There have been a lot questions posted here and elsewhere looking for more information on the development of the RDAC. It makes sense, this is a surprising product, and I’m glad to share : )
In the earliest development stages of our R-2R DAC, we were inspired by existing high quality implementations such as the MSB and TotalDAC products. These were some of the most beloved DACs in the audiophile community, and we wanted to achieve similarly excellent sound but at a more affordable price. The first DAC that let us believe this was even possible was the Soekris, which remains one of the best sounding DACs I’ve heard.
So we eventually started working on our own R-2R implementation with our partners at Massdrop, and as with most development processes, that meant studying the existing implementations to understand why they’re great and draw inspiration. One common thread between all of the great R-2R DACs is using sign-magnitude. By nature of how digital circuits work, the typical representation of binary numbers (called 2’s complement) leads to a large number of bits switching at the same time whenever you go from a positive to a negative number. This is called zero crossing distortion. Using sign-magnitude avoids this type of distortion, but doubles the number of precision resistors required per channel. This significantly increases the cost to produce, but we decided that we could not compromise here. The input also has to be converted from the native 2’s complement representation to sign-magnitude, which is done by the control logic in the PLD.
The rest of the basic R-2R circuit comes from this decision to use sign-magnitude. The output from each of the four ladders is rather weak, so we had to give it a bit more current before joining the two halves of the waveform back together and applying a reconstruction (low-pass) filter. These are the four op amps per channel on the output of the RDAC top board. The separation of the input into positive and negative halves requires the ladders to operate from a stable reference voltage, which we achieve with linear series regulator circuits, which are located next to the PLD.
The circuit elements in place, we had to put everything together on a board. R-2R DACs can be sensitive to noise, so it’s important for the component layout to have short paths. We worked with a consulting engineer for this and he looked at a number of existing boards for a baseline, including the MSB, Soekris, and several more that he researched. This included an early version of the Hibiki that he came across on the Chinese audio forums, as well as various open source boards. From there, he laid out our board with a ground layer to further manage noise in the circuit. This 4-layer PCB eventually became the top board of the RDAC.
And then, as it usually goes, we started showing it to some people and not everyone was a fan. We paired the prototype board of our R-2R DAC and some basic off the shelf input modules, and it didn’t sound anything like a great R-2R DAC. There was none of the gentleness or detail we were hoping for. We adjusted the filters, tightened the precision of the voltage regulators, and used better components for some minor improvement. The basic R-2R structure was a start, but it wasn’t enough yet. It was quite a while before we turned to digital signal processing, and that finally unlocked the full potential of the RDAC for us.
The DSP in the bottom board uses carefully designed FIR filters and upsampling to give the R-2R converter a smoother digital input, which then gives us a smoother analog output. It also let us relax the reconstruction filters, which gave us back the details we were missing. Now it was coming together, and the final touches were getting the right set of inputs, fine tuning the DSP, and getting everything production ready and at the price point we needed. And now, with all of that done, the RDAC is ready for your enjoyment.
Thanks for reading and reply below with your questions!