Chord Electronics Owners & Discussion Thread

[Bunno77]

 

 

[legend]

7 hours ago, Ittaku said:

Without hacking the software itself, you can also make the filter steeper and have more taps by increasing the rejection depth. It allows up to 200, but that tends to overflow calculations so usually only 198 works, so try adding "-R 198".

Thanks.  It increased the number of taps to about 11k - so better than many DACs but still well short of M-Scaler's 1M taps.  Looks as if I will have to go your full monty of 'extreme upsampling' as in the other forum!

[Anthony1]

The current M Scaler matches the Hugo series boxes,
so wondering if Chord will bring out a M Scaler that matches the Dave, or better a Dave with built in Scaler.

[Guest]

The M Scaler is built for all 4 current dacs equally. It’s not just built for the Hugo range, it’s built for the Dave also. 

[davewantsmoore]

On 23/02/2019 at 6:49 AM, legend said:

Project S2 DAC that I think does not itself upsample

Yes, it does.

 

First it resamples the audio to a rate of 1.5mhz ... and then modulates it at 40mhz (or more, would need to check for this model).

[davewantsmoore]

On 23/02/2019 at 7:14 PM, Ittaku said:

That's 629 million taps.

Not sure, how you are compiling SoX but noticed that they're pretty generic.    A quick google shows people getting from 2 to 10x increase optimising for their architecture.

 

 

.... but of course, you'll never be able to process 1 million taps on a computer.   /s

[legend]

11 hours ago, davewantsmoore said:

Yes, it does.

 

First it resamples the audio to a rate of 1.5mhz ... and then modulates it at 40mhz (or more, would need to check for this model).

Thanks for the correction - I said 'I think' deliberately because I was not sure what happened inside the Project S2 DAC.

 

More generally I wonder if what happens inside a DAC affects (even undoes partly) the improvements of the M-Scaler or other external extreme upscaling with large number of taps to impulse times etc?  Unless the pre-upscaling matches the DAC's maximum rate  as when the Qutest was connected to the M-Scaler through its 2 SPDIF inputs - and when I heard the major magical improvement when Sime brought the M-Scaler into my system.

Edited February 25, 2019 by legend

[Ittaku]

1 hour ago, legend said:

More generally I wonder if what happens inside a DAC affects (even undoes partly) the improvements of the M-Scaler or other external extreme upscaling with large number of taps to impulse times etc?  Unless the pre-upscaling matches the DAC's maximum rate  as when the Qutest was connected to the M-Scaler through its 2 SPDIF inputs - and when I heard the major magical improvement when Sime brought the M-Scaler into my system.

These are good questions because hearing improvements going 44->88->176 could be a combination of making the most of extra taps, and decreasing the amount of oversampling within the DAC, and it's impossible with our hybrid setups to really be able to tell the two apart. Yesterday I did some upsampling experiments on a different DAC (Gieseler) and going from 44->88 was good but 44->176 was substantially better, and these were with identically matched numbers of taps (30M), so it wasn't "extra taps" that improved the sound further per se. This is consistent with what you experienced with the samples I sent you, but on yet another DAC. What we need to do that extra comparison really is a NOS DAC, or a DAC that allows you to switch into NOS mode and can natively play 44/88/176 etc, and then mess with upsampling.

Edited February 25, 2019 by Ittaku

[Ittaku]

13 hours ago, davewantsmoore said:

Not sure, how you are compiling SoX but noticed that they're pretty generic.    A quick google shows people getting from 2 to 10x increase optimising for their architecture.

I always build with optimisations for my architecture except during debugging, however it's clear that sox is a generic tool and not optimised for this one function I'm using it for and a dedicated software application that is should be significantly faster and more memory efficient. My offline conversions are usually using 400-600 million taps and it takes 90 seconds to build the filter alone and half hour for a minimum sized file, though after that progression is quicker but obviously not realtime. A 30M taps filter which is "more than enough" from my experimentation happens at 5x real speed, so very much usable on the fly, though introduces something like 18s latency so can't really be used in an ordinary live chain.

[legend]

36 minutes ago, Ittaku said:

These are good questions because hearing improvements going 44->88->176 could be a combination of making the most of extra taps, and decreasing the amount of oversampling within the DAC, and it's impossible with our hybrid setups to really be able to tell the two apart. Yesterday I did some upsampling experiments on a different DAC (Gieseler) and going from 44->88 was good but 44->176 was substantially better, and these were with identically matched numbers of taps (30M), so it wasn't "extra taps" that improved the sound further per se. This is consistent with what you experienced with the samples I sent you, but on yet another DAC. What we need to do that extra comparison really is a NOS DAC, or a DAC that allows you to switch into NOS mode and can natively play 44/88/176 etc, and then mess with upsampling.

It may also be a problem with my off-line sinc upsampling attempts rather than buy an M-Scaler.  Sox will do upsampling to 8X 44.1 but does not seem to want to do upsampling to 16x 44.1 that I think is a native rate with the Qutest - but a rate which the Qutest will accept through its USB input (but not a single SPDIF input).  Not sure if it is a limitation of the Sox sinc algorithm or just a defined ,imitation?

[davewantsmoore]

2 hours ago, legend said:

More generally I wonder if what happens inside a DAC affects

Yes.

 

2 hours ago, legend said:

Unless the pre-upscaling matches the DAC's maximum rate  as when the Qutest was connected to the M-Scaler through its 2 SPDIF inputs

... but it's only matching the input rate.    "Unless" implies that if you match the input rate, then you are bypassing the issue... .and that is far from the truth for almost all DA converters.

 

 

Let's say you're using an M-scaler to resample a CD to 705.6khz  (16fs) .... You send that to the Qutest.    Which is it's maxium input rate ... but that doesn't mean it's not resampled further ..... The Qutest architecture relies on more resampling steps.

 

Inside the qutest ....  Yes, its skipping the first resampling  (resamples to 16fs, ie. 705.6khz for CD).

 

... The second resampling stage is another 16x to 256fs .... and the third is 8x to 2048fs.... and then the inerpolator, which resamples the data into its final 5bit precision points @ 104mhz, that are converted into the analogue signal.

[davewantsmoore]

20 minutes ago, legend said:

16x 44.1 that I think is a native rate with the Qutest

That's the highest rate it will accept as an input  (and it resamples all other input rates to this same 16fs rate).....which then connect to further stages.

 

Not that you can do anything about them ..... but if the strategy is to "alter the input data" (eg. using an Mscaler), then they can be a relevant (theoretical) consideration (eg. if they are not "perfect")

Edited February 25, 2019 by davewantsmoore

[Ittaku]

42 minutes ago, legend said:

It may also be a problem with my off-line sinc upsampling attempts rather than buy an M-Scaler.  Sox will do upsampling to 8X 44.1 but does not seem to want to do upsampling to 16x 44.1 that I think is a native rate with the Qutest - but a rate which the Qutest will accept through its USB input (but not a single SPDIF input).  Not sure if it is a limitation of the Sox sinc algorithm or just a defined ,imitation?

Sox limitation. If you watch carefully, once you get above 5x upsampling, it will start chaining sequential filters so you're no longer actually upsampling with the maximum number of taps. You can upsample to 16x but only with wav files I believe, as sox's flac container management won't allow you to do it. Either way though, you're getting chained filters so I wouldn't recommend doing that much upsampling with sox; it seems optimal to only go to 176/192 with the existing code. Maybe in time I'll try to see if I can hack that in the code as well, but I don't have a way to use anything higher sample rate than that so it's not high on my priority list.

[legend]

43 minutes ago, davewantsmoore said:

Yes.

 

... but it's only matching the input rate.    "Unless" implies that if you match the input rate, then you are bypassing the issue... .and that is far from the truth for almost all DA converters.

 

 

Let's say you're using an M-scaler to resample a CD to 705.6khz  (16fs) .... You send that to the Qutest.    Which is it's maxium input rate ... but that doesn't mean it's not resampled further ..... The Qutest architecture relies on more resampling steps.

 

Inside the qutest ....  Yes, its skipping the first resampling  (resamples to 16fs, ie. 705.6khz for CD).

 

... The second resampling stage is another 16x to 256fs .... and the third is 8x to 2048fs.... and then the inerpolator, which resamples the data into its final 5bit precision points @ 104mhz, that are converted into the analogue signal.

I did wonder if the Qutest 'senses' the M-Scaler when it is sending the 16x through the dual cables and then switches off other upsampling - because there seemed a dramatic improvement in SQ when the 16x was invoked from the M-Scaler - much greater than the 8x or 4x improvement.  But from what you are saying this is unlikely and so there is some other cause of the sudden 'magic' increase in SQ at 16x?

[Ittaku]

26 minutes ago, legend said:

I did wonder if the Qutest 'senses' the M-Scaler when it is sending the 16x through the dual cables and then switches off other upsampling - because there seemed a dramatic improvement in SQ when the 16x was invoked from the M-Scaler - much greater than the 8x or 4x improvement.  But from what you are saying this is unlikely and so there is some other cause of the sudden 'magic' increase in SQ at 16x?

Unlike my software approach, the mscaler doubles the number of taps going from 8x upsampling to 16x.

[legend]

I have just tried upsampling in my JRiver MC 23.  It allows 16x upsampling and also uses sox if you tick a box in their DSP options.  But I could hear little improvement if any over my sox previous attempts (11 k taps with 8X). Maybe to keep the process inline and with relatively little latency JRiver also uses a relatively small number of taps?  In an exchange on their web site soon after they added sox in 2016 Jriver would not discuss its implementation saying that they were not an 'audiophile' site.  Looks as if I will have to explore HPplayer while it gives a trial license,

[legend]

3 hours ago, Ittaku said:

Unlike my software approach, the mscaler doubles the number of taps going from 8x upsampling to 16x.

On reflection, although my hypothesis that if the M-Scaler is connected to a Chord DAC through the dual SPDIF leads at 16x upsampling with 1M+ taps then it disables any further upscaling in the DAC to prevent messing the M-Scaler's impulse response, makes some theoretical sense it would not make much commercial sense if all the Chord DACs then had the same SQ despite differences in prices - unless there were other differences such as the analog stages etc.

[Ittaku]

12 minutes ago, legend said:

On reflection, although my hypothesis that if the M-Scaler is connected to a Chord DAC through the dual SPDIF leads at 16x upsampling with 1M+ taps then it disables any further upscaling in the DAC to prevent messing the M-Scaler's impulse response, makes some theoretical sense it would not make much commercial sense if all the Chord DACs then had the same SQ despite differences in prices - unless there were other differences such as the analog stages etc.

The oversampling component is but one small part of a DAC's performance. There is no way to make a qutest sound like a DAVE

[legend]

4 minutes ago, Ittaku said:

The oversampling component is but one small part of a DAC's performance. There is no way to make a qutest sound like a DAVE

Even if both DACs are based on FPGA technology and not off-the-shelf chips?  As I understand it which probably very little but always hope to be enlightened!

[davewantsmoore]

5 hours ago, legend said:

I did wonder if the Qutest 'senses' the M-Scaler when it is sending the 16x

The first re-sampling stage converts everything to 16fs ..... so if it arrives at 16fs, then nothing happens.

5 hours ago, legend said:

through the dual cables and then switches off other upsampling

No, there's still many other stages in this type of DAC.

5 hours ago, legend said:

But from what you are saying this is unlikely and so there is some other cause of the sudden 'magic' increase in SQ at 16x?

Just because there is "more resampling" after what the M-Scaler has done, doesn't mean you won't hear the benefit of the M-Scaler.

 

It is just depending on what that resampling is, you might want to do something slightly different from what the M-Scaler does.    This is what people are doing with HQPlayer (for example).   It gives you the choice of lots of filters to use when producing the higher rate (eg. DSD512, DSD1024) data.   Which is the best filter?  Depends on which ones works best with whatever your DA converter does - for example your DA converter might do nothing by low pass filter the digital signal .... or your converter might internally remodule the data to a much higher rate again.

[rand129678]

2 hours ago, legend said:

Even if both DACs are based on FPGA technology and not off-the-shelf chips?  As I understand it which probably very little but always hope to be enlightened!

I've shared this below link/post a few times before in this thread. Well worth a read...

 

From the late and very well respected Charles Hansen of Ayre Acoustics:

 

https://audiophilestyle.com/forums/topic/35106-how-does-a-perfect-dac-analog-signal-look-different-than-a-cheap-dac/?page=7&tab=comments#comment-713189

 

Qutest and Dave have different power supply and analogue section designs...

Edited February 26, 2019 by Music2496

[legend]

23 minutes ago, Music2496 said:

I've shared this below link/post a few times before in this thread. Well worth a read...

 

From the late and very well respected Charles Hansen of Ayre Acoustics:

 

https://audiophilestyle.com/forums/topic/35106-how-does-a-perfect-dac-analog-signal-look-different-than-a-cheap-dac/?page=7&tab=comments#comment-713189

 

Qutest and Dave have different power supply and analogue section designs...

Thanks.  Yes I thought Dave would very probably have better (and more expensive) power supplies and analog sections than my Qutest.  However I have ameliorated this to some extent by connecting my Qutest to a DIY external linear power supply (that I grossly over-engineered) and its fixed output  to Sony P9000ES pre-amp that has discreet components (dual input fets etc) as well as over-engineered power-supply, more copper cladding that you can shake a stick at etc. 

 

But as @davewantsmoore says there may still differences between the Dave and Qtest if there is subsequent upsampling even after connection to the M-Scaler at its best output (x16 with 1M taps)??

Edited February 26, 2019 by legend

[rand129678]

14 minutes ago, legend said:

But as @davewantsmoore says there may still differences between the Dave and Qtest if there is subsequent upsampling even after connection to the M-Scaler at its best output (x16 with 1M taps)??

Yes, again... the analogue section designs are different, right? Have a read of the Charles Hansen post to see the importance of this, from a real expert designer...

 

Rob has shared plenty of notes about differences between his DAC designs over on Head-Fi Forum... best to get the correct info straight from the horse's mouth....

 

And as you know, power supply section design refers to the power supply designs for various sections of the DAC board/s (digital section, analogue sections)... Power supply section design doesn't just refer to the quality of the external PSU, in the case of Qutest... Qutest and Dave vary here too...

 

Add the two togther (power supply section design and analogue section designs) and you may get a big clue why Qutest may sound  different to Dave... even if the digital filtering is very similar with M-Scaler used with each.

 

The Charles Hansen post is well worth a read and re-read every now and then....

 

Charles Hansen designed consumer DACs and amps and pro audio DACs and A-to-D converters and as he says in the post, designed his own digital filters.

 

Edited February 26, 2019 by Music2496

[legend]

Yes you are correct - the analog section and PS of my Qutest system are different to Dave but am not sure how much difference it will make to SQ.  Must read Bob's notes on HFF.

 

I guess what I am more interested is their difference if at all in the digital sections - and whether with their FGPA technology they really do upsample again after the M-Scaler's best output and for what benefit?

[rand129678]

13 minutes ago, legend said:

I guess what I am more interested is their difference if at all in the digital sections - and whether with their FGPA technology they really do upsample again after the M-Scaler's best output and for what benefit?

I've already shared Rob's posts and links to posts earlier in the thread, that M-Scaler replaces WTA 1 of his DACs... The 1 million tap action all happens in WTA 1... 

 

There is further up-sampling after WTA 1 for noise shaping. I've shared those links earlier too.

 

You can ask Bob these questions directly too over on Head-Fi... you get the most accurate info from the designer.

 

 

 

Edited February 26, 2019 by Music2496