Transistor buffer with x2 gain *and* high input impedance?

[ElectricDruid]

Hi All,

I'm trying to tweak a buffer circuit on the front of my current delay prototype. At the moment I have a BJT voltage follower (emitter follower), which works fine. But checking the levels through the circuit, I realised I could reduce the noise a bit if I boosted the signal on the way in. I've got more headroom available than I'm using.

I could alter the emitter follower into a common emitter amplifier, but the input impedance of those is typically pretty low because of the biasing resistors.

Is there some simple scheme for a single transistor (bjt or fet? I don't mind) amplifier that could give me a gain of 2 *and* a high input impedance? Inverting is a bonus, since the rest of the circuit inverts too, so an inverting amp gives me no change overall.

(I'm avoiding op-amps because I'm currently on an even number in the circuit overall and adding one more would required an extra chip.)

Thanks for any ideas,
Tom

[antonis]

Any n-JFET biased with a 1M resistor (from Gate to GND) and Rd/Rs = 2 (or gm x Rd = 2, if you prefer..) should be more than fine...

Generaly, FETs have considerably lower transconductance than bipolar transistors, which makes them less suitable as amplifiers and followers, but there isn't such a point here...

[R.G.]

Antonis is right as long as you can stand adjusting the value of the drain and source resistors for each JFET.

This is because you need the JFET drain to be on approximately the center of the available power supply (since headroom is a problem already) and so for each JFET you will need to juggle Rs up and down to get the right bias point, and changing Rs means you also have to change Rd to match and get 2x Rs.

As an alternate, you could use a single-package Darlington, like the MPSA13, use Re = 1/2 Rc, and bias it with two resistors, then bootstrap the input with a capacitor from the emitter. The stages in the Univibe do this (excepting they have a gain of 1 to both emitter and collector) and achieve input impedances in the 1M range.

You're going to be losing 1/3 of your available power supply to the source or emitter resistors, so that may or may not be a problem.

If it is, you will need to go to a N-JFET/PNP compound stage. This can easily give 1M input impedance and gain, without losing much of the power supply. It's effectively a special-purpose opamp made up for the single case, as it's gain is determined by feedback resistors. But it will take a lot more pins, resistors and capcitors to do than an other opamp package.

Let's talk about "costs" for a moment. The "cost" in a design of putting in another part is largely a function of how many pins get soldered into the PCB. Transistors sound like a great deal and opamps too much work, until you realize that two transistors is six pins, and TWO opamps are six pins plus two power pins. The opamp will need fewer resistors and caps than the two transistors for the same gain, etc, and may in fact have fewer parts - and pins.

Don't fear the IC unnecessarily.

[ElectricDruid]

Antonis is right as long as you can stand adjusting the value of the drain and source resistors for each JFET.

I can't. This needs to be plug'n'play, not tweak'n'twist. I was hoping to build a digital delay that beginners could put together.

As an alternate, you could use a single-package Darlington, like the MPSA13, use Re = 1/2 Rc, and bias it with two resistors, then bootstrap the input with a capacitor from the emitter. The stages in the Univibe do this (excepting they have a gain of 1 to both emitter and collector) and achieve input impedances in the 1M range.

You're going to be losing 1/3 of your available power supply to the source or emitter resistors, so that may or may not be a problem.

No, that's not a problem necessarily. I'm feeding an ADC input of a dsPIC that does the digital delay, and it's running on a 3.3V supply. I was thinking 1V maximum input, boosted to 2V, going into a 3.3V input. That makes reasonable use of the headroom and resolution, and gives me a bit of clear space if you push it a bit (e.g. it won't actually hard clip until you get to 1.65V input)

Let's talk about "costs" for a moment. The "cost" in a design of putting in another part is largely a function of how many pins get soldered into the PCB. Transistors sound like a great deal and opamps too much work, until you realize that two transistors is six pins, and TWO opamps are six pins plus two power pins. The opamp will need fewer resistors and caps than the two transistors for the same gain, etc, and may in fact have fewer parts - and pins.

Agree, but in this case, I'd be adding a single op-amp to a design that already has four. Now, four is a nice number in op-amp world, but five is just...well, rong.

Don't fear the IC unnecessarily.

Oh, I don't, don't worry. I love 'em. This delay design is all ICs apart from this one transistor buffer on the front, and that's just me avoiding using another chip package.

Thanks,
Tom

[samhay]

As an alternative, do you have an active filter between input and pic? If so, can you not add gain here...

[R.G.]

Antonis is right as long as you can stand adjusting the value of the drain and source resistors for each JFET.

I can't. This needs to be plug'n'play, not tweak'n'twist. I was hoping to build a digital delay that beginners could put together.

OK. JFET is out. Too variable. The JFET/PNP combo will work fine, properly designed, but will eat up more board area and parts than another opamp, as well as being less "predictable" than an opamp.

Agree, but in this case, I'd be adding a single op-amp to a design that already has four. Now, four is a nice number in op-amp world, but five is just...well, rong.

There can be no disagreement about matters of taste, but since it's a matter of taste, and since you are designing for beginners, my personal taste would be to make it much more predictable, and in that case, I'd use the opamp. I don't consider the number five to be any worse than the number four, especially when it's hidden inside a PCB, and there are more ways for beginners to get FET/transistor stuff "rong" than opamps. Opamps are certainly easier to troubleshoot by email, as they only have to have very predictable voltages on their pins to be confirmed right or rong.

Oh, I don't, don't worry. I love 'em. This delay design is all ICs apart from this one transistor buffer on the front, and that's just me avoiding using another chip package.

OK.

The one thing that transistor circuits have over opamp packages is that there are more inner parts and so it's easier to squash the parts around on an already fixed-in-place PCB layout (IMHO), so if that's where you are, I sympathize. But as to the matter of taste, I would make a terrible face, perhaps mutter some words left unmentioned, and use the opamp. That's just me. Otherwise, use the darlington and bootstrap the input from the emitter.

[Phoenix]

Why not flip around the order of your active devices so you can best take advantage of their individual properties?
Presumably, one of the four opamps is being used as an output buffer, depending on the topology of your design, you may be able to replace that with an emitter follower/source follower, and not have to worry about super-high input-impedance, or tweaking the bias. Then you could use that spare opamp as your input stage with high input-impedance and predictable gain.
Without seeing your design I don't know if this is practical, but it's something to consider.

[Transmogrifox]

Otherwise, use the darlington and bootstrap the input from the emitter.

Is bootstrapping a darlington even necessary in this case?  Even suppose moderate gain darlington, Beta 8,000.

Not certain what input impedance to next stage, but suppose you can handle 10k collector resistor, gain of 2 you could use a 4.7k to 5k emitter resistor.

For sake of discussion assume 4.7k reflected up through emitter at beta 8000:  Zin = 36 Meg.  And due to the small base currents you can use about any "big 'ol" bias resistor (4.7 Meg)

Hell, for that matter a general purpose transistor with beta near 250 against 4.7k gives you >1M input impedance (but you still have the bias resistor problem).

Using the darlington eliminates your bias resistor problem but eats into headroom.

I would be willing to wager you can do this with a single high-ish gain BJT (2N5089 comes to mind).

The next stage input impedance might make the crucial difference here.  Can you afford to use a 50k collector resistor? More?  The higher the input Z of the next stage, the larger value of emitter resistor you can use for 2x gain, the lower value of base bias currents...it all might meet in a happy place.

[R.G.]

Is bootstrapping a darlington even necessary in this case?  Even suppose moderate gain darlington, Beta 8,000.

Not certain what input impedance to next stage, but suppose you can handle 10k collector resistor, gain of 2 you could use a 4.7k to 5k emitter resistor.

For sake of discussion assume 4.7k reflected up through emitter at beta 8000:  Zin = 36 Meg.  And due to the small base currents you can use about any "big 'ol" bias resistor (4.7 Meg)

All true - but the bias resistors wind up adding significant noise. It makes for better noise performance to use very high gain and a "noiseless biasing" setup, with a high value resistor from a divider to the base. From there, you're only one capacitor away from a bootstrap.

It is possible that a high gain ordinary transistor would be good enough. But darlingtons are a slam dunk.

Hell, for that matter a general purpose transistor with beta near 250 against 4.7k gives you >1M input impedance (but you still have the bias resistor problem).

Using the darlington eliminates your bias resistor problem but eats into headroom.

I would be willing to wager you can do this with a single high-ish gain BJT (2N5089 comes to mind).

The next stage input impedance might make the crucial difference here.  Can you afford to use a 50k collector resistor? More?  The higher the input Z of the next stage, the larger value of emitter resistor you can use for 2x gain, the lower value of base bias currents...it all might meet in a happy place.

Yeah, Keen's Second Law still applies - when in doubt, whip in a 2N5088. Well, or a 2N5089, or "big brother" MPSA18. Any of them might work, as might cascading two ordinary NPNs, as in the Univibe. But bootstrapping is pretty magic.

An NPN/PNP compound would work great as well. A variant of the "Onboard Preamp for Guitar" at geofex would work as well, with some tinkering with the biasing and getting rid of that current mirror stuff I used to keep current down. Easy enough to set its gain to 2.

[ElectricDruid]

Thanks everyone. There's all sorts of good ideas here.

One comment about the "bite the bullet - use the op-amp!" solution. Whilst it has a lot going for it in many ways, the best op-amp buffer is non-inverting. An inverting op-amp with gain of 2 is never going to have such a high impedance as the non-inverting one. Ok, I could use some pretty big resistors and get the impedance up (1M input resistor, 2M2 feedback, say) but then noise starts to be a problem again. Of course, I could bite another bullet and give up on having a non-inverting signal path through the pedal, but I'd rather not if I don't have to.

So unless there's something I'm missing there, I think the transistor version is preferable.

Thanks again. The discussion is enlightening.

Tom

[antonis]

But bootstrapping is pretty magic.

IMHO, he could modify the existing BJT emitter follower to a CE Amp, lower the bias resistors for an even "stiffer" & "noiseless" voltage divider (optional..) , place a capacitor (big enough for the frequencies of interest) in series with a resistor from Emitter to Base and , of course, move the bias point of the voltage divider to the junction point of bootstrap cap & resistor..

Something like this:


I think it should work fine even with "humble" transistors like 2N3904..

[samhay]

>One comment about the "bite the bullet - use the op-amp!" solution....

Well you could use 2 op-amps - a non-inverting buffer followed by low impedance inverting amp with gain of 2....

[ElectricDruid]

IMHO, he could modify the existing BJT emitter follower to a CE Amp, lower the bias resistors for an even "stiffer" & "noiseless" voltage divider (optional..) , place a capacitor (big enough for the frequencies of interest) in series with a resistor from Emitter to Base and , of course, move the bias point of the voltage divider to the junction point of bootstrap cap & resistor..

Something like this:


I think it should work fine even with "humble" transistors like 2N3904..

Ok, let's run with that for a minute. Let me slap a few values on things and tell me if I'm going wrong.

R1 & R2: Say 10K.
Rc: Dunno. Say 100K.
Re: Dunno. 4K7? (Did I mention I'm not a transistor designer?)
Rbst: This looks like the input impedance to me. 1M.
Cin: Given the 1M, this only needs to be >10nF
Cbst: I don't know. What does this value get based on? Is it a HPF with Re? That'd make it about 3.3uF or so.

How did I do?!

Thanks,
Tom

[antonis]

Sam, you force Tom to proceed in an official  public confession as an "Op Amp Hater"...!!! 

[ElectricDruid]

Sam, you force Tom to proceed in an official  public confession as an "Op Amp Hater"...!!! 

I've actually got another reason for not doing what Samhay suggests. Otherwise, he's right - if I'm going to add one op-amp (an 8-pin package) I might as well add two op-amps (same 8-pin package). It doesn't really use more resistors.

But if I'm going to add more op-amps, I want one of the two as an extra filtering stage after the delay. Currently I have a 12dB/oct filter there, but more would be better to clean up all the digital noise. If I had an extra op-amp, I could go to a 24db/oct filter. If I don't use an extra op-amp, I'm thinking of redesigning it to add an RC on the front and get a 18dB filter (every little helps, right?!).

[antonis]

<R1 & R2: Say 10K.>
This SHOULD be OK in case of an Emitter follower (Ve almost at Vcc/2)
Now we have to bias BJT as CE Amplifier..

<Rc: Dunno. Say 100K.>
It depends on DC Collector current..
(Vc should be near Vcc/2 - actually at (Vcc+Ve)/2 for equal swing)

<Re: Dunno. 4K7? (Did I mention I'm not a transistor designer?)>
No need to mention it - It's obvious from your need for gain of 2 when your Rc/Re ratio is about 21..

<Rbst: This looks like the input impedance to me. 1M.>
It looks the same to me but it's value shouldn't be greater than, say, 4k7..
(it's actual value isn't related to the value "seen" from the signal because of the same -and in fase - signal voltage variation between Base & Emmiter..)

<Cin: Given the 1M, this only needs to be >10nF>
See above...

<Cbst: I don't know. What does this value get based on? Is it a HPF with Re? That'd make it about 3.3uF or so.>
It should allow all frequencies of interest to pass (bootstrapped) from Emitter to Base to make the two legs of Rbst always "seen" at the same voltage - no current flowing through Rbst --> Rbst = Infinite resistance..!!

<How did I do?!>
Plz allow me to keep it for myself...


P.S.
Would you like some "ready" values or you prefer to calculate them step-by-step ..??

[ElectricDruid]

<How did I do?!>
Plz allow me to keep it for myself...

Yes, I think that's wise! It can't be pretty! I've just been reading up on the topic here:

http://www.tedpavlic.com/teaching/osu/ece327/lab1_bjt/lab1_bjt_transistor_basics.pdf

The circuit you posted is on page 10, and going through the equations made me realise my guesses were well wide of the mark. Oh well!

I'd be interested to see your process for working something like this out too, if you don't mind and have time. Transistor amps is a learning process for me. Op-amps and microprocessors is my usual domain.

Thanks,
Tom

[samhay]

>Sam, you force Tom to proceed in an official  public confession as an "Op Amp Hater"
Sorry - I am quite fond of using BJTs as input buffers and I like a good bootstrap, but I am not convinced they are the right tool for this job.

>(same 8-pin package)
That's what I was thinking.

Do you reaaaally need the effect to be non-inverting? If not, things get easier and you get to add your 24 dB filter.

[antonis]

I've just been reading up on the topic here:
http://www.tedpavlic.com/teaching/osu/ece327/lab1_bjt/lab1_bjt_transistor_basics.pdf

Very well, indeed...!!
(but you may ommit complex formulas with indegrations and other nasty terms.. )

I'd be interested to see your process for working something like this out too, if you don't mind and have time.

Plenty of time but completely out of coffee..

Maybe you can use some general guidelines and rules of thumb, like:
1mA Q current, Ic=Ie (for hFE greater than, say, 100), Ve = about 5 - 10% of Vcc (between, say, 500mV and 1V), Vc = (Vcc+Ve)/2, Vb-Ve = 600-700mV, R1//R2 = 10 x hFE x Re (greater doesn't mind), Vce > 200-300mV and anything else mentioned by another willing and more experienced guy..

According to your needs, you don't have to place Emitter bypass capasitor nor take in mind re (25mΩ/Ie) (this will be a further step in a need for higher gain with some current/resistor restrictions..)

You may "round" your calculations to commercially close values and, perhaps, verify that you still are between "margins"..
(and be happy until you realise that Input & Output impedances have come univited and your calculation sheets are dangerously ignited..)

P.S.
Transistor biasing isn't "rocket science" (or, is it..??) but for sure it can't be embeded in a few example sentenceses...

[ElectricDruid]

Do you reaaaally need the effect to be non-inverting? If not, things get easier and you get to add your 24 dB filter.

No, I guess not. It's just "good practice" is all. I think perhaps you're (and RG) are right and I should stick the op-amp in, thereby saving the complications and making a much better buffer, and then I get an extra op-amp and can improve the filtering too.

Tom