- Blog entry posted in 'Test equipment', January 18, 2013.
Designing this thing is a bit like writing a program - you get something basic that works, then you start improving it, and it makes it not work. Or maybe it's just me.
The successes so far (still untested because I can't build anything yet since moving house - so they're sort of virtual successes at the moment) are:
Solving the problem of matching a scaled input of other than a base of 2.5, with the rms converter's requirement for maximum peak input of 1v, it's maximum output of 1vdc and the adc's input being ±2.5v. I did this by setting the fully differential amp at the adc's input to gain of 2.5, having an amp connected to the input divider with a gain of 2.5, and letting the rms converter be able to be switched in between them.
The two amps have a combined gain of 6.25, and the input is always divided by at least 10. So the full scale input is based on 4 now - a 4 volt input is divided to 400mV, multiplied by 6.25, giving 2.5 volts. A 400 volt input still puts 40 volts across the relay for the first range, which is higher than the rating of the relays I have, but at least it's trivial to deal with.
It also scales for the rms conveter, at least as far as peak voltage is concerned. 0.4v peak entering the first amp gives 1v peak to the rms converter, the output of that is then increased 2.5v for the adc. I eventually decided it was meaningless to try to scale the ranges to rms values when I learned about crest factors - that is the ratio between the peak value and rms value of a waveform, which is different for different waveforms.
I found a precision rectifier circuit to detect peak voltage. The output of this can be used to trigger the range-up function, so there's no messy trimming to assumptions about rms values. 0.4 volts peak at the input amp, and the range goes up. The output from the rectifier goes to a comparator with a 1v reference on the other leg, this then goes to the micro.
I worked out a way to do four wire resistance measurement, and learnt a little bit about this type of measurement in general. The problem I found was that with standard ohms converter circuits, the unknown resistance forms the return to ground for the constant current. So whilst it's easy to put a sense wire to the high end of the device, there isn't a high impedance way to make a connection to the ground. The solution I found was to use an instrumentation amp - as far as it's concerned, ground is just another wire.
Couple of problems with this, however. The in-amp I picked was based on what I could get on eBay, which was an INA128. Seems good, however the problem is that the input bias current, even if its a few nano-amps, is way too high. With a low value of current setting resistor this isn't a problem, and with a reduced number of digits it's not a problem, however with 1 and 10 megohms setting the current, and with 6 digits, it skews the results unacceptably. And that's just in a simulation, with perfect resistors and ideal conditions!
So I've been bashing my head trying to find a way to offset the error - otherwise I need to give the in-amp a buffer, which seems silly, or find an in-amp with a higher impedance input.
Been trying to find the best way to create a pcb for this. Toner transfer would be ideal, but isn't an option - apart from the whole business of having etchant to deal with. The Sharpie method isn't very tidy, but I'm damned if I'm getting rub on transfers. So I thought I'd try my hand at milling by hand, since I have mini-drill with some milling bits. It worked well to cut traces on some Vero Board for the voltage reference I built a while ago. Will see how it goes anyway.
