Well, I’m back from unipolar output testing. (It was a bit of a slog, so bipolar testing will happen later and probably output-by-output as I go to use each. I’ll post back up if that uncovers anything particularly interesting.)
I created NerdSEQ projects that helped streamline things, and have a detailed spreadsheet with the results, but I’ll spare huge amounts of detail for now and just try to summarize my findings, with some key bits bolded:
My setup, by virtue of other modules involved, isn’t affected by this to the point of creating major issues, but hopefully the testing and results thereof can nonetheless help result in some improvements (or at least some caveats) for ourselves and others.
I created NerdSEQ projects that helped streamline things, and have a detailed spreadsheet with the results, but I’ll spare huge amounts of detail for now and just try to summarize my findings, with some key bits bolded:
- Just to get it out of the way, it is very rare that I make an order-of-magnitude calculation or interpretation error, and I’m still feeling rather embarrassed. I won’t bury the lede: My workflow will NOT be adversely affected by this, at least not beyond being a bit more careful with output jack selection for some duties, additional calibration and tweaking, etc.
- I have tested 72 CV outputs from five XOR modules, those being two NerdSEQs and three CV16 expanders. This seems a representative number but they’re all admittedly in my cases and connected to my power supplies and my other modules, so a good data set but just a single data set.
- Each output was calibrated and then tested at eleven whole-volt points from 0V through 10V inclusive, for a total of 792 measurements.
- Before calibration, the system was allowed to warm for at least an hour. Surprising or not, this can cause enough drift to affect the measurements, whether that drift is on the NerdSEQ module side, the Sinfonion side, or a combination I do not know. This drift was at times as large as some of the error in clak’s measurements, and of the lesser error in mine collected after becoming warm and stable. Considered differently: many outputs showing inaccuracy could zero or double their error simply by running the system cooler or warmer than usual.
- Yes, I’m using the Sinfonion’s built-in volt meter for measurements. This has the downside of not being a professionally-calibrated high-precision multimeter but has the upside of being the actual main destination for the voltages whose accuracy usually matters most, and the Sinfonion’s ADCs are generally quite well regarded.
- All outputs were calibrated to within one millivolt at their 0V and 10V ends.
- In terms of accuracy, the NerdSEQ module outputs were average to good in comparison to the CV16 expanders.
- The CV16 expanders each faired differently. One CV16 was the best of all sets of outputs, including the two main modules. The other two CV16s were the worst two sets of outputs amongst all five modules.
- Interestingly, the two CV16 expanders showing the worst issues did so in a way exhibiting physical patterns. By this I mean that the outputs on the “rows” with output 12 and/or 13, and the row with output 16, especially output 16 itself, were the worst on both of those two modules. I will likely attempt to move them within my cases, or to at least move ribbon cables around underneath them.
- 35 of 72 outputs had voltages between 5 and 10 millivolts out of line at one or more measured points.
- 7 of 72 outputs had voltages between 10 and approx 14 millivolts out of line at one or more measured points.
- 7 of 72 outputs had voltages 15 or 16 millivolts out of line at one or more measured points. This is approximately one fifth of a semitone, on 10% of all outputs.
- The vast majority of the outputs noted above do not experience something as simple as a curve and instead can vary up and down, volt by volt.
- The worst of the outputs (thankfully) experience their greatest error somewhere around 5V, and these at least somewhat resemble a curve. This will lead to a suggestion below.
- A great many of the outputs experience their greatest contributing error near 0V and/or 10V. This will lead to a secondary suggestion below.
- My workflow is not going to be largely impacted by this, as I have quantization downstream of the inputs whose voltage accuracy truly matters, though I will be making efforts to select and calibrate outputs as best as I can for certain uses (eg. to use the best outputs for those summed pitch offsets I mentioned earlier, to use the worst outputs for things I can calibrate to produce good results in a much narrowed required range. etc.)
- However, if this is audible to a given person then with these patterns of error up and down they’re not going to be able to compensate for it downstream short of some form of multi-point voltage remapping, suggesting that more detailed calibration within the NerdSEQ could still be of benefit.
- If my outputs are any indicator, though, even just the addition of an adjustment to help accurately hit 5V would bring most error down into the range where it could almost as easily be caused by temperature changes.
- And if my outputs are any indicator, adding adjustments for 1V and 9V target voltages would clear up most of what error remains.
- Please consider adding an adjustment to help accurately hit 5V. This could significantly clean up my worst 10-20% of outputs and turn the other outputs with measured issues into rounding errors. It would also be greatly beneficial to bipolar uses as 0V could be targeted much more accurately.
- Perhaps consider also adding a couple further adjustment points. Points a volt or two away from 0V and 10V would likely do nicely, at least for my outputs.
- Alternatively, consider adding a more involved calibration to help really zero in on things and/or to help users experiencing greater error than I. (Though, having now tested in much more detail, this does seem like overkill in relation to the simpler suggestions above.)
My setup, by virtue of other modules involved, isn’t affected by this to the point of creating major issues, but hopefully the testing and results thereof can nonetheless help result in some improvements (or at least some caveats) for ourselves and others.
