Just some meta commentary for what follows here: I make mistakes every now and then and I always welcome corrections, questions, and friendly banter from subscribers. That said, in this instance, after a spirited exchange in the comments section below, I do stand by the article :-)
The TL;DR is that a subscriber expressed concern that the non-inverting summing amplifier presented in the circuit may not work correctly. I think that in the general case, it should. If you're looking for independent references to the circuit, see the relevant section in https://www.ti.com/lit/an/snla140d/snla140d.pdf
Now, there is an important distinction between the non-inverting architecture presented in the article and the "textbook" inverting summing amplifier that you will, for example, find in "The Art of Electronics". The distinction is that in the non-inverting design, the current is necessarily flowing in through some of the input legs and flowing out through others. This can cause problems if one of the sources can only source currents, but not sink any (e.g., if it includes a series diode). The inverting variant, on the other hand, keeps the junction point near 0 V. Because of this, for positive input voltages, all the currents are flowing in the same direction and the drive constraint is eased.
In other words: I agree that in some situations, the inverting architecture may be preferred. That said, I believe the circuit presented in the article is reasonable. I did not use the inverting layout because it works as advertised only in a dual-supply circuit. The inverting summing amplifier *can* be converted to to single-supply operation, but that necessarily involves putting the summing point at a higher voltage and breaks the symmetry with the difference amplifier.
Noah Fact is (more) correct. It's preferable to do the summing at the inverting input of an inverting amplifier, where - because of feedback - each input resistor is referred to the voltage at the non-inverting input (usually ground or virtual ground). In this way the current through each input resistor is simply equal to V(in)/R(in) and independent of other input voltages. Of course, an additional inverter stage is needed for the adders output polarity to be correct.
The non-inverting adder can work if source impedances are negligible in comparison to the input resistance.
Noah asserted that the circuit doesn't work and then kept digging his heels in with a theory that there's zero current, and then that it will average voltages instead of summing them.
You are correct that the inverting circuit can be preferable. Specifically, the big distinction is that the inverting circuit works with asymmetric drive capability on the inputs (i.e., if the source can't sink currents as well as it can source them). But it's a deeper level of detail that's just not covered here. The inverting version is generally a dual-supply circuit, and if you make it in a single-supply version, it ends up having a flavor of the same problem.
> The non-inverting adder can work if source impedances are negligible in comparison to the input resistance.
I think that's a non-sequitur. That's a general assumption for a lot of circuits. You can use the same critique for R-C filters, etc. It must be true for the inverting circuit too.
>Noah asserted that the circuit doesn't work and then kept digging his heels in with a theory that there's zero current, and then that it will average voltages instead of summing them.
Uh.... what? Now I'm confused. You're right, it's an averager with a fixed division factor (the number of parallel resistors at the + input) which is then undone by an equally-fixed stage gain. It's a perfectly valid thing to do, as long as the application permits.
I just don't agree that it should be called a "non-inverting summer," as if it were somehow a fundamental building block like the more traditional inverting summing amplifier with independent currents adding at a virtual ground point. With the inverting topology, the input impedance is strictly defined by the input resistors. With a passive filter, the input impedance is also solely a property of the components used. With whatever this thing is, the input impedance (and consequently its actual transfer function) is subject to whatever else is connected to it.
We're all "correct" here. Smoke a peace pipe, already. :-P
Yeah, sorry, I probably reacted too forcefully when people started showing up and taking sides, and summed up your take somewhat uncharitably.
Maybe what I didn't really explain is the "why" behind the approach I've taken. I think that most people, in most circumstances, are trying to build single-supply op-amp circuits, but a lot of the textbooks rely on dual-supply topologies that either don't work at all with single supply, or get very clunky to use because of non-overlapping input and output ranges. So, whenever I write about op-amps, I try to provide a single-supply solution.
This is also why, for the final and most interesting circuit - the integrator - I went through the effort of analyzing the non-inverting variant, which is almost never taught anywhere. I learned about it here:
...but I don't think that page does a particularly good job of explaining it to novices (for one, it assumes that the reader will immediately recognize a "negative resistor" constructed with an op-amp and will know what it means).
Heh, to the extent it triggered me, it was because I was flashing back to the first time I tried doing this exact same thing with an opamp, about 500 years ago. ("WTF, why isn't this working? The opamp inputs are supposed to be the same except for their sign, right?")
There are no pills available at any pharmacy in the Matrix that can compare with the noodle-baking you get when you come face-to-face with the virtual-ground concept. Like losing your virginity or joining a new religion, it can only happen once, and the person who gets up from the bench is not the same as the one who sat down. This particular hack, on the other hand.... not exactly the same thing.
The impedance at the + input is ~infinity, so the resistors will have either no effect at all (if the sources being summed have low impedances) or you'll end up with a Thevenin tangle of interaction between source impedances and series resistors that might result in a sort-of sum, but won't work meaningfully in the general case.
TL,DR: no summing junction exists at the + input. Summing requires negative feedback to establish a virtual ground.
Imagine putting a couple of 9V batteries at the inputs. The voltage at the + input is... 9 volts.
I don't think you're right. I added an additional explanation in the article, but assume that the junction is at Vjct and then write Ohm's law equations for each resistor branch (where the input voltages are VA, VB, VC):
I1 = (VA - Vjct) / R
I2 = (VB - Vjct) / R
I3 = (VB - Vjct) / R
You're correct that there's no current flowing into the op-amp, but there are currents flowing through the resistors. From Kirchoff's current law, these need to balance out: I1 + I2 + I3 = 0 A. Combining the equations:
(VA - Vjct) / R + (VB - Vjct) / R + (VC - Vjct) / R = 0 A
Multiplying both sides by R:
(VA - Vjct) + (VB - Vjct) + (VC - Vjct) = 0 V
VA + VB + VC - 3*Vjct = 0 V
3*Vjct = VA + VB + VC
Vjct = (VA + VB + VC) / 3
Of course, this assumes well-behaved voltage sources with impedance much lower than R, but that's a given for the inverting architecture too.
You can use a circuit like that for something like an audio mixer, for instance, in the algebraic sense. +1 volt at one resistor and -1 volt at another will yield zero volts at the + input, so in that sense it's a summer. But if you put +1 volt at both resistors, the result will be +1 volt.
The only currents flowing through the resistors will be flowing between the sources, not into the opamp.
The resistor section is not the summing circuit, it's an averager. The average of 1 + 1 is 1, so the result you're describing is correct. You then have a 2x gain section with the op-amp.
I see what you mean, but it's still kind of a screwy circuit. You basically never want one input to influence or depend on the others. Again, only if the source impedances are very low will it ultimately behave as a general-purpose voltage summer.
I could see it making sense with a quad opamp package, where three unity-gain buffers are used to isolate the inputs.
Just some meta commentary for what follows here: I make mistakes every now and then and I always welcome corrections, questions, and friendly banter from subscribers. That said, in this instance, after a spirited exchange in the comments section below, I do stand by the article :-)
The TL;DR is that a subscriber expressed concern that the non-inverting summing amplifier presented in the circuit may not work correctly. I think that in the general case, it should. If you're looking for independent references to the circuit, see the relevant section in https://www.ti.com/lit/an/snla140d/snla140d.pdf
Now, there is an important distinction between the non-inverting architecture presented in the article and the "textbook" inverting summing amplifier that you will, for example, find in "The Art of Electronics". The distinction is that in the non-inverting design, the current is necessarily flowing in through some of the input legs and flowing out through others. This can cause problems if one of the sources can only source currents, but not sink any (e.g., if it includes a series diode). The inverting variant, on the other hand, keeps the junction point near 0 V. Because of this, for positive input voltages, all the currents are flowing in the same direction and the drive constraint is eased.
In other words: I agree that in some situations, the inverting architecture may be preferred. That said, I believe the circuit presented in the article is reasonable. I did not use the inverting layout because it works as advertised only in a dual-supply circuit. The inverting summing amplifier *can* be converted to to single-supply operation, but that necessarily involves putting the summing point at a higher voltage and breaks the symmetry with the difference amplifier.
Noah Fact is (more) correct. It's preferable to do the summing at the inverting input of an inverting amplifier, where - because of feedback - each input resistor is referred to the voltage at the non-inverting input (usually ground or virtual ground). In this way the current through each input resistor is simply equal to V(in)/R(in) and independent of other input voltages. Of course, an additional inverter stage is needed for the adders output polarity to be correct.
The non-inverting adder can work if source impedances are negligible in comparison to the input resistance.
Noah asserted that the circuit doesn't work and then kept digging his heels in with a theory that there's zero current, and then that it will average voltages instead of summing them.
You are correct that the inverting circuit can be preferable. Specifically, the big distinction is that the inverting circuit works with asymmetric drive capability on the inputs (i.e., if the source can't sink currents as well as it can source them). But it's a deeper level of detail that's just not covered here. The inverting version is generally a dual-supply circuit, and if you make it in a single-supply version, it ends up having a flavor of the same problem.
> The non-inverting adder can work if source impedances are negligible in comparison to the input resistance.
I think that's a non-sequitur. That's a general assumption for a lot of circuits. You can use the same critique for R-C filters, etc. It must be true for the inverting circuit too.
>Noah asserted that the circuit doesn't work and then kept digging his heels in with a theory that there's zero current, and then that it will average voltages instead of summing them.
Uh.... what? Now I'm confused. You're right, it's an averager with a fixed division factor (the number of parallel resistors at the + input) which is then undone by an equally-fixed stage gain. It's a perfectly valid thing to do, as long as the application permits.
I just don't agree that it should be called a "non-inverting summer," as if it were somehow a fundamental building block like the more traditional inverting summing amplifier with independent currents adding at a virtual ground point. With the inverting topology, the input impedance is strictly defined by the input resistors. With a passive filter, the input impedance is also solely a property of the components used. With whatever this thing is, the input impedance (and consequently its actual transfer function) is subject to whatever else is connected to it.
We're all "correct" here. Smoke a peace pipe, already. :-P
Yeah, sorry, I probably reacted too forcefully when people started showing up and taking sides, and summed up your take somewhat uncharitably.
Maybe what I didn't really explain is the "why" behind the approach I've taken. I think that most people, in most circumstances, are trying to build single-supply op-amp circuits, but a lot of the textbooks rely on dual-supply topologies that either don't work at all with single supply, or get very clunky to use because of non-overlapping input and output ranges. So, whenever I write about op-amps, I try to provide a single-supply solution.
This is also why, for the final and most interesting circuit - the integrator - I went through the effort of analyzing the non-inverting variant, which is almost never taught anywhere. I learned about it here:
https://www.analog.com/en/resources/technical-articles/consider-the-deboo-singlesupply-integrator.html
...but I don't think that page does a particularly good job of explaining it to novices (for one, it assumes that the reader will immediately recognize a "negative resistor" constructed with an op-amp and will know what it means).
This just reminds me how little I ever really understood of this area, and the fact that I forgot everything. Interesting, nevertheless.
Heh, to the extent it triggered me, it was because I was flashing back to the first time I tried doing this exact same thing with an opamp, about 500 years ago. ("WTF, why isn't this working? The opamp inputs are supposed to be the same except for their sign, right?")
There are no pills available at any pharmacy in the Matrix that can compare with the noodle-baking you get when you come face-to-face with the virtual-ground concept. Like losing your virginity or joining a new religion, it can only happen once, and the person who gets up from the bench is not the same as the one who sat down. This particular hack, on the other hand.... not exactly the same thing.
Your adder might not work the way you expect. Those three input resistors normally drive the virtual ground at the - input.
Hmm? AFAIR, non-inverting summing amplifier is a fairly standard architecture, here are some places that discuss it:
https://www.electronics-tutorials.ws/opamp/opamp_4.html#:~:text=terminal%20as%20shown.-,Non%2Dinverting%20Summing%20Amplifier,-So%20what%E2%80%99s%20the
https://analogcircuitdesign.com/summing-opamp/#op-amp-adder-formula:~:text=)-,Non%2Dinverting%20summing%20amplifier,%C2%B6,-Non%2Dinverting%20summing
https://developerhelp.microchip.com/xwiki/bin/view/products/amplifiers-linear/operational-amplifier-ics/introduction/non-inverting-summing/
https://www.petervis.com/Education/non-inverting-summing-amplifier-formula/non-inverting-summing-amplifiers.html
The impedance at the + input is ~infinity, so the resistors will have either no effect at all (if the sources being summed have low impedances) or you'll end up with a Thevenin tangle of interaction between source impedances and series resistors that might result in a sort-of sum, but won't work meaningfully in the general case.
TL,DR: no summing junction exists at the + input. Summing requires negative feedback to establish a virtual ground.
Imagine putting a couple of 9V batteries at the inputs. The voltage at the + input is... 9 volts.
I don't think you're right. I added an additional explanation in the article, but assume that the junction is at Vjct and then write Ohm's law equations for each resistor branch (where the input voltages are VA, VB, VC):
I1 = (VA - Vjct) / R
I2 = (VB - Vjct) / R
I3 = (VB - Vjct) / R
You're correct that there's no current flowing into the op-amp, but there are currents flowing through the resistors. From Kirchoff's current law, these need to balance out: I1 + I2 + I3 = 0 A. Combining the equations:
(VA - Vjct) / R + (VB - Vjct) / R + (VC - Vjct) / R = 0 A
Multiplying both sides by R:
(VA - Vjct) + (VB - Vjct) + (VC - Vjct) = 0 V
VA + VB + VC - 3*Vjct = 0 V
3*Vjct = VA + VB + VC
Vjct = (VA + VB + VC) / 3
Of course, this assumes well-behaved voltage sources with impedance much lower than R, but that's a given for the inverting architecture too.
You can use a circuit like that for something like an audio mixer, for instance, in the algebraic sense. +1 volt at one resistor and -1 volt at another will yield zero volts at the + input, so in that sense it's a summer. But if you put +1 volt at both resistors, the result will be +1 volt.
The only currents flowing through the resistors will be flowing between the sources, not into the opamp.
The resistor section is not the summing circuit, it's an averager. The average of 1 + 1 is 1, so the result you're describing is correct. You then have a 2x gain section with the op-amp.
I see what you mean, but it's still kind of a screwy circuit. You basically never want one input to influence or depend on the others. Again, only if the source impedances are very low will it ultimately behave as a general-purpose voltage summer.
I could see it making sense with a quad opamp package, where three unity-gain buffers are used to isolate the inputs.