[Rob Watts]

  Could you provide us with some signals where the timing errors reduced by your product should be detectable in other products?
 
For reference, here is the full paper linked in the comments:
http://arxiv.org/pdf/1208.4611.pdf
 
See also this rebuttal:
http://arxiv.org/pdf/1501.06890v1.pdf

Demonstrating the timing uncertainty via measurement is not simple. But I have been working on a new test technique to allow this; but it requires a near perfect decimation filter, but this is something I will be doing next year with the pro audio ADC project.
 
In the mean time here is the stop band performance from Mojo. As you can see the filter is almost an ideal brick wall filter, with no aliasing products visible.
 

 
This is done using random noise at 0 dBFS with 48 k sample rat sing my APX555.
 
Rob

 

[Rob Watts]

   I'm a huge supporter of oversampling, not sure if we need to go that far, but if there are no huge drawback, that's fine by me.
is it zero padding? if it is, what's the attenuation as a result? (I guess RRod you can answer that just as well). not that it actually matters, it's pure curiosity. I only have seen the model for the most basic delta sigma and never thought about what would happen with a lot more.
edit: forget this, first because I left half of the question in my head so it looks like I'm asking about the output signal when I was asking about dtft. and second reason to forget it, is that after looking it up a little, I had a wrong idea from looking at some PDF about another DAC where I just didn't see that the dtft graphs were showing linear values instead of DBs(thus my idea that they were doing something special when they really didn't

).
 
personally I don't see portable amps as voltage gain or current gain tools, I'm a hardcore IEM user so unless the source really sucks with current, power is usually not a concern in my life. my actual amp is used as a literal de-hissers for my overly sensitive IEMs. shigzeo mentioned some low hiss and I trust his hiss estimates with my life, so I know it's not for me. but for slightly less sensitive IEMs, portable headphones and apparently given the specs it would also drive my hd650 very well and very loud, then for those uses, it does look like a great pocketable product. 
 
I'm also curious about the impedance output, it's really great to have amps that are actual near zero ohm at the output, at least on paper, but it rarely happens even for gear with negative feedback. so as almost everybody else sticks above 0.5ohm or so, I wonder what are the drawbacks/difficulties of such a choice?

Mojo shares the output stage topology from Hugo - a single gain stage with only one negative feedback path. This gain stage is a discrete Class A (300 ohm) rail to rail op-amp hybrid. The only problem I found was making the OP stage fast enough so that stability margins were not compromised. This gain stage does the DAC I to V, analogue filtering, and output drive in a single stage. I get 75 milli ohm output impedance, with a total THD and noise of 0.00017% at 3v RMS 1 kHz into a 300 ohm load.
 
In terms of design philosophy I treated the gain stage as if it was a power amp - indeed Mojo will drive efficient horn loudspeakers to surprising loud levels. When I first became involved in headphone amp design I was perplexed as to why existing units were so poor technically. It is not difficult getting very low output impedance but is essential for damping factor and to retain low distortion with non-linear headphone loads.
 
Rob

 

[analogsurviver]

  Demonstrating the timing uncertainty via measurement is not simple. But I have been working on a new test technique to allow this; but it requires a near perfect decimation filter, but this is something I will be doing next year with the pro audio ADC project.
 
In the mean time here is the stop band performance from Mojo. As you can see the filter is almost an ideal brick wall filter, with no aliasing products visible.
 

 
This is done using random noise at 0 dBFS with 48 k sample rat sing my APX555.
 
Rob

As good as the above result is, I am infinitely FAR more interested in the forthcoming ADC project. Which formats are going to be supported - particularly being interested in DSD256 and above.
 
DACs outnumber ADCs by a factor of ??? - yet even a perfect DAC can not but reproduce the ADC errors.

 

[RRod]

  Demonstrating the timing uncertainty via measurement is not simple. But I have been working on a new test technique to allow this; but it requires a near perfect decimation filter, but this is something I will be doing next year with the pro audio ADC project.
 
In the mean time here is the stop band performance from Mojo. As you can see the filter is almost an ideal brick wall filter, with no aliasing products visible.
 
 
This is done using random noise at 0 dBFS with 48 k sample rat sing my APX555.
 
Rob

 
The question is still "what are the audible time benefits from having a reconstruction that is closer to the Whittaker-Shannon ideal?" So we'll just have to wait for your test technique.

 

[prot]

I feel like I'm playing the devils advocate here, because deep inside I'm with you. but about myself, every time I get a new device to toy with, I use it casually for a few days, and sure enough, just like anybody else, I come up with soundstage this, bass that, less than when I use XX but close to when I use YY...
I might not call it night& day, but I could make a review and be just as ridiculous as most stuff we read everyday.
then I take out the switch and try to match the volumes(not always easy depending on the device, but I try). and of wonders of wonders, differences melt like ice in my mouth. some differences usually stay, but the magnitude goes way down 100% of the time. which makes me think that I'm a living exaggerating machine when I use no control.
already I'm way more moderate and I've done my old reviews that way, sometimes just matching the loudness by ear . but it was enough already to make me look like a killjoy compared to other reviews.
nowadays I have a few low-fi measurement gear, so matching gear, recording, or doing an almost proper blind test(no double) has become my new normal when it comes to test gears. and the conclusions obviously are way more down to earth.
what if I didn't have all this? what if I had never bought my first switch? IMO, I would still believe that my O2 had way more bass and soundstage than my leckerton (when I fail to tell them apart in blind test). and most likely I would claim it every time the subject would come up. 
so really what saves me from making a fool of myself are those gears and knowledge of my very human limitations.

about internet and knowledge, again I agree with the idea, and people who spent 10years in the hobby and still don't know that volume matching matters, they just have no excuse.
but because the audio hobby is such a mess, it's easy to find answers, but pretty hard to be sure they're not BS. I had to unlearn a good deal of preconceptions I got from reading audio reviews by guys who looked like they knew their stuff. so I blame the guy that doesn't even try, but not the guy who tries and get mislead.

You're doing a very good job as the devil too .. but the night&day people still have no escuse .. it's not 1975, it's 2015 and free info is everywhere, waiting to find a door to your brain.

 

[moshen]

  To conclude; 500 pages in 8 weeks and dozens of awards and 5 star reviews is not magic - just thirty years of very advanced engineering.
 
Rob

 
That was a lot of very awesome info, thanks for detailing it. Have you done well controlled blind studies that the improvements are audible? if so, it seems it would mean quite a bit in adding to our scientific knowledge of where the limits of transparency is. Also, it'd be the first for a DAC or Amp designer to do this, it seems most rely on customer subjective bias. All of the awards and reviews I've seen so far are very subjective and not done in an objective way..

 

[analogsurviver]

   
That was a lot of very awesome info, thanks for detailing it. Have you done well controlled blind studies that the improvements are audible? if so, it seems it would mean quite a bit in adding to our scientific knowledge of where the limits of transparency is. Also, it'd be the first for a DAC or Amp designer to do this, it seems most rely on customer subjective bias. All of the awards and reviews I've seen so far are very subjective and not done in an objective way..

 
Objective crowd is doing the greatest of disservice in this matter. Why?
 
1.) There is absolutely no set of measurements that can correctly describe any piece of audio gear with a decent correlation in subjective domain.
2.) There is absolutely no commercially available  instrument or set of instruments which could do the 1. )
3.) There is no denying that any simple single analog stage can sound different. The problem is that most people use inferior associated equipment to test such an advanced component with, both in objective and subjective domains -  which is NOT transparent enough, therefore they can not reach but the false conclusion "no difference heard" .
4.) The burden of the proof is on the side of those who claim an improvement. In principle very true. However, making the measurements and equipment  capable enough to scientifically prove the claim(s) may well require MUCH more work and finance than making the improvement itself - which can be heard and DBT ABXed under favourable circumstances.
5. Making these very measurements is bound to be unconventional, is bound to be associated with great delays (on order of approximately a decade) - until peer reviewed and approved. No manufacturer today can afford to pour labour and money into something that has so slow turnaround. This is one of the main reasons progress is not faster - that and patent protection. As competition is unlikely to be willing to pay the required fees to legally use such an improvement, it will resort to any tactics to slow  the progressive manufacturer down - including requiring peer approved evaluation, which definitely can be also used as means of buying time. Either to postpone the inevitable as much as possible or allow the time during which they can themselves come up with something at least comparable.
 
It does not matter whether the proposed improvement is really good or not - if it means a threat to their profits, it will be fought with any means at their disposal.
 
Historically, the most prominent recipient of such a "treatment" was Nikola Tesla:
https://en.wikipedia.org/wiki/Nikola_Tesla

 

[jcx]

Psychoacoustics is the Science of exploring perception of sound in reference to "Objective" descriptions of sound from Physical Acoustics, Signal Theory
 
in the EE domain we have ridiculously finely detailed, verified working theories of Signals, Noise, Digital Analog representation, interconversions
 
linking those with well established Psychoacoustics lets us predict lots of things in the EE domain can be designed to give errors way below any tested human hearing limits when converted to sound
 
 
 
Pshychoacoustics isn't "complete", many open and unexplored dimensions remain
 
but new results seldom invalidate broadly tested previous "facts" - in fact its usually required that a new theory explains previous results at least as well as the old, and gives new Testable predictions
 
perhaps you need to read : http://chem.tufts.edu/answersinscience/relativityofwrong.htm and other works on Engineering/Science/Technology Epistemology
 
not knowing everything is very different from knowing nothing
 
metals were smelted, cast, forged before modern chemistry, Phlogiston Theory came and went without changing the facts of metallurgy
 
 
 
in Audio once you know you can hear differences from properly performed Double Blind Tests, then there are options for ranking subjective impressions than go beyond AB/X:
 

  there is a listening test "Bible" - and a presentation from it: http://www.delta.dk/imported/senselab/AES125_Tutorial_T4_Perceptual_Audio_Evaluation_Tutorial.pdf
 
http://www.moultonlabs.com/main/ has a few interesting articles on listening and mastering
 
http://seanolive.blogspot.com/ is another source for commentary by pros experienced in perceptual audio evaluation
 
 
lots of research has been done on room and loudspeaker interaction, impact on listening experience

 

[castleofargh]

 

   
That was a lot of very awesome info, thanks for detailing it. Have you done well controlled blind studies that the improvements are audible? if so, it seems it would mean quite a bit in adding to our scientific knowledge of where the limits of transparency is. Also, it'd be the first for a DAC or Amp designer to do this, it seems most rely on customer subjective bias. All of the awards and reviews I've seen so far are very subjective and not done in an objective way..

 
Objective crowd is doing the greatest of disservice in this matter. Why?
 
1.) There is absolutely no set of measurements that can correctly describe any piece of audio gear with a decent correlation in subjective domain.
2.) There is absolutely no commercially available  instrument or set of instruments which could do the 1. )
3.) There is no denying that any simple single analog stage can sound different. The problem is that most people use inferior associated equipment to test such an advanced component with, both in objective and subjective domains -  which is NOT transparent enough, therefore they can not reach but the false conclusion "no difference heard" .
4.) The burden of the proof is on the side of those who claim an improvement. In principle very true. However, making the measurements and equipment  capable enough to scientifically prove the claim(s) may well require MUCH more work and finance than making the improvement itself - which can be heard and DBT ABXed under favourable circumstances.
5. Making these very measurements is bound to be unconventional, is bound to be associated with great delays (on order of approximately a decade) - until peer reviewed and approved. No manufacturer today can afford to pour labour and money into something that has so slow turnaround. This is one of the main reasons progress is not faster - that and patent protection. As competition is unlikely to be willing to pay the required fees to legally use such an improvement, it will resort to any tactics to slow  the progressive manufacturer down - including requiring peer approved evaluation, which definitely can be also used as means of buying time. Either to postpone the inevitable as much as possible or allow the time during which they can themselves come up with something at least comparable.
 
It does not matter whether the proposed improvement is really good or not - if it means a threat to their profits, it will be fought with any means at their disposal.
 
Historically, the most prominent recipient of such a "treatment" was Nikola Tesla:
https://en.wikipedia.org/wiki/Nikola_Tesla

 
open mind is always needed, we can't go anywhere without it. and knowledge that usually is a great help, can sometimes become a blindfold of it's own when we neglect to test something just because it doesn't fit in some other model we know of. I've been a victim of this just this month. and only unhealthy curiosity made me test it in the end when I was sure what to expect(and was wrong^_^). so I look like a jerk when I ask stuff to people, and I look like I have zero self confidence when talking about my stuff, but systematic skepticism is my open mind.
 
 I've thought about it long and hard in my small brain, but blind test is the only thing I will trust when it comes to audibility. maybe we will move on from ABX and use MRI, retinal scans and blood pressure to assess differences, instead of asking the subject for his opinion. and move on from the conscious elements to the subconscious elements. that's an exciting field of research IMO. but the person listening should never ever know what he's listening to. that is a sine qua non condition of audible testing!
 I often ask people blaming blind tests or simple ABX to offer us a viable alternative. I would really be glad to have alternatives that I could use myself. the more the merrier. but sighted evaluation will never ever be accepted as one. so better concentrate on tests that can get some legitimacy when positive.
 
IMO, AFAIK, TBH etc. ^_^

 

[Rob Watts]

   
The question is still "what are the audible time benefits from having a reconstruction that is closer to the Whittaker-Shannon ideal?" So we'll just have to wait for your test technique.

I can tell you the audible benefits of having an interpolation filter that is closer to the ideal - but its just my opinion. My listening tests are simply to allow me to design better audio products, not to give scientific proof.
 
But its a fact that the timing of transients is a very important perceptual cue - it gets used by the brain for a number of things:
 
1. Imagery via the interaural delay - time delays from left to right is partially used for left right image location. This delay has a gross resolution of about 4uS (some literature puts it at 10uS).
2. Timbre. The timbre of instruments is partially determined by transients.
3. Pitch - bass pitch is determined by the starting transient.
4. Perception of a note starting and stopping - obviously transients are crucial.
 
Increasing the tap length and optimizing the algorithm does indeed give subjective improvements in all of those areas. But even if I gave you a number for a notional time domain error test - lets say it was -90dB - you would still need to do careful listening tests to confirm if that number was significant to the brains processing of timing errors. And people would still argue...
 
What I can say is that as you increase the tap length, the sound quality changes get smaller, its absolutely converging. My current record is 164,000 FIR taps and this is about 14 bit accurate against ideal coefficients.
 
Rob
 
Rob

 

[uchihaitachi]

  I can tell you the audible benefits of having an interpolation filter that is closer to the ideal - but its just my opinion. My listening tests are simply to allow me to design better audio products, not to give scientific proof.
 
But its a fact that the timing of transients is a very important perceptual cue - it gets used by the brain for a number of things:
 
1. Imagery via the interaural delay - time delays from left to right is partially used for left right image location. This delay has a gross resolution of about 4uS (some literature puts it at 10uS).
2. Timbre. The timbre of instruments is partially determined by transients.
3. Pitch - bass pitch is determined by the starting transient.
4. Perception of a note starting and stopping - obviously transients are crucial.
 
Increasing the tap length and optimizing the algorithm does indeed give subjective improvements in all of those areas. But even if I gave you a number for a notional time domain error test - lets say it was -90dB - you would still need to do careful listening tests to confirm if that number was significant to the brains processing of timing errors. And people would still argue...
 
What I can say is that as you increase the tap length, the sound quality changes get smaller, its absolutely converging. My current record is 164,000 FIR taps and this is about 14 bit accurate against ideal coefficients.
 
Rob
 
Rob

Sorry to raise this again, but what you have posted so far is all familiar to me, but your comments on listening tests had my interest piqued. Did you have any statistically significant results comparing your products against other products that meet audibly clear thresholds. What do you also mean by 'subjective' improvements?

 

[Rob Watts]

No I don't listen to other products as a listening test - there are too many variables. So a listening test would be very specific with as little variance as possible - say 100,000 taps against 164,000 say, with all other things being constant (even the place and route on the FPGA). I employ listening tests to evaluate specific aberrations, then reduce those aberrations so that you can no longer hear a change - the goal is to be able to hear no difference. From this I will know what the requirements are for specific errors - for example, noise floor modulation, or for small signal linearity requirements for noise shapers.
 
One of the benefits today is that Verilog simulation tools are so advanced that one can do digital domain measurements that give numbers that you could never measure in the real world. You can then evaluate through subjective listening tests to see if these errors have a consequence - that's why I know with some confidence that some errors that are well below the threshold of audibility are significant as these errors interfere with the brains processing of ear data.
 
Rob    

 

[krismusic]

that's why I know with some confidence that some errors that are well below the threshold of audibility are significant as these errors interfere with the brains processing of ear data.

Rob    

That's a very interesting claim. How do you assess these subjective improvements?
I've often thought that placebo and expectation bias must play havoc for an audio designer. How do you screen for them?

 

[RRod]

  I can tell you the audible benefits of having an interpolation filter that is closer to the ideal - but its just my opinion. My listening tests are simply to allow me to design better audio products, not to give scientific proof.
 
But its a fact that the timing of transients is a very important perceptual cue - it gets used by the brain for a number of things:
 
1. Imagery via the interaural delay - time delays from left to right is partially used for left right image location. This delay has a gross resolution of about 4uS (some literature puts it at 10uS).
2. Timbre. The timbre of instruments is partially determined by transients.
3. Pitch - bass pitch is determined by the starting transient.
4. Perception of a note starting and stopping - obviously transients are crucial.
 
Increasing the tap length and optimizing the algorithm does indeed give subjective improvements in all of those areas. But even if I gave you a number for a notional time domain error test - lets say it was -90dB - you would still need to do careful listening tests to confirm if that number was significant to the brains processing of timing errors. And people would still argue...
 
What I can say is that as you increase the tap length, the sound quality changes get smaller, its absolutely converging. My current record is 164,000 FIR taps and this is about 14 bit accurate against ideal coefficients.

 
Of course we agree that timing is important, but again it's all about what's audible in controlled, blind conditions. Subjective arguments for more and more FIR taps smack a lot like arguments for higher and higher sample specs: more is more, and more sounds better, even if only a little bit.
 
Just as example of the kind of thing we're talking about, here's the result of a little example of taking impulses that are just under 3µs apart, taking them down to 44.1ksps, and then simulating 16x oversampling, eventually going back up to an "analog" rate of ~11MHz using two FIR lengths out of SoX. At 32767 taps, the difference in timing of the peak maximum is maintained exactly, whereas at 1024 taps there is an error of 2 samples (0.18µs) at the pseudo-analog rate. Such results naturally lead to questions like:
a) Are such errors in peak timing audible?
b) Are other features of the two filtering processes (ringing, rise times, etc.) audible?
 
I know I certainly wouldn't want such questions answered via sighted, subjective tests.

 

[Rob Watts]

That's a very interesting claim. How do you assess these subjective improvements?
I've often thought that placebo and expectation bias must play havoc for an audio designer. How do you screen for them?

Absolutely - and I have been guilty of making mistakes before. I have developed techniques to be as neutral an observer as possible, and to accurately characterise the sound quality differences. But actually the real problem with listening tests are not these - the differences are very easy to hear to a skilled listener; the problems start with making value judgements. Far too many designers like a particular sound and are designing to taste. Now this is OK if that is your intention; but I am trying to make something sound as transparent as possible, without ever having the perfect product available - so there is no absolute benchmark. The problem is making value judgements from the listening test as its very easy for a listener to prefer the added distortion or noise modulation.
 
This is one reason why my listening tests are generally very specific - the ideal is to be able to hear no difference with the adjustment of a specific aberration. Once you get to this position you know you have hit the bottom of the barrel with that particular aberration under test.
 
Sometimes you hear something very extraordinary - like an impossibly small error, or something that does not make sense technically. Then I use blind listening tests to confirm something unusual. 
 
Rob