[jeremya]

No worries - we are dealing with a complex situation....

Thank you so much for the detailed tutorial, Rob! Definitely learned a lot there, although I'm still trying to fit the mental model together.

Here's my layman's takeaways from what you wrote... feel free to correct them!

1. Despite what Andreas Koch implied, not all PCM (perhaps even not much PCM?) is converted from a 1-bit DSD stream, because n-bit delta-sigma ADCs are more commonly used when PCM is the target.
2. "Pure DSD" recordings (which are optimized for that 1-bit stream) still have timing errors due to how long it takes the modulator and the noise shaper to "notice" a rise or a fall in the input voltage (significant enough to warrant the output bit being flipped).
3. Errors in DSD are a function of sampling rate (and some other factors), so higher sampling rates (like quad-DSD) are less error-prone, but the fundamental issue with a 1-bit modulation remains.
4. multi-bit PCM sourced from an n-bit delta-sigma ADC process has both the headroom to deal with a wider range of input values and the ability to be dithered effectively, which a 1-bit DSD signal lacks. This affords your process the ability to reconstruct the original analog input with much greater fidelity from PCM than from DSD, as there are fewer timing errors 'baked in' to the bits.

Am i close?

 

[Rob Watts]

Yes agreed, but the vast majority of recordings are via N bit delta-sigma ADC modulators - even most DSD recordings are converted from PCM to do the editing, with a DXD file used as the master.

But I would add that a 5 bit 128FS ADC with say 5th order noise shaping would have significant amplitude related timing errors, with a different delay for small amplitude steps against large ones - but would of course be better than a DSD 1 bit modulator.

 

[jeremya]

Thanks again, Rob. I really appreciate you taking the time!

I am now left to wonder why a majority of my DSD files sound quite a bit better than their PCM counterparts when played back through more pedestrian DACs like, say, an ESS 9018K2M via DoP. I'm specifically thinking of various MoFi releases where you have the same master tape playback converted directly to PCM and DSD (so provenance isn't the problem).

I suppose that either the timing errors in the DSD are largely euphonic (that should be an oxymoron, right?), or perhaps the greater sample density (even with more errors in tow) produces a moderately better experience on less capable hardware?

Either way, you've convinced me that DSD is not all it was touted to be by Sony's marketing team in the early 2000s.

 

[514077]

Thanks again, Rob. I really appreciate you taking the time!

I am now left to wonder why a majority of my DSD files sound quite a bit better than their PCM counterparts when played back through more pedestrian DACs like, say, an ESS 9018K2M via DoP. I'm specifically thinking of various MoFi releases where you have the same master tape playback converted directly to PCM and DSD (so provenance isn't the problem).

I suppose that either the timing errors in the DSD are largely euphonic (that should be an oxymoron, right?), or perhaps the greater sample density (even with more errors in tow) produces a moderately better experience on less capable hardware?

Either way, you've convinced me that DSD is not all it was touted to be by Sony's marketing team in the early 2000s.

I noticed the same thing, in that I seem to like DSD over PCM releases of the same album, on my QP1R DAP. Whereas PCM sounds much better on my Hugo2, no doubt, due to Rob's magic.

 

[miketlse]

Thanks for the link, Ragnar! It's the proverbial exception to the rule.

Actually I'm pretty well served in terms of good recordings within my preferred genres. Especially with Contemporary Classical and Jazz I find a lot of audiophile-grade recordings, some of them in hi-res. And e.g. Radiohead generally offer high recording quality. The Beatles' Anniversary Editions (in 24/96) are worth mentioning as well.

Since a few weeks I'm subscribed at Bandcamp, and I'm fascinated by the creative diversity there. Recording quality isn't always great, but most of it can be fixed to a degree making for a satisfying listen. But I also found audiophile delight beyond typical audiophile euphony. My album of the month – at least. I hope you (don't) feel shocked.

My weekly divergence into alternative and electronic music, via internet radio, plays quite a few tracks discovered on Bandcamp - there is some interesting music there, mostly not available via the major labels.

 

[kennyb123]

I'm specifically thinking of various MoFi releases where you have the same master tape playback converted directly to PCM and DSD (so provenance isn't the problem).

Interesting. I’ve yet to actually compare PCM to DSD ripped off the MoFi discs. I rip only the DSD and then put the disc in storage. I will have to rip the CD layer of a few of these to see how they compare through HMS/TT2.

 

[Babysnake]

I have a Blu2/DAVE combo and I'd be interested what settings on the DAVE do people use?

 

[Jiffi32]

I have a Blu2/DAVE combo and I'd be interested what settings on the DAVE do people use?

Another Blu2/Dave owner in suffolk woohoo

 

[JamesB797]

Do Intel's new Stratix FPGAs open up any possibilities? Or is a larger leap in performance still necessary.

 

[Rob Watts]

Last time I checked the Stratix price/performance ratio had similar numbers too Xilinx; the trouble with FPGA companies are they are fighting for the data server/internet/5G backbone territories, where the issue is all about raw performance and power dissipation - not about silicon cost. The death of Moore's law means that high performance but low cost commercial sector is stagnating. Its 7 years now since Xilinx released vapor ware data on the 7 series (the Artix range which has the best price/performance ratio), with no hint of a replacement currently in sight.

 

[MWeston]

Hi Rob,

Long time listener, first time caller...

I've gotten into the higher end of audio in more recent times and have been paying more attention to Chord products as I search for a new DAC. I remember the Hugo and Mojo from years back, but never really investigated them much. Now that I am, I find the technology very interesting!

I've tried my best to read through quite a lot of posts across multiple threads, including this one, and it has been very enlightening! As a more technical person, there have been a few questions that go through my head when I read this stuff, so I though I might take a chance and write them down, in the hopes you might have more to say to those with very curious minds. I apologize if any of this has been answered before. This is a big forum!

First, I've been trying to get a better understanding of the pulse array and how all the numbers you throw around about the internal processes translate to the physical interface. For example, you talk about a 5-bit output which is something I see on the DCS design as well. Why not 6-bit or 8-bit? How did that become the decision on bits?

DCS has that ring DAC which also talks about 5-bit output translated onto 48 latches. The explanation I have read is that it's really 4.6 bits as 24 outputs. I guess the 48 latches is for positive and negative swings with 24 resistors on each but I don't really know. Maybe it's for true differential output all the way to the balanced output. I'm not that deep into it at this point.

So you have 4E, 10E and 20E pulse arrays and you talk about some of those outputs being overhead to avoid clipping during processing. Mojo probably doesn't have that so the data here might be 4 bits on 4E, 8 bits with 2 bits overhead on 10E and 16 bits with 4 bits overhead on 20E. None of those numbers make the 5-bit thing obvious. That ring DAC seems so linear in its understanding but how do you translate so few outputs (by comparison) to the 5-bit noise shaper output? Do you do everything on one power supply and no swinging of voltage below zero like everyone else seems to do?

Thanks so much and you are real legend to me, as silly as that may sound. Your dedication to produce quality products but also listen to them exhaustively during development seems a cut above. There's some stuff out there that you have to wonder if anyone did more than a power on test before approving it and mass producing it.

 

[Rob Watts]

So assume we have 16 elements - and they all can go on, hence we have a value of 16. All they all can go off, hence 0. Encoding 0 to 16 requires 5 bits, with the 16 an overload bit, as 17 to 31 isn't actually used as real values.

Now that's the simple case - and we can actually encode more info (more resolution) into pulse array by having the PWM switching frequency to be say 64 master clock cycles - then we are actually looking at a noise shaper truncation resolution of 7 bits (65 to 127 being overload) - and this is the situation with Mojo for example on the 4e. So in reality the actual truncator resolution will vary depending upon the specific implementation - and it's a trade off between PWM switching frequency (a low PWM frequency means higher truncator resolution but poorer noise shaper performance). So each specific pulse array design is actually optimized for the best overall SQ performance, and the raw truncator output resolution is no indicator of overall performance - it's a great deal more complex than simple noise shaper OP resolution!

 

[MWeston]

Hi Rob,

Thank you for replying. Phew, that just adds so many more questions to the mix if you care to answer. Would using 6-bits to encode 0 to 32 with 32 being an overload be better or just overkill in some way? Maybe it takes more FPGA logic that works its way down all the filtering processes and would never fit? Your statement about the high switching frequency might be why DCS needs 48 elements. They only switch at 3MHz. I know you don't like talking about other people's technology but it seems the closest comparison I've ever seen. I like Chord pricing a lot better!

I'm stuck on one other thing that this last reply really messed me up about. You have said in other posts that you keep out the noise (and noise floor modulation I think it was) by having one element switching up when another is switching down. That sounds very symmetrical but based on what you say here, there would be no way to guarantee it as musical data is basically 'random'. I had thought your pulse array was a symmetric thing that a 10e was actually 5 real elements so that you could make sure one was going up when another was going down, but I think I am way off!

Oh the other thing was whether you use dual supplies and have to deal with the crossover distortion or if you have a way to use a single supply and just avoid all of that. I know about class-A outputs but I have no idea how the pulse array handles these voltage changes, if it is even a concern at all.


Thanks again!

 

[Rob Watts]

I thought this might open up a can of worms!

The extra resolution encoding is for the smaller elements pulse array. Also, the pulse transition aspect, where a rising edge cancels a falling edge, is also very much more complex - as this occurs with all the elements when the noise shaper is outputting a constant noise shaper value - which is the majority of the time. So an important aspect is to make sure that the noise shaper changes in value is not signal correlated, then the noise due to clock jitter becomes fixed, and so does not contribute to noise floor modulation; and this is very much noise shaper design related - in particular making sure that the integrator coefficients produce a well behaved output under all circumstances - well behaved means that the noise shaper activity rate (this is how rapidly the noise shaper requests a different output value) is signal independent - and my design of the noise shaper ensures this.

The choice between extra encoding or not is also horribly complex too - it's too complicated to go through in detail. But I optimize these decisions based on measurements and listening tests.

Single supply means huge class A bias currents, capacitive coupling which degrades SQ and measurements. To eliminate crossover distortion I use my 2nd order analogue noise shaper amplifier, which completely eliminates HF and crossover distortion, even when driven hard in Class B.

 

[IgeNeLL]

Here is the cover art:











The first (44.1 24b) and last one (96k) I guess is the best of the bunch.

I also got this which has one Richter track too, and is 96k:



Also to strongly consider are these box sets:













As these box sets are typically 4GB of RBCD standard and only £9.75 for each box set. The SQ is excellent too, with (to my limited musical expertise) fine musicianship.

I have bought a lot of Brilliant Classics box sets, and they have never disappointed. I also got some CD box sets too, which are a bit more expensive and a pain to rip but worth the effort:







I also got 3 Naxos CD's of James Whitbourn, of which the best is:

Thanks I will get this album. Start to follow this thread I should have known this thread sooner