Op-amp made from Discrete Components

RE: Op-amp made from Discrete Components

HaroonQazi — Fri, 29 Nov 2024 11:27:13 GMT

How to get inverting input from this circuit ? I connected function generator +ve to inverting input and com port to ground but the output is not inverting sine wave but its non inverting

RE: Op-amp made from Discrete Components

jc2048 — Sat, 20 Jun 2020 21:43:42 GMT

I mentioned above trying a mirror for the load. I've now tried it on the real circuit with this result

[keep in mind that I don't really know what I'm doing here and some of it is a bit random].

Here's the circuit I have on the board at the moment

[View:/resized-image/__size/620x352/__key/commentfiles/f7d226abd59f475c9d224a79e3f0ec07-021c4451-857c-43c6-8399-8e48ff079aee/6622.contentimage_5F00_207111.jpg:620:352]

and here are the two extra transistors installed on the board

[View:/resized-image/__size/600x450/__key/commentfiles/f7d226abd59f475c9d224a79e3f0ec07-021c4451-857c-43c6-8399-8e48ff079aee/4578.contentimage_5F00_207112.jpg:600:450]

The small step is now like this

[View:/resized-image/__size/480x234/__key/commentfiles/f7d226abd59f475c9d224a79e3f0ec07-021c4451-857c-43c6-8399-8e48ff079aee/8345.contentimage_5F00_207113.png:480:234]

It hasn't completely fixed the input voltage offset, but you can see it's much closer than the 50mV

that I had before. But it's more on edge. When I first power it up, the ringing extends all the way

along the top or bottom of the waveform, before it finally settles to what you see here. This is with

the Miller cap increased to 22pF [it's no longer stable with 10pF]. Compensating the amplifier is now

more difficult. Increasing the Miller capacitor dramatically works for stability, but results in the

slew becoming ridiculous [the capacitor is an integrator, so if you push it too high, large square waves

become triangular]. The 470R and 22n were where I was experimenting with trying to tailor the open loop

frequency response, but probably aren't the way to do it.

The slew now looks like this

[View:/resized-image/__size/480x234/__key/commentfiles/f7d226abd59f475c9d224a79e3f0ec07-021c4451-857c-43c6-8399-8e48ff079aee/2577.contentimage_5F00_207114.png:480:234]

the symmetry that the mirror gives to the drive of the compensation capacitor [compared to a transistor

collector and a pull-up resistor] sorts out the unbalanced slewing.

Anyway, it's something else to experiment with if you're interested.

BTW I picked the two most closely matched transistors I had in the packet for the mirror.

RE: Op-amp made from Discrete Components

aswinvenu — Sun, 14 Jun 2020 22:42:29 GMT

Thank you for posting such an informative blog. Going through the discussion was also a learning experience.

RE: Op-amp made from Discrete Components

Workshopshed — Sun, 14 Jun 2020 07:57:56 GMT

A cool experiment and the followup discussing is just as interesting

RE: Op-amp made from Discrete Components

Andrew J — Thu, 11 Jun 2020 14:35:49 GMT

Ever thought about doing a set of Spice tutorials Jon? I'd be first in line.

RE: Op-amp made from Discrete Components

jc2048 — Wed, 10 Jun 2020 11:00:36 GMT

I've made one. Not sure why, but I have. Here it is sitting on my battered copy of Walter Jung's cookbook:

[View:/resized-image/__size/620x465/__key/commentfiles/f7d226abd59f475c9d224a79e3f0ec07-021c4451-857c-43c6-8399-8e48ff079aee/8508.contentimage_5F00_207104.jpg:620:465]

Didn't have enough 2N3904s, so I used BC549C for the NPN and BC559C for the PNP. The BC parts have about twice the

gain of the 2N devices, so my open-loop gain will be higher than Frank's. [Simulation gives around x15k.]

Other difference is that I couldn't find a 10pF, so it's got an 8.2pF on it. I have both input-pair emitters

connected together, so no attempt to trim the input voltage offset.

To test, I've wired it as a follower with a light 10k load to ground on the output. Here's what it does with a small

step [the noise is coming from my cheap-and-cheerful function generator]. Yellow is input, blue is output. The

offset is about 50mV. Rings a bit on the step, but there's enough margin to keep it stable.

[View:/resized-image/__size/480x234/__key/commentfiles/f7d226abd59f475c9d224a79e3f0ec07-021c4451-857c-43c6-8399-8e48ff079aee/3731.contentimage_5F00_207105.png:480:234]

Here's a large step. This shows the slew in both directions.

[View:/resized-image/__size/480x234/__key/commentfiles/f7d226abd59f475c9d224a79e3f0ec07-021c4451-857c-43c6-8399-8e48ff079aee/2235.contentimage_5F00_207106.png:480:234]

RE: Op-amp made from Discrete Components

jc2048 — Sat, 06 Jun 2020 18:52:20 GMT

This is what I'm seeing in the simulator for the open loop gain. The low frequency gain is about 76dB. A 741 would be 100db (x100k). A precision op amp might be several million. The red line is input to output. The green line shows the gain from the differential pair. The pair have some gain [in this case x10], but it's quite low. As you pointed out, the bulk of the voltage gain is coming from the common emitter stage.

[View:/resized-image/__size/583x517/__key/commentfiles/f7d226abd59f475c9d224a79e3f0ec07-021c4451-857c-43c6-8399-8e48ff079aee/7462.contentimage_5F00_207102.png:583:517]

The phase response nicely shows the effect of the compensation capacitor. It holds the phase at 90 so that when we reach the point where the gain falls to one (0dB), the phase is still well away from 0. [There's about 45 degrees of phase margin. That's ok, but it will ring a little on small-signal steps.] Don't worry that the phase is shifted like this. It seems a bit counter-intuitive [at least to me it does], but the closed-loop response, once you get the feedback operating, isn't the same.

The closed loop bandwidth can be obtained by following across until you hit the red line. So if you had x100 closed loop (40dB), the bandwidth would be 150kHz. [In theory, on your real breadboard circuit it will be different.]

This is the circuit I used to get the plot. It tricks the simulator by making it closed-loop at dc, so it can get the bias points right, and then going open-loop for the ac analysis.

[View:/resized-image/__size/620x398/__key/commentfiles/f7d226abd59f475c9d224a79e3f0ec07-021c4451-857c-43c6-8399-8e48ff079aee/6866.contentimage_5F00_207103.png:620:398]

RE: Op-amp made from Discrete Components

jc2048 — Sat, 06 Jun 2020 18:22:58 GMT

C1 (using your idents) is the compensation capacitor and is there to give a reasonable level of stability with simple feedback networks. It sometimes gets called a Miller capacitor because it uses the Miller effect to increase the effective value [the value is multiplied by the gain of the gain element it goes across - you can kind of understand that in a hand-wavey kind of way by thinking how one end is going to be forced by a magnified version of what happens at the other end]. The integrated circuit designers liked it because capacitors of any size are very expensive in IC area and this was a way of keeping the capacitor physically small whilst having the effect of a much bigger one.

RE: Op-amp made from Discrete Components

jc2048 — Sat, 06 Jun 2020 18:02:47 GMT

That's nice. It gives a reasonably good balance between the different attributes you might be interested in.

I had a play with it in a simulator [probably slightly different to what you're seeing because I'm using two 10V rails]. Swing is within about a volt of the rails, good GBW (about 10MHz), reasonably stable. Against that, slew is a bit unbalanced and max open-loop gain is low (about 4000).

Worst thing looks to be the input offset voltage. I'm seeing about 40mV. (I think I've got the tail ccurrent about the same as you had.)

[View:/resized-image/__size/620x493/__key/commentfiles/f7d226abd59f475c9d224a79e3f0ec07-021c4451-857c-43c6-8399-8e48ff079aee/0020.contentimage_5F00_207100.png:620:493]

It's quite a long way from an ideal op amp, and even makes a 741 look like the height of precision, but for 7 transistors it's very good. I

couldn't have done that well.

I've just had a play with a current mirror in place of the resistor load for the differential pair, but though it immediately improves the

offset to 4mv it then has a real knock-on effect on the overall small-signal stability [it's painful to get it stable again and anywhere near

the same transient performance as before].

A quick and dirty approach to the offset is a small emitter resistor like this.

[View:/resized-image/__size/620x493/__key/commentfiles/f7d226abd59f475c9d224a79e3f0ec07-021c4451-857c-43c6-8399-8e48ff079aee/3252.contentimage_5F00_207101.png:620:493]

Maybe a 1k preset across the emitters, with the slider going to the current source, if you wanted to adjust it.

RE: Op-amp made from Discrete Components

DAB — Wed, 03 Jun 2020 18:47:14 GMT

Well done Frank.

Very well explanation of the OpAmp implemented with discrete components.

DAB

RE: Op-amp made from Discrete Components

shabaz — Wed, 03 Jun 2020 18:44:02 GMT

Hi Frank,

This is a great experiment!! Very cool topic. And it's really nice that you've recorded it all, in viewer-friendly short snippets, it's perfect!

The audio and video seems fine to me, there's some equipment/fan noise but we're viewing you in a lab doing lab experiments, so lab sounds are natural! Also, the video is in good quality, everything is clear to see. This could easily form part of an elec-eng undergrad lesson..

RE: Op-amp made from Discrete Components

jw0752 — Wed, 03 Jun 2020 17:23:33 GMT

Hi Frank,

This is a great experiment and an excellent learning opportunity for all of us. Thanks for posting it.

John

RE: Op-amp made from Discrete Components

three-phase — Wed, 03 Jun 2020 15:34:31 GMT

A great delve into the inner workings of an op amp - thanks for posting, I enjoyed working through it.

Kind regards.

RE: Op-amp made from Discrete Components

genebren — Wed, 03 Jun 2020 14:37:52 GMT

Very nice work Frank. Your clear video/audio make it very easy to follow along with your findings. Well done!

RE: Op-amp made from Discrete Components

Andrew J — Wed, 03 Jun 2020 08:38:02 GMT

This is excellent - I'm going to give it a go as well as I'm on a bit of a roll with op amps. Thanks for posting Frank.