Another discrete (germanium) op-amp

[samhay]

There doesn't seem to have been a discrete op-amp thread for a few years, so here's another take on the idea.
I had a couple of design goals:
The option of using germanium transistors in places that matter (more below).
The use of current mirrors instead of source/emitter resistors (like a real IC op-amp).
Reasonable, but not audiophile, behaviour (input and output impedance, noise and open-loop gain, etc.).

This is what I came up with:

(larger version: http://i.imgur.com/KI5cveF.png)

It was somewhat/largely motivated by the 'Simple Op-Amp' described here,
https://www.allaboutcircuits.com/textbook/experiments/chpt-5/simple-op-amp/
However, as I only have PNP Ge transistors, it was designed so that the differential pair (Q1,2) are PNP.
I also added an output stage as the output impedance of the 'Simple' op-amp is too high to drive any sensible load.

It meets most of the criteria for an op-amp:
It has differential inputs.
It can be DC coupled, although there is a little error - maybe 10s mV depending on transistors and R1.

Thus, in principle you could drop it into your favourite op-amp circuit and/or use the same tricks - Tube Screamer style negative feedback clipping, etc.
However, it doesn't have an open-loop gain of 10^6, and max gain is set largely by the hFE of Q1 and Q2  - think character / mojo!

It works and I built a test circuit to do some transistor sniffing:

(larger version: http://i.imgur.com/8a5r1Og.png)

You can mix-and-match Ge and Si transistors without having to make any adjustments - although I haven't tried this with really leaky Ge transistors.

Making R4 variable (via R5 or a variable resistor) allows you to play with the current through Q1-3. As you decrease current, you lose headroom - specifically, you get further away from the positive rail, giving quite asymmetric clipping when overdriven.

Here's a perf layout, which can be considered to be verified (but only by me).
It is laid out to comfortably fit in a 1590B.

(larger version: http://i.imgur.com/aqP37QK.png)

Enjoy...

p.s. As Q1 and Q2 are responsible for voltage gain, I built it with Bulgarian GT1322 Ge transistors in Q1, Q2. These are low leakage (few uA) and have hFE ~ 50-ish. All other transistors are BC327 or BC337s.
However, I expect you could use just about any BJTs, including Darlingtons, without major issue.

[Passaloutre]

Really cool! How does it sound?

[samhay]

Thanks.
Sound depends on gain. Low gain gives clean boost and with the gain maxed, it sounds quite fuzzy.

[Passaloutre]

I have a bunch of 2N404s (http://www.nteinc.com/specs/original/2N404.pdf) with hfe ranging 50-100. Do you think they will work well? Do Q1 and Q2 need to be matched? Also 2N3904 or 2N2222 for the NPNs? If I wanted to leave the switch off, what would be the "normal" configuration?

Sorry for all the questions, I'm really intrigued by this, sounds like it could be a grittier Micro Amp or something (a dirty clean boost?). I might breadboard it tonight, will let you know how it goes.

[samhay]

As far as I can tell, it should be pretty tolerant of transistor choice (other than channel - PNP vs NPN) and transistors do not have to be matched.
Easy enough to find out on the breadboard.

You do need some resistance between Q4 gate and ground - I suggest you use 10k if you don't want it variable.

[Passaloutre]

Excellent. I'm trying to understand how this works, so forgive my naive questions.

Resistor R4 (or the series combination R4 + R5) essentially determines how "open" Q4 and Q5 are?

So if the resistor were shorted, Q4 base is at 0V, Q4 is fully open, and all current flows from VA to ground through Q4.

By adding some resistance here, you "close" Q4 a little bit, the current flows through Q5, which gives the signal transistors something to work with.

With a lot of resistance there Q4 and Q5 close even more, limiting the current available to the other transistors, reducing headroom?

Something like that? I don't really know much about the inner workings of opamps

[samhay]

The description of the 'Simple Op-Amp' is quite helpful, but note that our circuit is the opposite polarity:
https://www.allaboutcircuits.com/textbook/experiments/chpt-5/simple-op-amp/

Q4-6 are a current mirror. See also Figure 1 here, but turn it upside down as we are using PNPs:
https://en.wikipedia.org/wiki/Current_mirror
Also note that we have 2 outputs (Q5 and Q6), which should behave identically.

In the op-amp circuit, the current is set by R1 (or R4 + R5). The Q4 collector sits at ~ 8V, so with a 9V supply and R1= 10k, the current = 8V/10k = 0.8 mA. As they are mirrored, Q5 and Q6 will now try to sink a constant current of 0.8 mA into whatever they are connect to.

Q7 and Q8 are another current mirror, which will try to balance the current through Q1 and Q2. As a result, 0.4 mA should flow through each transistor when R1 = 10k.
Q3 will see 0.8 mA as Q6 collector is not connect to anything else.

[Passaloutre]

Ok, I think I can grok that. So in AC land, how do the + and - input signals arrive at the output, and where does the differential bit happen?

[samhay]

Q1,2 does the voltage gain and Q3 buffers the output and does the current gain.

Q1 and Q2 form a differential amplifier with a single-ended output. See figure 3 (and flip polarity):
https://en.wikipedia.org/wiki/Differential_amplifier

Q3 is an common collector / emitter follower amplifier. Q6 is doing the job of the emitter resistor and the base is DC coupled (and thus biased) to the output of the differential amp.

[Passaloutre]

Perfect. Thank you for your kind replies. I wasn't familiar with the "common base" (https://en.wikipedia.org/wiki/Common_base) topology, so I couldn't see how the V+ signal was getting from Q1 to the output. It just took finding the right explanation that put it in words I could understand. I look forward to breadboarding it.

[Passaloutre]

Well I got it running! It sounds pretty good, definitely reminds me of the MXR Micro amp. I tried to squeeze more gain out of it by upping the feedback ratio, but it gets VERY noisy with a 500k gain pot. I replaced the output pot with a 10k resistor to ground (like a pot always at 10), so everything is controlled by the feedback pot. I think the values you chose are very good. And with 6 Germanium transistors, it's full of "mojo". Thanks for the design, and all your help!

I have to ask, why are the caps so big? Particularly the 10uf caps in the feedback filter and output sections. It's too late in the evening for me to go messing with them, but maybe I'll try tomorrow.  What other values might you suggest?

By the way I'm using 2N404 for all PNPs, and 2N2222 for all NPNs

[samhay]

Excellent - nice job getting it going and very cool to see it does actually work with Ge transistors in the current mirrors.

Why are the caps so big? I designed this to pass essentially full audio (I play bass), with a bit of a vintage vibe:
C4 works with R6 as a high pass filter on the amplified signal. 10u vs 2k2 = 7 Hz.
C2 is big enough to work with a similar load - my audio interface has an input impedance of about 2k.
C3 works with the feedback pot as a low pass filter on the amplified signal. With feedback at 100k and C3 = 100p, we get a corner frequency of 16 kHz, which is close enough to full audio for me.
Of course, you are free to tweak these values to taste.

Feedback and why Q2 isn't quite a common base:
Note the feedback pot feeds from output back to the Q2 base. This provides both DC bias and AC signal; note signal is out-of-phase with signal at the emitter, so is providing negative feedback.
If you make the feedback resistance too large, you will get significant voltage drop across this resistor/pot and Q2 won't bias properly any more. I found 100k was a reasonable upper limit, and should allow fairly low-gain transistors to plug into Q2 without problem; note, the requirement for DC bias via the feedback resistor also means that this won't work very well (e.g. as a comparator) without a feedback resistor.

[Passaloutre]

So the whole thing is sort of a flat bandpass from 7Hz to 16kHz? I mainly play hollowbody guitars, which have enough bass to overwhelm pretty much anything you plug them into. If I wanted to tighten this up, should I lower C2 and C4 in tandem? Or would just one suffice?

May also be a prime candidate for the Timmy "bass cut" control...

By the way, something I learned through clumsy breadboarding: if you feed the output back into the input this makes an excellent signal generator, with frequency controlled by the gain pot.

[samhay]

Here is the output from an LTspice simulation of the circuit using 2N3904/6 transistors and R4+R5 = 10k.

The frequency response as you increase the feedback from 1R to 100k:



The output waveform when feeding it a 0.1 V peak signal at 440 Hz:



If you want to tighten up the bass, I suggest you reduce the value of C4 only - leave C2 so that you get predictable results regardless of what you plug the circuit into downstream.
C2 = 100n will take you into Tubescreamer territory, so try C4 values in the range 100n to 1u.

p.s. The Timmy tone control won't work very well without a buffer on the output.
Edit - wrong tone control. Yes the Timmy/Zen Drive/whatever bass control should work nicely.

[Passaloutre]

Dude, you're awesome! I'll try out the bass cut tonight. I think making C4 = 200n and making R6 variable from 1k to 10k should work. My calculations suggest this will give a high pass with variable corner frequency between 80-800Hz (http://www.muzique.com/schem/filter.htm). I worry messing with that resistor might reduce the overall gain too much... Would I have to raise the value of the gain pot to make up for this?


I might also try a Big Muff-type tone control, but I'm guessing that'll be too lossy without a recovery stage after.

[antonis]

I see Q1/Q2 have bigger h_FE than max feedback gain (about 33db)..

It should be interesting a graph of voltage gain dominated by transistors h_FE.. 

[samhay]

>I think making C4 = 200n and making R6 variable from 1k to 10k should work

That should work nicely. Zen Drive uses the same arrangement with a 100n cap.
Only thing to point out is that this will only affect the amplified signal and you won't be able to cut bass below unity -  much like in the Timmy, Zen Drive, etc. I don't expect this will be a problem.

This circuit is pretty loud, so you might get away with a Big Muff tone stack on the end.
If I was going to do this, I would scale the resistance down 10x (and the caps up 10x), to keep the output impedance nice and low.

[samhay]

>I see Q1/Q2 have bigger hFE than max feedback gain (about 33db)..

Yes, I puzzled over that for a while. You'll note I was a little cagey in the OP:
'max gain is set largely by the hFE of Q1 and Q2 '. 'Largely' was a bit of a guess, so would be interesting to find out.

> It should be interesting a graph of voltage gain dominated by transistors hFE..

I was/am going to do this experimentally using 'Piggybacked' transistors, but haven't quite gotten around to it yet.
I guess there is no excuse not to run the simulation - keep an eye out.

[amptramp]

Q4, Q5, Q6, Q7 and Q8 are current mirrors and if you want to make the circuit more tolerant of transistor variation, you could add equal emitter resistors for the current mirror pairs.  Q6 may be a different story - you could vary this resistor to split more of the current and make it a multiple of the current sent to the differential amplifier.  Integrated circuit amplifier amplifiers have the benefit of all circuit features being deposited in deposition processes that are uniform across the device.  Discrete amplifiers do not have this so it may help to add some tolerance for variability.  Resistors do exist in integrated circuits and they can be used for exactly this purpose - reducing the effect of variations.

The added emitter resistors tend to equalize the current even with variations in Vbe.

[samhay]

^Indeed, there is much that could be refined.
I was a little surprised that it works as well as it does.
As drawn, it will work with 3 very different transistors at Q1-3. They won't be as well-balanced as they could be, and certainly won't match those on a single piece of Si, but then we aren't aiming for something indistinguishable from an IC op-amp either (well I'm not with this design anyway).