Nigel, I looked at your video again and it seems to me it's confusing as to
whether you mean 'don't dither the 24b final output' or 'don't ever dither at
24b'. You make statements several times that imply the former, but in your
discussion about 24b on all digital interfaces, sends and receives etc, you
clearly say to never dither at 24b. Several people in this thread have pointed
out the difference between intermediate stage truncation and final stage
truncation, and the fact that if truncation is done repeatedly, any distortion
spectra will continue to build. It is not noise-like, the peaks are coherent
peaks and correlated to the signal.
You don't say in the video what the processing history is for the files you are
using. If they are simple captures with no processing, they probably reflect
the additive gaussian noise present at the 20th bit in the A/D, based on
Andy's post, and are properly dithered for 24b truncation. My point is that
at the digital capture stage you have (S+N) and the amplitude distribution of
the S+N signal might be fine for 24b truncation if N is dither-like. After
various stages of digital processing including non-linear steps, the (S+N)
intermediate signal may no longer have an adequate amplitude distribution to be
truncated without 24b dither.
I think the whole subject of self dither might be better approached through FFT
measurement than by listening. Bob Katz shows an FFT of truncation spectra at
24b in his book on 'Itunes Music, Mastering for High Resolution Audio Delivery'
but he uses a generated, dithered pure tone that doesn't start with added
gaussian noise. Haven't thought about it but I can imagine extending his
approach into a research effort.
Offhand I don't know anything that would go wrong in your difference file ("
...if the error doesn't sound wrong). It's a common method for looking at
residuals.
Vicki
On Feb 8, 2015, at 6:11 PM, Nigel Redmon wrote:
>> Beyond that, Nigel raises this issue in the context of "self-dither”...
>
> First, remember that I’m the guy who recommended “always” dithering 16-bit
> (no “always” as in “alway necessary”, but as in “do it always, unless you
> know that it gives no improvement”), and to not bother dithering 24-bit. So,
> I’m only interested in this discussion for 24-bit. That said:
>
>> ...In situations where there is a clear external noise source present,
>> whether the situation is analog to digital conversion or digital to digital
>> bit depth change, the external noise may, or may not, be satisfactory as
>> dither but at least it's properties can be measured.
>
> For 24-bit audio, could you give an example of when it’s likely to not be
> satisfactory (maybe you’ve already given a reference to determining
> “satisfactory")? Offhand, I’d say one case might be with extremely low noise,
> then digitally faded such that you fade the noise level below the dithering
> threshold while you still have enough signal to exhibit truncation
> distortion, and the fade characteristics allow it to last long enough to
> matter to your ears—if we weren’t talking about this distortion being down
> near -140 dB in the first place. I’d think that, typically, you’d have
> gaussian noise at a much higher level that is needed to dither 24-bit; that
> could change with digital processing, but I think that in the usual recording
> chain, it seems pretty hard to avoid for your "analog to digital conversion”
> case.
>
> I’m still interested in what you have to say about my post yesterday (“...if
> the error doesn’t sound wrong to the ear, can it still sound wrong added to
> the music?”). Care to comment?
>
>
>> On Feb 8, 2015, at 8:09 AM, Vicki Melchior <vmelch...@earthlink.net> wrote:
>>
>> I have no argument at all with the cheap high-pass TPDF dither; whenever it
>> was published the original authors undoubtedly verified that the moment
>> decoupling occurred, as you say. And that's what is needed for dither
>> effectiveness. If you're creating noise for dither, you have the option to
>> verify its properties. But in the situation of an analog signal with added,
>> independent instrument noise, you do need to verify that the composite noise
>> source actually satisfies the criteria for dither. 1/f noise in particular
>> has been questioned, which is why I raised the spectrum issue.
>>
>> Beyond that, Nigel raises this issue in the context of "self-dither". In
>> situations where there is a clear external noise source present, whether the
>> situation is analog to digital conversion or digital to digital bit depth
>> change, the external noise may, or may not, be satisfactory as dither but at
>> least it's properties can be measured. If the 'self-dithering' instead
>> refers to analog noise captured into the digitized signal with the idea that
>> this noise is going to be preserved and available at later truncation steps
>> to 'self dither' it is a very very hazy argument. I'm aware of the various
>> caveats that are often postulated, i.e. signal is captured at double
>> precision, no truncation, very selected processing. But even in minimalist
>> recording such as live to two track, it's not clear to me that the signal
>> can get through the digital stages of the A/D and still retain an unaltered
>> noise distribution. It certainly won't do so after considerable processing.
>> So the short
>> answer is, dither! At the 24th bit or at the 16th bit, whatever your output
>> is. If you (Nigel or RBJ) have references to the contrary, please say so.
>>
>> Vicki
>>
>> On Feb 8, 2015, at 10:11 AM, robert bristow-johnson wrote:
>>
>>> On 2/7/15 8:54 AM, Vicki Melchior wrote:
>>>> Well, the point of dither is to reduce correlation between the signal and
>>>> quantization noise. Its effectiveness requires that the error signal has
>>>> given properties; the mean error should be zero and the RMS error should
>>>> be independent of the signal. The best known examples satisfying those
>>>> conditions are white Gaussian noise at ~ 6dB above the RMS quantization
>>>> level and white TPDF noise at ~3dB above the same, with Gaussian noise
>>>> eliminating correlation entirely and TPDF dither eliminating correlation
>>>> with the first two moments of the error distribution. That's all
>>>> textbook stuff. There are certainly noise shaping algorithms that shape
>>>> either the sum of white dither and quantization noise or the white dither
>>>> and quantization noise independently, and even (to my knowledge) a few
>>>> completely non-white dithers that are known to work, but determining the
>>>> effectiveness of noise at dithering still requires examining the
>>>> statistical properties of the error signal and showing
>>
>>> th
>>>> at the mean is 0 and the second moment is signal independent. (I think
>>>> Stanley Lipschitz showed that the higher moments don't matter to
>>>> audibility.)
>>>
>>> but my question was not about the p.d.f. of the dither (to decouple both
>>> the mean and the variance of the quantization error, you need triangular
>>> p.d.f. dither of 2 LSBs width that is independent of the *signal*) but
>>> about the spectrum of the dither. and Nigel mentioned this already, but
>>> you can cheaply make high-pass TPDF dither with a single (decent) uniform
>>> p.d.f. random number per sample and running that through a simple 1st-order
>>> FIR which has +1 an -1 coefficients (i.e. subtract the previous UPDF from
>>> the current UPDF to get the high-pass TPDF). also, i think Bart Locanthi
>>> (is he still on this planet?) and someone else did a simple paper back in
>>> the 90s about the possible benefits of high-pass dither. wasn't a great
>>> paper or anything, but it was about the same point.
>>>
>>> i remember mentioning this at an AES in the 90's, and Stanley *did* address
>>> it. for straight dither it works okay, but for noise-shaping with
>>> feedback, to be perfectly legitimate, you want white TPDF dither (which
>>> requires adding or subtracting two independent UPDF random numbers). and i
>>> agree with that. it's just that if someone wanted to make a
>>> quick-and-clean high-pass dither with the necessary p.d.f., you can do that
>>> with the simple subtraction trick. and the dither is not white but
>>> perfectly decouples the first two moments of the total quantization error.
>>> it's just a simple trick that not good for too much.
>>>
>>> --
>>>
>>> r b-j r...@audioimagination.com
>>>
>>> "Imagination is more important than knowledge."
>>>
>>>
>>>
>>>
>>> --
>>> dupswapdrop -- the music-dsp mailing list and website:
>>> subscription info, FAQ, source code archive, list archive, book reviews,
>>> dsp links
>>> http://music.columbia.edu/cmc/music-dsp
>>> http://music.columbia.edu/mailman/listinfo/music-dsp
>>
>> --
>> dupswapdrop -- the music-dsp mailing list and website:
>> subscription info, FAQ, source code archive, list archive, book reviews, dsp
>> links
>> http://music.columbia.edu/cmc/music-dsp
>> http://music.columbia.edu/mailman/listinfo/music-dsp
>
> --
> dupswapdrop -- the music-dsp mailing list and website:
> subscription info, FAQ, source code archive, list archive, book reviews, dsp
> links
> http://music.columbia.edu/cmc/music-dsp
> http://music.columbia.edu/mailman/listinfo/music-dsp
--
dupswapdrop -- the music-dsp mailing list and website:
subscription info, FAQ, source code archive, list archive, book reviews, dsp
links
http://music.columbia.edu/cmc/music-dsp
http://music.columbia.edu/mailman/listinfo/music-dsp
