The guys are telling you the truth about measurements versus audibility. AP analyzers give 'figure of merit' numbers. 'Merit' is not exactly a measure of how something sounds. Its more of a type of 'goodness' factor that is reasonably practical to measure.
Most things yes, like "top quality USB receiver", "top quality LDOs", "well engineered output stage".
P.S. Just what I want to say - using "dual AK449x" or dual ES9038 (8ch each for Pro version) is a marketing trick and give nothing in SQ (if all other things are done right).
That is correct. Dual dac chips may result in slightly lower noise, THD+N and crosstalk. Those benefits are not audible but may be useful in measurement applications.
I sort of got the impression that the practical benefit of dual dacs can be a slight improvement in stereo imaging. I am told that with at least some types of speaker systems, the effect can be audible. Is that categorically false?
Crosstalk can impact stereo imaging. But with properly implemented single dac crosstalk is already too low to be audible.
When a DAC chip and surrounding circuitry costs 20-30€ (worst case for AK4493), I'd rather put two of them on the board and be safe.
Plus "Dual Mono" sounds way cooler.. 😛
Plus "Dual Mono" sounds way cooler.. 😛
IIRC John Westlake said that the problem in single dacs has to do with correlated noise crosstalk between the channels. According to some people, say, for example, ESS, correlated noise is much more audible than uncorrelated noise. And of course Westlake visited the AKM factory and was shown an early prototype of AK4191/AK4498. He said AKM told him the reason for going to a two chip set was deal with substrate coupled noise. Maybe that type of noise crosstalk is related to human perception of imaging. I mean, do we know for certain one way or the other?
This is MY opinion. It can be considered as "IMHO".
The measurements, like SNR should be betterwith paralleling DACs (and even this is not always!).
But is it not enough SNR with the single DAC? (better than -110-120dB for modern chips)
Of crosstalk like -120dB @10kHz is not enough?!
THD? Usualy THD does not improved with paralleling.
But these are "the measurements", SQ is a little bit different thing.
I made a lot of different DACs (including the AK4490-old99, ES9038, BD34xxx), the only DACs where I heard the benefits of the paralleling, were AD1853, PCM1794 ~15 years ago.
I can express another blasphemous thought - all knows that ESS has H2 and H3 THD compensation, but I have heard a lot of opinions, that SQ is better when THC Compensation if off. (Which means worse THD).
IIRC that's what John Westlake claimed. Something about the sound of it seemed to bother him.
EDIT: Also I was very surprised to find that very small changes in harmonic correction were audible to different people, with some variation in sensitivity between people. Surprised me that it was audible it all for small changes. Could it be some quirk of the DSP engine?
Last edited:
I have made the same dac with single AK4493 and dual AK4493s. Same circuitry but partly duplicated in dual. SNR/DR improved from about -122dB to -124dB but surprisingly also THD+N improved (-117dB to -118dB). Crosstalk improved from about -120dB to -140dB. No audible difference in SQ.
SNR can be improved by 3dB for 2 DACs in parallel, in theory, as I remember.
THD - 1dB improvement can be related to the measurement tolerance.
Crosstalk - yes, -140dB is very important. What is at vinyl's crosstalk at 10kHz, -10-15dB? (rhetorical question...).
Well, the onus is on you to prove it. You are claiming Russell's teapot exists. Give me a teapot.
Well, the onus is on you guys to prove it. You are claiming Russell's teapot exists. Give me a teapot.
From a crappy USB receiver that is very sensitive to external noise to a decently designed one, I can believe it, in fact I can claim I have heard the difference myself, even though it was sighted. It just have a decent PS, maybe galvanic isolation, filter noise from the incoming power line, and after that there should be something to reduce jitter, like a buffer and reclocker.
From a badly regulated power supply to a good regulation, same – anybody can probably hear a difference in "hiss" or "hash" with sensitive headphones or speakers. Now ESS has even little single chip regulators that are apparently wonderful. One does not have to spend inordinate amounts of money, the little LT3045 LT3094 chips are fantastic. I use them myself, by the way, not that this proves anything.
As for the well engineered output stage, what we want is that it has a quite high slew rate, it has good noise PSRR, that is as linear as it gets, that it can provide more voltage swing that necessary to drive most power amps directly, and it also has a very low output impedance. This is nowadays easy to design - and even if the output impedance is in the 200Ohm rage (balanced), it is not a tragedy provided we use the DAC with a well designed preamp or power amp - i.e. with a high input impedance.
But beyond that?
So you are telling me that if two DACs have output signals that differ by at most -120Db this can be audible?
Assuming you are referring to ES9311 IME they are far from wonderful. Not-so-low lf noise and very unreliable. In fact the most unreliable IC I have ever used.
Wait a minute. It should be obvious to anyone that steady state HD and or IMD measurements do not and cannot fully characterize a physical audio device. To make matters worse, most audio FFTs discard phase information at all frequencies including LF where human sensitivity to phase is uncontroversial.
Therefore the onus is on you to prove that standard measurements completely define SQ.
Of course, even AP knows you can't prove that because it isn't true...
Differ in what way? Fixed sine wave harmonic distortion at 0dbFS, or what? Are you saying that's the only possible way to differ?
I am not talking of HD and IMD only, but why should steady state not characterise it? That it should not would be “obvious” only to those that do not understand the mathematics behind it. Indeed, there are higher order non linearities that would result as an error, but can we really imagine that they result in anything than a very low order one?
As for phase, since phase information errors are caused by delay, essentially, the effect on LF is minor. Also, the brain is sensitive to that, but as a difference between the signals that reach the two ears. And the delays in two channels of the same chip are the same, ergo the difference wrt the original signal pair is… zero.
Some mathematics helps.