Like the previous 4 versions, the concept for the HPDAC5 happened a little by accident. I was tinkering around with an old TDA1543 based DAC and it occurred to me that I didn't know how to flip phase on it. I don't remember why I wanted to flip phase, but it seemed like something worth knowing. After figuring it out (turns out you put a 74*04 inverter on the data stream), the idea of putting them back to back seemed like a good one.

Now, I like the TDA1543. By no means is the resolution, or the distortion spectrum, or the frequency response, or any of the usual measurement criteria of this chip good. Indeed, it is terrible by these standards. It is even worse when you use it NOS. But, there is something about it that just sounds good. Moreover, unlike your PCM1794's or your Buffalo chips, or whatever the kids are in to these day, it is easy to play with on breadboard, it is cheap, and it is fun. In short, it is everything DIY should be.

So, I dug around through the random parts bins, and turned up a SN74AVT04N, an extra LM7805, a CS8412, and a pair of Lundahl LL1930 output transformers. So, I plugged them together and low and behold, music. And, not just music but good sounding music. Many of the worst sins of the TDA1543 are eliminated here both because of the balanced nature eliminating some distortion, but also, I am guessing, because there are no output caps.

The Lundahls, or something like them (with high primary inductance), are necessary without the mosfets. In that arrangement, driving phones directly is probably not an option without stacking lots of chips (8 per side provides the Zout, though not the voltage, for Senns -- 16 per side and a stepup transformer might work, though). With the mosfets, low impedance (600:600) transformers will work -- this is good as they generally sound better, and they are much cheaper.

The biggest benefit, to me, of the differential design is that while there is offset, it is symetrical allowing a non-gapped transformer to be used for coupling. The chips are not matched exactly. However, I matched the fets pretty closely (within about 5mV) and then by putting trimmers on both DAC chips Vref pins, the offset as a whole can be reduced pretty well. I think I have 3mV one one side and -2mV on the other. If one were really clever, one would put the matched mosfets into the circuit such that the higher Vgs one of each pair were on the same side (+ output for instance) which would allow even tighter control of the offset. Alas, I was not that clever. Still, it is pretty good.

For the final version, I included an ASRC. With these chips, this only seems to work at 44.1KHz, so I am just using the recovered clock from the receiver chip into the ASRC as a bit of a jitter reducer. Trying it with and without the chip, it does seem to help.

In case anyone is interested, I have a small number of extra circuit boards available for this project. I also have a few TDA1543's.

Last Update:Friday, 04-Dec-2009 21:52:18 EST