Author 
microchip DACs: MCP4728 vs MCP4922 
Dimitree 
in the past I used the MCP4922 and it worked fine as 1V/oct CV generator. Now I have a project going on where I would prefer to use an I2C DAC instead of SPI like MCP4922, so I think I will use MCP4728:
http://www.microchip.com/wwwproducts/en/en541737
can anyone tell me if the linearity/precision of this I2C DAC is as good as the MCP4922? I'm still not good at reading DAC datasheets.
I need it for a narrow range, just 4 octaves, so 1V to 5V CV or 0V to 4V 

Synthiq 
The MCP4922 and MCP4728 have very similar performance according to the datasheets. The integral nonlinearity is typically 2LSB for both but worst case 12LSB for MCP4922 and 13LSB for MCP4728. What that means is that if you use it for a 1V/oct signal and calibrate the DAC at the endpoints, the error is typically not more than 2.3 cents but can be off by 14 cents for the MCP4922 and 15 cents for the MCP4728. This difference probably doesn't matter. The settling time when the code changes is 4.5us for the MCP4922 but 6us for the MCP4728, but again the difference shouldn't matter if it is just used for CV signals.
The MCP4728 has an internal reference voltage which the MCP4922 doesn't so it will most likely be more stable than the MCP4922 using Vdd as reference at least.
The main drawback with the MCP4728, besides the slower update rate, as I see it is the package with 0.5mm pitch pins. 

TheSlowGrowth 
15 cents is pretty significant, I'd say. It's probably not going to be as much in real life, but I'd go for something with more resolution. See MI Yarns. 

Dimitree 
thank you for the suggestions!
actually the package is fine for me (I like smd) and the update rate is not a problem, since the CV is only coming from the keyboard scanner and there is no modulations, so it's static
Synthiq wrote:  the error is typically not more than 2.3 cents but can be off by 14 cents for the MCP4922 and 15 cents for the MCP4728. 
what's the math to determine this? 

guest 
so INL is the measure of the difference between the ideal output and the actual output, usually specified LSBs. so, if you have a DAC that is scaled to produce 010V for a full audio range VCO, and its 12b, thats 10V/(2^12) = 2.44mV per LSB. if the error is +/13LSB, thats +/31.7mV. if you are using 1V/octave, thats 83mV per halfstep, and 0.83mV per cent. so +/31.7mV is +/38cents. if you want, you can calibrate these errors out with a lookup table and calibration routine. 

Synthiq 
Dimitree wrote:  what's the math to determine this? 
I guess guest already answered your question in general terms, but in your specific case you want to cover only 4 octaves, or 48 semitones, which is 4800 cents. Dividing that with 4096 steps for a 12bit DAC and we get 1.17 cent per lsb step ideally. Multiplying the 12 or 13 lsb INL figures by 1.17 cent should hopefully give 14 and 15 cent INL errors in this case. 

Powered by phpBB © phpBB Group
