Yea, I get that - totally. I have to be in the right mood, but when I am I enjoy it. Of course at this point, I have a rather significant investment in gear (huge magnifying work light, Metcal & JBC soldering stations, tons of tips for every size/pad, etc.). Nothing like the joy of your very last 10k 0603 .5% thin film flying away to places unknown :-)

I'll keep you all up on their progress. I suspect that laying out a JG that will work with their stack is going to yield further sonic improvements as they'll be able to match the LR/Corner exactly to the reconstruction filter output.

Some day I'll share the story of my daughter's dog passing some small components :-) Well, I guess that's the story.

Here is mine... currently stock regulators (lm317) and shortly Salas Shunts...



Uploaded with ImageShack.us

Initial listening shows to be relaxed, large sound stage,very good stereo separation, lots of detail but without harshnes, but i feel the dynamics are a bit on the shy/slow side...

Salas shunts will improve things for sure...

Thank you MisterRogers for bringing this dac to our attention, and many thanks to Lyuben and all the jlsounds team.

Looking forward to new developments, and will keep you posted on my progress!!

Nice stuff, congrats on the build! I'm working on a I/V buffer and filter too, first time seeing the JG buffer.

I might be missing something, but I think you'd get better phase response with a high order bessel? Here's a sim of the JG's filter section vs a 7th order bessel. The JG goes on Vout, while the bessel has to be used after a buffer on an Iout DAC. The filter's load impedance acts as the I/V resistor, so changing it would cause ringing - the impedance needs to be scaled to use it with a Vout DAC.
Filters: http://i.imgur.com/7kXKs0U.png
http://i.imgur.com/QjSLZXW.png
http://i.imgur.com/BMo48lX.png

Phase response is better on the JG, but group delay and stopband attenuation better on the bessel.

The JG seems to ring a lot more too, even with ideal components. It rings at 100KHz though, so maybe it doesn't matter?
http://i.imgur.com/qFxasIz.png

It's a few more components, and you need to unwind the inductors to get the exact values but even that's not really necesary (part tolerance isn't critical).

edit: better pics

I think I'm set on the Bessel. Here's the latest iteration, exclusively for (buffered) current source DACs:



This is a differential/balanced 7th order Bessel with an output impedance of 3k. All the values will need to be scaled for a different I/V value. This brings a pcm63 just about to line level. The caps are all standard Wima values, but L1/4 will need to be unwound from larger values, ideally within 2% tolerance. I've done it and it's not that bad with a decent L meter. No harm in rounding L2 and L3.

Here is the result: http://i.imgur.com/a026d8j.png
And against the JG buffer: http://i.imgur.com/QpCxYQP.png

The Bessel's biggest weakness is frequency response. It's at -0.1dB at 15kHz and -0.2dB at 20kHz. Not quite up to industry standard of -0.1 at 20kHz but it could be fixed by shifting the corner frequency up a bit- it's at 78kHz now. I picked it because it was possible to get the C values by combining just two standard caps with a filter impedance of 1.5K but it could stand to be optimized with. On the other hand, the maximum group delay is -6.6ns relative to the passband. Stopband attenuation is >50dB at 352kHz.

Since all filters necessarily introduce delay, it makes sense to optimize for constant delay rather than constant phase (i.e. minimize phase distortion). I've played with all-pass lattices as well, and they'd probably be great with a Butterworth but I haven't managed to optimize phase response. The Butterworth could easily reject 90dB of sampling noise but has bad delay characteristics.

It didn't pop up in simulations, but I'm also wondering if filter ringing couldn't be a source of IMD. High frequency AC heating of DHTs has been shown to increase IMD (through heterodyning) so it seems like risky business.

Will build and report back!