So, is it bye-bye NDX?

Conrad, the FPGA is not the actual DAC in the Hugo.. The two DACs in the Hugo are custom Chord Electronics / Rob Watts devices... Each with their own local regulated PSU. The FPGA does all the DSP work, and creates the bit/word stream to feed the two DACs.

The DSP / reconstruction filter programming takes advantage of the relatively immense power of the Xilinx FPGA chip chosen for the Hugo... 

Simon

On the contrary, Simon, as the FPGA is in fact an integral part of the Pulse Array "DAC" itself - the confusion may arise from the fact that, in older Rob Watts / Deltec designs, there are multiple FPGAs (4 PAMs, 2 noise shapers, 2 digital filters and the controller), with some optoisolation (between the 236-element pulse-array modulators [PAMs] and everything else), but these are further combined into a single device presently, along with high-precision resistors and capacitors. I believe that the lack of need for optoisolation (combining everything onto one massive* FPGA) is the real coup de grace in the Hugo, relative to the other Chord devices (but I could be wrong, there could in fact be smaller FPGAs still serving as the PAMs, with optisolation).

The pioneering work that allowed this to mature was performed by the University of Glamorgan, in conjunction with DPA Digital at the time - there is a white paper out on the interwebs, if you look carefully enough.

It is an academic exercise in most MSEE-level computer engineering coursework to create a DAC of arbitrary design using an FPGA and high-precision support parts.

* - massive in gate count, not in physical footprint

David, I beg to differ, as the FPGA performs the DSP and the reconstruction filtering according to Mr Watts. As we know the reconstruction filtering is done typically on board with many mass  produced ASIC DAC chips or seperately as with Naim and Chord Electronics and many other high end manufacturers. The reconstruction filtering is an integral part of the DAC process, and according to Naim and Watts is a key driver to the DAC performance.. So yes  the overall digital to analogue conversion requires a converter or raw DAC chip and a filter... This filtering is often done in the digital as well as analogue domains. It is this filtering and over sampling that is done in the FPGA according to Chord Electtonics block diagram... Just as it is done with Naim using the off board DSP microcontroller..

The other matter with Chord Electrnoics where I think it differs from Naim (currently) , is that in the Hugo they use a bespoke discrete DAC converter (Pulse Array), where as Naim use an ASIC DAC with the DSP disabled. Watts argues and demonstrates graphically he can achieve better linearity and noise control using his bespoke Pulse Array discrete DAC coupled to the FPGA rather than using an ASIC DAC chip with FPGA. ( but then he would wouldn't he)

PS my DACs when I was a Computer Engineering undergraduate were built with physical converters with programmable logic arrays (PLAs) which preceded FPGAs and were hugely less capable.. but I am going back 20 years or so..but I would argue the principle architecture is the same.

Simon

PS IOS 8 makes some editing difficult. So easier to delete and re post

You mean my view that the Hugo *does not* use conventional ASIC DAC chips... Yes that is correct.

Tony which Amarra product are you using?

Thx

S

It's someone you know well. He emailed me after seeing the Oppo post. At least he's heard the Oppo. Obviously it's subjective and as always one should listen to a piece of kit and decide for oneself, as I suggested Harry should. 

Steve, got you... I am now guessing to whom you may be referring..   

On that we can agree - my point was that the PAM design is in fact (on public record, accounting for the research funding) an FPGA itself (in the older Deltec / DPA designs, four FPGAs, due to limitations in gate and pin count), and, may in fact be in the current Chord devices a pair of offboard FPGAs (if not fully integrated into the main FPGA itself).

The Oppo, I believe, uses the ESS Sabre32 DACs, though these can be incredibly customized and actually re-programmed at near-gate level to support custom FIR filter profiles inside the DAC - it's not a fit-and-forget, one-size-fits-all design as it appears from the outside. And one has no idea what processing actually precedes the DAC ICs themselves....

David, over at Head Fi look at post 5927, there Watts explains why he uses seperate discrete PAM (Pulse Array) DAC with his Xilinx FPGA.

Simon

A statement of the obvious which nevertheless can't be repeated too frequently.

Seems that we are in interesting times. It can only be good.

I presume you mean the remark: "the FPGA is only the digital part, the DAC element is done with discrete components"–I read that to mean a simple 256-element RC summing array*, not, therefore, any FPGAs or digital ICs at all other than the (single) FPGA that does everything, according to the designer.

My post above reads:

I believe that the lack of need for optoisolation (combining everything onto one massive* FPGA) is the real coup de grace in the Hugo, relative to the other Chord devices (but I could be wrong, there could in fact be smaller FPGAs still serving as the PAMs, with optisolation).

It could also be the case that existing Chord devices also use exactly one FPGA and no other digital ICs in their DAC arrangement, and that the Hugo simply represents an implementation using a larger number of filter taps.

The point I was making was that the Deltec / DPA digital devices used PAMs in the form of multiple FPGAs only because the FPGA technology at the time was insufficient to combine the functions onto a single FPGA device.

Does that make sense?

* - not unlike an analogue summing bus, as might be found in a mixing console

Lastly - if this remark from Rob Watts in the same context doesn't put the state of the specialist audio industry, R&D, and its future as a self-sufficient and self-sustaining entity in glib perspective:

it is only about performance (measured and subjective) not cost, and finally there is no way Chord (or any other high end audio company) could afford my design fees. My "day" job is to create IP, and supply design consultancy to large silicon companies for audio.

David, if I am honest I have kind of lost the thread in your post... I was simply saying that the Hugo digital to analogue converter is a bespoke discrete device and is seperate from the DSP and data / noise shaping FPGA DSP processor. I subsequently pointed to the designer's public postings with graphs of why he he used a seperate discrete DAC as opposed to ASIC based DAC coupled with the FPGA proceasor.

As you point out there are FPGA ASIC DAC chips out there, that is a DAC chip with programmable DSP, but the Hugo and as far as I am aware Naim don't use them.. But perhaps the Oppo does.

Simon

There is/are no separate discrete DAC IC(s) in the Hugo - if the PAMs were implemented in FPGAs in older designs, and there is only one FPGA in the Hugo with no other digital components, it stands to reason that the PAMs are implemented on the same FPGA. The digital filter and DAC are one IC, in the same manner that other off-the-shelf multifunction DAC ICs with built-in digital filters are implemented in silicon.

The original DPA DACs actually used (at one point) a Philips digital filter IC and a discrete DAC implemented as PLDs - so it seems plenty plain to me that the discrete DAC (came first!) and digital filtering (came second!) exists as exactly one FPGA IC in the Hugo.

Wolfson, TI, and ESS (among others) all offer integrated digital filter / DAC ICs; perhaps the Chord exercise is a discrete (no on-board RC components) approach to implementing a complete, no-holds-barred solution without the commercial constraints...

Dave, ok I can't post the link as it is against Naim policy, but there is a document that Chord have on their site that shows a marked up photograph of the PCB and components of the Hugo. It identifies two 4e Pulse Array DACs (left and right and appears to show 4 chip/devices per channel that I can see within the markup) and also identifies elsewhere the single Xilinx FPGA chip. Additionally the block diagram makes this clear as well, and explicitly shows the DAC pulse array converters occurring outside the dotted area of the FPGA Xilinx data processing functions.

So in the absence of being able to show the seperate FPGA and DAC components here of the Hugo we will have to agree to disagree.

Simon

Ahh found a public domain photo rather than from Chord Electronics web site.. It just has no markup

The Xilinx Spartan FPGA is the large chip in the middle left. The discrete 4e DACs are to the top right of the FPGA chip.. I assume left and right are side by side.

Simon

Those are RC networks and dual-transistor packages, Simon (six pins are not enough for a DAC) - if you open the PowerPoint presentation, you can scale the picture to its original size (and almost read the values off the SMD components).

On the same diagram it reads:

FPGA –handles SPDIF decoding, USB timing, DPLL,WTA filtering, DSD decoding and filtering, volume control, cross-feed, control, noise shaping and DAC

Dave, ok as I say lets agree to disagree. Having said that I would  interested in your view of how to interprete the block diagram and its labelling that shows the Pulse Array DACs seperate to the Spartan FPGA with their own decoupled regulation and I am scratching my head about your 4e Pulse Array DAC 'RC networks' I guess they may have RC filters in there...not quite sure the relevance other than some sort of RF low pass filtering and part of the discrete DAC design.

Simon

Dave and the block diagram that clearly shows the Pulse Array DACs seperate from the Spartan FPGA? How should I interprete that please?

Simon

Strike that, by my guess it's a Sykes transistor integrator, two transistors per six-pin package, four pulse integrators per channel, eight ICs total, and fully discrete. Nice and simple!

Right, similar to how a discrete delta sigma DAC would work? Well I it does look like the Pulse Array is some sort of Delta Sigma variant, but I don't know for sure... But the noise shaping that takes place in the FPGA looks similar to what I would expect to see for a Delta Sigma device.

So we agree the discrete D/A conversion function and the FPGA are indeed seperate as indeed Chord Electrnics state In their master class presentation.

Phew I thought I must have been going mad...  

Simon

I appreciate your rigor - though we should probably have this moved to one of the Hugo threads...the key difference from the old Deltec / DPA designs is the requirement for DSD support.

Sorry Shaun, missed your question. It's Amarra Symphony with iRC Room Correction. I wouldn't be without it now.

Conrad

Can you explain your (pre-tinkering) Pi set up in simple terms?  I'm thinking of getting my son a Pi to play with, and wondered what else he'd need to link it to his Denon mini-system.  What would he need to store his music files on?