[sdiy] Linear response VCOs?

[brianw]

The Prophet 5, Rev 1 and Rev 2, use a 7-bit DAC made from hand-picked resistors. There is a note in the Service Manual that you should *not* replace these resistors because of the challenge of matching a new one to the network. 1 LSB is calibrated to 1/12 V (0.0833 V) for easy use in 1V/8va scaling. CV ranges from 0 V to 10.583 V (127/12), but the Prophet 5 only uses the lower 6 bits for pitch, limiting the range to 5.333 V maximum and thus 5 octaves. All CV were 7-bit, but the pitch combined coarse and fine with the scale of the DAC changed so that there were 64 steps in the coarse range plus another 128 steps in the fine range. This wasn't quite as accurate as a 13-bit DAC, but still quite accurate for the time.

The Prophet 5 Rev 3 simply used a 16-bit DAC, but maintained the firmware design with 7 bits per CV, so the pitch did not enjoy a full 16-bit precision. The 13-bit pitch values still have 16-bit accuracy, though, just not 65536 steps of precision.

One thing to note, Mark, is that a 6-bit DAC has an LSB that's 1.56% of the total range, so 1% resistors would be quite awful. Then there's the fact that a 1% error in the MSB could throw the whole binary scale off enough that the values are not monotonic (i.e. an increase in the code could actually cause a decrease in voltage!). A 7-bit DAC has the LSB at 0.78% so you definitely need better than 1% precision. These manufacturers were not making a custom resistor array so much as hand-selecting individual resistors that were matched well across the whole group.

Today, not only are 1% resistors more readily available than they were in the seventies, but you can even get 0.1% tolerance resistors at a reasonable. Still, that doesn't even get you to a full 9-bit DAC. This illustrates how impressive DAC chip technology is. One of the fasted DAC chips I've designed with can run at a sample rate of 125 MHz (yeah, MHz, not kHz) based on current switching rather than voltage, but it stops at 14-bit precision because the smallest current is only 0.0061% of the largest, and it's difficult to be precise enough at such a large scale factor. Larger DAC precision requires a different technique than binary-weighted digits. Fortunately, there are many ways to implement a DAC.

Brian


On Mar 18, 2026, at 4:34 AM, Tom Wiltshire wrote:
> Roland had form for this. SH-101 uses a simple DAC built from a few resistors too.
> 
> Like Roman said, it doesn't really make sense nowadays when DACs are cheap, but it was worth it then.
> 
> Tom
> 
> On 18 Mar 2026, at 11:31, mark verbos wrote:
>> Like a TR-909.
>> But, surely it is cheaper to use 1% resistors rather than a custom resistor array made.
>> 
>> Mark
>> 
>> On Mar 17, 2026, at 18:44, David Manley wrote:
>>> It's interesting to see how PAiA's John Simonton solved some these issues in the 1970's by having a custom laser trimmed resistor network built for their 6-bit "Equally Tempered DAC" to be used with linear VCOs.  See the bottom of the schematic on page 18, the resistor values are on the last page.
>>> 
>>> https://paia.com/wp-content/uploads/2024/05/8780pgs.pdf
>>> 
>>> As is typical for PAiA a very low cost solution: build your own DAC with a few components.
>>> 
>>> -Dave