Looks like a digital version of the 3M Dynatrac. Not much new but sorting out the bits isn't easy.
This has 24 bit accuracy: LTC2442 - 24-Bit High Speed 4-Channel Delta Sigma ADC with Integrated Amplifier - Linear Technology but it samples at 8 KHz. And this is a real 24 bit DAC: http://us.flukecal.com/products/electrical-calibration/electrical-standards/720a-kelvin-varley-divide But the sample rate is specified in minutes. It IS all discrete.
Last edited:
The crux of the dither argument seems to be semantic. What one person calls "fuzzy distortion" is what I would call noise. To be certain, what you put into a digital system is not what you get out, and therefore it is "distorted." However, what you put into an analog system is not what you get out, and thus analog systems are also distorted. In general, the digital systems get much closer to the truth, especially when discussing recording systems and/or systems that go through a large number of generations of reproducing the same signal.
I don't see the point of complaining about the "fuzzy distortion." If we could literally see the pattern of magnetized molecules on a tape, I'm sure it would look just as ugly as the computer displays of digital samples. Even if there were some subtle beauty to the patterns, the bottom line is that the signal coming out of a tape recording is much "fuzzier" than the signal that went in, and thus there is a undeniable amount of distortion.
I think some folks are deluding themselves by assigning loaded terms to the inaccuracies of one system versus friendlier terms for the inaccuracies of another system. Everything has noise, everything is distorted. It's just a matter of degree.
Keep in mind that the pictures you see on your computer monitor are not the waveforms that your amplifier is sending to your speakers. Digital audio always passes through some amount of non-ideal analog circuitry before you can hear it, and this non-ideal analog circuitry does not recreate the ugly waveforms you see on your screen. In a properly designed DAC, the waveforms coming out are much nicer. I'd prefer to see photos from an analog scope connected to a proper DAC, then we'd have a realistic image of any "fuzzy distortion" (or noise).
I don't see the point of complaining about the "fuzzy distortion." If we could literally see the pattern of magnetized molecules on a tape, I'm sure it would look just as ugly as the computer displays of digital samples. Even if there were some subtle beauty to the patterns, the bottom line is that the signal coming out of a tape recording is much "fuzzier" than the signal that went in, and thus there is a undeniable amount of distortion.
I think some folks are deluding themselves by assigning loaded terms to the inaccuracies of one system versus friendlier terms for the inaccuracies of another system. Everything has noise, everything is distorted. It's just a matter of degree.
Keep in mind that the pictures you see on your computer monitor are not the waveforms that your amplifier is sending to your speakers. Digital audio always passes through some amount of non-ideal analog circuitry before you can hear it, and this non-ideal analog circuitry does not recreate the ugly waveforms you see on your screen. In a properly designed DAC, the waveforms coming out are much nicer. I'd prefer to see photos from an analog scope connected to a proper DAC, then we'd have a realistic image of any "fuzzy distortion" (or noise).
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.
EXACTLY!
Most of the people who discuss digital audio do not understand that at any technological state, the product between speed and accuracy is constant. In our case there is a tradeoff between sample rate and how many bits can accurately represent the signal.
We can have 24 real bits if the bandwidth is 10Hz, for audio bandwidth we can achieve 21bits and if we have gigahertz we have a maximum of 4-5bits.
So even if we not take in consideration other limitations, bandwidth is one of the factors that influence accuracy.
In this context all the discussions about how to increase sample rate to 192Khz, 384KHz or even 768KHz looks absurd. You trade accuracy for speed, speed that will not give you anything. 96KHz is plenty enough to include all the info in the audio bandwidth.
chrissugar
Hi,
This is quite obviously untrue.
Using classic PCM tech there is no such link.
With a given pure multibit DAC or ADC I will get the same accuracy fundamentally at any given sample rate the converters logic can handle.
E.g with a PCM1704 I can operate at 44.1KHz sample rate with 22.05KHz bandwidth or at 768KHz with 384KHz bandwidth.
While the incrase of bandwidth will increase johnson noise, you will clearly find that the product between bandwidth and accuracy will rise significantly every time we bump up the sample rate and is NOT constant.
QED.
Of course, within the given system I COULD trade off bandwidth for more apparent resolution at lower frequencies, I believe abraxalito covered this a bit back and this is a given...
In our example case it is unlikely to help as the available analogue SNR is too low, though with 16 * Oversampling at 48KHz we could theoretically add four more bits to the existing 24...
Ciao T
This is quite obviously untrue.
Using classic PCM tech there is no such link.
With a given pure multibit DAC or ADC I will get the same accuracy fundamentally at any given sample rate the converters logic can handle.
E.g with a PCM1704 I can operate at 44.1KHz sample rate with 22.05KHz bandwidth or at 768KHz with 384KHz bandwidth.
While the incrase of bandwidth will increase johnson noise, you will clearly find that the product between bandwidth and accuracy will rise significantly every time we bump up the sample rate and is NOT constant.
QED.
Of course, within the given system I COULD trade off bandwidth for more apparent resolution at lower frequencies, I believe abraxalito covered this a bit back and this is a given...
In our example case it is unlikely to help as the available analogue SNR is too low, though with 16 * Oversampling at 48KHz we could theoretically add four more bits to the existing 24...
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).
Oh my god! That is the fuzzy distortion you are talking about? The noise that decorelate the wordlength reduced signal from the steps? You really do not clearly understand digital audio.
You should measure and listen after the reconstruction filter, but I can only think that you use some non oversampling no filtering DAC, so that would explain your conclusions.
chrissugar
You know what?
At this point I stop to answer to any of your post because it is useless. I will take Scott Wurcer's advice "if any amount of evidence will not change someones mind, just walk away".
You constructed your virtual world about how digital audio works and it would not matter if even God would tell you how things work it would not change your mind.
So for me it is waste of time, I have more important things to do. But interesting is on your part that you give credibility to flawed tests like Ethan's ones but decades of theoretical and practical experience from some of the guys who really understand digital audio (Keith Johnson, Dan Lavry, Paul Frindle and many others) do not matter. It is sad.
chrissugar
Hi,
Yet classic analogue noise and digitally created "dither noise" are usually never the same.
Tape noise is heavily noise shaped with a strong roll off starting at 50Hz, for example. Just as saying "distortion" in itself is meaningless, so is saying "noise".
Different distortions and different noises have dramatically differing audibility.
Unless and until we can agree that and then proceed to analyse the detail we remain at a point where many here cannot see the trees for the woods and hence mistake the larch for a firs, firs for redwoods, birches for oak and so on...
http://www.youtube.com/watch?v=5zey8567bcg
I agree that when you say "noise" and I say "fuzzy distortion" we are talking about the same process, but we have differing views as to what it constitutes and how it effects things, which as I pointed from the beginning of the debate was the crux of the matter.
As someone from a primarily analogue background I cannot, but see this as unnecessary added distortion and I for one do not see why suddenly something "BAD" becomes "GOOD" because someone claims "digital is different" without the slightest shred of proof.
Now many want to believe very much that what I consider "BAD" and which from an absolute view (e.g. information theory) is and remains bad, as something "GOOD".
They have faith that it is "GOOD".
They even have "miracles" (64K FFT, Averaging - miracles of the "now you see it, now you don't" kind) to underscore their faith in the goodness of it all.
Meanwhile I stand there and say "But the emperor has no clothes!". Well, I am after all only a little boy...
So, show me the clothes of the emperor and I shall recant...
Ciao T
Yet classic analogue noise and digitally created "dither noise" are usually never the same.
Tape noise is heavily noise shaped with a strong roll off starting at 50Hz, for example. Just as saying "distortion" in itself is meaningless, so is saying "noise".
Different distortions and different noises have dramatically differing audibility.
Unless and until we can agree that and then proceed to analyse the detail we remain at a point where many here cannot see the trees for the woods and hence mistake the larch for a firs, firs for redwoods, birches for oak and so on...
http://www.youtube.com/watch?v=5zey8567bcg
I agree that when you say "noise" and I say "fuzzy distortion" we are talking about the same process, but we have differing views as to what it constitutes and how it effects things, which as I pointed from the beginning of the debate was the crux of the matter.
As someone from a primarily analogue background I cannot, but see this as unnecessary added distortion and I for one do not see why suddenly something "BAD" becomes "GOOD" because someone claims "digital is different" without the slightest shred of proof.
Now many want to believe very much that what I consider "BAD" and which from an absolute view (e.g. information theory) is and remains bad, as something "GOOD".
They have faith that it is "GOOD".
They even have "miracles" (64K FFT, Averaging - miracles of the "now you see it, now you don't" kind) to underscore their faith in the goodness of it all.
Meanwhile I stand there and say "But the emperor has no clothes!". Well, I am after all only a little boy...
So, show me the clothes of the emperor and I shall recant...
Ciao T
Last edited:
I agree with what you're saying, Chris, but I think that it's important to realize that this tradeoff comes into play only when you're at the upper reaches of what we can achieve. Thorsten tries to say that the old converters did not experience any kind of tradeoff like this, but that's because they are not operating against similar limits. If you're willing to throw out potential accuracy, then you aren't forced into a tradeoff situation. It's only when you're up against the limit that increasing the sample rate forces a reduction in bits, or vice versa.
As for the PCM1704, it's not rated above 96 kHz. It may run internally at 768 kHz, but we do not have access to that data. At least the data sheet doesn't speak to anything beyond 96 kHz. I don't see how it can be claimed that this chip is not trading off various factors internally.
Nice example !
But it seems to me that you're demonstrating the usefulness of dither IN FRONT of an ADC, not inside a DAC ? Unless there is something I'm missing ?
By the way, what is a good book that presents the modern ADC and DAC theory and practice including dithering, filtering, etc. ? Preferably one at undergraduate or low graduate level, but a more advanced level would also work.
For me, it is not faith. I proved that dither is necessary for ADC by building actual circuits and listening to the results.
Before I performed actual circuit design experiments for myself, I believed what you currently believe: "How can it be any good to add noise?"
But the truth is that all analog recording systems "add" more noise at higher amplitudes than dither. These are not optional, so we accept them. With digital, the dither seems optional until you see and hear the distortion. Then you add dither and the input-dependent distortion and noise modulation is eliminated. Yes, you still have distortion and noise, but it is no longer input-dependent.
What I find fascinating is that there is a mathematical and scientific explanation for what I proved to myself experientially. That's a powerful tool that I can use to fine-tune the process and learn how to take a useful tool from perhaps a crude state to a much more refined state. In my case it cannot be dismissed as blind faith or belief in magic, although sometimes those terms are used to save trudging through the math.
Any time the bit depth is reduced, there should be dither. Going from analog to digital is just a unique case of reducing from infinite bit depth to finite bit depth. Thus, the reason for dithering in from of an ADC is exactly the same as dithering between a 24-bit master and a 16-bit CD. Finally, if a DAC employs digital processing, then it's probably handling the internal calculations at 48-bit or even 80-bit, and thus it needs dither.
There's a lot to be said for avoiding dither in your DAC, but the only way to do that properly is to first do away with the digital processing in the DAC so that dither is no longer necessary. As Thorsten points out, it's increasingly difficult to find a DAC without digital processing and dither. This is why you'll find people making 16-bit DAC products to this day (I only wish they wouldn't expect me to buy a nonlinear tube output stage along with the 16-bit DAC, since I'm no fan of the particular distortions introduced by tubes).
Hi,
Incorrect. RTFDS.
It is specified for 8 * OS at 96Khz so it's maximum input to the Chip is I2S at 768KHz Sample Rate. And yes, WE HAVE access to this data.
This is because you do not expect an old-fashioned pure multibit chip, but one of these newlyfangled things that use fuzzy distortion as fundamental operation principle.
The PCM1704 is a simple R2R DAC that will load 24 Bit of data into the output register and will update the output for each full wordclock cycle and the limit is 25MHz bitclock, which, if we have BCK = 24 * WCK allows theoretically around 1MHz actual, real sample rate (this is outside some of the datasheet suggestions but works)...
So no, this chip does not trade off anything internally, it is up to the user to do so.
Ciao T
Incorrect. RTFDS.
It is specified for 8 * OS at 96Khz so it's maximum input to the Chip is I2S at 768KHz Sample Rate. And yes, WE HAVE access to this data.
This is because you do not expect an old-fashioned pure multibit chip, but one of these newlyfangled things that use fuzzy distortion as fundamental operation principle.
The PCM1704 is a simple R2R DAC that will load 24 Bit of data into the output register and will update the output for each full wordclock cycle and the limit is 25MHz bitclock, which, if we have BCK = 24 * WCK allows theoretically around 1MHz actual, real sample rate (this is outside some of the datasheet suggestions but works)...
So no, this chip does not trade off anything internally, it is up to the user to do so.
Ciao T
Hi,
Not if we consider the amount of Dither in low bit Delta Sigma Systems. They add MUCH, MUCH less than that does.
Ciao T
Not if we consider the amount of Dither in low bit Delta Sigma Systems. They add MUCH, MUCH less than that does.
Ciao T
True, but one classic analog noise is usually never exactly the same as another analog noise, and digitally created dither in one design is not always the same as the digitally created dither in another design.
My point is that the output signal is distorted when compared to the input signal any time noise is added, whether it is added by choice or by unavoidable natural processes. Your postings often imply that all digital has fuzzy distortion and that it's all the same and yet somehow analog is always better.
You can shape the dither noise with a strong 50 Hz roll-off, too, if that's what you prefer. The amplitude of the dither may need to exceed 2 LSB if you alter the EQ, though. In fact, TPDF is available in flat, high-pass, inverse Fletcher-Munson, and a variety of shapes.
I have this book which is a great introductory text to DSP.
http://www.amazon.com/Digital-Signal-Processing-Practical-Scientists/dp/075067444X
http://www.amazon.com/Digital-Signal-Processing-Practical-Scientists/dp/075067444X
Hi,
You can buy Pedja Rogics DAC, he does not use tubes. It's cheaper too.
I find that as long as you use speakers to listen to your source, I would not loose sleep about the non-linearties that result in around 0.2% 2nd HD (and a little 3rd and nothing much else) a 0dBfs I would by far more worry about the speakers distortion.
No doubt, given speakers distort in a very similar fashion to tubes and to a by far larger levels you are then also not a fan of the particular distortions introduced by speakers (or headphones), so I am curious what you use to listen music.
Oh, I suspect you are also not a fan of the distortion introduced by the ears behaviour, right?
Ciao T
You can buy Pedja Rogics DAC, he does not use tubes. It's cheaper too.
I find that as long as you use speakers to listen to your source, I would not loose sleep about the non-linearties that result in around 0.2% 2nd HD (and a little 3rd and nothing much else) a 0dBfs I would by far more worry about the speakers distortion.
No doubt, given speakers distort in a very similar fashion to tubes and to a by far larger levels you are then also not a fan of the particular distortions introduced by speakers (or headphones), so I am curious what you use to listen music.
Oh, I suspect you are also not a fan of the distortion introduced by the ears behaviour, right?
Ciao T
Thanks for the correction. My thought processes were in ADC mode, not DAC mode, and in an oversampling A/D you do not have access to the intermediate data. You've been trying to convince us that everything is sigma-delta these days, and I didn't expect you to point to something like the PCM1704. Sorry I didn't spend more time on the data sheet.
Unfortunately, DAC is much easier to design this way than ADC. If know of an A/D chip that employs the equivalent quality 24-bit conversion, then please share!
Oh please, if you want to point out something I missed in a data sheet, then I welcome your corrections, but don't try to tell me what I expect or don't expect. I'm old enough that my default thought process is laser-trimmed pure multibit as created by Analog Devices for military applications.
Thanks again for the pointer to this chip. I already had the data sheet, but you have to realize that I have 432 data sheets from Texas Instruments alone, and over 2,500 data sheets total. It's really difficult to keep those all in mind.
Frankly, you usually come across as someone who doesn't understand what he's talking about, at least when it comes to anything digital audio. Most recently, you've claimed that nobody makes chips that are R2R any more, and you even went so far as to say most all converters today are sigma-delta internally. You totally caught me off guard by suggesting an actual chip that fits your own requirements. I wish you hadn't waited until I put my foot in my mouth to say something useful. If you're only going to use your knowledge for evil, then how do you sleep at night?
Oh, now I remember why I have the PCM1704 data sheet. Sadly, I also remember why I dismissed it: It does not have balanced current outputs.
What's the point of having two 23-bit DACs internally if you don't carry their outputs to individual pins?
By the way, I think Texas Instruments (Burr Brown) was foolish not to call this a 768 kHz DAC on the front page. I'm currently designing a product that uses a 125 MHz 14-bit DAC with balanced current outputs, so they're clearly not shy about documenting the actual sample rate on the first page of other data sheets.
Sorry to sidetrack, but is anyone aware of a 24-bit DAC with balanced outputs (current or voltage) and 192 kHz or greater sampling rate capabilities?
What's the point of having two 23-bit DACs internally if you don't carry their outputs to individual pins?
By the way, I think Texas Instruments (Burr Brown) was foolish not to call this a 768 kHz DAC on the front page. I'm currently designing a product that uses a 125 MHz 14-bit DAC with balanced current outputs, so they're clearly not shy about documenting the actual sample rate on the first page of other data sheets.
Sorry to sidetrack, but is anyone aware of a 24-bit DAC with balanced outputs (current or voltage) and 192 kHz or greater sampling rate capabilities?
Last edited:
- Status
- Not open for further replies.