as a thermocouple amplifier. After construction, I used two thermocouples, both screwed to the same place of a heated surface with temperatures going form 25 degree C to 350 degree C. One of the thermocouple is connected directly to differential analog input of NI9219 directly configuring in labview as a thermocouple (the device has special signal conditioning and CJC compensation included), the second one is then connected to my signal conditioning circuit (U1A & U1:D) with a gain of 240 and the output is connected again to another analog channel of NI9219 configured as a voltage input. I used ITS-90 coefficients to convert the measured voltage to temperature (Thot - Tcjc). Then ambient temperature is separately monitored and added to the calculated temperature value. When I monitored the temperature from both channels using NI9219 as a a reference read out device, the response of my signal conditioning circuit has some errors when the temperature goes above 200 degree C. The worst case error is 10 degree C when the temperature reaches around 300 degree C and also when the temperature falls back. When I practically pumped in the voltage to the input of my amplifier circuit, I am surprised to see that the gain is not really that linear. The ideal case gain is 240, however,at 3.9mV input the amplifier gives the output of 931mV (gain is 238.719 in this case), at
12.3mV the output is 2.909V (gain becomes 236), and at 13.5mV the output is 3.254V (gain is 241.03 then), I suspect this contributes to the error at the higher temperature.
I understand the affect of CMRR, Offset, Offset Drift, and ADC errors, but now it seems that I need to consider the Op-amp circuit non linearity in my calculations.
"Myauk" wrote in message news: snipped-for-privacy@pg2g2000pbb.googlegroups.com...
If you look at the differential voltage error, I think you'll find it's not too big. Crudely speaking, anything within 1mV of equality (+in to -in on the op-amp's inputs) is "close enough". The actual band depends on the op-amp's offset, gain, drift, noise and load condition (voltage and current output).
Cheap op-amps are in the single digit mV offset range, historically ~2mV for bipolar, ~5mV for JFET and ~10mV for MOSFET. Among the jellybeans you'll see today, the bipolar LM358 has 1mV, the JFET LF411 has 1mV, which is pretty good; the TLV2371, a rail-to-rail CMOS amp, is also 1mV typical, though only 5mV guaranteed.
Precision amps all have finely tuned input stages, so as to minimize input offset and bias, and high open-loop gain, so that errors due to distortion and output condition are minimized.
What's happening at the output matters, because no amp has a perfect voltage source output. If the output is loaded, it requires slightly more differential input voltage (a few microvolts, perhaps) to compensate for the change. Output impedance varies by design, so some amps have more difference than others. An example is in the LT1806 datasheet:
formatting link
See page 13, the "Open-Loop Gain" plots. The axes are input vs. output voltage for different loads, so these are simply the DC transfer function of the amp -- you can see it's fairly linear for most of the range, but the input is off by a few tenths of a milivolt towards the ends, particularly at higher load currents.
Now, your observations are in the first two columns. If one were to assume the output voltage is, in fact, what the op-amp "intends" it to be, based on the set gain (which, I suppose, is a true statement -- if it's in a stable condition, it's going to be in equilibrium with a gain of exactly the feedback ratio), then one can extrapolate the input offset error that you're seeing.
Now, the appnote specifies an MCP619, which has quite low offset voltage and quite high gain:
formatting link
Input offset is guaranteed within this range (+/-150uV, p.2), so there isn't really much you can say about it.
A transfer curve is provided here as well (p.7), showing the voltage gain is quite high. The output goes roughly between supply rails over a span of just 5-10uV, certainly not accounting for the error you see. Transconductance is not defined in any way (i.e., different load conditions), so it's hard to say what role load resistance will play.
It's not obvious how input offset varies with temperature, only the typical value is given: +/-2.5uV/C offset, which, if it's a constant slope at the "typical worst case", blows the input bias spec over only a 60 C range.
LF noise is given as only 2.2uVpp (0.1-10Hz, p.3). The lower frequency components (fractional Hz, down to DC) will manifest as slowly varying offsets. This measurement is performed at constant temperature -- weak air currents (convection or otherwise) can easily produce more than this. After all, it might only take a 1 C change to push it 2.5uV one way or the other.
As for your data, three points isn't much to go on. To really dig deep into the behavior of this op-amp, a microvolt scale is really needed to put a few extra digits on Vin and Verr. What's more, multiple measurements are required at constant operating conditions, because noise and drift will vary all the time. Thermal fluctuations can be a big driver, especially in microvolt circuits, and it might help to put your circuit inside a cardboard box, or under a styrofoam cup, or something like that. Let it sit in place, powered, for several hours to wait for everything to reach thermal equilibrium. Then take measurements.
Finally, as for the circuit, I'd recommend splitting up the gain stage into two amps, which will keep (output referred) noise down -- instead of 240 gain in a single stage, try 10 followed by 24 (or any other combination with the same product). Consider adding a negative voltage supply rail, to improve the common mode range on the inputs and outputs -- this will help them function better near 0V, keeping distortion down. Considering how important low drift and offset is in this kind of application, you might also consider switching to a chopper-stabilized amplifier. OPA335 is an example:
formatting link
Guaranteed 5uV max offset, hard to beat that. Despite the chopper stabilization, noise is still comparable -- 1.4uVpp over a slightly wider bandwidth (0.01-10Hz, p.3), and much worse at higher frequencies (obviously, you want to filter out the ~10kHz trash the chopper stabilization produces). It does, however, eliminate the 1/f drift characteristic (p.6). Although noise is rather high (50-60nV/rtHz in the LF band, and 13-17nV/rtHz in the HF band except for switching harmonics at ~100nV/rtHz), it has the enviable characteristic of being flat (not 1/f), so it's easily filtered out (can be made arbitrarily low when only DC response is required). Other ideas show up in the datasheet -- a pull-down resistor to a negative supply can be used to bias the output stage, enough to pull it slightly below ground when necessary, ensuring linear operation.
Tim
--
Deep Friar: a very philosophical monk.
Website: http://webpages.charter.net/dawill/tmoranwms
Unlikely, but must always check. any very high frequency oscillations [outside your range of observation] that make the bias 'walk'? Any external pickup sensitivities from local AM radio station? that willmake the bias 'walk'?
Possible to simulate using free PC Tools, like LTspice? makes for an easy calculation.
ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here.
All logos and trade names are the property of their respective owners.