I'm experimentally studying the self-heating of a special type of resistors. Typically I'm using a DT800 to record differential DC voltages as a "V" channel between, for example, 1+ and 1- terminals, or 3+ and 3-, et cetera, in a simple resistor network. In my experiment I gradually reduce the power supply voltage driving the entire network, over time. Unfortunately my approach is fairly vulnerable to the measured voltages approaching something a little different from zero, as the true physical voltages approach zero. A certain figure of merit in my system varies as the cube of the error in measuring true zero voltages. I get less than a millivolt error when the true voltage is zero, but if practical I'd like better. The data sheet says there are 12 auto-scaling ranges starting at 10 mV full scale, so I think the least significant of the 16 bits should represent a third of a microvolt and I'm seeing a hundred times that or so as my error. Is that to be expected?
Therefore I started some side experiments with zero voltages on the DT800. I am posting because I'd like more insight into what I'm seeing.
When I daisy-chain a bus bar wire across 1+, 1-, 2+, 2-, 3+, 3-, 4+, and 4-, and also tie it to Ch or Sr, I believe each of these 4 differential inputs is physically getting a true zero voltage, so ideally 1V, 2V, 3V and 4V should read 0 mV. I do averaging within records like this:
RS01T ' Subsampling at 100 Hz for later averaging
RA1S ' Schedule "A" runs at 1 Hz
and I run this overnight or over the weekend to get lots of points.
Oddly, the 1V channel reads about 0.05 to 0.1 mV, the 2V channel reads about 0.01 mV, the 3V channel reads about 0.003 mV, and the 4V channel reads about 0.0008 mV. That is, each channel number is several times more accurate than the channel number below it. It isn't just a fluke. I'm doing this with four different DT800 loggers (SN082868, SN082893, SN092777, and SN094769) and they all do this. Can anybody give me insight as to what is going on here?
In these results, the range of values I get excludes 0. That is, a measurement of 0 is many many standard deviations away from the mean.
I notice that sometimes I see a pattern that looks very much like the HVAC cycling in my lab space, including the cutback feature that sets less stringent temperatures for overnight or weekend periods. From the manual notes around page 188 about the autozero correction I understand that changing temperature is a possible source of error.
Also, I wondered if input impedance limited accuracy, so I set one DT800 up with its inputs shorted with a 100k resistor, a 1M, a 10M, a 100K, 1M, 10M and also a wire short. I tied the - side of all these inputs together to Ch. There is no tendency for the higher impedances to correspond to higher (or different) errors.
I experimented a little with the VNC channel type and learned that the autozero correction makes things way better. All my above results are with the V channel type which includes autozero correction.
Perhaps the reading accuracy is best at the ends of the scale, which in the DT800 are at plus and minus voltages, never at 0. I might be able to bias my physical experiment to put the readings at the low end of the measurement scale, so I'm measuring there and not in the middle of the scale. Is there an easy way to do that?
I've seen a bunch of other things but haven't spotted a pattern or made sense of it. Any help you might give me in understanding this would be very appreciated.
Thank you!
P.S. I guess I should ask, is there a market for antique DT800 loggers?
I'm experimentally studying the self-heating of a special type of resistors. Typically I'm using a DT800 to record differential DC voltages as a "V" channel between, for example, 1+ and 1- terminals, or 3+ and 3-, et cetera, in a simple resistor network. In my experiment I gradually reduce the power supply voltage driving the entire network, over time. Unfortunately my approach is fairly vulnerable to the measured voltages approaching something a little different from zero, as the true physical voltages approach zero. A certain figure of merit in my system varies as the cube of the error in measuring true zero voltages. I get less than a millivolt error when the true voltage is zero, but if practical I'd like better. The data sheet says there are 12 auto-scaling ranges starting at 10 mV full scale, so I think the least significant of the 16 bits should represent a third of a microvolt and I'm seeing a hundred times that or so as my error. Is that to be expected?
Therefore I started some side experiments with zero voltages on the DT800. I am posting because I'd like more insight into what I'm seeing.
When I daisy-chain a bus bar wire across 1+, 1-, 2+, 2-, 3+, 3-, 4+, and 4-, and also tie it to Ch or Sr, I believe each of these 4 differential inputs is physically getting a true zero voltage, so ideally 1V, 2V, 3V and 4V should read 0 mV. I do averaging within records like this:
RS01T ' Subsampling at 100 Hz for later averaging
RA1S ' Schedule "A" runs at 1 Hz
and I run this overnight or over the weekend to get lots of points.
Oddly, the 1V channel reads about 0.05 to 0.1 mV, the 2V channel reads about 0.01 mV, the 3V channel reads about 0.003 mV, and the 4V channel reads about 0.0008 mV. That is, each channel number is several times more accurate than the channel number below it. It isn't just a fluke. I'm doing this with four different DT800 loggers (SN082868, SN082893, SN092777, and SN094769) and they all do this. Can anybody give me insight as to what is going on here?
In these results, the range of values I get excludes 0. That is, a measurement of 0 is many many standard deviations away from the mean.
I notice that sometimes I see a pattern that looks very much like the HVAC cycling in my lab space, including the cutback feature that sets less stringent temperatures for overnight or weekend periods. From the manual notes around page 188 about the autozero correction I understand that changing temperature is a possible source of error.
Also, I wondered if input impedance limited accuracy, so I set one DT800 up with its inputs shorted with a 100k resistor, a 1M, a 10M, a 100K, 1M, 10M and also a wire short. I tied the - side of all these inputs together to Ch. There is no tendency for the higher impedances to correspond to higher (or different) errors.
I experimented a little with the VNC channel type and learned that the autozero correction makes things way better. All my above results are with the V channel type which includes autozero correction.
Perhaps the reading accuracy is best at the ends of the scale, which in the DT800 are at plus and minus voltages, never at 0. I might be able to bias my physical experiment to put the readings at the low end of the measurement scale, so I'm measuring there and not in the middle of the scale. Is there an easy way to do that?
I've seen a bunch of other things but haven't spotted a pattern or made sense of it. Any help you might give me in understanding this would be very appreciated.
Thank you!
P.S. I guess I should ask, is there a market for antique DT800 loggers?
