A PM I received from an interested DIYer asked me my view of PhiDAC SE's sound against the original lingDAC. I was piqued too so set up a comparison. The result was that PhiDAC SE won on bass (all those caps in the 'pants' aren't for nothing!) but lingDAC's top-end was still sweeter. So not an outright win for either design. Seeing as lingDAC's power supply is an extremely modest one due to the choice of a 5*5cm form-factor for the PSU board, I wondered whether lingDAC could be improved with the addition of more caps. Turns out it can - so, with a considerably improved PSU, it wasn't a big surprise that lingDAC became the clear winner in the shoot-out.

This result has led me away from using ICs and back to the discrete-based I/V stage of lingDAC for my 'next generation' DAC. I've also decided to introduce a new element into the design, a transformer. In the past I've used transformers for bal-SE conversion (or even just SE-SE isolation) of the output on various prototypes but this time the primary proposition is a little bit different - impedance transformation.  Isolation is of course handy to have but here its coming along for the ride.

As with any discrete I/V stage, input impedance is a major design focus. With a filter prior to the I/V as I'm using, the I/V stage doesn't see the high impedance current source output of the DAC, except at the lowest frequencies. Rather it sees the shunt caps of the filter and these have a falling impedance with frequency. The presence of the shunt caps leads to increasing HF distortion whose overall magnitude is related to the ratio of the filter's Zout to the I/V's Zin. Its for the lowest HF distortion that we need the lowest possible Zin for the I/V. Given the I/V stage's input topology is a grounded base transistor, its Zin is a function of the collector (or emitter) current. In other words, higher gm for the transistor is better as then the Re is lower. To get a 1ohm input impedance this necessitates 25mA collector current, which is a bit on the high side - lingDAC operates with single digit mA and has around 5ohm looking into the first transistor's emitter. If the filter were matched to this impedance we'd get a substantial amount of distortion at all frequencies seeing as the input impedance varies with the signal. So I use a degenerating resistor in series with the emitter to give a higher overall input impedance, the variation of the Re with signal then is a smaller fraction of the total input impedance. Still, the HF distortion isn't too pretty being of the order of 1%.

In order to aim for lowest distortion it turns out we want the filter to have as high an output impedance as we can, and the I/V to have the lowest Zin. Lower Zin can be obtained through the use of a CFP (complementary feedback pair) - a composite transistor which uses local feedback to give a much lower Re for the same Ic as a single transistor. I've played with CFPs and they're a bit tricky to get stable, particularly when both transistors are bipolars. A bipolar/MOS partnership I've found best from the pov of stability issues. I've already got a re-design in the works of lingDAC using CFPs in the I/V stage, however its only at the PCB layout stage and has been languishing there for months as other avenues have looked more interesting, in particular working on raising the filter output impedance.

To raise the Zout of the filter requires higher valued inductors (together with smaller caps). Larger valued inductors get more tricky to source - physically they end up bigger and most likely SMT versions won't exist. The TDK 7mm range lingDAC uses tops out at 1mH for example which would permit less than a 2-fold increase in Zout of lingDAC's filter. There's a considerable downside to raising the filter impedance and that's that all current output DACs prefer to have their outputs loaded with the lowest possible impedance. An increased filter Zout brings along an increased Zin. So what's good for the DAC is bad for the I/V, and vice-versa. What to do?

This is where the transformer comes along to resolve the dilemma - it offers impedance transformation so that the DAC can have its low impedance load while the I/V can still see a high impedance. In theory, everyone's happy. In practice its not quite so simple but the transformer does bring something genuinely useful to the party. I'll add more details in a subsequent post, time to refresh my cup of black tea....