Goosebumping headphone DAC-amp

The Intersil D2 chipset based amp arrived and while it sucked pretty badly on speakers I had a hunch that like the STA333BW boards I've been playing with recently, it would acquit itself admirably on headphones, given the much lower demands on the supply.

The mods I've done to this are fairly straightforward - changing the output filter to work more optimally with headphones (via a 7:1 step-down transformer) and tweaking up the power supply a bit.

To run with cans I went for an output load impedance of 1100ohms. This was determined primarily by the choice of inductor (MSS1210 8.2mH from Coilcraft) which in turn was chosen so that the carrier frequency would be below the SRF of the coil. This choice is fairly arbitrary in that the inductor would still work fine above its SRF just it's starting to look like a capacitor. However it turned out that with 8.2mH the cap value required was fairly simple, 5nF so I stuck two 10nF NP0s in series.

Mods in the two pics I've taken (a green PCB in my edition) - on the top side note the four 470uF/35V caps (Nichicon HE) came out in favour of 1500uF/35V Nichicon HD with about half the ESR and 3X the capacitance. The 15uH inductors have been subbed for Coilcraft 8.2mH MSS1210. Six yellow polyester caps also extracted (4 * 100nF, 2*470nF) and substitute SMT caps along with damping networks fitted on the underside. Star grounding is mandatory.

The pics from Taobao of what this looks like show a yellow PCB. A very well made and attractive looking piece of kit (although the display operation is rather tacky).

There's an analog input on this so I decided to give it a whirl. Amazingly undynamic compared to the sound when fed in digitally, so I'm figuring it must be something to do with the supply to the ADC chip, an AKM5381. A Nichicon 1800uF/16V very low ESR cap on the analog rails brought some life to the sound but its still a gulf that's probably too wide to bridge. Unless I go in and change the regulator and give all the rest of the circuitry a careful once over which I'm not too likely to do in the near future - having too much fun listening to it!

Here's how to make a 7:1 step down transformer to complete this amp. Using a PQ4040 core+bobbin, wind on 2000 turns of 0.19mm diameter magnet wire. This is the primary, the secondary is made from 290 turns of 0.65mm diameter wire. You can create various taps on the secondary to use if your headphones are very high sensitivity (like IEMs). If you have DT880s at 600ohms you'll need a secondary with about 1000 turns of 0.25mm wire.

The ferrite cores are here :
https://www.mouser.com/ProductDetail/...7dbYHNXn9NI%3d
The bobbins here:
https://www.mouser.com/ProductDetail/...%2fbqsxQ%3d%3d
I linked to the Fair-rite cores as they're so much cheaper than TDK/Epcos.

Update - I've bought a couple more of these beauties as I want to dedicate one to headphone listening and still have others to play around with. So with the second one, I am at present trying out driving speakers but through step-down transformers (roughly 2:1) and although the volume is lower, I'm now getting the kind of quality I am looking for in an amp. By which I mean depth of soundstage and ambience retrieval. The HF is a bit rolled off though, probably due to the either the transformers being lossy or their leakage inductance being a bit high. They are just re-purposed 15VA EI core transformers from the local electronics market, very bog-standard. I dismantled them and re-wound the primary (which was 220V) for much lower voltage (around 24V). The secondary remained as 12V. For this experiment I'm using the amp completely stock. I shall go on to improve the PSU and see where that leads me, after all its just a very standard black box of the kind used for laptops, the rating is 2.5A max.
Loaded up a 20-20k sweep waveform in Audacity and it shows there's significant roll-off of the FR, beginning about 2.5kHz, this is due to the leakage inductance of my OPTs (verified by seeing a flat FR unloaded), after all they're split bobbin which has the highest leakage. So split bobbin EI trafos I think should be limited to just bass/mid drivers in fully active systems.

Since there's a fair amount of interest in these amps, I've opened a thread - https://www.diyaudio.com/forums/digit...ml#post4879667

Update 2 - 2:1 transformers didn't get rid of all the artifacts on piano so I swapped them out for another couple I'd made earlier which have about 3.6:1 step down. These got me in the ballpark on piano but the roll-off bothered me so I made some leakage inductance measurements. Turns out the effective leakage inductance in series with 8ohms (my speakers) is 1.5mH so the low pass kicks in about 800Hz, quite a bit lower than I'd estimated previously. But then these trafos are different, larger (30VA). I imagined re-winding them with interleaved pri and sec windings (too much hassle) so wondered what a toroidal would measure like. I have a handful of 150VA donuts which used to power my active speakers, turns out the step down ratio is 4.8:1 so I figured they'd be worth trying. Their leakage inductance I found to be impressively low, only 2.7mH on a primary shunt inductance of 1.8H so under 0.2%, whereas the EIs have about 10% leakage. Substituting these and seems I'm heading for audio nirvana now as the FR should be much flatter (not actually measured mind, but estimated from the leakage : 2.7mH turns into 120uH effective on the secondary). The subjective effect is the contents of the soundstage has all shifted forwards, I'm going to play a considerable number of familiar tracks before giving this a verdict but its jolly promising....

Update 3 - suggest no-one buys one of these until I get to the bottom of a problem with it I've just found. When loading the output I found there's considerable sag, so that it looks like the output impedance is rather high. Turns out the P-channel FET in the output stage isn't staying in saturation. Best guess at present is the fitted FETs are fakes, with too high an on-resistance. Update later when I get to the bottom of it.

Update 4 - fixed this issue now. It wasn't the FETs rather its the caps doing the level shifting between the ACT540 driver chips and the FET gates. Bottom line is 2.2nF is just too small, given the amount of charge that needs transferring to turn the FETs on fully. I paralleled 3.3nF with the existing 2.2nF and all seems good so far. There are 8 of these 2.2nF caps that need their value increasing, on the schematic that's C61,63,64,65,108,116,117 & 118.

Update 5 - added pic of where I've paralleled the caps to fix the FET drive issue - 0805 4n7 ceramics are strapped on the underside of the 2n2 caps with the aid of 0805 0ohm resistors as jumpers. Having only done this mod on a straight-out-of the-box unit it definitely doesn't suck so much with speakers any longer. Its half-decent, I can hear the effect of the PSU noise on piano from inadequate supply impedance, but played at modest levels (say under 68 on the volume display whereas unity digital gain is 80) its not bad at all - it has some dynamics in the mid and decent depth.

Update 6 - I've modded a third one now for headphone duty primarily, but I wanted the option to convert it into a DAC without major surgery. So I'm going to out-board the trafos and I have fitted only the first stage LC (8.2mH and 10nF fitted differentially, no damping Rs) inside the case. I've also beefed up the supply some more with a total of 12,000uF across the rails. I shorted out the 120mohm current sensing resistors to lower the supply impedance since with the 8.2mH output inductors the amp's essentially short-circuit proof - they have a DCR around 5ohms. I also discovered a major reason why the analog input sucks - the ADC is fed direct from the LM2596 buck with only a ferrite bead between for filtering. So I've installed an LC to attenuate the switching ripple (150kHz) by about 60dB - this then feeds the analog 5V to the AK5381. Haven't listened to the analog in yet to check on the improvement as I want to do more mods to the input conditioning like raising the input impedance. Its sounding excellent through the lashed-up trafos at present, I'll find a case to suit them soon.

Update 7 - the pic just added shows transformers installed in a top-box, fed from below by the Popu amp. Here I'm using PQ3535 to be low enough profile to fit inside a case no taller than the amp itself. The trafos have a couple of taps, the lower one for lowZ 'phones (output impedance about 1ohm) and the other one drives higher impedances nicely, even delivering sufficient for DT880s at maximum volume. Winding details - primary 2400T of 0.15mm, secondaries of 0.47mm, up to 248T.
From further playing in LTSpice, including modelling the leakage inductance I've optimized the input filter details now, the previous 10nF shunt cap is too high, this becomes 4nF to get more flatness at HF and the damping is 2k wired between the two phases, in series with 330nF. Turns out also that the DC blocking cap feeding the trafo I selected was a bit low, it leads to a resonance below 20Hz (in conjunction with the primary inductance of 29H) which has effects above 20Hz, so I'm increasing from the 4u7 I used before - hence those extra yellow rectangular caps in the photo.

Attached Thumbnails