynamic music a truncated file will sound more harsh or distorted than a dithered one.

However, the difference between adding dithering to a 24 bit bounced file and not - that is very, very small. Almost theoretical I would say, though at least one person here claims he can hear the accumulated differences. I prefer not to add colored dither or noise shaping to a 24 bit file since it will likely be processed again. AFAIK since Logic self-dithers with TDPF when bouncing to 24 bits there really no use in adding dither at that stages.

[Ashermusic]

Originally Posted by Lagerfeldt

Did you mean that you import 24 bit files into WB and dither down to 16 bit (or truncate) when bouncing, then burn and import into Logic? You said "burn" 24 bit files.

I can understand if the difference for you is not big with #1/2/3 but the difference is clearly noticeable when leaving the dither off (truncating). Especially on reverb tails, fades, low level passages, etc. Even with less dynamic music a truncated file will sound more harsh or distorted than a dithered one.

However, the difference between adding dithering to a 24 bit bounced file and not - that is very, very small. Almost theoretical I would say, though at least one person here claims he can hear the accumulated differences. I prefer not to add colored dither or noise shaping to a 24 bit file since it will likely be processed again. AFAIK since Logic self-dithers with TDPF when bouncing to 24 bits there really no use in adding dither at that stages.

No I was referring to going from 24 bit to 16 for audio CDs.
I will have to run more tests, but I just do not remember un-dithered sounding that noticeably worse.

[SRS]

This may shed some light...

POW-R - Wikipedia, the free encyclopedia

"Since noise audibility depends on the nature of the audio being processed, the POW-R algorithm is made available in three variants, optimized respectively for simple program such as spoken word, limited dynamic range program such as rock music, and wide dynamic range program such as orchestral music[5]. Each algorithm moves the noise to those frequencies where its audible effect for the particular audio type is minimized."

And another great read from Mr. Katz... Which talks specifically about the 3 POW-R types.

Weiss Engineering POW-R Dither

[Lagerfeldt]

Originally Posted by Ashermusic

No I was referring to going from 24 bit to 16 for audio CDs.

But that's what I said. What you wrote orignally was something else, i.e. burning 24 bits, which is not possible. ;-)

Hearing the difference between a dithered or truncated 16 bit file is audible in a good listening environment at normal levels - depending on the material. On low dynamic, smashed material the difference is negligible. However, you can also try to boost the signal level/playback level in a fade out and compare. That's very audible. At least that gives you an idea of how truncation noise sounds. It also shows that dithering is a compromise if you compare a 16 bit dithered file to the 24 bit source.

[okydoky]

Originally Posted by Lagerfeldt

AFAIK since Logic self-dithers with TDPF when bouncing to 24 bits there really no use in adding dither at that stages.

hi,

i had been wondering what logic does what reducing bit depth. however, i have read some papers stating that floating point systems cannot actually be dithered "perfectly" as a fixed point system.

do you know anything about how all of that is handled by logic?

i use logic along with pro tools but there are things about logic that i have not really had a chance to research yet, and that probably limits my willingness to use it in some cases.

i am also wondering about the placement of tracks when delay compensation is being used. do you know if tracks that are being recorded are aligned with "advanced" or "delayed" tracks, or recorded with strict placement on the timeline [when delay comp. is on]? i heard that editing could get kind of screwy in some cases due to tracks being slightly "off the grid".


right.

[bob katz]

Originally Posted by oky****

hi,

i had been wondering what logic does what reducing bit depth. however, i have read some papers stating that floating point systems cannot actually be dithered "perfectly" as a fixed point system.

That is true theoretically but really not a worry in practicality. As far as I know, "normal range" signals can be easily converted to fixed point and dithered down to 24 bits properly.

BK

[okydoky]

Originally Posted by bob katz

That is true theoretically but really not a worry in practicality. As far as I know, "normal range" signals can be easily converted to fixed point and dithered down to 24 bits properly.

BK

hi,

ah ha!

if i understand correctly, you convert them to 32 bit fixed first, and then go to 24. correct?

how about sample rate conversion, fixed vs. float? do you see any advantages with either? thanks.


right.

[bob katz]

Originally Posted by oky****

hi,

ah ha!

if i understand correctly, you convert them to 32 bit fixed first, and then go to 24. correct?

how about sample rate conversion, fixed vs. float? do you see any advantages with either? thanks.


right.

The right approach (if your DAW permits it) is to stay in float for all your operations (including SRC) until the very end. My system permits auditioning 16-bit dither while working all the processing in real time without committing to it as well. We do offline sample rate conversion using Weiss Saracon and generally I use Saracon's output in 32 bit floating point when reinserting into my DAW's EDL. Saracan can also SRC and dither in one step and can be used as an excellent format converter as it understands and outputs just about every file type known.

[okydoky]

Originally Posted by bob katz

The right approach (if your DAW permits it) is to stay in float for all your operations (including SRC) until the very end. My system permits auditioning 16-bit dither while working all the processing in real time without committing to it as well. We do offline sample rate conversion using Weiss Saracon and generally I use Saracon's output in 32 bit floating point when reinserting into my DAW's EDL. Saracan can also SRC and dither in one step and can be used as an excellent format converter as it understands and outputs just about every file type known.

hi,

thanks for the information, bob. if you're using the weiss stuff, i'm sure he's got it handled.

but i still don't understand exactly where you are converting 32 bit floating point to 32 bit fixed point, if that is what you are doing. did i misunderstand?

i'm going to post a quote from an article comparing fixed vs. float. there seem to be a couple of other issues as well, such as the dither thing.

to be honest, i am not so sure that floating point is such a great idea compared to high bit depth fixed point. i'm sure a good-sounding result can be obtained with both if one knows ones way around, so i'm not exactly up in arms about all this, but i am interested in the whole story, you know. i have both types of daws.

The following might be interesting to you or others. You may have already seen it. Its from an "Electronic Musician" magazine :


"Both numbering systems pre-sent unique challenges and require unique solutions for handling overflow, resolution, and other quirks. The sound that a particular DAW has is due in part to the skills of the programmers who wrote the algorithms that decide how signals are summed, scaled, and otherwise manipulated internally. Given sufficient resolution, however, the differences between the two fade into relative obscurity. Modern DAWs often use double-precision arithmetic, meaning that fixed-point systems allow 48 bits to process 24-bit signals and that some floating-point systems allow 64 bits to handle 32-bit signals.

The other big distinction between the two systems is how they handle quantization error. Quantization error is the unavoidable result of the finite resolution of digital samples. When the measured voltage at an A/D converter or the result of a processing algorithm is between two numbers, you must round the result up or down to the system's resolution. Rounding very small fixed-point numbers (representing very quiet signals) results in very large errors relative to the overall signal level: rounding from 4.5 up to 5 is a 10 percent error! The louder the signal, the less significant quantization error becomes: rounding from 99.5 up to 100 is a 0.5 percent error and rounding from 999.5 up to 1,000 is only 0.05 percent.

Things are different, however, with floating-point numbers. Rounding is still more problematic with very small significand values than with large significand values. But each time that the exponent increases or decreases, the pattern starts over, just as the rising pitch of an engine drops and restarts every time you shift up. Quantization error no longer fades into obscurity with rising levels. It can be quite large in a loud floating-point signal compared with an equivalent fixed-point signal, but because the signal is loud, how apparent will the quantization distortion be? Some observers suggest that the absolute level of the quantization distortion is less significant than the fact that its level “pumps” with each change in exponent.

Resolution

The advantages of either system diminish with increased resolution, when more bits are available to buffer overflow and retain resolution. To put everything in perspective, you will get much more mileage from wise mic selection and placement than obsessing over the differences between fixed- and floating-point systems, and neither type of arithmetic will save a bad song."


right.

[It'sJoeAgain]

Enough already!! If i correctly understood all of uu's in the last dozen friggin' dithering threads or so , there is hardly any need for dithering working at 24 bit and certainly is useless working with 32 bit floating point and 48 bit fixed DAW's..yeah we all like to dither to avoid quantization distortion aight but the discussion as to why u'd want to use dither working with HD systems in the first place is good for someone who wants to record an ant's fart to then normalize it...we are talking sh*t (sound particles) at the atomic level and in reality serves no practical purpose ....

[bob katz]

Originally Posted by oky****

hi,

thanks for the information, bob. if you're using the weiss stuff, i'm sure he's got it handled.

but i still don't understand exactly where you are converting 32 bit floating point to 32 bit fixed point, if that is what you are doing. did i misunderstand?

I didn't mean to say it if I did! Ideally you want to remain in floating point until the last step. For example, I receive a 24 bit 44.1 k file from a client. Let's say I want to upsample it to 96k to go into an EDL that has other 96k material in it so I can master all at the same sample rate, or just to upsample because I believe that I'll get better results that way. I upsample it to 96 kHz 32 bit float. This preserves the entire calculation. I stay in floating point for as long as possible, which isn't for very long, admittedly, since I use outboard analog and digital processors, so it isn't long before it gets dithered to 24 bits on its way out the system for external processing. So I'm a commitophobe :-). The principle is there, though, storage at 32 bit float preserves the calculations that come out of the SRC in their purest form, and if I apply any EQ or other processing on it and want to keep an intermediate file product, it should ideally be 32 bit float.

I am already very aware of the issues of 32 bit floating point versus high depth fixed. Andy Moorer wrote the definitive article on that subject years ago. But if your DAW is "only" native at 32 bit float, then that is the most efficient wordlength that you should work in and it has its advantages (it will never overload until/unless you convert to fixed). If your DAW is 48 bit fixed point (which is very rare, for reasons I won't get into here) then you should insist that DAW manufacturer make the 48 bit product available and useable as an input and output file. Then and only then can you take full advantage of the intermediate product and store the intermediate product.

Someday all the floating point DAWs will find they have so much idle CPU power sitting around they'll start working in double precision (64 bit) float and give Andy Moorer's 48 bit fixed a run for the money :-). But most of the time, unless you are doing complex calculations, the transparency advantage over 32 bit float will not be audible. I do agree that high quality equalizers like the Algorithmix Red and Blue internally calculate at higher precision and this results in a very transparent sound. But their output product is 32 bit float and they don't seem to suffer. So we are in a situation currently where many processes can produce lower distortion by internally calculating at a higher precision, but cannot communicate to their plugin brethren without reducing the wordlength. Someday we may learn if this has any sonic advantage. I personally doubt it.


Best wishes,


Bob

[okydoky]

Originally Posted by bob katz

I didn't mean to say it if I did! Ideally you want to remain in floating point until the last step. For example, I receive a 24 bit 44.1 k file from a client. Let's say I want to upsample it to 96k to go into an EDL that has other 96k material in it so I can master all at the same sample rate, or just to upsample because I believe that I'll get better results that way. I upsample it to 96 kHz 32 bit float. This preserves the entire calculation. I stay in floating point for as long as possible, which isn't for very long, admittedly, since I use outboard analog and digital processors, so it isn't long before it gets dithered to 24 bits on its way out the system for external processing. So I'm a commitophobe :-). The principle is there, though, storage at 32 bit float preserves the calculations that come out of the SRC in their purest form, and if I apply any EQ or other processing on it and want to keep an intermediate file product, it should ideally be 32 bit float.

I am already very aware of the issues of 32 bit floating point versus high depth fixed. Andy Moorer wrote the definitive article on that subject years ago. But if your DAW is "only" native at 32 bit float, then that is the most efficient wordlength that you should work in and it has its advantages (it will never overload until/unless you convert to fixed). If your DAW is 48 bit fixed point (which is very rare, for reasons I won't get into here) then you should insist that DAW manufacturer make the 48 bit product available and useable as an input and output file. Then and only then can you take full advantage of the intermediate product and store the intermediate product.

Someday all the floating point DAWs will find they have so much idle CPU power sitting around they'll start working in double precision (64 bit) float and give Andy Moorer's 48 bit fixed a run for the money :-). But most of the time, unless you are doing complex calculations, the transparency advantage over 32 bit float will not be audible. I do agree that high quality equalizers like the Algorithmix Red and Blue internally calculate at higher precision and this results in a very transparent sound. But their output product is 32 bit float and they don't seem to suffer. So we are in a situation currently where many processes can produce lower distortion by internally calculating at a higher precision, but cannot communicate to their plugin brethren without reducing the wordlength. Someday we may learn if this has any sonic advantage. I personally doubt it.


Best wishes,


Bob

hi,

well, from what i can grok, i don't see any real advantage in 32 bit floating point over the digi 24 bit with 48 bit double precision calculations thing. or they each have their own strengths, or something.

i agree that either 100% 48 bit fixed or 100% 64 bit float would be happening, but since the aes and converters are going to be fixed point, i think the hammer comes down on the fixed point side.

i think weiss uses 40 bit fixed for dsp in his converters, for some reason.

it seems to me that the various daws are using whatever they are using simply because that is what the hardware requires, or makes easy, and there is at least some room for improvement all around.


right.

[bob katz]

Originally Posted by oky****

hi,

i think weiss uses 40 bit fixed for dsp in his converters, for some reason.


right.

Actually, Weiss is 40 bit float, with Sharcs. Well-engineered 32 float sounds very good. Beyond that you have shades of "excellent."

BK

[okydoky]

Originally Posted by bob katz

Actually, Weiss is 40 bit float, with Sharcs. Well-engineered 32 float sounds very good. Beyond that you have shades of "excellent."

BK

hi,

yes, your right. i thought he told me it was 40 bit fixed, but i looked at the the manual. it sure says floating point, doesn't it?

[manual (n): a booklet of instructions for using equipment, widely used before the advent of posting internet questions].


right.