T+a dac 8 dsd

[Miska]

I just received a mail from T&A with some information on their filtering:

 

"For DAC8 DSD we use a combination with an analogue filter, 3rd ordinal and a digital filter structure in analogue performance."

 

It looks like it has four traditional DAC chips for PCM and then similar conversion stage as DSC1 for DSD, but with 8 elements per side for balanced (16 total):

http://tapeconnection.dk/Resources/Files/Produktbilleder/TA/TA-DAC8-inside.jpg

 

How much does it cost in EU?

[Miska]

I'm not seeing the 8 elements (shift registers?) per side. Where? What would the sonic implication be of doubling the number of shift registers (FIR filter)? Assuming we are upconverting to DSD256-DSD512 the noise is already pushed way out?

 

OK, my bad, I think that was a resistor ladder analog volume control then, same as on the new one on both sides of the relay set. In this new one it seems to be between the four canned DAC chips. First two chips that may be a shift registers and then two chips that may be switches. I counted six resistors on both sides of those chips.

 

Not so easy to hunt for right inside pictures on the net and then trying to figure out from small resolution picture what there is on the board...

 

Here's straight link to the full picture of the DAC8-DSD:

http://www.rutherfordaudio.com/media/catalog/product/cache/3/image/1200x1200/9df78eab33525d08d6e5fb8d27136e95/d/a/dac8-dsd_innen.jpg

 

 

P.S. Curiously they seem to have transformer isolation between the DAC and digital side, see the component that says "Pulse" on top. Next to it may be another isolator, either capacitive or optical, possibly for other slower speed signals.

[Miska]

Also from the HFN review:

 

"OK, so there’s not quite the absolute bass extension and conviction of the Chord DAVE [HFN Apr ’16], nor quite the ability of that DAC to reveal the finest nuances of the recorded ambience, but then the T+A is a fraction of the price of its British rival (£7950), and very impressive for the money."

 

I'm pretty sure this refers to data being converted using it's PCM path (TI PCM1795 chips), not through the DSD path. Remember there are two DACs in one box...

[Miska]

If someone has $$$ to compare T+A DAC8 DSD head-to-head against PBD Merlot, I'd be happy to hear results. Merlot is also discrete DSD DAC. For comparing only the conversion stage, one could use for example HQP to upsample with same settings. Merlot is about 2x price from European point of view, but less difference in the US.

[Miska]

Rob on DSD:

 

He is representing DSD as if it would be somehow different from the delta-sigma modulation he is doing in Chord DACs. For some companies "DSD" seems to mean "DSD64", but it's like saying RedBook PCM is all PCM there is.

 

Chord is using 4-element conversion with 5th order modulator. While HQPlayer for example with DSC1 is using 32-element conversion with 7th order modulator. Seems like Chord applies proper filters only up to 705.6/768 rates.

 

And I can do much better than 120 dB for DSD64... If you input 24-bit PCM, then PCM is the limiting factor.

 

AFAIK, Chord converts DSD inputs to 352.8/384k PCM and then up-converts it again.

 

With DSD DAC capable of higher than DSD64, one can also upsample DSD64 to DSD512 as an example.

[Miska]

Chord calls their dac technology "pulse array". I'm not sure how that differs from standard delta-sigma.

 

Just yet another name for same old stuff... dCS calls theirs "Ring DAC". And ESS has something like "HyperStream modulator + Revolver DEM". Other manufacturers didn't bother to invent marketing names, or then I'm not aware of those.

 

But in engineering terms it is delta-sigma modulator combined with DEM conversion stage. Companies come up with different marketing names for their variant. The key however is in the details how all the DSP and conversion stuff is actually implemented.

[Miska]

The extreme resolution and detail via DAVE was as if the "pixels" of DSD 64 became too obvious in some cases.

And it was even more obvious than via Hugo that 24/192 was more realistic, than the less resolved DSD 64.

 

I would suggest to compare to DSD128 or DSD256 instead.

 

But I was less impressed with DSD via DAVE.

 

Maybe they have not spent much time and effort to make best out of DSD...

[Miska]

Chord DACs are also delta-sigma so he's not against it, on the contrary; Rob's issue is with DSD's 1 bit nature.

 

There's no issue with 1-bit, but I've covered the technical reasoning behind this in other threads on this forum.

 

Dave has a 20 elements DAC, 5 bits at 104 MHz, 17th order modulator. His noise shaper has 350dB THD and noise digital domain performance.

 

It is still quite low on number of elements, but if you are into numbers ESS Sabre can do 64 elements at 100 MHz and if you run the chip in mono-mode you get 512 elements at 100 MHz. For 104 MHz 17th order modulator I could do much much better than 350 dB THD. However, as you increase clock speed of oscillators, also jitter increases. This is also one of the issues with ESS. Doubling frequency from 24.576 MHz increases low frequency jitter by about 10 dB.

 

I can also produce any order of modulator, but as you increase the order it also increases demands for the analog section. So such DAC needs to have conversion/analog section that runs fine to GHz range.

 

It can be done, but the cost of doing such conversion and analog stages skyrockets, and I'm not really interested on things that only few people will ever use. I'm more interested on making 500€ DAC perform like 5000€ DAC by utilizing what is available in typical computer being used anyway, rather than even thinking about DACs in >10k€ price range. It is just too niche to be interesting.

 

Physics puts thermal limits somewhere between 120 - 140 dB range. HQPlayer already has >320 dB accuracy which has pretty decent safety margin. I could easily bump it to >386 dB if someone really thinks it matters.

[Miska]

I am repeating myself, but DAVE internal upsampling of 16/44 sounds better than HQ P upsampling to 24/352, so Kudos to Rob Watts

 

I have Mojo, and to me it definitely sounds better with HQP upsampling to 705.6/768 32-bit compared to it's internal upsampling. Using DSD256 instead doesn't make much difference, probably due to the way it is handled (converted down to 705.6 PCM).

 

It's biggest culprit is the built-in volume control, so it sounds best booted up in volume control disabled mode.

 

Since it is a mobile DAC, positive thing is that upsampling to 16x PCM consumes very little CPU time so it doesn't affect laptop's battery life.

[Miska]

The only true native 1 bit ADC used today that I am aware of is the Grimm Channel Classics uses. But it is limited to DSD 64.

 

The TI's PCM4202/PCM4204 chip I've been using and that is used in Korg and TASCAM DSD-recorders is true 1-bit ADC chip. Goes up to DSD128. With a small trick these can be made to record DSD256 too.

[Miska]

Do you know any ADC that actually records 1 bit DSD 256 natively?

 

You can find some DSD256 recordings from nativedsd.com

 

Judging by your other post you seem a lot more interested in compromising for monetary reasons than Rob Watts who seems to apply no limits in his approach for "digital veritas extremis"?

 

I don't make any compromises in HQPlayer. Objective of HQPlayer is to replace compromised DAC's DSP processing with better one.

 

But I would be interested to know if your "interpolation magic" if I may call it so, tries to solve the same problems in digital audio as Rob Watts tries to do with his approach to the problems or not?

 

Probably those, and bunch more. I have more DSP things in scope that for example DAVE doesn't do. Such as running digital room correction and X-over filters, also for DSD content and such.

 

Personally I am MUCH MORE interested in the highest possible transparency and accuracy possible than making a 500 dollar dac sound like a 5000 dollar one.

 

Since HQPlayer doesn't make any compromises (unlike for example Chord Hugo which seems to have compromises in the DSP), you can run no-compromises upsampling even with $500 DAC, or $5000 DAC. I think that is much more interesting. Of course the analog sections and such don't get replaced by doing this and thus you may still have some compromises with the $500 DAC.

 

Think it from this point of view; what if you would leave out all digital processing from DAVE and leave only the actual conversion and analog sections. All digital filters, modulators and processing could be left to player software running on a computer. Then it could leave out FPGA altogether and be implemented with just the Atmel chip that deals with USB interface. That way it could be much cheaper DAC with at least the same or possibly better technical performance!

 

I have very simple taste: only the very best is good enough.

 

Nice to know money is not an issue for you, but many other people have limited budgets.

 

Chord already have the Hugo and Mojo which are both very good indeed at showing a lot of the many times more expensive conventional "off the shelf chip" DAC competition the door provided one uses suitable source material.

 

I have Mojo and it can be improved with external upsampling. If it would have a direct mode bypassing all DSP processing it would be even better.

[Miska]

Mola-Mola does 1 bit 100MHz conversion on a SHARC DSP, so 1 bit 22.5792 MHz (DSD512) shouldn't be that difficult.

 

 

Yes... Since that one is still compromised in the DSP chain, it would be nice to get cheaper variant of the same DAC with the "PCB1" removed and replaced with simple DSD-only input capable of DSD512/DSD1024/DSD2048.

[Miska]

Rob on DSD:

 

The babble was too-long-didn't-read for first iteration for me. But it contains some annoying errors.

 

Well any signal that is below the noise floor of the noise shaper is completely lost - this is completely unlike PCM where an infinitely small signal is still encoded within the noise when using correct dithering. With DSD any signal below the noise shaper noise floor is lost for good.

 

This is complete bullshit.

 

The point I am making over this is that DSD noise shapers for DSD 64 is only capable of 120 dB performance

 

This is also bullshit. I can do better for DSD64, and as a result 24-bit PCM input is limited by the 24-bit PCM noise floor rather than the modulator.

 

For DSD256+ I can do better than 32-bit PCM.

 

 

Rest of the story is just funny, because DSD is just delta-sigma and Chord DACs are delta-sigma DACs.

 

DSD is only limited by the modulator implementation quality. Theoretical maximum dynamic range of DSD64 is 385 dB, for DSD128 it is 771 dB, for DSD256 it is 1541 dB and for DSD512 it is 3083 dB.

[Miska]

In any case don't assume that an FPGA can compete with a CPU/GPU approach in a cost effective fashion.

 

I'm actually quite happy about how latest Intel CPU combined with latest nVidia GPU performs. CPU is good at doing few very complex non-parallel things at high speed. And GPU is good at doing many simple things in parallel at high speed. This combined with several gigabytes of very high speed RAM provides tremendous possibilities for things like complex adaptive/look-ahead algorithms. Entire album can be pre-analyzed to know upfront what is coming up and when. Combining the two gives best of both worlds. And nVidia's new Pascal microarchitecture seems to give a huge boost for HQPlayer-style applications

 

Why not replace the Atmel with Zynq and get the best of both worlds. ? Much better handling of multiple clock domains.

 

Sure, I just said Atmel because Chord uses Atmel MCU for the USB interface implementation. It can deal with two clocks fine, but of course with more flexible choice there are more possibilities and flexibility (for things like DPLLs and such).

[Miska]

2:Your answer regarding "interpolation magic upsampling" seems more like you are guessing than really know what "your interpolation magic" actually does?

"Probably" doesn't sound very convincing to me.

And "a bunch more".What more?

 

I know what my stuff does, but I don't have very clear picture of what exactly Chord DAC does. So I cannot say if the goals match 100% or if there are differences. You are never going to get clear answer if you ask me to compare my stuff against some else's, because I cannot get inside someone else's head.

 

I said that the "bunch more" is things like:

- Digital room correction (convolution engine)

- Digital cross-over filters for speakers

- Digital re-routing and mixing

- Multichannel speaker distance and level adjustments

- Sampling rate conversions for DSD, both up and down

 

All this for DSD sources of any sampling rate too, to either DSD or PCM output.

 

How does your DAVE play 2L's 5.1 channel DXD content? You know, 2L is all about multichannel audio. I'm playing it all fine upsampled to DSD256 through my exaSound e28 DAC.

 

3: I have never said that money is not an issue for me, only that my taste is for the very best possible SQ without compromises.

If that could be achieved without rip off pricing I would be a very happy music lover indeed.

But as you say cheaper DACs will still be compromised by the cheaper less transparent analogue parts used.

 

So my objective is to make cheaper DACs less compromised by providing uncompromised upsampling and delta-sigma modulators. And in addition allowing cheaper no-compromise DACs because DAC doesn't need to perform any digital processing, just convert digital input to analog at highest possible quality - which is actually the connection to T+A DAC8 DSD in this thread!

 

4: As far as I know Rob Watts has categorically adviced against any external upsampling with his DACs.

He claims the opposite of what you are saying here. He claims it will deteriorate transparency and ultimate SQ with his products to do so.

 

Of course he does, but it is precisely type of blanket statement I just ignore because he possibly cannot know if any external upsampling is better or worse than what he's budget-limited DSP implementation does.

 

But with such statement, I would just advice people to primarily look at other DACs where manufacturer doesn't make such blanket statements.

 

It would be interesting to know what music and audio system you used to come to the conclusion that external upsampling of Mojo was an improvement?

 

Music is all kinds of stuff I have in my library. Headphones are Sennheiser HD800. Plus of course bunch of measurement gear.

 

 

P.S. I don't really understand WTF this Chord Dave discussion has to do with a T+A DAC8 DSD thread? I would suggest to start a new thread about Dave and suggest Chris to move related discussion there. Of course it is in a sense interesting that DAC8 DSD seems to be able compete straight against much more expensive DAC. Maybe because Chord is compromised in a way that it is not able to do native unprocessed DSD512...

[Miska]

I don't see a lot of advertising of PCs and Laptops with CUDA architecture. But it looks like a tempting combination. And perhaps an alternative to strictly focusing on higher powered Intel CPUs for upsampling to Multichannel DSD 256 on your exaSound e28 or the NADAC.

 

Are there examples on the market of CUDA-equipped machines you would recommend with HQ Player - and DSD 512 playback on the T+A?

 

CUDA is so much geared towards technical computing, that it is not advertised much to consumers. Practically most of the currently available nVidia GPUs are CUDA capable. At the moment nVidia's Maxwell-generation GPU architecture is best choice - usually meaning GeForce GTX 9xx series in consumer products. Also the professional Quadro Mxxxx series falls into the same category at higher price.

 

For a normal PC desktop there are many options and since the graphics cards are normal PCIe 16x cards, can be fitted on most modern desktop PCs. For laptops, there are number of gaming oriented laptops with GeForce 9xx GPUs and number of business oriented laptops with Quadro Mxxx GPUs.

 

nVidia just launched their newest Pascal architecture that promises huge performance improvement for HQPlayer-type computing tasks. First products rolling out during the summer and more towards Q1'17. The consumer product like is likely to be called GeForce 1xxx (1000-series).

 

Tesla-series is their HPC (HighPerformance Computing) product line, and the just announced

is pretty amazing.

[Miska]

Nope, they can also process in the analogue domain.

 

Seems to be trend these days that studios go back to analog desks with DAW automation. Like for example one of the well known mastering studios here:

Chartmakers Mastering Studio expands into mixing with an SSL AWS 948 | Solid State Logic

 

With desks like:

http://rupertneve.com/products/high-voltage-discrete-mixer/

AMS Neve | 88RS

AWS

[Miska]

It is sometimes possible, even very generous cooling, to cause overheating in a very small part of a CPU using a carefully crafted loop of instructions. It's highly unlikely, however, that anything actually useful should ever do this.

 

All CPUs have thermal management these days and OS is also partially responsible in participating it. There is also thermal protection hardware on the dies these days.

 

Most typical reason for overheating is dust collected in the internal heatsinks, blocking proper air flow. Result is involuntary thermal shutdowns/halts. Yearly internal cleanup for machines running 24/7 in non-clean air conditions is usually sufficient to keep the machines happy.

[Miska]

How can running closed form algorithms destroy a cpu? Do we even know for sure it's the cpu that's dead? Would love miska's input on this :-D

 

HQPlayer is pretty light on CPU and GPU compared to modern computer games at 2.7k resolution with full details on...

 

For example something like Far Cry 4 or others based on CryEngine or Unreal engine.

[Miska]

PCM at 88.2/24 is by any reasonable measure better than DSD64 while needing 25% less space.

 

Not in my opinion. DSD64 can reach significantly higher SNR in audio band, while having much better temporal resolution due to wider bandwidth (no inherent need for bandwidth limiting).

 

You can get nicely detectable 250 kHz sine out from a DSD64 DAC, although attenuated by the reconstruction filters.

 

[Miska]

Do they use the XMOS interface?

 

Someone said it is Amanero...

[Miska]

There is a reason why "one of the older stable firmware versions" is being used :-)

 

I think the spec Amanero released contains needed info also how to make the older firmware work on Linux. I was planning to try it out at some point when I have time. Doesn't take much time or effort...

 

For me personally, Linux support is very important because Linux is my primary development platform.

[Miska]

There is some activity this year on the Github pages to adapt things to the new firmware versions, so I would say there is hope:

 

I flashed one of my Amanero boards with the 1099rc2 firmware and swapped it on my DSC1. I will try it out hopefully tomorrow to see if/how it works.

[Miska]

AFAIK work on this issue is in progress - so there is hope to see DSD512 Linux support in the near future....

 

I tried the newest rc2 firmware with the Linux driver support, but the endpoint seems to just get stuck in DSD mode (PCM & DoP work fine). So no success yet. I haven't got time yet to debug further what is the problem exactly.

[Miska]

It's rather neutral/real sounding to me. The chord dave surely sounds "leaner/more clinical". The obvious other end of the voicing scale is a lampizator dac. Actually I'd say tonally it sounds pretty close to the PS Audio DS. All these dacs are in a different pricerange, im sure you can see the value proposition here. The T+A does sound clean at dsd512, cleaner then the Dave, I could see someone needing some acclimatization time to get used to that which for me adds to the realism of the performance. The voicing is definately warmer, more colourfull though. Its quite special in both sounding cleaner and having a more colourfull midrange and saturated low end at the same time. Where the Dave shines is high frequency detailing. Its unbelievable how much detail it's able to retrieve from a plain 16/44.1 source. Anyway Michael's personal tastes wouldve been better served with a lampizator. Hope this helps...

 

If you would like to move HQPlayer sound perhaps closer to Dave (I have not heard it so cannot say for sure), you could try poly-sinc-ext or closed-form filters. The actual D/A and analog sections of course still play a big role to the sound, something that is not easily switched.