Seeing as building hexacaps with wire is very time consuming, I'm investigating ways to get the production cost down. Here's my first attempt at a PCB hexacap, using 2oz copper - I did the layout on EasyEDA - Web-Based EDA, schematic capture, spice circuit simulation and PCB layout Online

Once I've ironed out the minor errors on the groundfill I'll make the PCB public so anyone can order up their own. There are 127 caps on this board giving a 50Hz capacitive reactance just under 8mohms.

Hexacaps have their limitations - they provide a great low impedance for lower freqs but according to my LCR meter, turn inductive for frequencies in the mid-single digit kHz. Which is about the lower edge of the passband for a tweeter. So what to do to get a low impedance power supply to a DAC chip from 4kHz and up? Here's my first attempt - there are 7 TDA1387s each sandwiched between two low ESR 4700uF lytics. Sandwiching them together is an attempt to get the inductance as low as possible - the loop area can't practically get any smaller than this. Getting the wires in has tested my patience and that's even before powering the beast up....

Update - well I could have been knocked over with a feather - this worked first time I plugged it in. So now I'm thinking about how to build a markII version, considering if I can get more decoupling close in....
Oh I'm now not sure that my bandpass tweeter filter is really delivering the goods as per its simulation. Connecting...

I'm prototyping this baby which will probably be the first Ozone board to go to PCB layout. It implements the 10th order Chebyshev filter shown earlier and is intended to be mounted atop a 'hexacap' passive shunt. The DAC stack in the centre is made up of six TDA1387s. Input will be I2S and output unbalanced audio 40mV RMS.

Update - over on a thread about AD815 Eldam asked about the inductors - how many I need to buy to get the values in the schematic - so I'll talk about that a little here.

I have bought a big reel of 1k pieces of these 680uH inductors. The price was around $100 but I think I probably should have paid a little more as although the reel is marked up as TDK, they might well be fakes. I suspect that because their appearance isn't as professional as some 1mH parts I bought earlier from another seller. Added to that is the spread of values - having measured at least 100, only one has shown up as 680uH (within 10uH). The rest have varied from 800uH...

I took this to the park this morning, it survived the journey in my backpack but it exposed a dodgy RCA connector. The case I used (temporarily) for it is a cylindrical tea carton - the red 'tower' behind. Battery power comes from 3 NiMH AA cells giving around 10hrs of listening.

I've just done a mental estimate of the BOM cost of this DAC (labour cost is way dominant, but this is DIY right?) and it turns out to be dominated by the cost of the caps in the top three tiers (the small black ones). These are real Japanese Rubycons, chosen for their diminutive height (16mm) vs the Chinese ones at 25mm height. So if you didn't mind adding 3cm to the overall neight you could build this with all Chinese caps and save a few dollars. As it is the total parts cost is under $30 including batteries. If you wanted S/PDIF input (currently its I2S) you'd add a couple more $.

I've munged the earlier 8 inductor filters down to only 5 - this one is using off-the-shelf TDK inductors, SLF7045-681, albeit they need to be hand-selected to get the correct values as the production tolerance is 20%. X7R caps sound just fine, surprisingly

FR is -3dB at 17,8kHz with around 0.25dB of passband ripple.

Here's my current bench lash-up DAC based on three hexacaps. Total capacitance of the order of 1.3F.

32fs EIAJ is fed in, balanced, from the left, the central hexacap feeds two TDA1545ATs. Filters are TDK inductor based, six stage balanced with X7R caps (gasp!). Caps in the centre provide the mid-rail power (2.5V) which the I/V resistors are fed from. The two hexacaps on the right power the post amps based on AD815ARBs, gain set roughly to 26dB with a little NOS droop correction. They're the postage-stamp sized boards roughly in the middle. Output is differential and fed to ferrite-cored transformers to convert to SE.

I'm blown away by the dynamics of this, given the right recording. Next up is an amp based on paralleling 8 or 10 AD815s. Each fed from its own hexacap, of course....

Update1 - the DAC supply has now been upgraded to two tiers of the 4700uFs, interconnected by six 1.5mm diameter solid copper wires. Capacitive reactance measures 2.5mohm...

Here's my latest crazy project, intended to answer the question 'How low can you go?' in power supply impedance. No listening tests yet, I've just finished assembling this stack of hexagon capacitor arrays out of cheap electrolytics. Total capacitance of the top 4 tiers is comfortably above 1F. The top tier is designed for lowest ESR at HF using Nichicon 1500uFs, the next tier is a trade between ESR and uF. The last two are the highest capacitance density I could find (a total of 182 4700uFs). I've yet to connect the bottom two hexagons to the upper tiers but the top two are connected by wires running through the gaps between caps.

The whole is intended to push the limits of low frequency noise in powering a couple of TDA1545s which will sit on a small PCB supported on thick wires from the centre of the top tier. I am aiming to get a capacitive reactance (at 50Hz) at the DAC supply pins below 3mohm and ESR below 1.5mohm.

Update1 : connected the bottom two tiers...

Last night I finished building the second channel of my dual mono approach to DAC building which I've called 'free radical'. Here's a picture of the second channel's build just prior to adding all the crapacitors (cheap shanzai 'Sanyos' which measure extremely well). While building the second channel I was listening to the first in mono and that was a spur to quick completion

From left to right there's the AD605 with its top hat array of MLCCs - outputs are isolated via ferrite bead chokes from the AD8017 under its own pile of SMT ceramics. In between the two active stages are the capacitors associated with power supply reference voltage filtering. I realized from the previous build that as the AD605's gain is controlled by DC voltages, these voltages need to be low noise to ensure gain stability. Hence lots of RC filtering with those Nichicon and Rubycon low ESR lytics. I'm using BC817 transistors as low drop-out regulators and the reference voltages (2.5 and 5.7V) come...

I've had the beta build Ozone (no input stage yet, fed from my QA550 via the I2S transcoder to down-convert to 32fs) playing out 24/7 for a few days now.

Overall I'm very happy with how it sounds, just a minor gripe about sibilance on some operatic vocals which I'd like to understand better. On the upside the jump-factor (read dynamics) and soundstage stability (holographic on the right disks) are about the best yet. I'm using a couple of Decca double CDs for this - 'La Traviata' and "La Boheme'. They're about the most transparent sounding and demanding disks I have. Demanding in the sense that they have lots of emotional drama which should positively engage my attention if the DAC's really up to snuff. I never much enjoyed opera until I got into building my own NOS DACs, but now I really enjoy my dramatic fixes and these two recordings are really top of the pile. They're about 50 years old but so far I've not found anything newer which touches them (not that I've looked...

Inspired by this thread https://www.diyaudio.com/forums/digit...-new-post.html
I've been giving a little thought for how to move beyond the 'Ozone' to a DAC able to handle more than 16bit inputs and up to 96kHz input rate. The target being 20bits and 120dB SNR (non-A weighted). The OP in that thread preferred a more marketable 'ultimate' DAC (with 192k and potentially 384k capability, along with DSD) - aims which to my way of seeing clearly conflict with a DAC having any pretentions to ultimacy.

The simplest solution - building on the digital part I already have - would be to add more TDA1545As in parallel and with a Cortex M4 direct the data to the respective chips. What's unclear though is how low the noise will go when the extra chips are added. From the DS, an A-weighted SNR is quoted of 101dB (typ) with 2mA, however this is a static noise (code = 0) and hence may well not translate to the noise achieved with bits toggling....