[x RELIC x]

@sm60

Please see this quote from Rob, or search for "digital truncation" quotes by @Rob Watts

Yes - as the digital data-path (from input right through to the pulse array outputs) is better than 350 dB, so all truncation errors are eliminated - thus a major change in volume will have no effect on detail resolution and (more critically) depth perception.

 

[Rob Watts]

Again, it comes back to what is more “accurate“. If you think digital truncation of bits is more “accurate”, more power to you. I don’t, knowing the inherent trade offs involved in DSP, and having taken grad level courses in DSP. You need to look at signal to noise ratio of digital audio not w.r.t. 0 dB, which is meaningless, but at the signal level in question. If I truncate the output of Dave to -40dB, the idea that it maintains the same S/N ratio that it does at -3dB or 0dB is absolute poppycock. You haven’t understood how digital audio works then. I suggest reading Alan Oppenheim’s books on DSP or similar grad level textbooks. There’s no free lunch in digital audio. If you believe that, their marketing propaganda has worked. Pay close attention to how distortion rises sharply in digital audio as signal level drops.

The idea that a balanced tube preamplifier with distortion levels around 0.001% is obviously coloring the sound is a bit far fetched. Take a look at the distortion measurements of any recent ARC Reference tube preamp in Stereophile. Or look at McIntosh’s latest model 1100 two chassis tube preamplifier. Their previous generation 1000 preamp came in both solid state and tube versions. For the 1100, they decided to only release the tube version, because, and I’m quoting here, “it is the quietest preamplifier McIntosh has ever designed”. Check out its specs on their web page.

if you think running Dave in preamp mode is the most accurate, I strongly disagree and challenge you to produce measurements that back up your claim (i.e., show me the Dave at -40dB maintains the same S/N ratio it does at 0dB). If you say I like the sound of Dave as a preamp, I can’t argue with you. That’s your personal choice and we are all entitled our choice. Just don’t claim it’s because it’s more accurate. That I’d like to see a proof for.

Regarding the use of external power supplies for Dave, that’s again your choice whether you want to plonk down 10 grand for an external power supply for a DAC that’s already hugely overpriced (The Topping DAC decodes more formats than the Dave, measures better, and costs 1/10th the price, just in case you think the Dave is priced reasonably). But it’s your decision and your money. Once again, I have seen no measurements of any kind that support using the external power supply for the Dave. Does it improve the S/N ratio? Can you measure the improvement? It’s a risky business in any case. If it blows up the Dave, you’re out of not only the money spent on the Dave, but the power supply as well. At least in the case of Naim, they design their products for external power supplies and support it in case something bad happens. Here you are completely screwed if there’s a power glitch and the Dave fries (don’t forget DSP chips are notoriously fickle about static and they are very sensitive to even slight glitches in a power supply).

Sorry but you are incorrect. The output from the volume control is simply not truncated - nor is it dithered - but aggressively noise shaped, so that the signals within the audio bandwidth are perfectly preserved. By perfectly, I mean it will reproduce a -301dB signal to an accuracy better than +/-0.001 dB and with a phase shift of within +/- 0.001 degrees; this level of accuracy is essential in order to maintain the perception of depth and detail resolution. To illustrate the power of this noise shaper look at the 16FS output of the M scaler's truncator to 24 bits:

You can see that the noise floor is at -390dB - and this is actually down to my test vectors. The noise shaper starts to kick in at around 16 kHz. Subjectively, this is completely transparent, in that there is no loss in SQ if a digital module can reproduce -301dB with zero amplitude or phase errors - from this perspective, then the volume control function can be considered lossless. All of my modules have to pass this demanding test, plus of course listening tests, before they can be incorporated into a design.

This approach is very radical, and to some would be considered excessive; but after many listening tests, it is the only way to accurately reproduce soundstage depth. Generally, digital volume controls are not done with such aggressive 11th order noise shaping - they are normally dithered when truncating, which is subjectively not lossless. Some don't even do that, just discard bits, which is just plain stupid/incompetent due to the severe small signal distortions introduced.

PS - just so you can see the -301.03 dB signal zoomed in:

The level is -301.030 dB, against -301.299957 dB ideal value

 

[MarkusBarkus]

...oops. Back to online grad school this time.

 

[miketlse]

...oops. Back to online grad school this time.

I don't think attending grad school was the problem here.

The problem arose because of forgetting to first check if the mathematical model for digital volume control studied at grad school (or described in the books), was the same as the mathematical model used in the DAVE for digital volume control - before posting and making incorrect claims.

Grad school can teach theory, but not common sense or experience.

 

[Triode User]

I don't think attending grad school was the problem here.

The problem arose because of forgetting to first check if the mathematical model for digital volume control studied at grad school (or described in the books), was the same as the mathematical model used in the DAVE for digital volume control - before posting and making incorrect claims.

Grad school can teach theory, but not common sense or experience.

Also, i knew I was comparing the sound of the Dave pre amp output to an almost totally transparent pre amp with virtually no noise and no distortion (the Music First Baby Ref V2).

All good fun.

 

[penguin69]

When you digitally reduce the volume down by -30 to -40 dB, you throw away a lot of bits and all this talk of transparency goes out the window.

I think @Rob Watts produced a slide deck a couple of years ago showing why his digital attenuation design does not compromise SQ. If I can find the link, I will post it. Agree, though, that digital volume controls degrade SQ in standard designs as you are throwing away information. (I've designed one or two myself.)

EDIT: apologies, I was reading the posts in chronological order and see now that Rob has already replied.

 

[jlbrach]

Great question! One I ask myself every week. . Each time I look at the clunky setup with the Blu Mk2 and the Dave, which my wife refers to as “the spaceship”, I view it as another of my insanities. Originally I got the Blu Mk2 with the Dave because I liked its compact form factor, its ability to be used as a preamp and to drive a pair of headphones. If I downsize my setup, which is a bit excessive, I thought the Blu/Dave would be great to have since they would fit in a small space. But the Blu proved so unreliable as a transport, the Dave is so underwhelming as a headphone amplifier and the software so glitchy that it produces a blast of digital noise every time I power down the Blu or the Dave that I realized the idiocy of owning this overpriced combo! If I told my colleagues I have a digital playback device that costs over 15 grand and is very unreliable, they’d suggest I urgently seek psychiatric help!

go figure, I love my dave/blu2 combo...it has been stable and performed beautifully for me and I think the dave headphone out is marvelous as long as you arent using a few of the hardest to drive headphones

 

[sm60]

Have u listened to a Topping? Or done A/B with a Dave? Measurements don’t mean much. We don’t eat chemical goop based on some lab report of what is nutritional. You cannot measure something when the quality itself in the end is mostly subjective. You can’t do crowd based statistical research either, because it doesn’t matter what percentage of the crowd says chocolate tastes better if my mouth always says different. Even if measurements of brain waves said that strawberry was worse and 99% of people agreed, decreasing production of strawberry ice cream and making it more expenisive, it would still be worth it to me if I got more joy out of eating it. Joy is end game here, not solid measurements. Enjoyment! That is the only standard worth anything. If engineering measurements help that, great. If they don’t, then they don’t matter. The variety of products out there in terms of engineering methods shows that there is no golden standard that can correlate measurements with human enjoyment across the board. We are too different from one another, listening to different music and looking for different things in that music/experience. If measurements are a part of what brings you joy and offers pride of ownership or a feeling that it was money well spent, then good for you. Manufacturers offer specs, so you can use them to map your way. I have no issue with that and even do it myself. But the measurement crowd is always gaslighting—always trying to tell me what I am experiencing and that my enjoyment isn’t real. Some of the subjective crowd do as well. But gaslighting people’s harmless enjoyment is being a partypooper, a bummer, a downer… if we are going to offer opinions here, it should be of the stuff we actually tried. We should discuss what we heard and why we liked it better. Why it was more enjoyable. That contributes something meaningful and helps me make better decisions about what to try myself…that’s my two cents.

IMHO the unreliability u r experiencing is maybe a good enough reason to look elsewhere. I understand budget. Everybody has to face that, or most of us at least. We are lucky that there are good products at almost any price range these days. I don’t think the Dave Is overpriced though. A lot more went into it than a Topping DAC, in a different county with way different overhead, research and development processes and corporate structure. It contains intellectual property beyond the materials, in other words. It also can compete with dacs way beyond it in price, especially if you add a DC4

I agree measurements don’t tell the whole story especially in digital audio. But they do dispel some myths. For example, if you feed the Dave a 24 bit signal, you might think the Dave can actually resolve all 24 bits. In fact, its resolution is less than 20 bits — see measurements in Stereophile. This is not a put down of the Dave. It is a very good DAC in terms of its measurements, better than many. No DAC actually resolves all 24 bits! It’s all marketing hoopla, now made worse by the fact that every cheap CD player uses 32-bit D-to-A converters. Of course, when you measure these converters, their linearity is often worse than even 24-bit converters. Recent Esoteric DACs claim to use 35-bit DACs. I’m waiting for some company to claim it is using 64 or even 128-bit DACs. You can make up any story you want in high end audio, but when you do simple measurements, you see it’s all marketing fluff. It’s like Roon advertising its HQ-player can upsample everything to DSD 512. Golly, now all your 16-bit recordings are magically transformed to DSD 512 remastered recordings. Of course, when you listen to these, often it sounds worse than just playing back the un-upsampled tracks. It’s like thinking if I take a NTSC 480i TV signal and upsample it to 4K or 8K, voila, all my old NTSC VCR tapes or TV broadcasts are 4K or 8K now!

 

[iDesign]

go figure, I love my dave/blu2 combo...it has been stable and performed beautifully for me and I think the dave headphone out is marvelous as long as you arent using a few of the hardest to drive headphones

Same experience.

 

[miketlse]

But they do dispel some myths. For example, if you feed the Dave a 24 bit signal, you might think the Dave can actually resolve all 24 bits. In fact, its resolution is less than 20 bits — see measurements in Stereophile.

Rob has already openly posted that DAVE + MScaler will perfectly resolve to 16.2 bits (i think that was the decimal place), and that to perfectly resolve 24 bits would require DAVE + 256MScaler.
I don't think anyone is being conned on this issue.

 

[Rob Watts]

I agree measurements don’t tell the whole story especially in digital audio. But they do dispel some myths. For example, if you feed the Dave a 24 bit signal, you might think the Dave can actually resolve all 24 bits. In fact, its resolution is less than 20 bits — see measurements in Stereophile. This is not a put down of the Dave. It is a very good DAC in terms of its measurements, better than many. No DAC actually resolves all 24 bits! It’s all marketing hoopla, now made worse by the fact that every cheap CD player uses 32-bit D-to-A converters. Of course, when you measure these converters, their linearity is often worse than even 24-bit converters. Recent Esoteric DACs claim to use 35-bit DACs. I’m waiting for some company to claim it is using 64 or even 128-bit DACs. You can make up any story you want in high end audio, but when you do simple measurements, you see it’s all marketing fluff. It’s like Roon advertising its HQ-player can upsample everything to DSD 512. Golly, now all your 16-bit recordings are magically transformed to DSD 512 remastered recordings. Of course, when you listen to these, often it sounds worse than just playing back the un-upsampled tracks. It’s like thinking if I take a NTSC 480i TV signal and upsample it to 4K or 8K, voila, all my old NTSC VCR tapes or TV broadcasts are 4K or 8K now!

It depends upon what you mean by resolution; your using the term to compare the fixed noise from Dave and using that as resolution, which in Dave's case is 21.3 bits. But that isn't what most people term as resolution in that resolution is defined as the ability to resolve small signals. In the case of Dave it can easily resolve a +/- 1 bit LSB from 32 bits, that is -186dB. Here is a plot from the analogue outputs of Dave showing it resolving -186dB at 2.496kHz:

Of course a -186 dB signal is well below the noise floor of Dave (noise floor is -180dB) and this was done by using the AP's 6kHz sampling and synchronous averaging to reduce noise.

I hadn't published this before, as all of my DACs can do it, and all of them show a +0.6dB amplitude error - since it's consistent error across all my DACs, I am assuming the problem is down to the ADC in the AP - so we will have to await my pulse array ADC to confirm this.

Defining resolution as equating it to in bandwidth noise is facile, as an inaudible fixed noise is inaudible; what is vitally important subjectively is how well small signals are actually resolved in terms of amplitude.

Rob has already openly posted that DAVE + MScaler will perfectly resolve to 16.2 bits (i think that was the decimal place), and that to perfectly resolve 24 bits would require DAVE + 256MScaler.
I don't think anyone is being conned on this issue.

Yes I used the term to indicate that 1M taps with WTA would give 16.7 bits of coefficients that were ideal Whittaker Shannon accuracy, and so would reconstruct the timing of transients to a better accuracy than this. Unfortunately, this is a bit of a simplified picture, as reconstructing transients perfectly is a great deal more complex than this!

My recent research is indicating that we need much better accuracy than 16.7 bits - the ear/brain is extremely sensitive to vanishingly small timing errors.

 

[Triode User]

My recent research is indicating that we need much better accuracy than 16.7 bits - the ear/brain is extremely sensitive to vanishingly small timing errors.

Rob, I have extracted just one line from your reply and I am going to read between it and sense a new product.

Reading between one line reminds be of the joke, What is the difference between a duck? One of its legs is both the same.

 

[Ragnar-BY]

My recent research is indicating that we need much better accuracy than 16.7 bits

How many taps do we need for “next level M Scaler” to get subjective improvement comparable to improvement between 500k and 1M in current model? 2M? 24M?

 

[miketlse]

How many taps do we need for “next level M Scaler” to get subjective improvement comparable to improvement between 500k and 1M in current model? 2M? 24M?

Buried in some of Robs historical posts, is the comment that doubling the number of taps, increases the accuracy by 1 bit.
So a doubling the taps via a 2MScaler would enable 17.7 bit accuracy.
A further doubling via a 4MScaler would enable 18.7 bit accuracy.
And so on, until a 256MScaler would enable 24.7 bit accuracy.

Of course that relates to the mathematical model, describing the technical performance.

The subjective performance, which involves the non-linear ear/brain processing - probably only Rob knows the reality (based on current testing).

 

[The Jester]

@rob,
With ever increasing small signal accuracy giving better depth perception and our own ability to perceive relative distances in sound, similar to the way we perceive direction from minor timing differences from ear to ear ?
Is it possible that this is caused by more and more reflected vs direct sound reaching our ears at minute differences in level and timing as the distance increases, and the more accurate those signals are replayed the better we hear depth in recordings ?