Watts Up...?
Aug 29, 2021 at 5:16 PM Post #2,656 of 4,530
Just in case anyone is interested. Explanation of MQA:



btw, I’m not a fan of MQA.
 
Aug 29, 2021 at 6:40 PM Post #2,657 of 4,530
Yes when I saw that announcement last month (I was just starting my 10km morning hike) I got very excited with the possibilities of 1200 dsp cores and much lower power. But when I got back and downloaded the data sheets, I was in for a disappointment; it may have 1200 cores, but has very slightly less memory than the 200T (the M scaler FPGA). So it will be memory limited. Also, given Xilinx track record, and the current severe problems with supply from silicon, it will be unobtainable for several (perhaps many) years. And probably expensive...

So my needs are simple and probably similar to others - more on chip memory, more dsp cores, more fabric, and much lower power - and lower cost! All of which together is not looking possible or probable.
Coming back to this post, Rob, I was wondering if Xilinx has miraculously delivered an FPGA that you can work with for xMS?

Am I right in saying that the original 16FS WTA1 FPGA algorithm e.g. in DAVE, depends mostly on DSP cores and doesn't have a requirement for "lots of memory"? My understanding is that the WTA1 FPGA algorithm in HMS consumes lots of memory in order to fit into the number of available DSP cores. I seem to remember that you had to completely re-do the form of the FPGA algorithm to achieve 1 million taps and in doing so the memory consumption went up substantially.

I'm wondering if it's possible to return to the older algorithm to solve the problem with having too little memory, even if there's enough DSP cores?

If that were true, then maybe that's how you're able to contemplate building xMS. Otherwise, it seems as if you would be stuck using multiple FPGA devices and networking them together.

Another thing I've been wondering is relating to the final noise shaper in DAVE, just before the pulse array modulator. In theory with the DX power amp you can build a noise shaper that's higher than the 17th order seen in DAVE, presumably because you will have loads of FPGA capacity since the DX amps will not have a WTA1.

I'm also wondering whether future DACs could have a switchable noise shaper just before the final stage. If the DAC is being fed by a 16FS signal, then the DAC's WTA1 algorithm can be removed and the noise shaper could be "upgraded", using more of the FPGA.

1637943
 
Aug 29, 2021 at 7:40 PM Post #2,658 of 4,530
If you've never heard state of the art digital room correction, you wouldn't know what I'm talking about. Needs to be experienced.

Of course if you have the perfect room acoustics already, then possibly not required.

I apologize that I missed it, but what is this solution you are speaking of?

There is no such thing as perfect solution. With everything we have to accept tradeoffs. Given the current state of ADCs, I don’t think one has to hear analog run through a state of the art digital room correction to know that there will be tradeoffs. Whether those will be acceptable to the listener will require it though.
 
Aug 29, 2021 at 8:54 PM Post #2,659 of 4,530
I apologize that I missed it, but what is this solution you are speaking of?

There's a few out there, considered 'state of the art' by the qualified experts in the field. I'm just a dumb end user though, not an expert.

There is no such thing as perfect solution.

Ya I don't believe I typed 'perfect solution' anywhere so I think there may be a misunderstanding here.

Anyway all the best solutions (currently!! doesn't mean perfect !) are subject to user error which can really lead to bad SQ results.
 
Last edited:
Aug 29, 2021 at 9:57 PM Post #2,660 of 4,530
There's a few out there, considered 'state of the art' by the qualified experts in the field. I'm just a dumb end user though, not an expert.
There was a debate on Audiophile Style a month or so ago. “Qualified experts“ needs to be considered within their field of expertise. Their experiences may not account for many of the qualities Rob thinks are important for music to come to life.
 
Aug 29, 2021 at 11:30 PM Post #2,661 of 4,530
There was a debate on Audiophile Style a month or so ago. “Qualified experts“ needs to be considered within their field of expertise. Their experiences may not account for many of the qualities Rob thinks are important for music to come to life.
The same can be said of Rob's opinions.

His gear measures superbly but there is other stuff out there too... so Rob is hardly the only D to A expert on planet earth...

For the APx555 measurements Rob shows for various products, there are other products with as good or better APx555 plots...

Talking about analogue outputs not the digital domain simulations...

People need to be open to looking at other products though. Some are iron fisted and loyal to Rob and that's fine. It's made him very rich so kudos to him :)

I love my Hugo2 btw but I enjoy and appreciate other gear out there too.
 
Last edited:
Aug 30, 2021 at 1:21 AM Post #2,662 of 4,530
@Rob Watts thanks a lot for the suggestions. Will pass that one to the electrician.

Using a battery could be the answer as well, I just don’t fancy invalidating the warranty as the M scaler is still covered.

and duly noted regarding the Dave BNC, will defer from that route!

Does the leakage current in the M Scaler affect SQ? And leakage current in general?

I spent a lot of time analysing the leakage current and building up a SPICE model to characterise it. It's not audibly important, as the bulk of it is 100Hz, and easily extinguished within the DAC by a low impedance ground plane. The HF components are negligible.

Coming back to this post, Rob, I was wondering if Xilinx has miraculously delivered an FPGA that you can work with for xMS?

Am I right in saying that the original 16FS WTA1 FPGA algorithm e.g. in DAVE, depends mostly on DSP cores and doesn't have a requirement for "lots of memory"? My understanding is that the WTA1 FPGA algorithm in HMS consumes lots of memory in order to fit into the number of available DSP cores. I seem to remember that you had to completely re-do the form of the FPGA algorithm to achieve 1 million taps and in doing so the memory consumption went up substantially.

I'm wondering if it's possible to return to the older algorithm to solve the problem with having too little memory, even if there's enough DSP cores?

If that were true, then maybe that's how you're able to contemplate building xMS. Otherwise, it seems as if you would be stuck using multiple FPGA devices and networking them together.

Another thing I've been wondering is relating to the final noise shaper in DAVE, just before the pulse array modulator. In theory with the DX power amp you can build a noise shaper that's higher than the 17th order seen in DAVE, presumably because you will have loads of FPGA capacity since the DX amps will not have a WTA1.

I'm also wondering whether future DACs could have a switchable noise shaper just before the final stage. If the DAC is being fed by a 16FS signal, then the DAC's WTA1 algorithm can be removed and the noise shaper could be "upgraded", using more of the FPGA.

1637943

Please don't talk about new Xilinx FPGAs as the old devices are on 55 week lead-times. Forget about new devices. Fortunately, Chord keep > 1 year of stock of semiconductors, as you always get burnt with semiconductors - its an extreme cyclic business.

For a given WTA filter, memory and DSP core requirements are fixed - there is no old algorithm to fall back on; but you can play around with different types of memory, and design your own DSP cores from FPGA fabric. Then the design has to place and route and meet timing, so it's a bit of an art to get optimum performance from a given device, as you may be constrained by routing, DSP or memory, or all three!

I don't think there is much mileage in improving Dave's pulse array noise shaper - but I do have something to try that would need vast amounts of gates. The real benefit in having lots more FPGA space is with the WTA2 and the other filters going up to 104MHz.

There was a debate on Audiophile Style a month or so ago. “Qualified experts“ needs to be considered within their field of expertise. Their experiences may not account for many of the qualities Rob thinks are important for music to come to life.

Particularly when these "experts" never conduct carefully controlled objective listening tests. They would have very different opinions if they did...

The same can be said of Rob's opinions.

His gear measures superbly but there is other stuff out there too... so Rob is hardly the only D to A expert on planet earth...

For the APx555 measurements Rob shows for various products, there are other products with as good or better APx555 plots...

Talking about analogue outputs not the digital domain simulations...

People need to be open to looking at other products though. Some are iron fisted and loyal to Rob and that's fine. It's made him very rich so kudos to him :)

I love my Hugo2 btw but I enjoy and appreciate other gear out there too.

I would like to see these so called plots that have better APx 555 measurements than Dave or TT2.
 
Aug 30, 2021 at 1:53 AM Post #2,663 of 4,530
its obvious from looking around that’s it’s not that difficult to produce a good sounding DAC, or one with a different “flavour” that will flatter various music genres but going for the best transparency is always going to lead to the better sound, if the listener doesn’t like what’s on the source reproduced as faithfully as possible it can be suitably tailored by the choice of amps and headphones, and measurements are always needed as a guide but would you buy a car solely on its measured top speed or 0-60 time ? … valid Measurements but they don’t tell you how it rides, handles or makes you feel …
 
Aug 30, 2021 at 2:40 AM Post #2,664 of 4,530
People need to be open to looking at other products though. Some are iron fisted and loyal to Rob and that's fine. It's made him very rich so kudos to him
You responded to arguments I didn't make and ignored the ones that I did make.

No we're not talking about analog - we've been talking about converting analog to digital and about room correction. I'll drop now as for some reason you've elected to argue in bad faith. No worries - it's not a big deal. Let's get back to the regularly scheduled programming.
 
Aug 30, 2021 at 2:41 AM Post #2,665 of 4,530
Particularly when these "experts" never conduct carefully controlled objective listening tests. They would have very different opinions if they did...
Exactly, Rob. I also don't think they use gear that's up to the level you choose to use.
 
Aug 30, 2021 at 3:41 AM Post #2,666 of 4,530
I would like to see these so called plots that have better APx 555 measurements than Dave or TT2.

Hi @Rob Watts , I'm a fan and really enjoy my Hugo2 as I mentioned. I love it and recommend it but I don't treat it like objectively the greatest D to A ever made...

And I'm not sure why you left off Mojo and Hugo2 from your comment here?

If we look at John Atkinson's measurements, there are some measurements where Dave is bettered by a Weiss DAC502 and Holo May... And of course some where Dave performs better. And that's not even using an APx555. Maybe better gear will show even greater differences than JA shows?

Dave's noise floor is one, no?

https://www.stereophile.com/content/chord-electronics-dave-da-processor-measurements

https://www.stereophile.com/content/holoaudio-may-level-3-da-processor-measurements

https://www.stereophile.com/content/weiss-engineering-dac502-da-processor-measurements

Mojo is claimed to not have measurable IM distortion but JA claimed it does in his measurements?

Figure 13: https://www.stereophile.com/content/chord-electronics-mojo-da-headphone-amplifier-measurements

I'm not an expert. Just reporting what I'm reading there by JA.

I would love to discuss this with you but some over zealous fans of yours may jump in thinking this is a personal attack on you (when it isn't) and we won't really be able to have a good discussion.

To re-iterate, there is no doubt Rob's gear is up there with the best objectively. No debate.

But I'm really not sure it is clearly better than everything else out there? At least John Atkinson has measured some better stuff.
 
Aug 30, 2021 at 5:36 AM Post #2,668 of 4,530
So where are these APx 555 plots that show better performance?
If JA's Audio Precision SYS2722 didn't show differences, then one could fairly say his AP is not sensitive enough / lacks resolution, and then we need to go to the APx555 right?

But isn't the point his AP has enough resolution to show the differences he has shown?

Apologies if I'm missing something obvious here.
 
Aug 30, 2021 at 5:42 AM Post #2,669 of 4,530
And I guess to say it another way, there are some measurements where JA's AP may have been a limitation with his Dave measurements and very clearly some measurements where his AP wasn't a limitation for measuring Dave and Mojo (and bettered by Holo May and Weiss DAC).
 
Aug 30, 2021 at 6:34 AM Post #2,670 of 4,530
so it's a bit of an art to get optimum performance from a given device, as you may be constrained by routing, DSP or memory, or all three!
Reminds me of the year or so I spent building state of the art matrix-matrix and matrix-vector multiplication software for a graphics processor:

HD5870 Matrix Factory

wrangling assembly code in the hope that the machine code produced by the compiler actually matched my intentions. Sigh. Looks so simple...

A few years later I discovered that people had worked out how to load machine code directly on to the GPU, entirely bypassing the compiler I battled with. I wish I'd had that capability, but it came too late and I'd moved on.
 

Users who are viewing this thread

Back
Top