@tubekiddo Graphs look good; are the 16-bit DACs and ADCs a material limitation on resolution?
2.0A / 2^16 = 0.03mA resolution, say around 0.05mA accuracy.
5.0V -> 0.076mV resolution, around 0.15mV accuracy.
2.0A / 2^16 = 0.03mA resolution, say around 0.05mA accuracy.
5.0V -> 0.076mV resolution, around 0.15mV accuracy.
The quality of your results are mainly impacted by the amount of noise introduced into your overall setup.
I get best results by using batteries for the +-24V supply. The resolution is defined by the 16bit DAC/ADC chips. The table below is with a linear power supply, you get a little better results with batteries, but they’re expensive.
- the quality of your PSU
- the way you contact the DUT
- and the cleanliness of you lab environment (emitting sources, static and electromagnetic)
I get best results by using batteries for the +-24V supply. The resolution is defined by the 16bit DAC/ADC chips. The table below is with a linear power supply, you get a little better results with batteries, but they’re expensive.
Last edited:
I don't understand what you mean by "quality", and I don't get how DAC/ADC resolution affects accuracy.
As I wrote that, I thought I'd probably get picked up on it...
I was thinking about how precisely Dr Borgmann's curve tracer can measure voltages and currents, and whether it matters.
16-bit ADCs allocate measurements into 2^16 buckets, so the minimum difference between analogue measurements in adjacent buckets is almost 0, and at the maximum, twice the size of the bucket. Hence my rough calculation of 0.15mV at 5V, or 0.075mV at 2.5V
Along with that comes noise and bias from the circuit and connections. Presumably - I haven't tried tubekiddo's software - the readings can be calibrated to reduce bias.
The 2.5V voltage reference LM4040DBZ-2.5 (voltage tolerances from A=0.1% to D=1%) has 'Average temperature coefficient of reverse breakdown voltage' ±20ppm/°C at 10mA and 25°C; ±150ppm/°C at –40°C to 85°C. Interpolate and 45°C might be around ±63ppm/°C at 45°C.
63ppm x 25°C x 2.5V = 3.9mV
So temperature variations in use could introduce 3.9mV variation/error outside the calibration?
Other voltage references:
If so, is there any benefit in improving the resolution of the curve tracer?
I was thinking about how precisely Dr Borgmann's curve tracer can measure voltages and currents, and whether it matters.
16-bit ADCs allocate measurements into 2^16 buckets, so the minimum difference between analogue measurements in adjacent buckets is almost 0, and at the maximum, twice the size of the bucket. Hence my rough calculation of 0.15mV at 5V, or 0.075mV at 2.5V
Along with that comes noise and bias from the circuit and connections. Presumably - I haven't tried tubekiddo's software - the readings can be calibrated to reduce bias.
The 2.5V voltage reference LM4040DBZ-2.5 (voltage tolerances from A=0.1% to D=1%) has 'Average temperature coefficient of reverse breakdown voltage' ±20ppm/°C at 10mA and 25°C; ±150ppm/°C at –40°C to 85°C. Interpolate and 45°C might be around ±63ppm/°C at 45°C.
63ppm x 25°C x 2.5V = 3.9mV
So temperature variations in use could introduce 3.9mV variation/error outside the calibration?
Other voltage references:
- LM4050 (eg LM4050CIM3X-2.5/NOPB) reduces the thermal effect to 20ppm/°C at 25°C, say 30ppm/°C at 45°C=+25°C: 30ppm/°C x +25°C = 0.075% x 2.5V = 1.9mV
- LM4030A-2.5 reduces the thermal effect to as little as 10ppm/°C at –40°C to 85°C, say 3.3ppm/°C at 45°C, or 0.2mV
If so, is there any benefit in improving the resolution of the curve tracer?
Last edited:
Ah, so you were looking at precision, not accuracy. Makes more sense now.
My experience with real-world precision in curve tracing comes from pypsucurvetrace, which uses (possibly cheap) programmable bench PSUs for curve tracing. I did some repeated measurements of the same part over the course of a few weeks / months. The differences between these replicate measurements were tiny, certainly much smaller than differences between extremely well matched parts.
I am pretty sure a proper curve-tracer board would be even better.
Last edited:
Yes.
Looking at @tubekiddo's numbers, to get 90% confidence in a ±10% range means a standard deviation of ±10%/1.65 = 6% of the reading. 95% confidence interval is 2sds so 5% for standard deviation.
A quick calculation gives SD/value:
PYActual
7.3%
10.5%
7.2%
6.3%
4.6%
UGv
23.6%
8.0%
6.4%
3.1%
It looks like it gives fairly reliable results sometimes.
Given care over external sources of noise, what else can lower the variance?
And does it matter - how precisely do we need to characterise the curves?
Looking at @tubekiddo's numbers, to get 90% confidence in a ±10% range means a standard deviation of ±10%/1.65 = 6% of the reading. 95% confidence interval is 2sds so 5% for standard deviation.
A quick calculation gives SD/value:
PYActual
7.3%
10.5%
7.2%
6.3%
4.6%
UGv
23.6%
8.0%
6.4%
3.1%
It looks like it gives fairly reliable results sometimes.
Given care over external sources of noise, what else can lower the variance?
And does it matter - how precisely do we need to characterise the curves?
Thank you! Great info!
IMO to improve overall performance/precision
- use the best voltage reference available (to minimize thermal effect)
- use batteries for the +-24V supply
- change PCB layout to place DUT adapter directly on the PCB as close as possible to the input OAmp
- Calibrate every time you prepare for measurement
- Improve the software?
Not sure how important a higher resolution would be. It might be of interest for measuring absolute values but usually not for matching components, IMO…
Last edited:
Great thanks!
PCB needs a little work to incorporate C1 and C2 in later additions, which can be smd components. And add a couple of blank pads for the better voltage references.
I'd also think about laying out to use a Faraday cage to control noise. Bit more expensive with a separate ground plane.
So revise with the DUT board on the PCB?
PCB needs a little work to incorporate C1 and C2 in later additions, which can be smd components. And add a couple of blank pads for the better voltage references.
I'd also think about laying out to use a Faraday cage to control noise. Bit more expensive with a separate ground plane.
So revise with the DUT board on the PCB?
I ran out of time...
Great thanks!
The PCB also needs a little work to incorporate C1 and C2 in the later revisions, which can be smd components. And a couple of blank pads for the better voltage references.
I'd think about laying out to use a Faraday cage to control noise. Bit more expensive with a separate ground plane.
Dr Borgmann:
What are your thoughts on connecting DUTs?
Great thanks!
The PCB also needs a little work to incorporate C1 and C2 in the later revisions, which can be smd components. And a couple of blank pads for the better voltage references.
I'd think about laying out to use a Faraday cage to control noise. Bit more expensive with a separate ground plane.
Dr Borgmann:
If the DUT is to be heated to operating temperatures, the DUT board will need to be remote with shielded cable attached to the board close to the input OpAmp U9 LTC1050. The current DUT connections look close to the input OpAmp.
What are your thoughts on connecting DUTs?
Well my current DUT adapter was already a big improvement in terms of noise reduction and convenience. It’s not perfect, rather a compromise until we get a PCB where it sits directly on the board. Running wires is not optimal IMO. Also there’s likely improvement potential in the software itself. I am at my capacity with what I provided…
Maybe a DUT adapter on the board and parallel sockets for a remote adapter to allow the use of a heater block and temperature control? Shielded BNC cables work pretty well for sensitive LCR meters.
Actually I was more wondering about the mounting hardware for different footprint DUTs, such as SMD.
Are there any other enhancements to add to the wish-list?
Actually I was more wondering about the mounting hardware for different footprint DUTs, such as SMD.
Are there any other enhancements to add to the wish-list?