[samsquanch]

So amongst my many other projects, I'm starting to think out a design for a dac, which for me is looking at what other people/companies have made and getting ideas...
 
While looking at the Audio-GD Reference series dac's, I notice that they use multiple dac chips in their design, the 7.1 uses 4 per channel (8 total).  Are they stacking these chips in parallel with each other, or are they using each one for each of the different output connections, 1 for RCA, 2 (1 normal, 1 inverted) for XLR, and 1 for ACSS (what ever that is)?
 
This one's got me scratching my head.

 

[nullstring]

From what I understand, they run the DACs in parallel in order to sum the outputs and average the difference.
 
http://www.dddac.de/ma_dac21.htm
 
Go to the bottom of the page where it says Paralleling DAC chips.
 
 
 
Quote:

[size=larger][size=larger]Paralleling DAC chips [/size][/size]
[size=x-small]OK, last but not least, parallel DAC's...... Why this? Actually the TDA1543 is not know for its great linearity and great performance. And this is already a huge understatement, haha !! Why use the 1543 then? Simply because it runs on almost 9 Volts and has a current output. This enables us to make directly 2 Volt Output (CD Standard) without any further tricks than connecting a resistor to the output........ Back to the 1543 quality, so it is lousy? yes it is, but thanks to statistical laws we have a way out !! If you run (ANY !!) process many times, after each other or in parallel does not matter, the uncertainty or errors in the output of the process will improve with the function of SQRT(n) where "n" stands for the number of events. This trick can be used for example to get very precise resistors or capacitors by paralleling them. Thanks to the current source output we can easily do the same for the TDA1543.......[/size]

 
 

 

[rds]

I think that is the idea.  However, that argument in blue is full of holes.  Generally averaging is used to get rid of electronic noise - it doesn't make bad measurement equipment (or in this case a bad DAC) good.
 
Quote:

From what I understand, they run the DACs in parallel in order to sum the outputs and average the difference.
 
http://www.dddac.de/ma_dac21.htm
 
Go to the bottom of the page where it says Paralleling DAC chips.
 
 
 
 
 

 

[akgfan]

That's the same with opamps. I've seen it in datasheet. It was because of lowering noise.

 

[nullstring]

I don't understand at all how they can figure that there is an "extra bit resolution" gained from paralleling dacs.
 
If someone wants to explain that, I'd appreciate it.

 

[rds]

Quote:

I don't understand at all how they can figure that there is an "extra bit resolution" gained from paralleling dacs.
 
If someone wants to explain that, I'd appreciate it.

How you could improve bit depth by reducing the noise at the analog output is ...a mystery to say the least.  Signal averaging cannot improve the resolution beyond the inherent limitation of the equipment; it can only reduce noise.

 

[cobaltmute]

For the PCM1704 is the Audio-GD, using two per channel allows you to generate a differential output (you invert the signal to one chip).
 
As for 4 per channel or parallel, it allows you create more current out of the output stage.  The chips only supply +-1mA each.  They are supposed very touchy with loading so this may be a way to get around that.

 

[jcx]

there are no "24 bit" output noise audio DAC - 144 dB is hard to do in any audio frequency analog electronics
 
but sigma-delta DACs can show differential linearity well down into the analog noise floor so lowering analog noise can provide "extra bits" with a "24 bit" DAC - if they were there to start with but hidden in the single DAC's analog output noise
 
as long as the analog noise is random you get "an extra bit" worth of S/N for every 4x paralleling of similar outputs
 
several of the "flagship" monolithic audio DACs reach 120 dB S/N without multichip paralleling, mono mode Sabre reaches -130 dB
 
 
so yes you can play numbers games with "extra bits" resolution due to lower noise having engineering meaning  - but it will not mean much for audio listening
 
 
there is of course no "naturally miced" recording with audio information there - really good mics have ~20 dB noise floor
 
you could argue that close miced instruments get large proximity gain so a studio mix could have a little better S/N - but that is not the way the industry is going - dynamic compression is far more common today - all of the "music" exists in the top 5-10 bits
 
 
another point specific to headphone listening is "microphonics" - headphones turn mechanical motion into sound right at your ears - muscle tremor, pulse, breathing, body motion all raise the noise floor when wearing headphones by 10-15 dB over the ultimate limits determined in anechoic chambers (with minutes of accommodation - you never have access to the lowest level perceptual limits while listening to music )

 

[nullstring]

Quote:

as long as the analog noise is random you get "an extra bit" worth of S/N for every 4x paralleling of similar outputs

Where does this 4x come from?
Whats the theory behind the number 4?

 

[jcx]

with uncorrelated random noise if you can average N noisy measurements then the the output noise is reduced by the square root of N - this depends on the statistics of the noise and being able to align the measurements with the signal - so that the signal is in the "same place" for every average
 
engineers can "see" the bits buried in the noise by averaging multiple waveforms - theoretically 100x averaging reduces random noise by 10x == 20 dB

 

[rds]

Quote:

there are no "24 bit" output noise audio DAC - 144 dB is hard to do in any audio frequency analog electronics
 
but sigma-delta DACs can show differential linearity well down into the analog noise floor so lowering analog noise can provide "extra bits" with a "24 bit" DAC - if they were there to start with but hidden in the single DAC's analog output noise
 
as long as the analog noise is random you get "an extra bit" worth of S/N for every 4x paralleling of similar outputs
 
several of the "flagship" monolithic audio DACs reach 120 dB S/N without multichip paralleling, mono mode Sabre reaches -130 dB
 
 
so yes you can play numbers games with "extra bits" resolution due to lower noise having engineering meaning  - but it will not mean much for audio listening
 
 
there is of course no "naturally miced" recording with audio information there - really good mics have ~20 dB noise floor
 
you could argue that close miced instruments get large proximity gain so a studio mix could have a little better S/N - but that is not the way the industry is going - dynamic compression is far more common today - all of the "music" exists in the top 5-10 bits
 
 
another point specific to headphone listening is "microphonics" - headphones turn mechanical motion into sound right at your ears - muscle tremor, pulse, breathing, body motion all raise the noise floor when wearing headphones by 10-15 dB over the ultimate limits determined in anechoic chambers (with minutes of accommodation - you never have access to the lowest level perceptual limits while listening to music )

 
I find the way this terminology is being used quite confusing.  When I think of bit depth in a DAC that means resolution to me.  That bit depth is a digital property and is independent from the analogue output.  At the analogue stage you have a signal: S + N, where S is the DAC's interpretation of the digital information and N is noise.  Even if you eliminate N, S is an imperfect representation of the digital information ...and it's analog! 

 

[jcx]

there lots of "moving parts" and nuances - signals, noise require bandwidth and frequency response specs to be really usable in calculations, detailed comparisons, then there is the reference level for dB - you have to determine by context whether analog, digital signals or SPL are being discussed
I did use both SPL and digital dB in the above posts
 
the simplest for casual conversations is to use the flat response in the full Nyquist bandwidth S/N approximation of 6 dB/bit for digital signals with "0" being the top of the scale
SPL has a reference pressure level that puts "0" close to human hearing threshold a our higher sensitivity frequency range
 
 
24 bit PCM words is just a digital "container" specification - DAW sometimes use 32 or even 64 bit wordlengths to allow many processing steps without worries about quantization errors
 
audio signal "information" is limited by Shannon-Hartley Channel Capacity Theorem; transducer, electronics (preamps, ADC input), room noise and by microphone, electronics bandwidth, ADC anti-alias filtering and sample rate
 
no one has ever recorded "24 bit", 144 dB S/N audio in a musical performance, no "24 bit" input noise audio bandwidth ADC exist ~120 dB S/N is about what the best monolithic audio ADC can do today ~"20 bits" of audio signal
 
but for testing a DAC it is possible to feed them a digitally synthesized 24 bit signal

 

[samsquanch]

Ok, so it boils down to prevent overloading and averaging out errors.  
 
Thanks!

 

[FallenAngel]

Pretty much

 

[qusp]

simply put, parallel dacs output a signal that is hotter with more dnr and pushing the noise floor down by comparison, so provided the iv or next stage can handle this input without clipping then more bandwidth is realized. errors are also averaged out to a lower level the larger pool of bits you pull from
 
the sabre itself uses multiple dacs in parallel internally to leverage this effect.
 
there are now a new range of silly resolution products and projects coming to market such as the exadevices usb-i2s interface, which is capable of 32 bit and 4 channels of 384khz lol a senseless waste of hd space imo as nothing is produced at this level and anything that comes close is probably just upsampled. the sabre is capable of 32bit input and the internal async easily covers that bitrate, but its only a 32bit dac, so some of the filter functions will not work i would imagine, being they usually use the top 8 bits for things like filter and volume control.
 
i do like the multichannel isolated i2s conversion feature though for possible dsp and crossover functions, which is high on my wishlist for finishing my speaker project