There are three different aspects to be discussed in DA conversion.
First is the Conversion process.
Second are the processes before conversion takes place (DSP), third is the timing of it all or in other terms - clocks. There are multiple books on each of the above subjects, so don’t expect an in-depth analysis but more a guide for what to look for in a DAC.
Multibit vs. Single bit (bitstream) converters. There were some 20-bit converter chips (burr brown) in production and they have legendary status among users. Even today there are new designs coming out built around the first Philips chips (40 year old tech) claiming audiophile performance and unbeatable sound. We have taken the road of multibit converters and have developed a solution around 1ppm precision industrial chips to meet our requirements. About as good as it gets today.
The hype about high sampling rates and the argument for non-oversampling DACs. Well in order to avoid this argument we decided to support both.
We support high sampling rates up to 384kHz accommodating all existing formats, yet we are not bats – it is the intermodulation and filter characteristics that are audible, not ultrasonic signals.
Again to overcome all arguments and to accommodate user preferences we support allmultiple options for upsampling, re-clocking and digital filtering. Any signal processing on a 24-bit word of data will generate extra data. Almost all other DACs and processors will truncate the data back to 24-bit. We let it flow at 32-bits keeping all information, as insignificant as it might be. Otherwise those errors accumulate and after 3-4 processes (re clocking, upsampling, dithering, energy filters) they become significant and are clearly audible.
Apart from data fed to the converter, the other major influence is the time it takes for the conversion to happen. You can google “jitter” and will be swamped with lots of meaningless pages. In essence you are interested in the cleanliness of your clock and short-term stability, also known as phase noise. You will see companies offering rubidium, atomic, GPS locked, tourbillon or hourglass clocks out there but a properly designed and manufactured quartz oscillator is pretty much unbeatable for audio. Having said that please note that less than 1% of the ones we have seen were up to standard.
The most interesting property for us is low phase noise at the very low frequencies and high s/n of the clock signal with no distortion.
All conventional DACs require buffers and filters. Those are always implemented with opamps. (even discrete opamps) They isolate the DAC from the output and sum the outputs of multiple DAC chips, convert I to V and other tasks, but it does not matter how good they are, they always severely limit the sonic performance of the DACs. Whatever you have done up to that point performs at the quality level of the opamp used. (isn’t this sad?). Fortunately on multibit DACs with low impedance resistor ladder you can take the output directly from the conversion resistors. This way we can sum the multiple DACs with a transformer with multiple primaries and by taking the output from the secondary of the transformer we have a complete solution.
Isolation of the output from the DAC, stopping any high frequency content passing trough the band limited transformer avoiding IM distortion and summing out of phase, removes all common mode noise and artifacts. You can call it „the magic of coils". Why nobody does that and why they do only a partial job? Transformers don’t measure that well. They have limited bandwidth and rising distortion at high levels and low frequencies. All those shortcomings are true, but would seem to be substantially better performing in dynamic conditions than ANY op amp. The effect is jawdropping not subtle.