Differential DACs?
Jun 26, 2002 at 5:30 PM Thread Starter Post #1 of 17

dlow

Head-Fier
Joined
Jan 3, 2002
Posts
65
Likes
0
I was flipping through the specs for a Denon 5803 and noticed that they have more DACs than output channels (16 DACS for 8 channels). Later on, they say that they put 2 DACs per channel in "differential mode" or optionally 4 DACS per channel in "dual-differential mode".

With all the discussion about better DACs, I was wondering if the more knowledgable people here could describe what "differential mode" is and whether it makes much of a difference in sound quality. Thanks.
 
Jun 26, 2002 at 5:44 PM Post #2 of 17
I would suppose that by `differential mode` they mean putting a differential amplifier on the output of two DACs -- the higher quality way convert balanced signal to unbalanced. The result would be increasing SNR by 3db (at least that's what should happen statistically). Dual-differential mode is probably the same thing but using 3 differential amps to bridge all those DACs to get another 3db in SNR (bringing it to a 6db increase).
 
Jun 26, 2002 at 7:33 PM Post #3 of 17
I think they dedicate each stereo DAC chip to one channel. One major benefit is extreme channel separation, which is probably quite desirable for high end stuff since some of the very best DAC chips such as CS4397 or CS43122 have only 90 - 100dB channel separation, which is not good enough especially in light of their noise and dynamic specs going over 110 and even 120dB. Having 100 times more crosstalk than noise in a channel is a lousy way to use a great chip so they dedicate single chip per channel. Then they decide to put the second channel on the same chip in use but I don't really know how. If they were fed the same signal, they should both be exactly the same so a differential amp over them would always return 0. Some DACs have differential outs, some have current outs, some have single ended outs so I don't know what strategy is used in each case. Maybe they use + of one and - of other channel to feed + and - of one opamp, and then - of one and + of other channel to feed + and - of another opamp, and then feed outputs of those two opamps to + and - inputs of the third one?
 
Jun 26, 2002 at 9:18 PM Post #4 of 17
My current player has 4 DACs, 2 mono DACs for every channel. I presumed that the Denon in question uses similar strategy.

Channel separation is an obvious benefit of having separate DACs for every channel. By using 2 or more DACs per channel the following could be achieved (one DAC gets the original signal, another DAC gets signal that was inverted in the digital domain):

1) any noise from internal circuitry or RF picked up on the way to the output is cancelled out when the difference of two signals is computed (assuming that what influenced one channel, influenced the other one similarly. it's an alright assumption, because even if noise is not cancelled out completely, it gets significantly reduced),

2) assuming DACs are the same (which they probably are) and their SNR (or relative error) is some epsilon, after the signals from DACs (V and -V each with absolute error of at most |epsilon * V|) go through differential amp, the result has twice the intensity of the original signal (2 * V) but the absolute error is only sqrt((epsilon * V) ^ 2 + (epsilon * (-V)) ^ 2) = sqrt(2) * epsilon * V, thus the relative error of the final signal is sqrt(2) * epsilon * V / (2 * V) = 1 / sqrt(2) * epsilon, or 3db smaller than output of each DAC by itself.

Bridging 4 DACs probably goes as follows:

(DAC (V) - DAC (-V)) - (DAC (-V) - DAC (V))

[- stands for diff. amp, DAC (V) -- DAC that processed the original signal, DAC (-V) -- DAC that processes the digitally inverted signal].

As I said, I don't know if Denon does exactly that, but this mechanism is used sometimes for the reasons I outlined...
 
Jun 26, 2002 at 9:35 PM Post #5 of 17
Ok, here it is: http://216.239.33.100/search?q=cache...hl=en&ie=UTF-8

Yes, they use both 2 channels of PCM-1738E and then probably feed them through diff. amp, not mono-DACs. When in `dual-differentiable`, two stereo DACs are bridged.

So yeah, mistery solved. Now if only `whoosh` was that simple...
tongue.gif
 
Jun 26, 2002 at 9:44 PM Post #7 of 17
Yes, that's what you'd do if your DACs have single-ended outputs. If they are differential, things are a bit more complicated. I just wouldn't know how to easily reverse digital signal in DIY, when you don't have a signal processing chip available.

As zzz pointed out, the actual benefit is likely much greater than just 3dB because in reality noise in each channel is not independent and there is in fact some common-mode noise e.g. from RF pickup or from digital rails. Differential amp will significantly reduce any component that exists in both of its inputs. However, many high end DAC chips have differential outputs for each channel anyway, and therefore get this benefit of common-mode filtering. 3dB by itself when you're talking about DACs with over 110THD+N is really not too important.

To me, channel separation is the biggest benefit. You WANT to use separate DAC anyway if its channel separation is not satisfactory. Using the rest of the channels in the fashion above, which would otherwise go unused, is (just) a plus.
 
Jun 26, 2002 at 9:50 PM Post #8 of 17
Differential DAC configuration is when you have two DACs of opposite polarity (by feeding one the negative signal of the other) for each channel, and you sum their outputs together to get a 3 dB reduction in noise. That's for uncorrelated noise, as someone pointed out. I don't think there's enough correlated noise to make it a big deal for that kind of noise.

Personally, I think it's a good technique, but it's not a huge deciding factor, as there are so many other things you have to get right in a DAC. Also, some silly people don't sum the DAC outputs, and instead use it to drive each phase of a balanced line. That's just stupid, and a bad design. The outputs must be summed immediately and as close as possible to the DACs to get the most benefit out of it. After that you can can then feed it to a balancer circuit to drive a balanced line.

--Andre
 
Jun 26, 2002 at 10:47 PM Post #9 of 17
Quote:

Originally posted by AndreYew
Also, some silly people don't sum the DAC outputs, and instead use it to drive each phase of a balanced line. That's just stupid, and a bad design. The outputs must be summed immediately and as close as possible to the DACs to get the most benefit out of it.


Why would you say that?
 
Jun 26, 2002 at 11:42 PM Post #10 of 17
I don't know what detrimential effect would that have on signal when it's finally summed - again, common mode RFI would get filtered out - but it is definitely not good for almost any DAC - and certainly none of high end ones. Because long wire it would present too much load on the very delicate output stage (opamp) of DAC chip, or its current out. You'd be driving a transmission line with something that can provide only couple of mA at best (likely less than 1mA), and you'd have lots of RMI and noise reflecting back and entering the DAC chip, plus the whole line would be capacitive load on the chip. All of that would probably cause distortion of the signal. You need to buffer DAC output immediately, close to chip and then you can abuse that one all you want.
 
Jun 27, 2002 at 12:18 AM Post #11 of 17
If the two DACs has different errors (+/- 1/2 LSB), summing the outputs together won't that create distortion?

Also if the signal path skew is different, shouldn't that also create distortion?
 
Jun 27, 2002 at 4:12 PM Post #12 of 17
Zzz,

The outputs of the differential DACs must be summed very close to their outputs so that the symmetry can be preserved (it's not affected by outside noise) so that the most cancellation happens. Otherwise, it's pointless to run differential DACs if you're not going to take advantage of their differential nature.

dvw,

Often, differential DACs are two DACs on the same die, so they have very similar behavior. The LSB error is only an issue if you don't dither properly.

--Andre
 
Jun 28, 2002 at 8:11 AM Post #13 of 17
It's very good that this question was brought up, even since I've read the article in Stereophile about "first American SACD player", one Classe, that uses this technique of using multiple DAC chips and combining outs, I wondered how it actually worked as I wanted to do it myself. Now I know
smily_headphones1.gif
. Now if I only knew how to invert digital signal
frown.gif
(without DSP that is).
 
Jun 28, 2002 at 7:10 PM Post #15 of 17
Ah, but DACs usually require data in second complement. That's invert all AND add 1. It's easy to invert but not so easy to add 1 to a serial stream.
 

Users who are viewing this thread

Back
Top