I am new to electronics and have been told in class to always add precision potentiometers to compensate for input offset voltages in opamps for signal conditioning. My concern is that this offset voltage is inherent to the opamp IC and is different on each instantiation of the circuit. How do professional designs made for massive production deal with the problem of offset voltage? Do they adjust potentiometers after manufacture for each board to check that they are adjusted? This sounds impractical. Do they use other tricks to get around the problem or perhaps simply use better opamps or tailor the design to avoid depending on individual IC parameters variance?
-
4\$\begingroup\$ Design so that the errors don't matter. There are very few precision analog computations any more : where DC accuracy is required, usually you convert to digital ASAP and do everything there. \$\endgroup\$– user16324Sep 14, 2020 at 20:43
-
\$\begingroup\$ I measured it as one phase of each measurement cycle and used the average in order to subtract. In the case of analyzing an exponential decay, you must know the offset or the decay curve fitting result is unusable. It's possible to develop more sophisticated algorithms that also develop the offset as an output, rather than an input. But (1) there are other consequences of these algorithms that made the result unusual and (2) offset values in a fitting algorithm are very noisy with respect to their average and shouldn't be taken with equal weight so the math gets very complex, fast. \$\endgroup\$– jonkSep 14, 2020 at 21:37
-
\$\begingroup\$ Calibrate the offsets out... and minimize them in the design \$\endgroup\$– Voltage Spike ♦Sep 15, 2020 at 21:15
6 Answers
Depending upon your application there may be low cost opamps that have negligible offset voltages.
A common approach in modern CMOS opamps is to use an arrangement called chopper-stabilization. This can give microvolt level offsets with no adjustment needed.
This technique uses an auxiliary amplifier with CMOS switches that alternately swap the polarity of the input voltage. The output of that amplifier is then averaged to negate the effects of any offset voltage. This output is not used directly but instead is used as an offset adjustment voltage for the main conventional amplifier so high-frequency performance is not affected. The static offset and any temperature-related drift can both be cancelled. Even low-frequency noise (often known as 1/f noise) can be cancelled.
This is an especially good match to CMOS technology with its ability to fabricate good extremely low offset switches but with relatively large offsets in linear stages. CMOS linear circuitry is normally at a disadvantage with respect to 1/f noise as well that can be reduced.
This is not without its drawbacks, in particular, it can result in noise at the switching frequency and have a long recovery time after saturation, so it is not usually suitable for circuits where this occurs.
In cases where Vos matters, gain also often matters.
You can use pot(s) and adjust the error out. This works well if you carefully calculate and limit the range of adjustment and (especially) if the adjustment can eliminate other errors at the same time. It does not deal with offset drift and poorly designed adjustment circuits may drift significantly themselves. On the plus side, users can easily adjust it for maintenance. On the minus side, users can equally easily misadjust it.
You can avoid the pot(s) and make adjustments digitally. This means throwing away some of the range of the ADC and front end. There is also the pesky problem of dealing with potential loss of nonvolatile data and it does not deal with drift of offset (or gain).
You can add switching to the front end to measure (say) ground and input alternately (or measure ground once in a while) and subtract the two. This adds a bit of complexity but requires no storage and works with inexpensive parts. This also deals with offset drift with temperature.
You can use a "zero drift" op-amp. That essentially implements the above on a chip in a relatively transparent manner. They are cheap enough for many applications these days, and it basically eliminates the Vos problem (and drift of offset) but if there is an offset from some other source they don't solve the problem. They may also add noise in the signal bandwidth and have relatively large transient currents coming out of the input terminals that can affect some applications.
At some levels of accuracy it may become uneconomical to purchase parts that are guaranteed good enough for total system accuracy - for example, a high gain amplifier, ultra-precise and ultra-stable resistors to set the gain, an ultra-accurate reference, ultra-stable divider and buffer for the reference, etc, so some kind of adjustment or calibration makes sense. We needed to pull most of the available tricks out of the bag to make a 1uK/sqrt(Hz) temperature control, for example.
I worked for a senior engineer who was developing a telemetry system for use on board various prototype Sikorsky Helicopters. The box was bright orange, to be grabbed and rescued (about $50,000 each) in case of crash.
There were about 1,000 channels; a few were high_level to record switch/relay settings.
The myriad analog channels had programmable_gain (from an EPROM UV erasable) from 1X to 1,000X. Given the accompanying 10_bit 10 microsecond ADC, with +-5 volt inputs, at highest gain a single Quanta was 10 microVolt and needed for strain gauge monitoring of structural stresses and vibrations. Notice the 10uV quanta provides about 1/4 degree C resolution of temperature from thermocouples.
The telemetry system was FACTORY CALIBRATED.
The analog multiplexors used expensive (early 1970s components) MOSFETS, in differential_sensing configuration to provide up to 140 dB common mode rejection of 400 Hertz (aircraft power frequency).
The input INA Instrumentation Ampllifier used
matched JFETs, with the Idd current being hand selected to be the ZERO TEMPERATURE COEFFICENT operating point for the diffpair
the JFET output currents went to Vishay matched resistors, for some gain, and then to a Harris HA2525 opamp that closed the precision_gain loop and drove some bootstrap points to improve the AC CMR.
Result was an instrumentation system, robust in the face of helicopter vibration (this company routinely provided boxes for launch into space, so design for vibration was not any bother), and robust over the military temperature range because the Helicopter Prototypes would be tested in extremes, and having dependable and ACCURATE telemetry was crucial to debugging any failures.
YES---- the senior technicians certifying each of the 3 production units were responsible for setting the 10_turn pots to the proper point.
=======================================
DKNguyen asks about determining the zero-temp-coefficient Idd for JFETs:
Yes. The purchasing department ordered some dual_JFET pairs from ADI. And the super-techs (people with 15/20/25 years progressive experience and demonstrated attention to detail ---- this company developed and sold equipment that was expected to function per spec for 20 years, after being in storage for 20 years and then launched, so highly responsible skip-no-steps staffers were very treasured) would characterize the dual JFET pairs per the modest Idd (3 -- 8 mA) of zero temperature drift.
Chances are the associated "repair manual" for the 3 boxes would include a DRIFT NULL adjustment. Once a year the customer test_people could plug in the CALIBRATE EPROM, and use a channel set to 1,000X to monitor the offset_drift from -55C to +125C. If not in spec, that 10_turn Bourns military_grade sealed-by-Oring pot could be adjusted to compensate for JFET aging.
When strain_gauge bridge readouts are crucial to detecting airframe flexing during flight tests, the telemetry system may be calibrated regularly, and the strain_gauge cables carefully positioned within the airframe to minimize the upsetting influence of 400Hz currents adjacent to the telemetry twisted pairs.
-
1\$\begingroup\$ How do you go about finding the zero temperature coefficient bias current for a bunch of JFETs? You have to sit with a bunch of them in an oven and find out? \$\endgroup\$– DKNguyenSep 15, 2020 at 19:00
-
2\$\begingroup\$ @DKNguyen way back when I built a system that tested 100 op-amps at a time for temperature drift, using a lab oven and a diy data acquisition system. We sorted them into 3 grades and used the drifty ones in less critical parts of the instruments. \$\endgroup\$ Sep 16, 2020 at 2:10
I don't think I have ever had to do a high volume design with offset issues. I have done a few odd boards that needed precision and I relied on high dollar precision parts.
If I did have to do a high precision, high volume design, I would try to either use an op-amp with guaranteed low offset voltage or alternatively, remove offset in software.
The idea is that the offset would be measured and stored in non-volatile memory during production in an automated process.
Using a trim pot would be a last resort.
-
3\$\begingroup\$ In one company I used to work for, production would skin you alive, if you were even to think about the word pot. \$\endgroup\$ Sep 14, 2020 at 20:54
-
\$\begingroup\$ Trim pots are very seldom needed or affordable due to the cost of labour required to tune them. Micros are cheap nowadays. If you don't have a micro on board, there are digital trim pots which can be trimmed by the factory test jig \$\endgroup\$ Sep 16, 2020 at 18:52
- You can use several amplification stages with the first one having a low offset voltage and the rest giving more amplification etc. This way you might end up paying less for the parts.
- Buy an expensive op amp that has several good qualities
- Try to calibrate the offset in SW :) and even have it adjust the offset automatically
- Use a trim pot if you really must, but usually the offset will drift so you'd be constantly turning the pot
The decision to add potentiometers to trim offset voltage is completely dependent on the application.
Sometimes a few mV of offset matter, for example in a preamplifier for a piezeo electric sensor. Often times it's so small it doesn't. It's all in the application.
When it does matter, it is usually better to pick a better op amp with lower offset voltage rather then add a POT. Pots can be a pain to set consistently in production.
To answer your question, no not every op amp needs a potentiometer to trim offset voltage. The big picture is that you have to understand offset voltage and think about whether it is a problem or not. It is a fundamental consideration when choosing a op amp.