[paulchiu]

   
I have looked at the chips used for ADC's and the best ones offer -122 dB SNR, and THD and noise of -110dB. I am actually targeting 135 dB SNR, and THD and noise to be the same as Dave.
 
But by far the biggest problems are:
 
1. Depth perception - the work on Dave indicates that you need 350 dB capable noise shaping to accurately preserve depth - and current devices are only about -140 dB
 
2. Noise floor modulation - a chip based ADC has a lot of noise floor modulation, and I have solved this problem by using Pulse Array and discrete integrators.
 
3. Aliasing. The conventional view is that aliasing products in the 20-22.05 kHz range is inaudible. But it is not, as this aliasing errors degrade the timing of transients - which are audible.
 
I hope Davina will be less than Dave in terms of price. But since it is a technology prove of concept, its not about cost.
 
Rob

 
Pardon.  What is this Davina product?  The next top end DAC?

 

[Rob Watts]

Davina is the code word for the ADC project notionally intended for Dave.It will be the first product to use my Pulse Array ADC. 

 

[paulchiu]

  Davina is the code word for the ADC project notionally intended for Dave.It will be the first product to use my Pulse Array ADC. 

 
ADC? Is that an analog back to digital converter for some digital pre-amp upcoming?

 

[Mediahound]

ADC? Is that an analog back to digital converter for some digital pre-amp upcoming?

ADC units are used in the studio to record digitally.

 

[paulchiu]

ADC units are used in the studio to record digitally.

 
In that case, could there be a consumer application for the Davina?

 

[Mediahound]

   
In that case, could there be a consumer application for the Davina?

I can't think of one except for perhaps recording vinyl LPs to digital files on your computer. 

 

[paulchiu]

 
I can't think of one except for perhaps recording vinyl LPs to digital files on your computer. 

 
That's interesting.  I would like some high bits tracks of old LPs.

 

[Kamil21]

Hi Rob, in the latest issue of The Absolute Sound, Robert Harley said ,"MQA is the most significant audio technology of my lifetime.".

I know hat you may have debated this before, but I would think now that if Tidal were to make their files MQA compatible , it would be a service to your customers to seriously consider introducing MQA upgrade to at least DAVE.

 

[x RELIC x]

Hi Rob, in the latest issue of The Absolute Sound, Robert Harley said ,"MQA is the most significant audio technology of my lifetime.".

I know hat you may have debated this before, but I would think now that if Tidal were to make their files MQA compatible , it would be a service to your customers to seriously consider introducing MQA upgrade to at least DAVE.

As far as I can tell MQA is simply another codec that handles high resolution formats with a bit rate that makes streaming of such files feasible. The DAC would not have anything to do with it as it would simply receive a decoded MQA bit stream from the source, just like the DAC is oblivious to ALAC, FLAC, WAV etc. PCM formats when it receives the bit stream. The source is where MQA counts (software decoding), not the DAC AFAIK.

 

[Kamil21]

As far as I can tell MQA is simply another codec that handles high resolution formats with a bit rate that makes streaming of such files feasible. The DAC would not have anything to do with it as it would simply receive a decoded MQA bit stream from the source, just like the DAC is oblivious to ALAC, FLAC, WAV etc. PCM formats when it receives the bit stream. The source is where MQA counts (software decoding), not the DAC AFAIK.

MQA files can be played back without a decoder, but one will not get the full benefit of the improvement in sound quality without the full decoding that has to be incorporated by the dac manufacturer.

http://www.theabsolutesound.com/articles/beyond-high-resolution/

http://www.theabsolutesound.com/articles/robert-harley-listens-to-meridian-mqa/

 

[x RELIC x]

MQA files can be played back without a decoder, but one will not get the full benefit of the improvement in sound quality without the full decoding that has to be incorporated by the dac manufacturer.

http://www.theabsolutesound.com/articles/beyond-high-resolution/

http://www.theabsolutesound.com/articles/robert-harley-listens-to-meridian-mqa/

Yes, I've read them. The decoder is essential if the DAC doesn't have MQA specific filters. It's a flexible format and you can use either software or hardware, or both, but as noted in the article the DAC will see the bit stream just as it would another format as long as the DAC can support the incoming sampling rate. Backwards compatibility is a key point in their marketing. Still, I'd also be interested to know Rob's thoughts on MQA.

 

[originalsnuffy]

Seems to me that for MQA to really make a difference recordings have to originally be recorded using the technology.  That will take quite a while to have an impact in the marketplace.
 
We really need yet another format.

 

[Mython]

I suspect accuracy of the ADC is a much more critical factor than MQA, even though MQA might (perhaps) make a worthwhile contribution to a well-recorded music file.
 
 
Rob is currently concentrating on the former, so I suspect his energies are already directed towards the most significant aspect.

 

[Rob Watts]

I was replying to a post in the Dave thread, and my reply started to get pretty big. So I have turned it into a blog post about how pulse array works....
 
Quote:
"Originally Posted by JaZZ 
 
Thanks, Jawed, for explaining!
 
I have always asked myself what exactly this «Pulse Array» could be. Now it turns out to be based on Pulse Width Modulation (PWM), which I happen to be somewhat familiar with."
 
 
Pulse array is much more than PWM, as a simple PWM system has a considerable number of problems.
 
Let's backtrack a bit to DSD so we can see why you would want to move to a PWM based system. Imagine a DSD 64, so you only have 64 unitary value pulses to encode a 22.05 kHz sine wave, and you want to encode a -140 dB sine wave to reasonable amplitude accuracy.
 
The answer is - forget it. There simply are not enough pulses to accurately encode high frequency information in spite of what DSD fan boys may tell you. The best you can get with DSD 64 is 120 dB THD and noise 20 to 20k. So if we want more resolution, we need to encode more data (n bits not 1 bit) or run at a much faster rate.
 
The second problem that DSD has is down to switching activity. Now you may say that the benefit of DSD is that it is a single element, so therefore no distortion (unlike R2R where it is impossible to have acceptable distortion performance due to resistor mismatch and switching activity). But examine this DSD sequence:
 
0101010101010101
 
Or this sequence
 
1100110011001100
 
They both are reproducing analogue 0, both legal sequences that a noise shaper may adopt. But if you run with those values for ever, you will measure very different DC values with a real DAC - even one with a perfect clock. The reason for this is switching activity - the first sequence is switching at double the rate. Now a real output has N and P ch FET's in the output buffer, and these devices have different propagation delays, different OP impedances, and physically different layout with different inductance - the result of this is that a real device will always have a mismatch or error when reproducing a pulse - ringing or glitch energy, differing rise times against fall times, and this creates a switching error that then will non-linearly demodulate the noise shaper sequence. The result - distortion, noise, and idle pattern noise - and this is all before we start talking about jitter and RF intermodulation effects. So although DSD is claimed to have one element and so no distortion - in practice that is not true, as it has two states and those states are different.
 
So how do we solve the resolution problem and the switching activity problem but retain the single element benefit?
 
Use PWM. Let's take our DSD 64 and use a 2048 FS clock. Run the noise shaper at 64 FS, and use the extra clock cycles to encode more resolution - in this case we have 0 to 32 possible states, so instead of DSD's 1 bit we now have 6 bits (to encode 32 we need 6 bits), so much more resolution. Let's also ban 0 and 32 as a legal state, now we have a system where it is always switching at exactly the same frequency so now the error introduced by switching becomes constant and signal independent - which just means a DC error, no distortion or noise!
 
Audio nirvana, we have solved the switching error and increased the noise shaper resolution by 32 times so now we will get 150 dB performance and no distortion from switching!
 
Err - no. We still have a jitter problem - the output swings from fully + to fully -, and any master clock jitter will create noise - now it's not as bad as DSD, as the jitter induced noise is not so signal dependent (with DSD we get maximum switching activity reproducing zero, none at 100% modulation) - but it is still an issue.
 
And PWM as a modulation process creates HF distortion. Now you can reduce that by running the noise shaper at 2048FS and feed the switching behaviour back into the feedback loop, so the feedback of the noise shaper will solve it. But we have another problem - timing. Any reader of my posts will know that the timing of transients is a very important concept for me - hence the WTA filter etc. But with PWM you innately have a timing of transients' issue, as the amplitude of the signal affects the timing of the output - simply a higher amplitude signal will be earlier than a low amplitude signal.
 
Oh dear. So now we have timing problems too. But we had timing problems with DSD anyway - a small amplitude transient has a bigger delay than a large amplitude transient. This is because a small value transient will not get thru the OP quantizer, so needs the noise shaper integrators to build up in value to trigger the OP. You can see this problem in simulation, and it's of the order of uS - in timing terms massive. It's this error that gives DSD its unnatural soft sound - because transients have amplitude errors, the brain can't process the transient information properly, and hence you can't perceive (hear) the transients. And when you can't hear a transient then things sound soft - it's the audible equivalence of an image going out of focus.
 
So PWM helps but has other issues. Where do we go from here?
 
Let's take a step back, and examine the scale of the problem. In order to solve the resolution problem, what do we need for perfect transparency? And what level do we need for perfect timing?
 
Ask that question, how much resolution do we actually need will depend on doing listening tests to find out - theory won't help you as we have no idea how the brain process's ear data to determine depth for example. You can only do it by designing noise shapers and actually listening to them. In the Dave project I did exactly that and was shocked to the core by what I discovered - there is no limit to how good the noise shaper needs to be.
 
That's actually quite a profound statement. I can substantiate this by saying that a 200 dB (THD and noise 20 kHz BW) has much better depth perception than a 220dB one. Furthermore, listening to a noise shaper with better than 350 dB has audibly deeper depth than a 330 dB one (albeit it's getting to be a much smaller change). These are staggeringly small numbers - better than 350 dB has a trillion times more resolving power than DSD 64 - no wonder DSD has such poor depth perception - the noise shaping is simply not good enough at resolving small signals. Indeed, the required performance is still 30 billion times better than the PWM example I talked about earlier. Now it's possible that the requirement for better than 350 dB performance on the noise shaper was simply a proxy for something else going on in the analogue section - but that when I  use this performance for digital noise shapers that are purely digital (truncation noise shaping) I still hear the benefits in terms of depth reproduction. So the evidence based on my listening is that the brain is sensitive to small signal amplitude error of any size - even if almost zero and we need > 350 dB noise shaping
 
What can we say about timing errors?
 
If you would have asked me this a few years ago, I would have said uS accuracy was needed. Now I make no such assumption - there is perhaps no limit to how good the timing of transients need to be. So how can I substantiate that bold statement? Unlike noise shapers, it's rather difficult to put a number to timing accuracy. I guess I ought to state what I mean by transient timing accuracy. I do not mean - unlike the rest of the audio business - ringing performance; this is absolutely not what I am thinking about when I talk about the time domain or timing accuracy. Ringing uses an illegal signal from sampling theory POV as it is not bandwidth limited, so you would not actually get a perfect impulse from a perfect legal bandwidth limited ADC. So why worry about a signal you will never get? So it is actually pointless talking about it. What I mean is the accuracy of the timing of transients. Imagine a bandwidth limited analogue signal that is being sampled in the ADC - it is fully negative, goes positive and at some time crosses through zero. Let us say it is sampled at 44.1 kHz, so every 22,676 nS it's sampled. Let us imagine that the signal is sampled, and then crosses through zero at exactly 20,155 nS after sampling. Of course, when it gets sampled again at +22,676 nS it will now be a positive value. The question is, when the DAC reconstructs the sampled data - converting sampled data back to a continuous analogue signal - when will the signal cross thru zero? Theory is completely clear and undeniable - if we use an infinite oversampling FIR filter with a sinc response at 22,676 nS and a perfect DAC we will reconstruct the time it crosses thru zero absolutely perfectly at 20,155 nS. But with a finite non sinc function reconstruction filter, it will not cross thru at exactly 20,155 - maybe at 19,000 nS or 21,000 nS. And it is these differences in the timing of transients, are what I am talking about. Now in the past I would have said that getting it right to a uS was perhaps OK (timing errors can be as big as 100uS in conventional filters) - now I know that instead of worrying about uS we need to worry about getting it correct to nS's.
 
What is the evidence for that view? In designing Dave, I wanted to discover what I had done in the Hugo design (it was a happy accident) to give me the timing performance that I so enjoyed with it. By this I mean the ability to hear the stopping and starting of notes. After trying different things, I chased down this quality to the interpolation filters after the WTA filter. Now with Hugo, I used a 16FS WTA filter, followed by a linear interpolator and a two stage IIR filter filtering up to 2048 FS. Changing this to a 256 FS WTA filter followed by my usual 3 stage filtering gave a massive change in sound quality - at this point Dave was sounding impossibly rich and smooth and almost soft sounding. By changing it to 256FS WTA gave a substantial change in character - it was still smooth, but very fast and you could hear the starting and stopping much more easily. It went in character from soft and smooth to fast and sharp - when the occasion demanded.
 
Now replacing the WTA from 16FS (data every 1,417 nS) to 256 FS (data every 89 nS) is technically very small in the sense that transient accuracy using a WTA against an IIR filter at this speed is not a vast change in the time domain - it is a very subtle difference, but was nonetheless extremely audible. What it tells me is that very small - impossibly small - timing errors are very significant for the brain's ability to process the ear data.
 
So we have timing uncertainty due to the filter reconstruction, and timing errors due to the noise shaper. To eliminate filter errors we need a filter that oversamples to the max possible, with as close as possible to a sinc response - hence why 256 FS WTA filters, and further digital filtering to 2048FS. But we also need a DAC that has no amplitude errors either - by this I mean small amplitude transients has exactly the same delay as large amplitude transients - otherwise the brain can't perceive the starting and stopping of notes and things sound soft. Now we can't use DSD - the 1 bit noise shaper errors amplitude timing errors are greater than a uS - we can't use PWM, as this also creates amplitude related timing errors - we can't use R2R or ladder DAC's as they can't go faster than 16FS so will have interpolation filter timing errors.
 
So you can see now why I wanted to solve these problems and others well over 20 years ago by inventing pulse array. So how does pulse array solve these issues? The problem with PWM is the switching activity of one element. If we have 16 PWM states, and then use 16 different elements with different timing, then we can arrange the system so that a rising edge is always balanced by a falling edge for a constant OP, so that when they are added together there is no net switching activity. In short, with 16 elements and 16 PWM time states we can completely balance switching activity - this in a nutshell is how pulse array works.
 
The primary benefit of the system are that when the noise shaper asks for a value 8 say, then the output will immediately give a value 8 constantly with no variation. With all rising edges being matched by a falling edge, for a constant noise shaper value, the clock jitter has no effect whatsoever, so the system is innately jitter insensitive.
 
The key benefits are:
 
1. Element mismatch. The value of the resistor makes no difference to distortion at all as each element carries the wanted audio signal equally. Measurement and simulation has shown that resistor tolerance only creates a fixed unvarying noise. With Dave (20 pulse array elements) this noise is small compared to the amplifier and Johnson noise of the resistors.
 
2. Clock independent. The output from the array has a rising edge exactly balanced with a falling edge for all possible outputs actually used. Thus when it is reproducing a constant noise shaper value, common master clock jitter has no effect. The only possibility for jitter to play a part is when the noise shaper changes its output - and here the error is 1/16 of the DSD or PWM error. Simulations of adding jitter to the elements revealed negligible output of a fixed noise - negligible as it was very small compared to the resistor and amplifier noise. 
 
3. Signal independent switching. The switching errors (rise time fall time and glitches) are constant so are signal independent - thus no noise floor modulation or distortion.
 
4. No transient timing problems. Since the analogue output is identical to the noise shaper output, the problems of PWM creating timing errors are eliminated. Since the noise shaper can run at 104MHz, I have failed to see any problems of transient timing errors; the output crosses thru zero with small transients and large transients with no consistent error (unlike DSD).
 
5. With the very fast clock and noise shaping running at 104 MHz, I can make noise shapers with better than 350 dB THD and noise performance - which I have found to be essential for realistic depth perception.
 
But talk is easy, measurements are the proof of the pudding. Take a look at this simulation. Now this is the actual analogue output from pulse array but with of course ideal analogue performance, but it proves that pulse array is innately capable of resolving a -301 dB signal perfectly, which is now my standard test for depth capability:
 

 
Here we have Dave's 20 elements reproducing a 6 kHz -301 dB sine wave. It is accurate to within 0.001 dB, and has no distortion whatsoever.
 
What is also very unusual is the high frequency performance - at 100 kHz the noise floor from the noise shaper is at -200dB. This is crucial, as it means that Dave needs very little analogue filtering, and it means the output amplifier does not have to deal with large amounts of out of band or RF noise. This is very important for two reasons - I do not need to have a complex analogue filter on the output, and the lack of RF noise innately gives much lower noise floor modulation.
 
Indeed, because the filtering needs are simple, I actually only need one amplifier, combining the roles of I to V, filtering, output buffering and headphone drive in one single amp and having very simple analogue electronics maximizes transparency.
 
Noise floor modulation is subjectively very important too as it adds hardness and grain to the sound when measurable; even when reducing it below measurable levels you can still hear it; as reducing it further makes things sound smoother and warmer. Pulse array is the only DAC technology that enables no measurable noise floor modulation - all other forms of DAC's I have seen have major noise floor modulation problems. To prove this, here is an actual measurement using an APX555:
 

 
The plot in blue shows 2.5v RMS 1 kHz, and red shows the AP with no signal. What you can see from this is the complete absence of any noise floor modulation at all - noise is just fixed and unvarying.
  
Of course with pulse array there are other things going on under the hood which are too complex to explain, in particular how I designed a noise shaper that can perfectly resolve a -301 dB signal. And of course, some will say that we simply do not need that level of performance. Indeed, before I actually listened to 350 dB capable noise shapers, I too would have agreed with that view!
 
Rob 

 

[originalsnuffy]

Rob ,
 
For us plebes here; what is your view on the relative merits of redbook PCM vs higher bit rate and sampling rates of PCM?