Multi-bit, but not R2R

[opus101]

Its not correct to say that 'the 154x family was DEM' since really its not a family at all. TDA1543 is a bipolar IC technology DAC without any fancy DEM tricks. TDA1545 has, as you correctly point out, 'continuous calibration'. TDA1547 is an altogether different beast - a 'Bitstream' converter back-end. There is also TDA1549 which looks to be a cut-down variant of TDA1305, one of the first multibit Bitstream devices, also using CC.

[opus101]

Without wanting to get excessively nerdy, there are more differences between 1387 and 1545 than just the digital interface. The output compliance range is a fairly significant one, the facility for increasing the Imax is another. I tried increasing Imax on 1387 and quickly ended up with a dead chip so that's not just a cosmetic difference.

 

Back to your original post, I agree some more attention in the mainstream for Philips' DAC designs would be welcome - why should R2R capture all the multibit limelight?

[opus101]

5 hours ago, tmtomh said:

As for older multi-bit designs like the 1980s Philips 14-bit and 16-bit DACs, I'd always assumed all of those were R2R and had no idea that such a DAC could be anything else.

 

So... could anyone explain more about how a non-R2R but still multi-bit DAC is actually implemented?

 

I'll have a stab at this, hopefully I'm not sacrificing too much accuracy for brevity here. The term 'R2R' refers to a ladder of resistors where each is either value 'R' or '2R' - each stage (two resistors, R and 2R) divides the voltage of the earlier stage by 2 to generate the binary bit weights for the DAC. Such a ladder requires very good matching of the Rs in order to get the necessary performance. Doing this on an IC is tricky as process variations mean matching better than a fraction of a percent can't be achieved directly, it needs trimming after the wafer has been produced. Laser trimming is expensive so Philips' with their consumer mindset wanted a way around such a requirement. Their engineers (Rudy van de Plassche in particular) came up with 'DEM' - a way to self-calibrate a DAC in-circuit, circumventing the need for any trimming.

 

The TDA1541 uses 'emitter scaling' to create part of the DAC's structure - a way of using transistors to divide currents successively by 2 (rather than using resistors). This accounts for 10 of the 16 bit total, the remaining 6 bits use the DEM technique. In later designs which used CMOS (for much lower power) the DEM technique was implemented as 'Continuous Calibration' and relied on charge storage on small capacitors within the IC itself. The earlier bipolar chips needed external caps as the values required are much higher.

 

There are other ways to create multibit DACs - for example a resistor string. It is a different structure from R2R and there have been commercial DACs using such (Metrum Octave is one I believe, though the manufacturer never confirmed the chip identity in their design).

[opus101]

I'm still trying to figure out what the market is for this chip - the fact that it has an SPI interface tells me its more industrial than audiophile. 'Lab and field instrumentation' it says but what about those applications needs R2R rather than (significantly cheaper) S-D ? Could it be performance inside a control loop, given there's an example shown on the first page? 'Professional audio amp (rack mount)' looks like an outlier.. The maximum supply and reference of 40V suggests some seriously impressive SNRs might be possible, the output noise figure of 7nV/rtHz  implies 1uV in the audio band, signal may go up to 14VRMS,

[opus101]

There's some technical background on the device from one of the guys at TI over here : https://www.diyaudio.com/forums/vendor-s-bazaar/349809-dac11001-20-bit-2r-precision-dac.html

 

Seeing as the maximum update rate is around 500kHz (faster if the deglitcher is bypassed but that comes with its own issues) there's room for some oversampling/noise shaping to improve on the quantization noise limit imposed by the part being only 20bits.