Cool, before we get carried away lets talk no-nonsense 24 bit INL and DNL with no gain ranging or other sleights of hand. That is a straight line from 1V to -1V with no deviation greater than 1/2^24 Volts. Instantaneous no averaging allowed, which means p-p noise in the signal bandwidth. That figures out to about 71 pV/rt-Hz not bad, I want one of those.
Last edited:
These guys have the highest DR ADC that, AFAIK, doesn't use gain ranging.
It achieves (specs) a full 24 bit DR at 48k.
They are an interesting company that has a mix of seemingly cutting edge
digital design with some traditional hi-end subjectively driven design.
An example is the abundant use of open loop circuitry. Even their power
amp is void of global or inter stage FB, so it won't win any numbers
races. Gets great reviews though.
Not the case, but it's irrelevant anyway.
I presume they would only get those numbers in low gain IP mode, 0dBFS =
9.3V RMS. I don't have to tell you this a bit over 4nV/rt Hz which is
achievable, if only on the test bench to verify the spec.
Regardless, if this is in fact a true spec then I say good luck to them - I'd
like a dollar for every stupid thread I've had to wade through that argued the
merits of 24 bits and the ability of ADC's to even come close.
Well I can't follow the way they keep throwing around equivalent measurement BW. That's a peak voltage of 12V or more and the answer is within 700nV at every code, thermals? They shouldn't need the hand waving.
I am not aware of any theoretical conflicts. For that matter, I am not even aware of any conflicts resulting from actual listener tests. If you can point to some conflicts, I'd love to read them.
The most I've seen is the claim that listener testing revealed no detectable advantage beyond triangular, but at the same time those tests certainly did not indicate that Gaussian or other dither of a higher order (>2) was worse than triangular. For those specific tests, my first reaction was to intuit that future tests on better equipment might make it possible to discern finer details that might put higher-order dither at an advantage. In other words, my take away was indications that people can clearly hear problems when there is less than 2nd-order dither, but above 2nd-order dither the results seemed inconclusive.
Do you have links to specific articles from Widrow, et al?
Ed,
I see Pavel is already volunteering, may I say Me2? I am looking at a working on a line of interesting pro stuff, so this interests me immensely.
Ciao T
I see Pavel is already volunteering, may I say Me2? I am looking at a working on a line of interesting pro stuff, so this interests me immensely.
Ciao T
Hi,
If by REAL we mean that they measure 24 Bit ENOB or more, no.
The ESS Sabre currently leads the field with around 22 Bit ENOB on the DAC Side. Plus, I kind of take exception to the use of lowish bit parts with dither to get there, though some disagree.
On a hardware level the ESS Sabre has enough hardware resolution to handle 16/44 directly (no noise shaping), which is probably why it reportedly sound so good and is so much better than competitors chips. They have given the Chip enough real resolution to take on old multibit chips, but the penalty is a very large die and a cost that makes you think of buying a pair of K Grade PCM1704 instead, if you are me...
If we mean with real 24 Bit's implemented in a multibit architecture, there is the TI/BB PCM1704, however it's ENOB is more like 18.5 Bits. You can theoretically parallel several and implement "analogue oversampling" into the bargain, with 16 pcs in parallel you should get 20.5Bit ENOB without any mathematical tricks.
Then there is the MSB DAC, but last time I looked you could not buy it (internal use by MSB only) which is what we called a hybrid integrated circuit in the old days and the long announced Arda Tech DAC, which to this day appears to be lacking shipping silicone.
If you are lucky, the PCM63 and AD1862 can still be found despite being long discontinued. Both are "only" 20 Bit but they offer close to 20 Bit Enob and as they can run at 8 or 16 Times Oversampling one could combine them with a PWM Modulator to gain 8 or 16 more levels, which is 3 or 4 extra Bits...
Ciao T
If by REAL we mean that they measure 24 Bit ENOB or more, no.
The ESS Sabre currently leads the field with around 22 Bit ENOB on the DAC Side. Plus, I kind of take exception to the use of lowish bit parts with dither to get there, though some disagree.
On a hardware level the ESS Sabre has enough hardware resolution to handle 16/44 directly (no noise shaping), which is probably why it reportedly sound so good and is so much better than competitors chips. They have given the Chip enough real resolution to take on old multibit chips, but the penalty is a very large die and a cost that makes you think of buying a pair of K Grade PCM1704 instead, if you are me...
If we mean with real 24 Bit's implemented in a multibit architecture, there is the TI/BB PCM1704, however it's ENOB is more like 18.5 Bits. You can theoretically parallel several and implement "analogue oversampling" into the bargain, with 16 pcs in parallel you should get 20.5Bit ENOB without any mathematical tricks.
Then there is the MSB DAC, but last time I looked you could not buy it (internal use by MSB only) which is what we called a hybrid integrated circuit in the old days and the long announced Arda Tech DAC, which to this day appears to be lacking shipping silicone.
If you are lucky, the PCM63 and AD1862 can still be found despite being long discontinued. Both are "only" 20 Bit but they offer close to 20 Bit Enob and as they can run at 8 or 16 Times Oversampling one could combine them with a PWM Modulator to gain 8 or 16 more levels, which is 3 or 4 extra Bits...
Ciao T
Pavel,
Merci beucoup, Thank you, Danke Schoen, Děkuji, Spasiba, Mille Gracie, Domo arigato!
Now what is all this "fuzz" we are seeing?
I know what the defenders of the orthodox faith will say, it is NOT Fuzzy Distortion (as such does not exist).
But it is dither and by use of a process that is a Genuine Miracle (TM) it makes this 12 Bit Device behave like it was a 32 Bit one. Well, I guess enough faith can make even a 12 Bit (actually even a 1 Bit) DAC into a 24 Bit one...
Ciao T
Definitely, the best reasoning would be the captured transient, rather than walking and laughing. Everyone is able to walk and laugh
Or just let's show something like this:
Merci beucoup, Thank you, Danke Schoen, Děkuji, Spasiba, Mille Gracie, Domo arigato!
Now what is all this "fuzz" we are seeing?
I know what the defenders of the orthodox faith will say, it is NOT Fuzzy Distortion (as such does not exist).
But it is dither and by use of a process that is a Genuine Miracle (TM) it makes this 12 Bit Device behave like it was a 32 Bit one.
Well, I guess enough faith can make even a 12 Bit (actually even a 1 Bit) DAC into a 24 Bit one...Ciao T
Now what is all this "fuzz" we are seeing?
Looks to me like the absence of an anti-imaging filter, but PMA didn't really provide any information on what he was measuring and how he was doing so, so it didn't invite much comment.
Scott,
Let us posit a microphone with a 1" Capsule and a second co-incident 1/4" Capsule (I know, no such exist, but it is possible to make such) so we can get 10dBA self noise, around 134dB Maximum SPL, 10mV @ 94dB sensitivity and using the classic Neumann Solid State circuit around 50 Ohm Differential Z Out.
Based on my trusty Johnson (noise) Calculator from Sengpielaudio this gives around -138dBV self noise for the microphones impedance and around 0dBV for maximum SPL.
So if the ADC is set to 1V full scale we can easily get these theoretical SNR, except for the around 10dBA noise contributed by our microphone.
Then again, maybe they use a ribbon mike with S&B TX-103 style transformers, if they do they can easily get such noise figures...
To make this whole more practical BTW, do you know any off the shelf vendor of good 1" or 1/2" low cost capsules that sound decent? I do feel like building a few Microphones, though it may be best to just buy a bunch of Behringer B2's...
Ciao T
Let us posit a microphone with a 1" Capsule and a second co-incident 1/4" Capsule (I know, no such exist, but it is possible to make such) so we can get 10dBA self noise, around 134dB Maximum SPL, 10mV @ 94dB sensitivity and using the classic Neumann Solid State circuit around 50 Ohm Differential Z Out.
Based on my trusty Johnson (noise) Calculator from Sengpielaudio this gives around -138dBV self noise for the microphones impedance and around 0dBV for maximum SPL.
So if the ADC is set to 1V full scale we can easily get these theoretical SNR, except for the around 10dBA noise contributed by our microphone.
Then again, maybe they use a ribbon mike with S&B TX-103 style transformers, if they do they can easily get such noise figures...
To make this whole more practical BTW, do you know any off the shelf vendor of good 1" or 1/2" low cost capsules that sound decent? I do feel like building a few Microphones, though it may be best to just buy a bunch of Behringer B2's...
Ciao T
Yep, if by 'music' you mean the majority of electronically regenerated music today. Fuzz/sibilance is par for the course.
Those guys don't have anything near a 24bit converter - they claim 124dB in 25.6kHz bandwidth. They get better only by going to a narrower bandwidth, looking like they use synchronous averaging. Been there, done that, back in the 1980s.
Definitely, the best reasoning would be the captured transient, rather than walking and laughing. Everyone is able to walk and laugh
Or just let's show something like this:
It may well be that a multibit converter does not show this type of behaviour in its tranfer function . . . but for a bitstream converter you have to include the filter to get the final result. This is like trying to compare two amplifers -with one of them missing the output stage.
- Status
- Not open for further replies.