No, I had in mind Figure 15. elements 207 and 208, which linearize the ADC 203 as well as providing subtractive dither.
Audio Science: Does it explain everything about how something sounds?
- Thread starter PeterA
- Start date
- Status
No, I had in mind Figure 15. elements 207 and 208, which linearize the ADC 203 as well as providing subtractive dither.
That's very interesting, thanks. I had never heard of this being mentioned as a possibility.
AD is a B2B supplier. They don't do much marketing, and they've made a LOT of DACs. It's a lot more likely that like the chart you show just above, they've looked at the measurements and have concluded that the effect is inaudible. It really doesn't matter if it is constant or if it is modulating with the signal, if it's below the hearing threshold. Of course, as your source above says, "some people might believe otherwise." Always. Some people hear almost anything, regardless of how audible it is.
Tim
Huh? Not from the measurements displayed & his pointing out which is the AKM (the green one on top) & ESS (blue graph at bottom)
Oh, I just got them mixed up. I thought the Blue one @ the bottom was the AKM one. ...Then from those graphs that particular AKM DAC with that particular implementation does indeed look "cleaner".
oh, oh, did you miss the bit where I said that the chief engineer in ESS said "it was audible to some (yes, in blind testing)" but not to him? And yes the measured effects were below the "currently accepted thresholds of audibility"
You conjecture "It really doesn't matter if it is constant or if it is modulating with the signal, if it's below the hearing threshold." is myopic - have you thought about how these hearing thresholds were established? What test signals, what conditions?
Some of the same people hearing these "differences" also hear differences when there are no offsets. So in my mind the offset vs no offset doesn't explain much of anything. I do know it will alter the noise floors of sigma-delta processes, but some people hear differences either way. This does not show it is or isn't true for audibility, just that the offset and its effect on noise floors is currently a guess at best.
Your criticisms of ESS DAC chips seem to be all over the web. That does not prove anything, of course, and your issues may or may not be on target. I do note that Merging Technologies also selected ESS for their new NADAC, presumably after extensive due diligence for this, their first audiophile product and one at a very high price point. In my own travels and unscientific comparative auditioning, I really have not heard better sound than what I get through my Exasound DAC, also reliant on ESS Sabre. I doubt it is perfect, but it is far and away the best my system has ever sounded.
Huh? In fact ESS recognised the problem of noise modulation & tried to do something about it. Other sigma delta DAC manufacturers haven't addressed this, AFAIK. This is hardly a criticism, is it? I'm simply pointing out well documented measurements & blind listening on the ESS DAC.
I did point out to you that your claim of "galvanic isolation" in that DAC was not correct & that it was using the ES9018's in-built ASRC, which is not the best use of that DAC for sound quality.
Sorry, if you take these facts as a personal criticism as you seem to?
John, the limits of human hearing have been tested and re-tested thousands of time over many, many years. I don't know what methodologies have been used other than the standard hearing tests I get when we go to an audiologist, but after all these years, i suspect the conditions and test methodologies are too many to mention. I understand that some people will question anything that disagrees with what want to believe, but what we're talking about here is not some relatively recent test, like the 30 years of Harman listening studies, that we can poke at with a blunt stick looking for flaws to help us rationalize our way to different conclusions. 20 - 20khz is the scientific community's long-held and rather broad, conservative, definition of the range of human hearing (most humans don't meet that standard). To say that acceptance of those limits is myopic is like saying belief in evolution is myopic. I mean we could have been left here by aliens; it could happen. That evolution stuff is just a theory, after all.
Worry over whatever you like. Go in search of imperceivable problems to wrestle with if it makes you happy. When it comes to audio reproduction, I don't intend to concern myself with the sounds of things humans can't hear. And that's not myopic, it's rational.
Tim
Is this noise modulation thing an artefact of 'electrical' imperfections, or are we saying that the DAC's fundamental design is the problem? If it's the latter, we don't need to build it and test it, merely simulate the algorithm on a computer to show what the noise will do. If it's at -110 dB, however, I can understand why some people may be sceptical that it can be heard. On the other hand, if it can be eliminated then it should be eliminated regardless.
If a theoretical 'dumb' algorithm can produce statistically-perfect random noise then so be it. But if not, presumably, these days a 'DAC' could incorporate a fully-fledged DSP that analysed the proposed output in advance and modified the processing to always produce the most statistically perfect random noise possible..? (maybe this was what someone was mentioning earlier).
It sounds like a nice problem to work on.
If a theoretical 'dumb' algorithm can produce statistically-perfect random noise then so be it. But if not, presumably, these days a 'DAC' could incorporate a fully-fledged DSP that analysed the proposed output in advance and modified the processing to always produce the most statistically perfect random noise possible..? (maybe this was what someone was mentioning earlier).
It sounds like a nice problem to work on.
Tim, audio equipment has "been tested and re-tested thousands of time over many, many years" using the standard battery of tests & your logic is that by this very fact we know all the tests necessary to characterise it's performance. Yet, the fact is, we don't. I'm suggesting the same applies to hearing - it's not as revolutionary/irrational a concept as you seem to portray it
What I think you are failing to fully comprehend is the difference between simple test signals & complex, (music-like) signals & how they are handled by audio equipment - a non-linear system. The same applies to our hearing - also a non-linear system.
EDIT: Anyway, I'm not talking about the frequency range of hearing, I'm talking about the lower thresholds established for amplitude level.
I'm just stating an hypothesis & the evidence & logic that supports the hypothesis. You may want to believe that your are being rational & I'm not but so far I don't find your counter arguments particularly convincing.
Last edited:
Probably electrical imperfections but the design of sigma delta DACs that use noise shaping probably doesn't help.
Sure, if people take the concept literally i.e noise at -120dB is audible but what I'm saying is that a fluctuating noise floor at this level seems to have an audible effect - it may be that it perceivably alters the timbre or some other aspect. Remember some people can't understand how jitter can be audible - they take it literally - thinking that nanoseconds of mistiming cannot possibly be audible.
Yes, that's what Tony Lauck was mentioning earlier - the decorrelation of noise from the signal - in other words if the noise fluctuates in synchrony (or nearly so) with the music signal we know that this is a much more perceptible issue because of the way our auditory system works. If this fluctuation can be broken it will be far less or even completely imperceptible. This is why we can listen & enjoy LP's Tape (if they have a constant noise floor) - we have the ability to listen through such constant noise once below a certain level.
DAC manufacturers have long recognised the practical issues of the limitations imposed by the physical world in generating the small signals necessary in handling the least significant bits of 16 bit DACs - that's one reason why multibit DACs were expensive to manufacture - the manufacturing tolerance levels were being pushed. Even with these steps taken there were still techniques & tricks needed to linearise & make the DAC output more accurate. Further techniques were developed with the move to >16bits DACs. But there are many different approaches to trying to ensure this low level accuracy - not just one single approach - which demonstrates that the problem is not solved completely & yet this has been worked on for many, many years. There are many real-world issues to be dealt with when trying to implement the theoretical world of scientific theory into the real world of electronic devices & most of these issues still revolve around the known properties of capacitors & resistors even when they are miniturised to microscopic dimensions on DAC chips.
Note that for audio DACs you wont find a graph of a multitone test in their spec sheets - you will in some DACs designed for instrumentation purposes - this multitone test gives an indication of how the DAC might perform with more complex signals like music. The best figure I've seen for this test (SFDR) is around 80dB - meaning the side spur distortions are within 80dB of the main tones
Last edited:
False analogy. First, you're comparing complex audio reproduction systems to hearing thresholds. Many parameters vs one; a system vs a singularity. Microphones would be a better point of comparison, though not perfect. And their amplitude and FR sensitivities, while highly variable, are easily ascertained.
Nope. Not missing that at all.
Which is even easier to measure than FR range.
The evidence is thin. The logic is convoluted. But you design, manufacture and market DACs. if you're right, you've discovered a potential path to a lower audible noise floor that your competitors, and even those priced way out of your range, are ignorant of, or ignoring. You could create a superior product, a paradigm shift. You should acquire the equipment and the expertise, and go to the top of the DAC food chain. Given the prices of very high-end DACs that are ignoring your discovery, I'm sure it would be worth the investment.
Tim
Oops, do you think the perception of hearing is a singularity? Not only do we have the non-linear aspects of the ear mechanism itself but we have the highly non-linear aspects of auditory processing. So, no, not a false analogy
See my above comment
Indeed but what test signals are used? Do you think a noise test tone on it's own will establish what the limits of hearing are for how fluctuating noise may perceivably interact with concurrent music?
There you go - that's good logic
I am honestly not losing any sleep over it. There is always something better. If you had an 8 channel DAC with balanced output, I might even want to do a comparative audition of your ideas.
Since we now know that jkeny (WBF and other places) = mmerrill99 (CA Forum), my suggestion is that you put your status as a manufacturer in your signature there as you do here. Your posts are often filled with insights, but I think full disclosure would be best all around. It might even help your marketing.
No. I think measuring the amplitude and FR thresholds of the mechanisms of human hearing are completely separate from perception. The ear is the microphone. Perception is the rest of the studio.
Yeah, I saw it.
I already answered this. Do you want to know what test tones were used to establish that the world is round?
There you go - that's good logic
Glad you agree. You believe you've discovered an unmeasured/unused parameter that could lead to the development of superior audio components. If you really believe, I can't imagine why you wouldn't put your money where your mouth is.
Tim
If all that was being measured were the nerve signals coming out of the ear then I would agree with you & this can be done. But really divorcing hearing from perception in a test that tries to establish the thresholds that are PERCEIVED is silly - it's already part of the definition of what's being tested whether you admit this or not.
If you think I am the only one who holds this view about "currently accepted thresholds"s then you are wrong
Right, And you think this is an easily solvable issue? There are many sources that give rise to noise floor modulation - it's case of firstly identifying the conditions & sources of them, the perceptible effect of each one & then finding a way of dealing with each one. This ain't no easy task but do you not think others aren't working on this too? Look at Uptone Audio's Regen device, for instance - the underlying operation theory for it's audible improvement is because it improves the signal integrity of the USB signal which reduces the self-generated (fluctuating) noise at the USB receiver which is in the USB audio device. This very low level of fluctuating noise is hypothesised to affect the DAC/clock/analogue output circuitry. There is certainly an audible effect whether this is the correct explanation for it is not fully confirmed, yet
Last edited:
There is format noise, dictated by the mathematics of the coding scheme. In the case of multi-bit codes there is a theory of dither. This can be used to show that the quantization noise can be decorrelated through the n-th order by adding the sum of n rectangular distributed white noise. The commonly used triangular noise is just this, with n = 2. With one bit code as in DSD it isn't possible to use dither of the correct amplitude, as this will leave no head room for the signal. However, for high enough sampling rates, noise can be shaped so that there is little that remains in the so-called "audible band". What remains is going to be modulated.
In practice, a DAC that uses R-2-R or other techniques to achieve multibit will have elements that do not have ideal values. This will necessarily introduce noise modulation. These are electrical imperfections. There are ways (e.g. dynamic element matching) to decorrelate these distortions from the music at the expense of (hopefully constant) increased noise.
If a DAC uses a 1 bit modulator, then it can, in principle, be immune to noise modulation in the implementation since there is only one element involved and hence no need for component matching. This leaves non-linearities as a function of the modulator design and this can be analyzed by analysis or simulation. Unfortunately, these modulators are non-linear devices with feedback and defy any easy analysis (or at least any that I've found in the published literature). Here there is room for research and "secret sauce" design principles. It is always possible to measure these modulators if one has access to their digital output, but it won't be possible to test them completely as a black box algorithm because of the exponential number of cases involved.
One can produce statistically random digital bit streams quite easily today, thanks to the existence of high quality cryptographic algorithms. Converting these bit streams into the appropriate digital audio format is straight forward. Converting these to an analog signal will require a DAC, of course, and it will have some imperfections.
If by "nice problem" you mean "difficult problem" then I agree with you.
Thanks but I wrote to Chris many times, many years ago after being told to have a break. This was because I reacted badly to a suggestion of Chris's that I share some of my technical background when I suggested that a $12 RF attenuator was worth trying on the output of a Hiface USB to SPDIF converter (I would still react badly to such a form of censorship). He never answered any of my emails to be reinstated after my break became a break of indefinite duration.
- Status
