Would this raw DSD version play PCM by converting through HQPlayer? I was interested in the original PCM one but as many said, the cost kept me away from it.
Also, since I’m right next to China, I could have the boards made a lot more affordable. I don’t mind the distribution for participants.
Also, since I’m right next to China, I could have the boards made a lot more affordable. I don’t mind the distribution for participants.
To also answer the implicit question I'll quote section 13 of my own article, which obviously relates to the original version with PCM input and FPGA board. It doesn't tell the whole story, though, because features such as headroom for intersample overshoots have been designed in but not measured. The -3 dB-bandwidth should be about 82 kHz, but I also never measured that. The measured results are with a balanced output without the optional output transformer.
In-band noise
The final noise floor in PWM8 mode and with the loud setting was -85.76 dB(A) on the left and -91.29 dB(A) on the right channel with respect to a full-scale sine wave. These are very decent values for such a simple circuit. The noise with a full-scale 1 kHz signal present was -75.37 dB(A) left and -76.74 dB(A) right, with 20 Hz full-scale it becomes -69.35 dB(A) and -69.39 dB(A) due to the reactances of the DC blocking capacitors. The noise gradually increases with the signal amplitude, which is more similar to the type of noise modulation found in an instantaneous companding system than to the periodic noise modulation found in an undithered or inappropriately dithered digital system.
Out-of-band noise
The out-of-band noise is about 10 mV RMS in PWM8 mode, 4 mV RMS in PWM4 mode and 400 µV RMS in chaotic mode. Although the 10 mV RMS of the PWM8 mode might cause subslewing TIM in an amplifier with a bipolar input stage, no input filter and a relatively small loop bandwidth, none of these values are likely to cause significant subslewing TIM in a valve amplifier because of the better open-loop linearity of valves.
Harmonic distortion
The distortion consists mainly of the third harmonic. In PWM8 mode, which is the mode with least distortion, its level is around -76 dB (0.016 %) at 1 kHz and at 10 kHz, with no significant difference between the channels. It increases to -63 dB (0.07 %) at 20 Hz.
Frequency response
The output level at 0 dBFS in loud mode was measured to be 1.2 V RMS at 1 kHz. At other frequencies between 20 Hz and 20 kHz, it remained within +0/-0.21 dB from the 1 kHz value using the steep (non-apodizing) filter setting. It is unclear how much of the variation was due to the measuring equipment.
Last edited:
Yes, that's what I described as my approach a few posts back and it has worked extremely well with my other DSD decoder implementations. My music library is held as FLAC files, they get converted/upsampled to DSD by HQPlayer which then get's passed to the DAC, either to a connected USB device (such as Amanero or I2SoverUSB) or, as I do, over the network to a BeagleBone based renderer. In simple terms, HQPlayer replaces the FPGA-based components of Marcel's original Valve DAC.
One thing to be aware of - I upsample my FLAC files to DSD256 or DSD512 and that needs a powerful computer to host HQPlayer.
Perhaps you could give some indicative costs? Assuming Marcel's original PCB with the specific layer stack requirements the best price I got was from PCBway at around 400USD for five boards (inc shipping to the UK) - affordable if five people were interested in purchasing but not a price I would want to pay singly.
Last edited:
Might you be tempted to try a reimplementation of your DAC using HQPlayer instead of the FPGA section?
Signalyst
Ray
Hi Dan, was this from EuroCircuits (at around 250 Euros)? Just wondering if going halves on the PCBway price of around 400USD (5 PCBs) might be a possibility?
Ray
Hi Ray and Dan, just to make sure there won't be any mix-ups, I gather that Dan is writing about the board for the original design (PCM version with FPGA module) and Ray about my proposal for a raw-DSD-only board (which has a much smaller digital section, no FPGA at all and a different mute circuit, but is otherwise similar to the original).
Thanks for raising this Marcel as you are correct that there is scope for confusion/disappointment.
Dan, I'm looking to build the DSD-only version, either via the PCB layout that Marcel developed from the board for his original version or with my own layout boards. If you're after the original version board our objectives are different so a joint purchase won't work using Marcel's boards. It may be possible with the board layout I've started working on, though using Marcel's boards is, I believe, lower risk.
Ray
Morning Ray, Marcel
Apologies for not responding in a timely manner, work getting in the way.
Good catch Marcel as, yes, I was thinking of the original boards.
I have the next 12 hours off so I will reread the thread and was halfway through an article on PCM v's DSD.
Currently the argument is for DSD.
@ Ray. Will be happy to contribute to Marcel's edited boards.
Dan
Apologies for not responding in a timely manner, work getting in the way.
Good catch Marcel as, yes, I was thinking of the original boards.
I have the next 12 hours off so I will reread the thread and was halfway through an article on PCM v's DSD.
Currently the argument is for DSD.
@ Ray. Will be happy to contribute to Marcel's edited boards.
Dan
Good morning Dan.
I've got a finger on the trigger to buy a batch of Marcel's DSD-only PCBs from PCBway but I won't pull it just yet - it's not as if there is a pressing target date for the project - so I'll leave you time to consider your options.
From my perspective, sharing the cost makes Marcel's board financially viable and saves me having to invest in designing new boards to reduce the cost. I'm also assuming that, should this project go well and some further interest is generated (and I don't make a mess of assembling the board), that I may be able to recoup some outlay by selling on the spare PCB/SMD stencil.
In terms of assembling the PCB, as it won't fit into my friends reflow oven, I'm now thinking that using a stencil to apply the solder paste and using a hot air solder station is the best option available.
Marcel might disagree, based on his comments about his ideas for another build, but I think the DSD-only board is the way to go even if you have PCM data; you can't play PCM data via the Valve DAC cores and have to convert it to what is basically DSD, which is what the ASRC/FPGA sections do in the original Valve DAC, but with the added complication of the 27MHz clock. Given the necessity of that conversion, if you move it upstream to, say, HQPlayer you get the DSD stream required and at multiples of its native sampling rate and the DAC is much simplified but you fundamentally achieve the same thing. Not only that but for native DSD data you avoid the 'convert to PCM then convert back to DSD' process. I also suggest that HQPlayer is better for the DSD conversion, being more optimised for the processing, and offers more choice with lots of filter/noise-shaping/delta-sigma modulator options to experiment with and a proper digital volume control is available. There are other options for upstream derivation of the DSD stream, for example I believe Daphile has the capability.
The downside of HQPlayer is that to exploit it's capability to derive high-rate DSD, such as DSD256 or DSD512 (to me the sound improves with the higher rates) you need a fairly heavy-lift PC - I use an HP workstation that benchmarks at about 12000 and does a good job. I run AudioLinux as the operating system BTW. You don't have to have a noisy (acoustically and electrically) PC in your listening room though as HQ player allows you to stream its output over the network to a lightweight renderer, for which I've used silent Intel Atom based devices for USB equipped DACs but now my default is to use a Beaglebone Black, running a derivative of the Botic kernel, equipped with an isolator/reclocker - that's why I asked Marcel to equip the DSD-only board with an Amanero pattern connector - it facilitates easy connection of an Amanero USB board (without reclocker) or the BBB/reclocker I'll use. If the BBB/reclocker is of interest to you I can send you a PCB or possibly an assembled board (at cost only).
Just some thoughts...
Ray
I've got a finger on the trigger to buy a batch of Marcel's DSD-only PCBs from PCBway but I won't pull it just yet - it's not as if there is a pressing target date for the project - so I'll leave you time to consider your options.
From my perspective, sharing the cost makes Marcel's board financially viable and saves me having to invest in designing new boards to reduce the cost. I'm also assuming that, should this project go well and some further interest is generated (and I don't make a mess of assembling the board), that I may be able to recoup some outlay by selling on the spare PCB/SMD stencil.
In terms of assembling the PCB, as it won't fit into my friends reflow oven, I'm now thinking that using a stencil to apply the solder paste and using a hot air solder station is the best option available.
Marcel might disagree, based on his comments about his ideas for another build, but I think the DSD-only board is the way to go even if you have PCM data; you can't play PCM data via the Valve DAC cores and have to convert it to what is basically DSD, which is what the ASRC/FPGA sections do in the original Valve DAC, but with the added complication of the 27MHz clock. Given the necessity of that conversion, if you move it upstream to, say, HQPlayer you get the DSD stream required and at multiples of its native sampling rate and the DAC is much simplified but you fundamentally achieve the same thing. Not only that but for native DSD data you avoid the 'convert to PCM then convert back to DSD' process. I also suggest that HQPlayer is better for the DSD conversion, being more optimised for the processing, and offers more choice with lots of filter/noise-shaping/delta-sigma modulator options to experiment with and a proper digital volume control is available. There are other options for upstream derivation of the DSD stream, for example I believe Daphile has the capability.
The downside of HQPlayer is that to exploit it's capability to derive high-rate DSD, such as DSD256 or DSD512 (to me the sound improves with the higher rates) you need a fairly heavy-lift PC - I use an HP workstation that benchmarks at about 12000 and does a good job. I run AudioLinux as the operating system BTW. You don't have to have a noisy (acoustically and electrically) PC in your listening room though as HQ player allows you to stream its output over the network to a lightweight renderer, for which I've used silent Intel Atom based devices for USB equipped DACs but now my default is to use a Beaglebone Black, running a derivative of the Botic kernel, equipped with an isolator/reclocker - that's why I asked Marcel to equip the DSD-only board with an Amanero pattern connector - it facilitates easy connection of an Amanero USB board (without reclocker) or the BBB/reclocker I'll use. If the BBB/reclocker is of interest to you I can send you a PCB or possibly an assembled board (at cost only).
Just some thoughts...
Ray
Hi Marcel, a couple of questions if I may...
Muting: with muting arrangements you've implemented on the DSD-only board, did you consider using the same arrangement as ppy on his DSC2 boards? Just curious why you chose a different approach?
Output Caps: with the 10uF output caps, does that value suggest that you have assumed a low(ish) load impedance in calculating these? I ask because I've more or less made up my mind to use a Broskie BCF on the output of a Valve DAC project (the BCF converts balanced to single-ended and includes it's own input coupling caps and the option for a low pass filter on it's nput) and I'm thinking it makes sense to take the cap values from the BCF schematic (they're 1uF IIRC). I can still locate the caps on the Valve DAC board, in fact doing so gives me more flexibility for placing the LP reconstruction filter.
Thanks
Ray
Muting: with muting arrangements you've implemented on the DSD-only board, did you consider using the same arrangement as ppy on his DSC2 boards? Just curious why you chose a different approach?
Output Caps: with the 10uF output caps, does that value suggest that you have assumed a low(ish) load impedance in calculating these? I ask because I've more or less made up my mind to use a Broskie BCF on the output of a Valve DAC project (the BCF converts balanced to single-ended and includes it's own input coupling caps and the option for a low pass filter on it's nput) and I'm thinking it makes sense to take the cap values from the BCF schematic (they're 1uF IIRC). I can still locate the caps on the Valve DAC board, in fact doing so gives me more flexibility for placing the LP reconstruction filter.
Thanks
Ray
It all depends on what you want, of course. I wanted to make a box that you can connect to any old S/PDIF, AES3 or Toslink source, I wanted to implement all critical analogue and mixed-signal functions with valves and I wanted to learn more about digital signal processing, Verilog and FPGA programming. As S/PDIF, AES3 and Toslink sources require the DAC to synchronize to the source and as I was not able to make a reactance-valve tuned crystal oscillator with a sufficient tuning range, that logically led to a design with asynchronous sample rate conversion.
Your requirements are different, so the solution is also different.
I haven't a clue whether the sigma-delta modulators in HQPlayer are better or worse than mine. What I understood from the noDAC thread is that the details of the HQPlayer sigma-delta modulators are trade secrets, so I don't know exactly what they are doing. I would guess that their quasi-multibit modulator is something with an embedded pulse-width modulator and dithered multibit quantizer, though, which would mean it is rather similar to my PWM4 and PWM8 modes.
Because a TS5A3167 could not work well with the negative voltages. It needs its analogue I/O voltages to be between the voltages at its pins 3 (GND) and 5 (VCC). My outputs are biased at 0 V (after the 10 uF capacitors, that is) and can swing negative. Hence, I would either have to change the output biasing of the DAC, which would be inconvenient when you need a single-ended output, or to make a -2.5 V and a +2.5 V supply and connect the TS5A3167 in between those, which would also complicate things. All in all, using a bunch of J109 JFETs seemed simpler.
The DSC2 has a 2.5 V common-mode voltage that fits nicely with the TS5A3167 voltage range.
The valve DAC requires a fairly low load resistance. In the original design, that resistance also terminates the reconstruction filter.
What comes out of the upper E88CCs is a pair of currents that need to be converted to voltages. That conversion is done by the anode resistors and the resistors on the other sides of the 10 uF capacitors. With 375 ohm from each output to ground, you get a maximum signal level of about 1.2 V RMS differential. With an infinite resistance from each output to ground, you get a maximum signal level of about 19 V RMS and much more severe noise modulation artefacts.
(By the way, it is the sum of the anode and the load resistances that determines the lower cut-off frequency together with the 10 uF capacitors, so you get a cut-off around 2.7 Hz with a low-impedance load.)
Have you got a link to a schematic of the Broskie BCF, so I can see what the most suitable method of connecting it would be?
Last edited:
Ooh, I'm a bit embarrased now
. Re-reading my earlier post I was a bit self-centred with my comments and I hope I haven't caused you any offence Marcel, sorry if I have.I do understand and appreciate the rationale behind the original Valve DAC project and I should have been more sensitive to that. My comments about HQPlayer were meant in the context of the DSD-only Valve DAC, for example it has delta-sigma modulators optimised for high rate DSD, such as I plan to use.
All I can say is thank you again for your patience and ongoing support with this project.
Ray
Last edited:
Here's some information about the Broskie BCF;
BCF PCB and Kit
BCF-2 & Aikido Single-Ended Output Stage
I have previously used a BCF in a conventional DAC project (AK4490 based) and it performed very well. Somewhere I have the manual for it, which isn't in the public domain as far as I know, let me dig it out...
BCF PCB and Kit
BCF-2 & Aikido Single-Ended Output Stage
I have previously used a BCF in a conventional DAC project (AK4490 based) and it performed very well. Somewhere I have the manual for it, which isn't in the public domain as far as I know, let me dig it out...
Hi Ray, Marcel.
I use Roon which, although it does not have the filters that HQPlayer has, will upscale my files. Failing that I could put HQplayer on my Roon server although it would be like doubling software to do the same job.
The more I think about it DSD is the way forward, my CD player is looking as tired as I am.
@ Ray, I am willing to put in the funds for Marcel's adjusted board design. As you say we are in no rush and it would be better to iron out any over-sites / challenges now. PM me and we can discuss terms.
@ Marcel, will you be observing from a distance?
It would be good to hear a subjective review of PCM v's DSD through your tube DAC design even though I suspect that different algorithms will shape (another subjective word) the sound differently and therefore difficult to judge.
Once I have plumbed it together perhaps I can send it to you so that you could give us a review....
Back to work....
I use Roon which, although it does not have the filters that HQPlayer has, will upscale my files. Failing that I could put HQplayer on my Roon server although it would be like doubling software to do the same job.
The more I think about it DSD is the way forward, my CD player is looking as tired as I am.
@ Ray, I am willing to put in the funds for Marcel's adjusted board design. As you say we are in no rush and it would be better to iron out any over-sites / challenges now. PM me and we can discuss terms.
@ Marcel, will you be observing from a distance?
It would be good to hear a subjective review of PCM v's DSD through your tube DAC design even though I suspect that different algorithms will shape (another subjective word) the sound differently and therefore difficult to judge.
Once I have plumbed it together perhaps I can send it to you so that you could give us a review....
Back to work....
OK Dan, let me check where I got to on the PCBway website and I'll PM you tomorrow so we can get things moving.
Interesting that you're a Roon user, I'm actually planning to adopt it for my control software but to retain HQPlayer. HQPlayer's user interface is pretty basic/clunky and at the moment I drive it from an Android tablet using HQPDcontrol, which is a lot better but I do like the richness of Roon.
How are you planning to connect the DSD Valve DAC to Roon? If it helps, the Beaglebone botic based software I use as an HQPlayer Network Audio Adaptor can be set up as a Roon bridge;
https://www.diyaudio.com/forums/twisted-pear/258254-support-botic-linux-driver-280.html#post5822865
Ray