I totally agree that dither in front of an ADC is useful and also when reducing the bit resolution.
My question is: is it useful to put dither in the output of a DAC ?
In more details: suppose you have an ideal non-oversampling DAC with an ENOB of 20 bits. Suppose that the DAC receives exactly 20 bits as digital input and outputs 2^20 levels. The error of the output signal is smaller than 1 LSB.
In this case you can either:
-let the DAC output the analog "equivalent" of the original digital signal with no processing (no dither)
or
-add dither to the digital input signal.
Is it useful to add dither to the digital input signal ? I would think it isn't, but I am not sure.
Yes, up to a certain point a lot of the discussion seems to be a misunderstanding, as ThorstenL and abraxalito have some concerns regarding wrt to the changeover from multibit DACs to Delta-Sigma-DACs combined with heavy dithering and noise shaping.
And because their concerns are perception based it should not be easily dismissed.
A good introduction overall is still:
John Watkinson, The Art of Digital Audio
@ rsdio,
Widrow and Kollar have written a comprehensive book:
Quantization Noise - A book on quantization
Excellent question. For the exact situation that you describe, dither would be completely unnecessary. It would not add anything but pointless noise. This assumes that your data sampling rate is the same as your DAC sampling rate.
However, there is one technical flaw in your description. "No processing" is not quite correct, because you need an anti-alias a.k.a. reconstruction filter at the output. That's arguably analog processing, but it is processing nonetheless. You cannot (correctly) just use the DAC output without further processing. Also, depending upon whether the DAC has some sort of sample-and-hold to isolate the output from the intermediate changes between sample codes, you might need to add a S/H before the low-pass filter. With a parallel multibit DAC, changing the 20-bit code can result in some bits affecting the output before or after the other bits, such that there is a settling time before the output is accurate. Sample and hold is generally used to prevent this noise from spoiling the output. For example, if the code were to change from 0x400000 to 0x3FFFFF, it might temporarily output a voltage for 0x000000 or 0x7FFFFF, which would represent an absolutely horrible spike in the waveform that is unrelated to the input.
Thanks for confirming my guess.
By "no processing", I meant no processing of the digital signal which is fed to the input of the DAC, I didn't mean what happens at the DAC output.
I also totally agree that the output filter is an integral part of a DAC, so any discussion about the output signal of a DAC should include the filter.
This also holds for any sample and hold circuit at the output of a DAC (and before the output filter).
Anyway, thanks for your polite and reasoned answer.
Last edited:
Hi,
First I always pointed out that ALMOST everyone does only DS these days, I have repeatedly mentions the PCM1704 as one of the very few exception. In fact, I pretty much always qualify things, as I am quite critical of those who do not. So I am rather surprised...
Sadly, I do not.
I have been thinking of bodging several S&H circuits and several PCM1704 into something that is fast enough, but it would be a mess, I'd probably be better off with much faster industrial 16 Bit DAC's in a different architecture...
Is that I because I refuse to swallow the Bull (and I mean the whole bull) that is used to sell DS systems?
The point is that I DO understand and hence naturally object.
I claimed that ALMOST nobody makes R2R for Audio today and I listed the exceptions I know. If you read more into my writing than what I write that is regrettable, but not my fault.
Because everytime I do for example suggest that Multibit DAC Chip X by Maker Y is a good part and recommended I get ten people jumping on my comment and pointing that Chip A by Maker O is so much better and the oldfashioned multibit is no good.
Then I point out that the modern miracle chips are not really 32 Bit, but just 12 Bit and the rest is just "fuzzy distortion" (which it is) and then more people come into the argument on the side of orthodox faith and tell me I know nothing about digital audio.
It gets boring after a few times, so I skip part of it nowadays.
I do like to use my knowledge to help people, but too many people do not want to be helped and others seems to feel the need to protect poor idiots that might follow unorthodox suggestions from such a horrific faith (like having actually a system who's sound satisfies them, even if in an orthodox sense it does not measure too well), so I get to do much less of that than I like.
But even so, I sleep very well, thank you.
Quem di diligunt, adolescens moritur, vivere pergam.
Ciao T
First I always pointed out that ALMOST everyone does only DS these days, I have repeatedly mentions the PCM1704 as one of the very few exception. In fact, I pretty much always qualify things, as I am quite critical of those who do not. So I am rather surprised...
Sadly, I do not.
I have been thinking of bodging several S&H circuits and several PCM1704 into something that is fast enough, but it would be a mess, I'd probably be better off with much faster industrial 16 Bit DAC's in a different architecture...
Is that I because I refuse to swallow the Bull (and I mean the whole bull) that is used to sell DS systems?
The point is that I DO understand and hence naturally object.
I claimed that ALMOST nobody makes R2R for Audio today and I listed the exceptions I know. If you read more into my writing than what I write that is regrettable, but not my fault.
Because everytime I do for example suggest that Multibit DAC Chip X by Maker Y is a good part and recommended I get ten people jumping on my comment and pointing that Chip A by Maker O is so much better and the oldfashioned multibit is no good.
Then I point out that the modern miracle chips are not really 32 Bit, but just 12 Bit and the rest is just "fuzzy distortion" (which it is) and then more people come into the argument on the side of orthodox faith and tell me I know nothing about digital audio.
It gets boring after a few times, so I skip part of it nowadays.
If you're only going to use your knowledge for evil, then how do you sleep at night?
I do like to use my knowledge to help people, but too many people do not want to be helped and others seems to feel the need to protect poor idiots that might follow unorthodox suggestions from such a horrific faith (like having actually a system who's sound satisfies them, even if in an orthodox sense it does not measure too well), so I get to do much less of that than I like.
But even so, I sleep very well, thank you.
Quem di diligunt, adolescens moritur, vivere pergam.
Ciao T
Hi,
You are correct, it is not I2S but EIJA.
Though both differ only in how data is aligned to wordclock and bitclock, the difference however has no impact on maximum frequencies and sample rates, which was the concern.
Ciao T
You are correct, it is not I2S but EIJA.
Though both differ only in how data is aligned to wordclock and bitclock, the difference however has no impact on maximum frequencies and sample rates, which was the concern.
Ciao T
Last nights comments were mis-understood I was refering to the old industry controversey about "real" bits. I was talking about NIST traceable standard cell and Kelvin-Varley divider bits. Audio A/D's generally don't make good instrument front ends. A 7 digit Fluke DVM on a chip for $10 would be neat.
I hear good things about K47 and CK12 Microphone Capsules | Microphone Parts they are selling both flavors of the classic Neumann capsules as clones. At $100 painless. Martin told me he does not like Pelusos which are considerably more.
You can coax 120dB SNR out of a 1/2" capsule, which is what we were playing with in Austin.
Last edited:
Hi,
Well, now precisely WHY would balanced current outputs be "better"? Balanced carries invariably a 3dB noise penalty...
But if you must do balanced because single systems are against your religion, it is quite trivial to use two PCM1704 in a balanced configuation and if you take care with the layout you can arrange things so that the two DAC's really cancel each others currents well into the MHz region.
To avoid the glitch at zero crossing. Well, that is what they say anyway.
Go, tell them. I would say they where foolish not to call it a 1MHz DAC, but I cannot be arsed to tell them.
Two PCM1704?
The Arda Tech AT1401 DAC is still vapour ware, unless I missed the samples shipped... It may have one DAC too many for you, it has two 24Bit/1.536 MHz DAC's with differential output. To be honest, I wish they could have NOT used differential outputs but single ended and with +/- Supplies.
Optimum est pati quod emendare non possis.
Ciao T
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.
Well, now precisely WHY would balanced current outputs be "better"? Balanced carries invariably a 3dB noise penalty...
But if you must do balanced because single systems are against your religion, it is quite trivial to use two PCM1704 in a balanced configuation and if you take care with the layout you can arrange things so that the two DAC's really cancel each others currents well into the MHz region.
To avoid the glitch at zero crossing. Well, that is what they say anyway.
Go, tell them. I would say they where foolish not to call it a 1MHz DAC, but I cannot be arsed to tell them.
Two PCM1704?
The Arda Tech AT1401 DAC is still vapour ware, unless I missed the samples shipped... It may have one DAC too many for you, it has two 24Bit/1.536 MHz DAC's with differential output. To be honest, I wish they could have NOT used differential outputs but single ended and with +/- Supplies.
Optimum est pati quod emendare non possis.
Ciao T
Yes, subtractive dither works and is used in places other than audio. I wanted to prove to myself the principle and shared the exercise.
Dimitri,
I see lots of people jumped on this.
But let us start with basics. The issue was a audio D/A convertor. This is a nice paper about A/D conversion. It does use the technique of processing to improve the signal to noise in the range under examination. There was a nice Analog Devices paper mentioned earlier that covers that.
It is quite possible to build A/D's that are bandwidth limited to meet 24 or more bits. The most common application is in digital scales.
T.
Let's wait and see what comments Pavel has on how to make a 24 bit audio D/A since he has the info.
Scott,
Those freakin capsules are way too cheap. If only someone would publish a nice preamp circuit to go with them. BTY I have a standard cell and surprisingly the modern semiconductor voltage references work better!
Y'all
On another thread there is a fellow who seemed to improperly identify an IC. So for fun I thought I would list partial part numbers and see who can correctly identify the part. For example 741 would be a uA741 the first commercial op amp with built in compensation introduced by Fairchild.
47 (More than one answer!), 072, 83, 107, 300, 555, 703, 709, 722, 923, 7400
Any takers?
Last edited:
Part 2, it's coming. BTW I hear they offer matching and invididual response curves as well as having great customer service.
- Status
- Not open for further replies.