fully-differential amplifier with discrete devices

[Zeyneb]

Hi dear peoples,

I hope you like this post. For fun I would like to see if I could make a fully differential amplifier out of discrete devices. You know, one that also have an inverting output.

TI has an Application Report titled: fully-differential amplifiers SLOA054E. From this report I've the following simplified schematic:



I think it would be cool if I can realize this circuit with RF BJTs, ones that have multiple transistors in the same SMT package like the HFA3046BZ from Renesas for example.

However it has been a while when I was doing analog circuits so this might turn out to be a naive question.
In order to start simple I think it will be best to see if I can get something reasonable working in LTSpice with devices like the 2n2222 and 2n2907.

Ok I did watch the great tutorial by w2aew: #193: Back to Basics: the differential amplifier, aka long-tailed pair, diff-pair

And put in LTSpice the following circuit:



But with this circuit I loose the voltage swing of Qn. So then I tried the following circuit:



But now node Q as well as Qn stay slightly below 9V supply. As Q4 and Q5 are in saturation. I couldn't nail down where the problem lies.

Questions
1) Are there more detailed resources to teach me to build the fully-differential amplifier with discrete devices?
2) How can I change my circuit to make progress getting the FDA to operate?

Best regards,
Zeyneb

[T3sl4co1l]

Right, you can't bias a fixed current source with a fixed current source.  This only works closed loop -- you need to add the Vcm / bias servo circuit.

You're also trying to measure the diff amp itself in open loop, which is sure to saturate from the slightest perturbation.  You can set DC-op using a huge inductor and capacitor, to close the loop at DC but leave it open at any meaningful AC frequency.

Tim

[Zeyneb]

Hi Tim,

Thanks for your suggestions. Can you please provide a link where I can learn about Vcm / bias servo circuit?

Will the Vcm / bias servo circuit close the loop with negative feedback? So that I can further experiment with the circuit. Or do I need to add that as well. Like Rf and Rg in the TI Application Report.

Ok but you think my network of pnp transistors is not wrong? As you didn't say something about that?

Thanks
Regards Zeyneb

[T3sl4co1l]

You need the two R's and Vocm Error Amp, which will control the current through R1, say by lifting one end from ground and driving it with the error amp output instead.

And probably make R1 half the... wait, R1, R3 and R7 are all the wrong values anyway, as shown you've got ~0.62mA tail current but 5.5mA collector current.  Of course the collectors will always be saturated to Vcc...

R7 = 3.6k, R3 = 680 and R1 = 4.16k would be typical, with R1 driven by an error amp.

Note that putting the CM sense R's on the collectors directly, will significantly reduce the differential mode gain.  You really need the buffers in there to get it back.  Just for setting up the CM loop, it's fine.

You can use a virtual (5 terminal) op-amp for the Vocm EA, or construct one from discretes if you like (but test it independently first, so you know you've got the voltage range and biasing correct!).

Tim

[Zeyneb]

Hi Tim,

I hope I do understand you. The two R's you mentioned are taken from this 1st stage Q and Qn?

If I fully make the rest of the circuit as shown by TI I would need a second Vocm error amplifier, with buffers?

Thanks a lot for your time!!

Zeyneb

[T3sl4co1l]

I'm referring to the only "R" on this diagram.



I don't know what "Qn" refers to.

Tim

[Zeyneb]

Hi Tim,

Q and Qn are the node labels on the collectors of Q1 and Q2 in my schematic.

I think I do understand you. I'm going to apply it tomorrow.

Thanks!

[T3sl4co1l]

Ah, yes, quite.

Tim

[Zeyneb]

Hi Tim and others!

I applied the Vocm Error Amp like you suggested and it is working nice. Below is the schematic:



However the FDA shown by TI has the Vocm Error Amp applied on the second stage. So I think I need to simulate the first stage in open loop. Also you told me that you can't bias a fixed current source with a fixed current source. So therefore I was studying bias circuits a bit and came with the following approach:



Here R1 not only sets the reference bias current but also a reasonable common mode collector voltage on Q and Qn. It's a bit delicate to set correctly.

I've two questions:
1) Although I got the bias approach with Q6, Q7 and R1 from a book. I'm unsure if I'm allowed to tie the sources and sink together with a long tailed pair in this case. It this ok? Or should I try another approach?
2) How about the next stage of the FDA shown by TI. Does it have a name so I can study this further? Can I just implement it as shown?

Thanks Zeyneb

[T3sl4co1l]

To do that, you need to use two BJTs in parallel for the tail current, hence setting twice the collector current so it balances normally.

But you cannot avoid the use of the Vocm error amp, even with perfectly matched SPICE BJTs I think you will find

The first and second stage really are the same, it's called a folded cascode.  This is a standard way to improve Miller effect and CM voltage range through level shifting, at the expense of more than doubling the bias current requirement (a fair tradeoff for very fast amps).

Tim

[Zeyneb]

Hi Tim,

To do that, you need to use two BJTs in parallel for the tail current, hence setting twice the collector current so it balances normally.

Have you seen my two Widlar resistors R2 and R3. I thought I already dealt with that in my design, or am I missing something?

But you cannot avoid the use of the Vocm error amp, even with perfectly matched SPICE BJTs I think you will find

Yeah, But like the FDA by TI I can apply it as shown on the 2nd stage and that will also deal with the voltage levels on the collectors of Q1 and Q2, right?

it's called a folded cascode

Lets see what I can find on folded cascode on the internet!

Thanks, Zeyneb

[T3sl4co1l]

Yes; I forgot to finish my thought.

The emitter resistors adjust the current, but again it will never quite match.

To control the current, you could put an emitter follower in series with the resistor, like used here:
https://www.seventransistorlabs.com/tmoranwms/Circuits_2008/Triangle.gif
but that's where I left off because you still have a degenerate case of two CCSs fighting.

The folded cascode works on the sums and differences of currents; it isn't a fight because the summing node is low impedance, not a high impedance gain node.

Tim

[magic]

A similar circuit invented by one audiophile vendor does away with the Vocm servo by weakly pulling emitters/sources of the diff pair to ground, see US patent 5,376,899 which by the way TI was forced to license