Hi Guys,

Really long post, but there's a lot of info. Sorry.

A multibit converter is the simplest to explain, and the simplest to draw, but pretty much the hardest to make. Each bit in binary is worth twice the one below it. A multibit converter is often made of an R-2R ladder, so every bit gets through a resistance proportional to its value. This gives you a current or voltage proportional to the whole lot. The problem is that you have to make a buffer and resistor accurate to 0.5 * the top bit / 16,777,216 for each of the 24 bits to get 0.5 LSB accuracy. Transistors are easy in silicon, resistors are hard. Even if you achieved it, you'd have to then achieve it across temperature.

Any inaccuracy in the resistors and buffers leads to that bit having the wrong value, which can lead to a non-monotonic DAC (a bigger number should always lead to a bigger output, but for some values it's smaller). Really bad.

A Delta-Sigma uses a multibit DAC, but run at a higher speed to get the effect of a slower but more precise DAC. I used a Burr-Brown with a 4 bit multibit DAC ran to emulate 24 bits. Imagine a 2 bit DAC running this sequence:

3, 3, 2, 3

If you analogue filter it to slow it down you end up with a value somewhere just above 2.75, which the original DAC couldn't have achieved.

Take that to extremes and you could have a 1 bit DAC running many times faster. This is bitstream or DSD.

I'd argue that there are really two categories - full multibit, then Delta-Sigma (which includes 1 bit/bitstream/dsd as a special case) as well as more standard multibit/DS implementations.

Cheers,
Mark.

861v8, DSP5200, DSP5200HC, DSP5000mk2, UHD722, MF10, Prime, Prime PSU