LP with better dynamic range than digital

[esldude]

Don,
I would like the tutorial because I understand how it is possible when already in the digital domain and with small signals, but not specifically when the 16bit ADC is converting a complex soundwave (say symphony crescendo of an orchestra) that has true dynamics of over 100db.
The problem is the PCM stream must rely upon n-bits encoding, not an issue until you are talking about huge amplitude complex sounds with many harmonics; such requirements are pretty small as I have only seen a few hirez records reviewed-measured that have been measured with the requirements of needing over 100db (if taking ALL the frequency range into consideration up to say 15khz).

That said this is something none of us should be losing sleep over because IMO if differences are being picked up it has nothing to do with extreme dynamic range potential of LPs/CD/24-bit.
IMO of course
Cheers
Orb

Don may explain it better than I. But you can have both a high level complex wave and the small below -96 db sound encoded at the same time. If you look at the FFTs you would still see the lower level tones or other sounds show up even while there is a big signal there. ADCs with quiet enough electronics and given such a signal would encode it without problem. Now upon listening you won't know whether it is there or not because the loud sound will mask the tiny little sounds below a certain size. What you are imagining is actually no problem for a good ADC to do.

 

[Orb]

I can give a real world PCM audio example.
If it was that simple then the telephony world would not had created mu-law and a-law; a way to increase dynamic range by using companding algorithm.
Appreciate the design concept of this goes back a fair few decades but added noise (dither) was understood even before then with telecom transmission theory.
Cheers
Orb

 

[FrantzM]

Orb

IIRCTcompanding was for increasing SNR not, dynamic range. . In Telephony you have noise on the lines and it is far from constant. Designing extremely low-noise electronics would have been costly. We do not share the same problems in an ADC for music encoding. So not sure your example applies here.

 

[esldude]

I can give a real world PCM audio example.
If it was that simple then the telephony world would not had created mu-law and a-law; a way to increase dynamic range by using companding algorithm.
Appreciate the design concept of this goes back a fair few decades but added noise (dither) was understood even before then with telecom transmission theory.
Cheers
Orb

The problem is the noise. The signal is there it is encoded in the noise. We can hear into the noise some, instruments can "hear" further. But for conversation over a phone such a noisy though understandable signal wouldn't be desired. Due to our inability to hear into noise much more than 20 db you would say we gain perhaps 20 db below -93 db noise floor. You could say you don't want to hear into noise and the floor of CD is -93 db. But doing the same for LP you would lose some usable dynamic range as well.

 

[Orb]

Don may explain it better than I. But you can have both a high level complex wave and the small below -96 db sound encoded at the same time. If you look at the FFTs you would still see the lower level tones or other sounds show up even while there is a big signal there. ADCs with quiet enough electronics and given such a signal would encode it without problem. Now upon listening you won't know whether it is there or not because the loud sound will mask the tiny little sounds below a certain size. What you are imagining is actually no problem for a good ADC to do.

I have only ever seen examples with simple sinewave (tone) and not at full amplitude meaning low signal, which I agree with all you and Don has said although I would not really say it is explicitly increasing dynamic range (just semantic difference I think we have there).
Doing the same with a complex soundwave with fundamental+harmonics covering 20hz to say 15khz that also utilises the full amplitude PCM bit depth of the ADC is something I have not seen done in practice for the reason I mentioned.

Cheers
Orb

 

[Orb]

Orb

IIRCTcompanding was for increasing SNR not, dynamic range. . In Telephony you have noise on the lines and it is far from constant. Designing extremely low-noise electronics would have been costly. We do not share the same problems in an ADC for music encoding. So not sure your example applies here.

Sorry sort of but remember it is to take 14 bits ADC and reduce it to 8bits to combine with 8khz sampling, giving us the 64khz channels.
It still involves dynamic range, although we have recently been blurring boundary as the talk has been about dynamic range AND SNR very recently.
Cheers
Orb

 

[Orb]

Orb

IIRCTcompanding was for increasing SNR not, dynamic range. . In Telephony you have noise on the lines and it is far from constant. Designing extremely low-noise electronics would have been costly. We do not share the same problems in an ADC for music encoding. So not sure your example applies here.

Just to add the G.711 was not that noisy as these were designed for digital network carrier E1 and T1 associated digital circuits and not the same as the traditional residential telephony lines.
There were/are legal regulations in place ensuring Toll quality in many ways.

Edit:
Frantz sent you a PM to go into a bit more detail offline as this is going off tangent to the point of dynamic range/noise floor/n-bit word length encoding.
Thanks

 

[Don Hills]

Don,
I would like the tutorial because I understand how it is possible when already in the digital domain and with small signals, but not specifically when the 16bit ADC is converting a complex soundwave (say symphony crescendo of an orchestra) that has true dynamics of over 100db.

Here's one I prepared earlier: It doesn't exactly answer your question, but I'll get to that in a minute.
Note that it uses a 4 bit ADC for simplicity, but works the same for more bits.

http://www.whatsbestforum.com/showt...e-than-digital&p=290823&viewfull=1#post290823

Now, your question: It works exactly the same for complex, loud signals as well. Since an ADC works by sampling the input analogue signal at discrete time intervals, it may help to think of the operation in the time domain. Regardless of the complexity (number of frequencies) and amplitude of the input signal, at any one instant in time it will be represented by a certain voltage. A snapshot. Like a single frame from a movie. All the ADC has to do is produce the digital value that represents that voltage. It doesn't matter whether we're digitising a single sine wave at -100 dB, or a tiny harmonic of a violin note that's 100 dB down from the orchestral crescendo, it will be treated the same by the ADC.

The problem is the PCM stream must rely upon n-bits encoding, not an issue until you are talking about huge amplitude complex sounds with many harmonics; such requirements are pretty small as I have only seen a few hirez records reviewed-measured that have been measured with the requirements of needing over 100db (if taking ALL the frequency range into consideration up to say 15khz).

That said this is something none of us should be losing sleep over because IMO if differences are being picked up it has nothing to do with extreme dynamic range potential of LPs/CD/24-bit.
IMO of course ...

As you say, if you can hear violin harmonics that are 100 dB down compared with the crescendo they form part of, you have better hearing than the average listener. Assuming you are listening to that crescendo at 120 dB SPL, definitely in pain territory, those 100 dB down signals might just be above the noise floor in your room. But you can't hear a pin drop when you're standing next to a jackhammer...

Any "analogue" medium (tape, LP etc) has a noise floor and a maximum practical recording level. The difference between these levels is usually called the "dynamic range". Very quiet signals start to get "buried in the noise" and overly loud signals get distorted (clipped). Hopefully that is well understood.
Now, the crucial point:
Properly dithered digital recording works exactly the same way.

 

[ack]

Mathematically? Yes. (But I hasten to add, I take the word of mathematicians on this, I'm poor at math.)

You appear to have misread his explanation.

Item 1: A 16-bit ADC with dither can encode signals of amplitude lower than that represented by the LSB. The signal never fades away as the level gets lower, it just gets buried in the noise. Exactly like analogue. Do you accept that, or shall I post a short tutorial?

Yes, we would like to see the tutorial, although I think I know what I will see: either a) severe band limiting (as you explain right below); or b) severe dither shaping, as esldude showed in his follow-up post, included below for clarity - in his custom noise-shaping example, the vast majority of noise is pushed toward 20kHz and it seems to be reaching 0dB - UGH. So, OK, yes if you manipulate things enough, sure, you can encode a -130dB test tone well below the noise band... BUT... as Orb says, how will you be able to encode such a low level signal ALONG WITH a high-enough one as in a symphonic program *at the same time*... The problem here is that we are talking about dynamic range of CD vs LP and the assumption is that we actually encode music, not test tones. Therefore, when Benchmark or you guys claim we can reach down to -140dB or whatever *somehow*, it's just not a realistic depiction of encoded *musical events*.

The question then becomes: what's the best dynamic range of CD when encoding an entire musical program, and what is the dynamic range of the LP under the same conditions. Therefore, although all the efforts to show how you can encode a -130dB tone are well-meaning, accurate and all, the solutions offered (bandlimiting or severe noise shaping) are just simply unrealistic for encoding musical programs.

Your band-limiting example:

Item 2: 16-bit dithered digital has a noise level of about -93 dB (a bit lower if we use shaped dither). 16 bit implies -96dB, but we add +-0.5 bit of dither. The important thing is that this is broadband noise (0 to fs/2, or 0 to 22 KHz for 44.1 KHz digital.) Now we record a 3 KHz tone at, say, -130 dB. It might appear that this is well below the noise floor and should be inaudible. But remember that the noise is spread across the whole 0 to 22KHz range. At any given frequency, the noise will be much lower. For example, if we filter out everything at the DAC output below, say, 2.9 KHz and above 3.1 KHz, our bandwidth decreases by a factor of 100. (20,000 / (3100 - 2900) = 100.) The amount of noise in that band is also 1/100 of the -93dB figure. If I've done my math right, this reduces the noise floor in that narrow band to the vicinity of -130 dB. So our -130 dB tone should be quite visible and audible in the filtered output.

Now our auditory system comes into play. Our ears have the equivalent of filters. They break up the range into bands and analyse the energy in each band, just like the filter example above. We also have discrimination: We can pick out signals that "don't belong". We can pick out the 3 KHz tone because it doesn't sound like the noise. We're hard wired to do this. You may have noticed if you listen attentively to white noise, after a while you may start to hear fragments of other signals (voices etc) in it.

Yes, I am quite familiar with these auditory behaviors in humans.

Item 3: When he said "amplified", he meant that the output of the DAC has to be amplified to a level where the noise (and tone) becomes audible, not that the tone had to be amplified before encoding.

Yes, I misinterpreted that.



@esldude:

Next is where I digitally created a 3 khz tone at -135 db. I saved that file as a 16 bit file with shaped dither. Then reopened it for this FFT. You see the noise floor with most of the noise at upper frequencies. You also should see a thin line at 3khz. I could have used a large FFT bin size to bring it out more, but the display of the picture makes it hard to see. I also zoomed in on the next picture to show it better.
View attachment 17956View attachment 17957

So though this is a 16 bit file with dither you can encode and recover signals below the -96 db noise floor. Now even amplified I could not hear this. However at -115 db for the tone I can. At -120 for the tone I can still hear it faintly. In both cases I need to amplify it 50 db which means I hear a low pitched whooshing noise and a tone buried in it.

Of course the point isn't we hear these effects without amplifying, but merely to show with dither digital has the ability to record below the basic noise floor of the least significant bit. Since our hearing splits things into bands as Don Hills explained we can hear into a noise floor somewhere between 10-20 db if the noise floor is mostly flat. With some noise shapes we can hear further into the noise.

So as I mentioned above, your noise shaping is customized for this particular demonstration; but won't apply to encoding an entire musical program. So let's pick a shaped dither used universally and stick with it, as we carry the discussion forward - is TPDF the one we all agree on? UV22?? Pick one. Regardless, thanks for the demonstration, albeit not at the practical level (e.g. live music). I assumed that when you said you could hear a -130dB tone that you used a typical dithering algorithm used for every day recordings and that dynamic range was calculated over the entire 0-fs/2 range.

 

[Orb]

Don, although that is theory and does not take real world considerations such as the actual PCM fixed word length encoding involving multiple hardware, which negates the fact of infinite output voltage level values-resolution.
Same way in theory regarding ideal transfer function,quantisation, dither and infinite precision; as I mentioned much earlier this cannot happen with real world PCM audio solutions.
Most theory examples (I have posted done this as well many times ) do not include the real world chain and all integral functions.
The DAC voltage correlates to a fixed PCM word length representation encoding, which in your example seems to not be shown and why Dynamic Range and SNR and THD+N are explicitly defined by ADC/DAC manufacturers
BTW I get what your saying regarding noise and dither associated with the lsb (hence why noise shaping improves SNR at the cost of it somewhere else); one system that did this for audio without using dither is Sony Super Bit Mapping to give CDs comparable SNR to 20bit and this also relies upon noise shaping.

Well properly dithered digital is to improve quantisation error/distortion and low signal precision IMO
Just so we are on the same page; when you mention properly dithered do you mean amount-level of dither applied in context of lsb or mean noise shaped dither or both, or mixture of that using the right dither type (not noise shaped) such as flat TPDF?

One aspect we should consider is the theory debate and difference between perceived dynamic range increased and the product's/technology dynamic range.
Some of what we have touched upon recently relates more to perceived dynamic range.

Sorry for delay and time taken in posting (so will be out of synch to some newer posts by others), very late here and decided to peruse some engineer papers while also double checking Xiph's approach.
Also I thought of someone on this forum who might be able to do a real world test for us, will ping them as it would be interesting either way.
Cheers
Orb

 

[Don Hills]

Yes, we would like to see the tutorial, although I think I know what I will see: either a) severe band limiting (as you explain right below); or b) severe dither shaping, as esldude showed in his follow-up post, included below for clarity - in his custom noise-shaping example, the vast majority of noise is pushed toward 20kHz and it seems to be reaching 0dB - UGH. ...

See my post #268 above. Is there any part of it you do not understand, or disagree with?

Regarding noise shaping: If a tree falls in a forest, and no-one is around to hear it, did it make a sound?
Did you listen to the 4-bit audio samples I posted a link to a few days ago?

The question then becomes: what's the best dynamic range of CD when encoding an entire musical program, and what is the dynamic range of the LP under the same conditions. Therefore, although all the efforts to show how you can encode a -130dB tone are well-meaning, accurate and all, the solutions offered (bandlimiting or severe noise shaping) are just simply unrealistic for encoding musical programs. ...

Digital: (16/44.1) about 93 dB dynamic range, 0 to 22 KHz. It doesn't change regardless of whether its a single tone or a full orchestra. This assumes TPDF, whereas a shaped dither is most commonly used for final distribution. (Mastering engineers can be just as obsessive as audiophiles, they have been known to argue over the best shaped dither to use for different music genres...)

LP: About 60 to 80 dB, 0 to 22 KHz. Sometimes more if exceptional care has been taken in the mastering and pressing. As has been pointed out, a master (cut acetate) can be very quiet. The main noise contribution is the pressing process. If you try to push the levels to increase the dynamic range the quality starts to fall apart due to incipient mistracking and inner groove distortion.

 

[Don Hills]

Don, although that is theory and does not take real world considerations such as the actual PCM fixed word length encoding involving multiple hardware, which negates the fact of infinite output voltage level values-resolution. ...

You will have to explain that more clearly. What "multiple hardware"? Are you trying to say that operations occurring between the ADC and the DAC may negate the "infinite resolution"?

Same way in theory regarding ideal transfer function,quantisation, dither and infinite precision; as I mentioned much earlier this cannot happen with real world PCM audio solutions.
Most theory examples (I have posted done this as well many times ) do not include the real world chain and all integral functions.
The DAC voltage correlates to a fixed PCM word length representation encoding, which in your example seems to not be shown and why Dynamic Range and SNR and THD+N are explicitly defined by ADC/DAC manufacturers ...

Indeed, the DAC converter stage produces a fixed voltage for each digital sample. But this does not imply a loss of resolution. It reproduces (hopefully) exactly the voltage (relative to 0dB FS) that was digitised by the ADC (including the dither). In my 4-bit example, where a 2.1 volt input (and +-0.5 volts dither) is digitised, the output of the DAC, after the (essential) reconstruction filter, will be 2.1 volts (with +-0.5 volts noise). ADC / DAC manufacturers know exactly the specifications of a theoretically perfect ADC / DAC. Their job is to make their real devices approach these specs as closely as practical for the desired price point. And they do a pretty good job of it these days.

BTW I get what your saying regarding noise and dither associated with the lsb (hence why noise shaping improves SNR at the cost of it somewhere else); one system that did this for audio without using dither is Sony Super Bit Mapping to give CDs comparable SNR to 20bit and this also relies upon noise shaping.

Sony SBM is a specific form of noise shaped dither - no more, no less.
http://pdf.textfiles.com/manuals/ST...o/Manuals/White Paper - Super Bit Mapping.pdf

Well properly dithered digital is to improve quantisation error/distortion and low signal precision IMO
Just so we are on the same page; when you mention properly dithered do you mean amount-level of dither applied in context of lsb or mean noise shaped dither or both, or mixture of that using the right dither type (not noise shaped) such as flat TPDF? ...

"It depends". No dither, or insufficient dither (of whatever shape), results in signal-correlated distortion. Excessive dither results in increased noise floor (reduced dynamic range). So "properly dithered" is the dither that is just sufficient to replace the quantisation distortion with the dither signal.

 

[esldude]

Yes, we would like to see the tutorial, although I think I know what I will see: either a) severe band limiting (as you explain right below); or b) severe dither shaping, as esldude showed in his follow-up post, included below for clarity - in his custom noise-shaping example, the vast majority of noise is pushed toward 20kHz and it seems to be reaching 0dB - UGH. So, OK, yes if you manipulate things enough, sure, you can encode a -130dB test tone well below the noise band... BUT... as Orb says, how will you be able to encode such a low level signal ALONG WITH a high-enough one as in a symphonic program *at the same time*... The problem here is that we are talking about dynamic range of CD vs LP and the assumption is that we actually encode music, not test tones. Therefore, when Benchmark or you guys claim we can reach down to -140dB or whatever *somehow*, it's just not a realistic depiction of encoded *musical events*.

The question then becomes: what's the best dynamic range of CD when encoding an entire musical program, and what is the dynamic range of the LP under the same conditions. Therefore, although all the efforts to show how you can encode a -130dB tone are well-meaning, accurate and all, the solutions offered (bandlimiting or severe noise shaping) are just simply unrealistic for encoding musical programs.

Your band-limiting example:




Yes, I am quite familiar with these auditory behaviors in humans.



Yes, I misinterpreted that.



@esldude:



So as I mentioned above, your noise shaping is customized for this particular demonstration; but won't apply to encoding an entire musical program. So let's pick a shaped dither used universally and stick with it, as we carry the discussion forward - is TPDF the one we all agree on? UV22?? Pick one. Regardless, thanks for the demonstration, albeit not at the practical level (e.g. live music). I assumed that when you said you could hear a -130dB tone that you used a typical dithering algorithm used for every day recordings and that dynamic range was calculated over the entire 0-fs/2 range.

Sorry ACK, but you misinterpreted me again. The noise shaping I used wasn't customized for this particular example. It is one of the standard ones that have been built on the hearing acuity of humans. It is the one used by Sox, I believe Adobe Audition, and some of the iZotope versions are pretty similar. And before you get too upset about the high frequency noise, notice that the very topmost line on those FFTs I posted was -90 db on the one with fewest frequency bands. For humans it functions much better than TPDF. We can hear lower signals because of its design. TPDF also has a bit of a boost to noise in high frequencies, but isn't shaped as much as this one. It is more often used when files may later see more processing because EQ changes could make shaped dither audible when TPDF would still be inaudible. If you know you aren't doing anything further shaped dither is fine.

There is nothing to prevent using shaped dither over the entire range using musical signals. I do it all the time. The very one I posted. It actually will perform better. We have better acuity thresholds around 3-4 khz, and less good acuity elsewhere. With TPDF I can't hear a minus -130, but can get around -114 or 115 db on such a tone. The effect of hearing below the -96 db 16 bit noise floor still functions.

We don't have dynamic range of even 96 db at some frequencies. Thresholds at 100 hz are some 40 db higher than at 4 khz, and some 20 db higher at 10 khz. It gets worse as you go below 100 hz and above 10 khz. For that reason shaped dither will give a better full bandwidth dynamic range for humans than TPDF. Either will give more than 96 db.

Again I point out, my FFT chart topped out at -90 db. None of the dithering was close to 0 level. If you want to stick with TPDF for comparison here is an image comparing TPDF to UV22. Notice UV22 also pushes noise to the upper range to lower it elsewhere.

http://www.playgroundstudio.com/images/apogee-uv22-dither.gif

Last post on this page, where they are discussing different types of dither is an image comparing shaped, TPDF and flat dither. You will notice at least two of them are pretty much the same as the shaped dither I used.

http://www.soundonsound.com/forum/showflat.php?Cat=&Number=979737&Main=979680

 

[Orb]

You will have to explain that more clearly. What "multiple hardware"? Are you trying to say that operations occurring between the ADC and the DAC may negate the "infinite resolution"?



Indeed, the DAC converter stage produces a fixed voltage for each digital sample. But this does not imply a loss of resolution. It reproduces (hopefully) exactly the voltage (relative to 0dB FS) that was digitised by the ADC (including the dither). In my 4-bit example, where a 2.1 volt input (and +-0.5 volts dither) is digitised, the output of the DAC, after the (essential) reconstruction filter, will be 2.1 volts (with +-0.5 volts noise). ADC / DAC manufacturers know exactly the specifications of a theoretically perfect ADC / DAC. Their job is to make their real devices approach these specs as closely as practical for the desired price point. And they do a pretty good job of it these days.



Sony SBM is a specific form of noise shaped dither - no more, no less.
http://pdf.textfiles.com/manuals/ST...o/Manuals/White Paper - Super Bit Mapping.pdf



"It depends". No dither, or insufficient dither (of whatever shape), results in signal-correlated distortion. Excessive dither results in increased noise floor (reduced dynamic range). So "properly dithered" is the dither that is just sufficient to replace the quantisation distortion with the dither signal.

Sony Super Bit Mapping is not dither but a noise shaping solution
I would need to do a search but it has been mentioned in a few forums by other engineers such as Bruno Putzeys.
By multiple hardware I am talking the whole chain and functionality that is the context for this thread in terms of audio we play back either from LP or CD or hirez PCM or DSD.
Regarding voltage output just to clarify I did not imply loss of precision, I implied the n-bit word coding must be correlated to voltage output so it is not possible to have a dynamic range beyond full amplitude as the ADC and DAC must support PCM n-bit word length coding implementation.

Thanks for clarifying proper dither and yeah we are in agreement (although we should mention relation to lsb), although proper dither also means not using shaped dither at the ADC or mixing for audio.
Cheers
Orb

 

[Orb]

Fun link (but does not answer really what we have been discussing over the last few pages) has real world graphs-measurements comparison flat dither,noise shaping, Super Bit Mapping, word length reduction systems.
http://audio.rightmark.org/lukin/dither/dither.htm

Cheers
Orb

 

[jkeny]

Here is what you are talking about: Care to elaborate on what you are trying to tell us here about noise modulation. I know you are big on this as a problem, just thought you could tell us what the chart is saying, and why only at that bit of level that this shows up and nowhere else. Could this be the result of the way they are attempting to measure this thing. The linearity chart is so brilliant, its awesome.

It's not my idea, I first encountered it in that Martin Mallinson video (since removed) about the ESS DAC - he being the Chief Technical Officer in ESS. In it he pointed out this issue with S-D Dacs & how ESS attempted to resolve it. In simplistic terms, I believe it has to do with the noise shaping aspects & how the low level noise becomes correlated to the data - in other words it varies from audibly benign white noise. This seems to be amplitude dependent based on the graphs in the Weiss measurements - & other graphs produced of measured performance of ESS Dacs (obviously they haven't solved the issue completely). Opus11 (a former forum member) did a better analysis of those measurements than I but I believe what that Weiss "THD Vs level" graph shows is a bump in the D+N of 6dB as the signal level goes up from -40dB to -30dB. This is accounted for by a bum in noise as the FFT plot shows that it's very unlikely to be distortion. Could this be an artifact of the measuring technique? I have seen it reported now on a number of different graphs so probably unlikely to be a measurement artifact.

Opus11, has a working hypothesis that the lack of perceived dynamics in S-D DACs compared to true multibit DACs has to do with this noise modulation in S-D Dacs squashing the perceived dynamics..

I'm extending this thinking to the comparison of LP Vs Digital (99.9% of digital audio is now S-D based) & a possible explanation why the measured superior DNR of DACs does not translate into perceived DNR.

http://designwsound.com/dwsblog/2012/09/weiss-medea-measurements/

By the way, a sine tone is fine for exercising a device, sine tones one at a time, of all the frequencies, or all at once, as long as the amplitude stays within the linearity of the device, no issues, as far as the test itself. But there are others, such as group delay, etc, that when all combined can characterize a device.

I'm not so sure sine tones are fine for exercising a device - they lack the asymmetry & crest factor characteristics of music & as a result exercise the internal process within DAcs in a different way to real music.

 

[FrantzM]

Still not having proved that LP has a wider dynamic range than CD.... Lowered it to 100 dB, 20 dB down from the previous 120 dB claim. Still no cigar...

 

[Atmasphere]

Still not having proved that LP has a wider dynamic range than CD.... Lowered it to 100 dB, 20 dB down from the previous 120 dB claim. Still no cigar...

You missed my post about that Frantz? LP can do 100db, using agreed upon numbers. Not as much distortion as Tomelex claims, as I reduced the maximum level to +10db, which can have a dramatic effect on mechanical devices, as the excursion is reduced quite a lot.

Here is the post:

http://www.whatsbestforum.com/showt...e-than-digital&p=291183&viewfull=1#post291183

 

[ack]

@Don Hills:

See my post #268 above. Is there any part of it you do not understand, or disagree with?

I need to revisit that; the thread is moving too fast for me... can you provide links again?

So then, to clarify in a single post, I asked:

The question then becomes: what's the best dynamic range of CD when encoding an entire musical program, and what is the dynamic range of the LP under the same conditions. Therefore, although all the efforts to show how you can encode a -130dB tone are well-meaning, accurate and all, the solutions offered (bandlimiting or severe noise shaping) are just simply unrealistic for encoding musical programs. ...

to which you responded:

Digital: (16/44.1) about 93 dB dynamic range, 0 to 22 KHz. It doesn't change regardless of whether its a single tone or a full orchestra. This assumes TPDF, whereas a shaped dither is most commonly used for final distribution. (Mastering engineers can be just as obsessive as audiophiles, they have been known to argue over the best shaped dither to use for different music genres...)

LP: About 60 to 80 dB, 0 to 22 KHz. Sometimes more if exceptional care has been taken in the mastering and pressing. As has been pointed out, a master (cut acetate) can be very quiet. The main noise contribution is the pressing process. If you try to push the levels to increase the dynamic range the quality starts to fall apart due to incipient mistracking and inner groove distortion.

Thank you for a CLEAR and complete answer to both. But two more succinct questions:

1) having just said the CD's dynamic range is 93dB over 0-22kHz using TPDF dithering, do you believe THIS 16-bit system can still encode any signal in the [-100dB to -130dB] range, or if not that range, what range would you say? (I think you answered this upthread, but I am looking for a clear Yes/No answer again, if I may, along with a clear range)

2) Now assuming a "shaped dither is most commonly used for final distribution" as you said, can you provide a similar clear and complete answer to my original question relating to dynamic range and the follow-up question relating to ability to encode in the [-100db to -130dB] range, including the commonly used noise-shaping algorithm in the industry that would give you that?



@esldude:

Sorry ACK, but you misinterpreted me again. The noise shaping I used wasn't customized for this particular example. It is one of the standard ones that have been built on the hearing acuity of humans. It is the one used by Sox, I believe Adobe Audition, and some of the iZotope versions are pretty similar.

OK, let's be more specific about all these details next time, because it's easy for anyone to start drifting...

And before you get too upset about the high frequency noise, notice that the very topmost line on those FFTs I posted was -90 db on the one with fewest frequency bands.

Frankly, I really couldn't tell the scale, because the pictures were too small. Can you post larger versions?

There is nothing to prevent using shaped dither over the entire range using musical signals. I do it all the time. The very one I posted. It actually will perform better. We have better acuity thresholds around 3-4 khz, and less good acuity elsewhere. With TPDF I can't hear a minus -130, but can get around -114 or 115 db on such a tone. The effect of hearing below the -96 db 16 bit noise floor still functions.

We don't have dynamic range of even 96 db at some frequencies. Thresholds at 100 hz are some 40 db higher than at 4 khz, and some 20 db higher at 10 khz. It gets worse as you go below 100 hz and above 10 khz. For that reason shaped dither will give a better full bandwidth dynamic range for humans than TPDF. Either will give more than 96 db.

Again I point out, my FFT chart topped out at -90 db. None of the dithering was close to 0 level. If you want to stick with TPDF for comparison here is an image comparing TPDF to UV22. Notice UV22 also pushes noise to the upper range to lower it elsewhere.

First of all, I understand everything you just said, but I highlighted a key point I want to press one... so here it goes: can you answer the same questions (my original and new follow-ups now) I asked of Don in this very post - relating to dynamic range of CD (and LP if you want), in the same clear and complete fashion as he has?

Thanks

 

[esldude]

@Don Hills:



I need to revisit that; the thread is moving too fast for me... can you provide links again?

So then, to clarify in a single post, I asked:


to which you responded:



Thank you for a CLEAR and complete answer to both. But two more succinct questions:

1) having just said the CD's dynamic range is 93dB over 0-22kHz using TPDF dithering, do you believe THIS 16-bit system can still encode any signal in the [-100dB to -130dB] range, or if not that range, what range would you say? (I think you answered this upthread, but I am looking for a clear Yes/No answer again, if I may, along with a clear range)

2) Now assuming a "shaped dither is most commonly used for final distribution" as you said, can you provide a similar clear and complete answer to my original question relating to dynamic range and the follow-up question relating to ability to encode in the [-100db to -130dB] range, including the commonly used noise-shaping algorithm in the industry that would give you that?



@esldude:



OK, let's be more specific about all these details next time, because it's easy for anyone to start drifting...



Frankly, I really couldn't tell the scale, because the pictures were too small. Can you post larger versions?



First of all, I understand everything you just said, but I highlighted a key point I want to press one... so here it goes: can you answer the same questions (my original and new follow-ups now) I asked of Don in this very post - relating to dynamic range of CD (and LP if you want), in the same clear and complete fashion as he has?

Thanks

Not sure exactly what you want here. Don's answers are correct.

If it is still about what might be heard, I'll try this approach. We are in the worlds quietest anechoic chamber. As in there is no extraneous noise. We have some horn speakers able to play up to our 120 db tolerance. The available dynamic range for us in that room is 120 db as we have set volume so a 0 db peak will be 120 db. With flat rectangular dither 16 bit playing back a very low level tone you would hear a bit of hiss at a -93 db level for the whole 20khz band. In that hiss if your hearing is relatively undamaged you would hear a 3 to 4 khz tone at -110 db or something like that. At other frequencies where our hearing is not as good you would have to raise the level of the tone to hear it. At 50 hz you would hear it only if the tone is raised to something like -70 db.

Now if you used shaped dither some frequency bands in the noise will be increased, and some lowered. The total level of noise over the entire 20 khz band would be higher. Most likely in this totally quiet environment you would hear dead silence. In the 3-4 khz range you would therefore be able to hear a -120 db tone or just barely above that if your hearing is good.

Without putting exact numbers on it if you did this same thing with an LP in vinyl as you normally can buy records the noise level will be higher. In that 3-4 khz range you probably will need a tone 20 to 30 db louder to hear it amongst the noise.