Choosing an op-amp for your project
lcamtuf.substack.com
lcamtuf.substack.com
In particular, any low voltage current sense circuit is going to require very precise Vos. Let's say you have a 0.01 Ohm current-sense resistor on a 5-Amp or so circuit.
Your current-sense is now in the range of +/- 0.05Volts (!!!!). So a Vos of 0.005 would represent a 10% error. Likely too much for most applications.
In effect, CMRR has become hyper-sensitive to Vos in this particular use case, to the point that Vos is suddenly the most important statistic.
Fortunately, there are specially designed low-offset chopper or auto-zeroing OpAmps like MCP6V26 or whatever out there.
MCP6V26 has Vos of 2uV, or in relatable terms... 0.000002 Volts (!!!!). Meaning it is more than sufficient at reliably making this current-sense application. Indeed, you can drop down to 0.0001Ohm resistance and still have high accuracy (and power savings compared to the earlier assumption).
------
Alas, nothing is ever free in life. Chopper Amps have noise issues and other designs have very very low bandwidth (which is truly an important statistic for most circuits).
Choosing a chopper amp specifically is making a Vos tradeoff with Bandwidth. So only choose if you know what you are doing (aka, dealing with very low voltages and needing the precise zero).
On CMRR, in some mathematical treatments it's modeled as a change in offset voltage with respect to common mode, which indirectly effects output voltage of course so at the end of the day it's the same result. (See: https://www.google.com/books/edition/Design_With_Operational... highly recommended )
It's also odd that the 741 was dismissed, as it should be, but the TLV9301 was not recommended. This part is specifically called out on TI's 741 page as what to use instead in 2025. Not only does it perform better in basically every possible spec, it's also a drop in replacement for most, if not all, applications.
https://www.ti.com/product/LM741 https://www.ti.com/product/TLV9301
TLV9301 ($0.5) is also cheaper than a MCP6272 ($0.88)
A 10% error can be calibrated out if it is constant (in practice it probably varies a bit with temperature).
But for measuring thermocouples or strain gauges, for example, 50 mV (your example) can be 100-500% of the signal, which becomes impractical to calibrate-out (due to maximum output levels, etc).
For these applications, Vos is one of the first things to look at. Another one is the temperature coefficient on the gain. High frequency noise metrics such as CMRR and PSRR are sometimes important if you're looking at high frequency signals, but most of the time mechanical phenomena don't have much interesting content above a few 100s of Hz, and high frequency PS or CM noise can be removed by a simple high pass filter.
Unfortunately, Vos on cheaper general purpose OpAmps is the kind of thing that varies by... voltage. Ick.
> For these applications, Vos is one of the first things to look at. Another one is the temperature coefficient on the gain. High frequency noise metrics such as CMRR and PSRR are sometimes important if you're looking at high frequency signals, but most of the time mechanical phenomena don't have much interesting content above a few 100s of Hz, and high frequency PS or CM noise can be removed by a simple high pass filter.
No. CMRR is about DC in the applications I'm talking about. It's weird because CMRR is listed in decibels but it's absolutely a DC spec.
If you have a high side current sense circuit with common-mode voltages of 24V +/- 0.05V (ex: 24V power supply that dips to 23.95V at 5Amps), CMRR tells you how accurate you are here.
Your typical 60db (btw I need to kill the engineer who decided db measures DC noise/errors....) means that the 24V of common mode voltage (which is the 24V DC power supply in this case) leaks into your measurements.
Or in other words: 60db * 24V == 3 decades or 24mV of 'Noise' aka your +/-50mV signal/measurement got completely wiped out by your DC errors. Like 50% error bars on your signal now, good luck with that.
That's the real issue with OpAmps. There's surely an OpAmp out there that solves your problems. But it requires knowing the general tradeoffs and picking-and-choosing different parts for different purposes.
Secondly, the specs are not intuitive. 60db CMRR sounds like a high frequency issue but becomes DC in this case.
You could of course go full isolation (optoisolators) that allows you to shift voltages down to near zero (removing CMRR issues) but that's money and additional parts.
You could go low-side voltage sense but this doesn't work for all circuits (most circuits are fine with Vcc error, not Ground errors). So high side current sense is the most flexible and generic well engineered solution. So long as you choose the correct OpAmps.
-------
As far as when this could be useful: consider Maximum PowerPoint Tracking for solar. 0-24 V and 0-1 Amps. And a need to accurately measure Amps and Volts from this entire range. (A variable load + voltage converter like a switch-mode power supply + battery can search for the optimal Voltage/Current combination to maximize the Solar Panels power).
Yes the microcontroller will do the bulk of the math. But the initial multiplies and subtract is best handled by an OpAmp.
If you get the Franco book equation 5.27 (my edition is the 3rd) explains why they do that. Long story short: It's a convenient form when CMRR = dVcm/dVos due to the orders of magnitude involved.
Decibels are useful for any intensity scale that occurs in a context where something responds in a logarithmic way, regardless of whether the power phenomenon is steady or pulsating.
Common mode gain can cause or contribute to an output offset. But if the amplifier circuit is biased such that + and - are kept very close to zero, that should not be the case. We normally care about the common mode rejection because common mode noise makes the + and - inputs swing together, causing the amplifier to amplify noise, like RFI/EMI.
In your current sense example, although we have a 10% error, it is a systematic error (a displacement), not 10% uncertainty in the value. It is an inaccuracy, not an imprecision.
You're assuming that the amplifier just buffers the output of the current sense resistor, so that the +/- 0.05 V across the resistor is translated to the output of the amp via unity gain.
We could configure the amplifier to amplify the signal, so that it becomes +/- 5V (or whatever). Then the Vos of 0.0005 will be a 0.1% inaccuracy.
For mA range, of course not.
https://www.st.com/en/amplifiers-and-comparators/tsc1031.htm...
Hall sensors drift.
I'd think the main benefit of using lower-value current-sense resistors in this application would be that the resistor would heat up less, so its resistance would be more stable?
The issue is that any circuit with 1 to 5 amps of current is a serious amount of power, meaning power efficiency is likely one of the top priorities.
A 5-Amp circuit with a 0.01 Ohm sense resistor wastes 250mW on the resistor alone, likely more than the entirety of your microcontroller!! You can actually run an entire Linux capable microprocessor + Low-power DRAM off of that kind of power!!
Dropping down to 0.0001 Ohms uses 1/100th the power or 2.5mW. which is likely a more reasonable cost.
In the contexts I'm thinking of, I would think that, if your load is drawing 5 amps of current at 3.3 volts, which is 16.5 watts, an extra 0.25 watts in the 10mΩ current shunt is not likely to be a big problem. And if it's 5 amps at 48 volts or 240 volts, it's even less of a problem, relatively speaking. I guess you're thinking of different contexts, contexts where the power-measurement system is paid for from a different budget than the load, but I can't figure out what they are.
The more I post on this subject, the more I'm "backwardsly-targetting" a solar-powered MPPT circuit.
Maximum Power Point Tracking circuits improve your solar-panel's efficiency by changing the current (through the use of a buck-boost converter, changing the voltage-and-current downstream). Or maybe you have excess current sunk ionto a battery of some kind. Either way, you have some kind of configurable-load and can therefore maximize the solar panel's Voltage/Current curve characteristics to seemingly magic energy out of nothingness.
If it costs you 250mW to just *sense* the current and run the calculations, it becomes much harder to justify the small gains of any MPPT circuitry.
------
But yes, I'm changing the target application to suit my argument style. Apologies on that but I think you can forgive me on this!! The point of MPPT is to magic more energy out of nearly nothingness so efficiency is of great concern here!
I feel like a car's transmission or a bike derailleur may be a good analogy to explain it to people, though an MPPT tracker is a ratcheting CVT.
So a 60W Solar Panel might be 12V 5-Amps in the best case scenario (directly pointed at the sun during clear skies). If you draw 5.1-Amps, suddenly the voltage drops all the way to 0.7V or some other nearly useless level.
What is annoying about solar panels is that this point changes depending on temperature, shadows, and other conditions. As the sun sets and hits the solar panel at a shallower angle with dimmer afternoon light, it might drop to 12V 2-Amps, and of course at night it will drop to 12V 0.01Amps or less.
---------
A MPPT circuit "tracks" this point where you can obtain the maximum amps at the maximum voltage (or really: the maximum volts * amps), and changes the current draw of the effective circuit to "find" this point.
It requires some kind of power-sink (ie: hooking it up to the wall power and assuming wall-power can take infinite amounts of energy). Or more likely, a lead-acid battery or LiFePo4 charging circuit that can "store the extra energy dump".
What I meant was that I wasn't confident that I was remembering the open-circuit voltage of a silicon solar cell, and in fact I had it wrong—it's not 0.7 volts, but 0.5 to 0.6 volts. And I couldn't remember if the diode junction was forward-biased or reverse-biased in normal operation. It's reverse-biased—the photocurrent goes the opposite direction from normal diode current. Now that I think about it, maybe that's why shade on one cell knocks out a whole string.
One quibble, though: if you maintain 12 volts in the usual direction across a series of 10 monocrystalline silicon solar cells at night you are just going to lose a subthreshold leakage current through them, because you're forward-biasing the pn junctions in the cells, just not quite by enough to turn them on. They'll emit a little bit of infrared light, a feature used to analyze solar panel failures (so-called "EL testing"). Illuminance at night is at best a million times dimmer than sunlight https://en.wikipedia.org/wiki/Orders_of_magnitude_(illuminan... and that subthreshold dark current is not going to be a million times lower than your normal current. Even under indoor lighting, which is only about 2000 times dimmer than direct sunlight, monocrystalline silicon PV cells will consume power rather than producing it.
Because of the intermittency you're describing, I suspect that thermal energy storage of various kinds (sensible heat, phase change materials, or especially TCES) is going to be important for the wide adoption of solar power, because according to my notes lead-acid batteries store about 20kJ/US$ and LFP a bit less, while industrial calcium chloride costs about US$300 per tonne (US$272/tonne according to https://derctuo.github.io/notes/desiccant-climate-control.ht...) and can absorb about its own mass of water from the air, liberating the water's enthalpy of vaporization, providing TCES.
I believe the heat thus stored is 408kJ/kg (see linked notes) which works out to 1500kJ/US$ at that price, roughly 1% of the cost of the same energy storage capacity in a battery. And, depending on the desiccant, it's plausible that you could reduce that by another order of magnitude, or two orders of magnitude for industrial installations. You can probably get by with impure calcium chloride or as-mined carnallite, for example.
Sudden memories of https://en.wikipedia.org/wiki/Autodesk_Animator , which was commercial software with exactly those limits (due to inheriting them from VGA). Despite the limited resolution it had a spectacular array of features.
Technically you probably could do it externally in most cases but it would require a bunch of extra stuff, and be a pain, so usually it's best to use the stuff built into the amplifier itself.
https://www.electronicsurplus.com/national-semiconductor-cor...
But the potential advantage that building a chopper (or auto-zero) out of ordinary opamps would be that you don't need to source, order, and await specialized chips. A long-discontinued bolt-on auto-zero for a regular opamp has almost all the disadvantages of just buying an auto-zero opamp.
I was recently looking at opamps as alternatives to the LM324 and found some interesting-sounding parts, in particular for a poor man's SMU application (precision, low current, and voltage requirements, but not much bandwidth). Haven't tried any of them yet. Comments would be welcome.
- LM324B: TI's improved LM324, with half the input offset voltage and otherwise improved ratings, and just as cheap, but still bipolar.
- OPA4197 and family: three dollars but it's a quad RRIO 36V 10MHz opamp in a 14-SOIC with ±15nA input bias current, ±100μV max input offset voltage, and 120dB min CMRR. The datasheet makes it sound amazing for the price. The OPA177 seems like it would be better but pricier.
- OP4177ARUZ: a 16-dollar quad 36V 1.3MHz opamp with ±2nA input bias current, 75μV max input offset voltage (at ±15V power supply), and min 120dB CMRR
Then I decided I'd screwed up my design sketch by requiring one of the opamps to sink significant current very close to the negative rail, which is something even "rail-to-rail" opamps can't do; I was planning to use millivolts from ground to represent measured nanoamps. If you want to look at a simulation with idealized opamps, it's at https://tinyurl.com/2aomvpn5, but don't take it as exemplary in any sense; it's a novice design with novice mistakes (and I would be grateful in the unlikely case that someone took the trouble to point some of them out). I think I need to redesign the circuit as a bipolarity-supply circuit or something, or use a differential output for the current measurement, or rethink it entirely.
(Distributors like DigiKey and Mouser have somewhat adequate search functions; I usually have to go to manufacturers' web sites like https://www.ti.com/amplifier-circuit/op-amps/general-purpose... to be able to filter by all important parameters. I'm mentioning TI because they have a large selection and a good search; even when you do not end up selecting on of theirs, you see what is possible.)
___
If you need only a small negative supply and have nothing else, the LM7705 charge pump can generate −0.23 V. (This is designed to fit into the typically allowed 5.5 V range of a nominal 5 V opamp.)
I do not know what a "significant current" is for you, but there are opamps with strong outputs. (When comparing opamps, you usually have to estimate the drive strength from the short-circuit current.)
What I meant by "requiring one of the opamps to sink significant current very close to the negative rail" is that, if you look at the schematic, the differential-to-single-ended op-amp that measures the voltage across the current-sense shunt resistor is using 10kΩ resistors in its feedback path, and the inverting input to that feedback network might be close to the positive voltage rail, say 12V, while the single-ended output is ideally millivolts from ground. So you have 12 volts across 20kΩ, which works out to 600μA, which has to be sunk into that op-amp's output.
600μA doesn't sound like a lot, and it certainly isn't going to strain the drive strength of any op-amp IC, but in this context we're hoping for millivolt precision down near the negative rail. The OPA4197 datasheet https://www.ti.com/lit/ds/symlink/opa4197.pdf figure 14, "Output Voltage Swing from Negative Power Supply vs Output Current (Maximum Supply)", shows what you might call a gently nonlinear output impedance roughly in the 40–80Ω range depending on temperature (2–4V at 50mA), which means 0.6mA of output current works out to tens of millivolts (24–48mV using those nominal impedances). Worse, even under no-load conditions, it's rated to swing only down to as much as 25mV from the negative rail (§6.7, "Electrical Characteristics: VS = ±4 V to ±18 V (VS = 8 V to 36 V) (continued)", p. 8, "Vₒ: Voltage output swing from rail, Negative rail").
In retrospect, it seems obvious that the op-amp's output isn't going to be able to reach beyond the input rails (unless it integrates a charge pump like the LM7705 internally) and is going to have trouble getting too close to them when it's sinking any current (for the negative rail, or sourcing for the positive). Because where is that current being sunk to? You need some voltage drop to get the electrons and holes to move in the desired direction through the silicon. A small negative supply might be the right solution. Or a differential output, which would be easy.
If you know more specific information about your circuit or it's application, the. You can specialize. But general purpose OpAmps are jack of all trades with specific known weaknesses to avoid.
In most cases, you calculate the error bars and none of the errors matter, so sticking with a cheap general purpose amp is best engineering.
This 100%. If you need a comparator get a comparator not an op amp. Current measuring? There are specialized chips for that as well, etc.
I live in a third-world country where importing chips from abroad is expensive, unreliable, slow, and sometimes dangerous. There are circuits I cannot build because I cannot get the very specialized parts they need. Obviously a linear power supply that can measure how much current it's supplying is not such a circuit, unless you have very stringent precision requirements.
It would be to my benefit to figure out a relatively small set of parts I can buy, ahead of time, in bulk, to cover a wide range of possible circuits. Better still if they're so popular that local distributors have them in stock. An analog comparator probably needs to be in that set. A chip specialized for current measuring probably does not.
If you're designing a product for mass production that needs to be competitive in the market, you can't do it that way. Super-specialized parts will always have better performance, and usually better price/performance than overpowered general-purpose parts. (Also, you need to live in Shenzhen.) But hobbyists have other priorities.
For someone working on the systems interactions, the failure point is making sure all the bits of the project are working together.
For someone working on optics, the failure point is finding cost-effective optics -- if you can't do that, then the project isn't going to go forward.
For the hobbyist? The failure point is <i>getting the project done</i>. Every other concern takes a back seat to this!
From my vantage point as someone living in America, however, I'm probably in a similar boat to you, because of my inexperience in electronics. If I want to take a deep dive into the subject, I'd be much better off getting a lot of generic cheap parts I can accidentally burn through, but would give me 90% of what I need, and I can worry about whether or not I need something highly specialized later, as my projects -- and my knowledge and skills! -- mature.
By the way, I'm living in America, too.
fwiw I searched and found "A newer version of this product is available Same functionality with different pin-out to the compared device LM27761"
Get an OpAmp specifically designed for current sense applications.
OpAmps for current sense applications have high accuracy near 0 Volts and Vos measured in single digit microvolts.
Oh, you WILL lose bandwidth with these designs. So make sure you are allowed to be much much slower.
Being slower is not a big problem, but needing specialized parts might be, due to supply-chain issues.
(More detail in https://news.ycombinator.com/item?id=42627042.)
Recommend Ch5 for Precision Design and Ch8 (or 9? Can’t remember) on noise.
However, it's not enough to say that it costs ~8.7x as much in local tokens because that doesn't reflect how much it costs as a percentage of the other components around it, many of which have a more profound impact on the outcome. This wouldn't stand as a good argument except that in many devices, you have a ceiling for how much your total BOM can cost before you've priced yourself out of competition.
In the end, a 8.7x cost bump is a lot to swallow for a feature that most consumers are physically incapable of distinguishing between. Every time I've raised the possibility of using the fancy new chips, vastly more experienced engineers than me have come out of the woodwork to tell me that in almost all scenarios, the tradeoff between price and quality isn't worth it.
Of course, if budget is not an issue, use the OPA2323. It really does sound great. Or more accurately, the degree to which it destroys good sound is as low as we can currently achieve.
(This comment originally stated a 12x factor, but I was terrible at math.)
Looking at Mouser Canada, it looks like the cheapest 071 is the TL071CDR, at $0.138/each (Canadian) in quantities of 5k. The OPA2323IDDFR is $0.49 in the same quantity.
> In the end, a 12x cost bump is a lot to swallow for a feature that most consumers are physically incapable of distinguishing between.
I think that the performance of an op-amp should very rarely have user-visible effects. The more interesting question is whether the more expensive chip can make for a simpler design elsewhere. For example, can a rail-to-rail amplifier save the extra cost of needing charge pumps and split-rail design elsewhere?
Also, not all domains should be cost-optimized. Hobbyist or prototyping work might best benefit from using a more expensive but more capable amplifier as a first choice, saving on the number of components that might need to be stocked in the home lab.
FWIW, here is what I was going off of, price wise:
https://www.digikey.ca/en/products/detail/texas-instruments/...
https://www.digikey.ca/en/products/detail/texas-instruments/...
In other words, the cheapest TL071 variant and the cheapest OPA2323 variant on Digikey Canada, in a quantity of 1 (ie wildly expensive). $0.31 vs $2.70 means that I shouldn't attempt math before coffee; 8.7x is still a big bump, although I acknowledge it's not the 12x I disinformationed earlier, with apologies to anyone reading.
So the BOM cost may not be a significant consideration for hobbyists, especially when weighed against things like familiarity or being able to keep a stock of a smaller number of less specialized components, as Majromax points out.
I'm surprised to see you say that a US$1 opamp is as good as we can currently achieve. Presumably there are Analog Devices chips that are better than the OPA2323 even for audio? Even if you can't hear the difference, you ought to be able to measure it.
Also valuable to look for is ESD protection. Many op amps have good 2kV static shock protection. Really sucks to have cascading failures due to touching your circuit and blowing an op amp.
For some uses it’s also good to checkout input current bias.
Don’t stick with 1% for analog circuits as 0.1% precision resistors are cheap and come with low temp drift as well.
where Dave discusses the "Jelly bean" op amps
None of the listed opamps would be an alternative to where an LM324 or TL071 are typically used - low cost, 20V-30V supply, infinite/guaranteed availability. They are 5V only and cost 10x as much.
At that point you are engineering with a completely different set of tradeoffs than would be expected for a modern ground-up design effort.
I replaced some of them with good opas and burrbrown opamps and for me it made , even if not big, audible changes to the better. But this could be my imagination.
whatever i prefere some opamps in the range from 2-4 bucks. even if its only for the feeling…didnt measure it.
> One thing is obvious — the 5532 is still one of the great opamp bargains of all time.
"Stuff not to worry about:
Despite what content-farmed articles imply, most of the other parameters in the spec can be usually glanced over. For example, the exact value of open-loop gain (AOL) is almost never of real consequence; the same goes for input offset voltage (VOS) — even in high-precision instruments, the absolute value is less important than drift over time. In any case, the parameters are usually only eyeballed in most specs, so if you’re building sensitive instrumentation, you will still need to calibrate the readings using a known reference."
I rarely think about fGBP and AOL as separate ideas, the author could make the point more directly by saying they are related and that one is easier to select from a single decimal value. AOL or fGBP is a primary design consideration.
https://e2e.ti.com/support/audio-group/audio/f/audio-forum/4... (this list only contains the op-amps, not stuff like the also-discontinued LME49810)
https://news.ycombinator.com/item?id=39944839
April 2023
Unity gain stable
Sure, the other characteristics are important, but a whole bunch of circuits that beginners are likely to use rely on opamps being unity gain stable. If they're not unity gain stable, the circuit will do very weird things, and a beginner won't know why.
Of course, debugging issues like that are how you eventually become an expert.