Valve DAC from Linear Audio volume 13

You could indeed remove L6 and use a separate 5 V supply for U17. Any jitter caused by ripple on the 5 V for U17 is supposed to be suppressed by the current steering gate made of E88CCs, but that suppression is imperfect, so maybe a separate supply or separate regulator could improve performance to some extent. I don't know, as I haven't tried it...


Excellent. Where on the pcb would you ground the extra 5V supply if the +ve went to the correct pad on L6 (or somewhere closer to the load if physically possible)? Certainly not going to do this upfront, but perhaps at a later date as a trial.


...it certainly makes more sense than using an ultra low noise regulator for supplying an FPGA board.

For suppressing really rapid current variations, physically small capacitors work best. Hence the use of 10 uF X5R SMD capacitors at various places.


Sweet.


Regarding noise, as reported in the Linear Audio article, I measured noise floors of -85.76 dB(A) left and -91.29 dB(A) right with respect to a full-scale sine wave in the PWM8 mode. That's low enough not to bother me at all, but not nearly as low as some solid-state DACs.


Yes, some SS dacs have excellent numbers but not necessarily excellent sound. I am firmly in the "good enough" camp in terms of measurements and the decent noise performance, relatively simple circuit and your approachability is what has drawn me to this project. My playback system is enormous with 110dB/w/m sensitive horns and I have zero tolerance for hearing any idle noise with my head in the horns so noise is quite important in this context.

Currently have 40 vacuum tubes in the system including DHTs with both AC and DC in the filaments and have managed to get things perfectly quiet almost straight away with well planned layout, generous space where required and use of good components, particularly the iron, and there is a lot of iron...just don't want this dac to sound so great I want to keep it but have my horns hiss at me with little potential to lower noise. Not sure if that will happen or not. Had to go to some extreme lengths recently to chase noise out of a SS phonostage but got there in the end with particular attention to the power supply. The ValveDac is not so simple as a phono and there is a lot more to dropping noise than improving a bipolar power supply, hence my questions here somewhat in advance of actually needing to try to lower noise...they are more aimed at trying to figure out viable options I may need to lower noise and how much extra room to leave in the chassis just in case (I make all my chassis' from scratch). Hence my questions about reconstruction filters, DC blocking caps, 5V supplies. I've even considered asking about dual mono power supplies, but looking at the KiCAD files that would be difficult to implement with the pcb...would also require a dedicated -300V clock supply.
 
On the subject of iron, Dave Slagle has posted the custom ValveDac output transformers today, wound on 82% nickel cores. It should not matter in terms of performance, but in my experience those nickel cores just sound better than anything else I have tried. There is Slagle iron in my valve LCR phonostage, preamplifier and the midrange channel of my power amplifiers, and soon the ValveDac.

Price is good too.
 

Attachments

  • Slagle ValveDac OPT.jpg
    Slagle ValveDac OPT.jpg
    125.7 KB · Views: 165
Must be a funny current waveform, roughly 1 kHz square wave with on average some 22 kHz on top of it, but frequency modulated by the square wave? It's not the megahertz signals I was thinking about.

I don't think that particular supply is aimed at the MHz of an FPGA, but is supposed to show the transient improvement when ultracaps are used on a generic, presumably lowish quality LPS. 325F is about 325 million times more capacitance than a 10uF X5R, with vanishing ESR, but they are quite large and would take some time to charge off the power supply before ready for first use.

EDIT: Just looked at the scope shots more closely...seems the power supply is the USB charger for a Rasberry Pi...not even a decent quality LPS to start with.
 
Last edited:
On the subject of iron, Dave Slagle has posted the custom ValveDac output transformers today, wound on 82% nickel cores. It should not matter in terms of performance, but in my experience those nickel cores just sound better than anything else I have tried. There is Slagle iron in my valve LCR phonostage, preamplifier and the midrange channel of my power amplifiers, and soon the ValveDac.

Price is good too.

Do you have a link? I checked and could not find it in the intact forum.
 
I just did what I normally do, sent Dave an email with the schematics, back-and-forthed a bit about about what I am trying to do and the specs it needs to hit and he talks me back to reality and then custom winds exactly what is required. If you send him an email I am sure he will be aware of the project and the transformers...the ones made for Anthony and the ValveDac.
 
Price is comparable to the Jensens. I think Dave might want to see how mine work out before winding too many more, not sure.

I'll be building as soon as the pcb's arrive which should be any day. Power transformers are the biggest wait, but I do have some here that I can make do with as an interim measure.
 
Probably not straight away, but I will likely give it a try once everything is working and verified. Across C18 looks the spot to do it, and it is easy to desolder a couple of SMD's to facilitate the change. Like Marcel says, maybe it will improve something, maybe not.

What I have done is order two complete sets of inductor bits and a load of polystyrene caps so that I can test various reconstruction filters just by switching them in and out. First decide between 6th order 0.05deg Linear Phase and Gaussian-to-6dB, then maybe try some lower order filters and/or lower cutoff frequencies. My hunch is that some experimentation there may yield some nice performance or sound quality benefits.
 
I think leaving it would be best unless the new power supply can offer something "faster" and very close-by, preferably only an inch or two away (i.e. no long wires), which is why I brought up the ultracaps earlier.


10uF X5R cap ESR circa 4.5mOhm...I think...tucked right in next to the load which is ideal.

vs

2 x 325F Maxwell Ultracap ESR circa 1.5mOhm each...from datasheet...3mOhm in series...a few inches of extra wire adding to the ESR...lets call it equivalent ESR to the 10uF X5R if we keep the wires very short. So no advantage to the ultracaps even though they cost $50 to implement compared to cents for the X5R.



Where the two implementations may differ is 325million times more capacitance with the ultracapacitor smashing down ripple/noise and the output impedance of the 5V power supply in parallel to the Ultracap/X5R, along with different FPGA noise which may or may not actually bring an improvement in performance of the clock/dac. That is the big question.

However, if I had your voltage regulator and wanted to use it then I would get it as close to C18 as possible without removing that cap (and of course would remove L6).

Please don't take this for gospel! I could be wrong...I'm a farmer/audio enthusiast not an electronics whiz like Marcel.
 
Last edited:
The main reasons for avoiding long wires and large-area loops are inductance and inductive coupling rather than resistance, otherwise I agree with acg. Definitively leave in C18 and mount the new regulator somewhere close, with short wires and only a small loop area from the regulator output to C18 and back to the regulator ground pin.
 
By the way, I added a section about the digital precorrection for the analogue reconstruction filter to the opening post:

Digital correction for the reconstruction filter roll-off:
The digital interpolation filters in the original valve DAC have a partial correction built in for the droop of the analogue reconstruction filter. No such droop compensation is possible in the raw DSD version.

Hence, for the flattest possible amplitude response, the reconstruction filter of the raw DSD version has to be flat by itself while the reconstruction filter for the original valve DAC has to have a roll-off that's about the same as the roll-off of the filter that the correction was designed for. Therefore, Butterworth filters give the flattest response for the raw DSD valve DAC and 0.05 degrees linear phase or Gaussian-to-6 dB filters give the flattest response for the original valve DAC. While the phase response of Butterworth filters is not as good as that of Gaussian-to-6 dB or 0.05 degrees linear phase filters, it is still pretty close to linear phase an octave or more below cut-off.

The digital droop correction filter has a magnitude response:

1 + 13/256 - (19/512) cos(2 pi f T) - (7/512) cos(4 pi f T) with T = 5 us.

This is only a small correction: 1.009680428 or +0.083678759 dB at 15 kHz, 1.016534293 or +0.142440687 dB at 20 kHz, peak of about +0.644660534 dB at 73.5 kHz.
 
In regards to the power transformers in the case, it depends. My experience is that they need lots of separation from low level circuits and most certainly from any inductors. Steel is the best shielding material by far (ferrous metals) as is distance but all transformers will couple to the steel in a chassis and cause vibrations which is not what you want with valves and clocks in this project. Toroids are supposed to have less stray flux than EI cores but this is not always the case, and they are very susceptible to DC on the mains which makes them vibrate and sing...I still prefer to keep all transformers far away and tucked away inside a heavy steel sarcophagus of sorts.

At the moment I am designing a chassis to finally put away my breadboarded DHT preamp. The preamp is currently sitting on a lump of MDF and the EMF meter needs at least one metre (three feet) separation for interference to fall to background levels. By contrast, the fully boxed up mono power supply for my 6 channel SET amplifiers is packed with power transformers and chokes, operates on much higher currents than the preamp project (and ValveDac project), weighs about 100k each with the chassis made from 4mm mild steel. That 4mm thick steel, along with some generous spacing for the EI transformers in the chassis (keeping them at least 40mm from anything ferrous) gives me background RF and EM readings just 30cm (1 foot) from the case, which is a much larger task than the fields for the ValveDac.

For the ValveDac, I will use the EMF meter to measure any issues before making decisions about single or dual cases. Toroids will be used although I usually prefer EI because of the DC gap and lower bandwidth but the local guy that normally winds them for me is snowed under.