Inverting FET openloop gainstage

[Onvinyl]

Hi,
I made a sketch for an openloop stage with ~30dB gain.
Sim only, for the moment.

This is how it works (at least, how I understand it does...):

The heart of it is complementary version of a grounded gate diff-amp with 'no' miller effect (or it would be with R1/R8 = 0).
The cascoding FET's (J4/J8) shouldn't show a miller effect as well, but without Q1/Q2, high-Freq rolloff starts very early at only some Khz. With Q1/Q2 present, it's no earlier than 70Khz (see next posting).

FET models sim a IDSS of 14.8mA, so I real world biasing would show different values for some parts.
J6 is present for convenience in the real world when finding the optimum base voltage for Q1/Q2.

This cascode connection has been used by borbely, curl, pass and many others, it is discussed in the A75 article and in the new borbely-papers, for instance. There is a patent of M.Noro which goes into details as well (all infos at diyaudio.com

Because the circuit depends strongly from any load, it needs buffering.

I'm looking forward to build it.

Unfourtunatly, I don't know how to perform square-wave tests oder distortion measurement in LTSPice to look for fundamental errors.

A few questions:
-Why is miller effect still present at J4/J8?
-How to calculate the optimum load (R12)?
- How can I adjust the fet models, so I can set them to my real world devices (a bunch of 2sk389/2sj109 around 6-9 mA IDSS)?

thanks,
Rüdiger

[Onvinyl]

<output bandwith>

[john curl]

Why include an output stage at all? Think about using it as a transconductance amplifier.

[Onvinyl]

I build the real circuit (without output follower), and while the sim with LTSPice showed wildly different dB-numbers for very near values of Rl, the real circuit is much more robust against loadchanges.
If all fits, it shall drive the 50/500 pole of a riaa network.
At 23.5dB, it puts out fine waveforms (no distortion measurement gear at my command, sadly), a 20 Khz square wave is, well, square.

Sadly, I have too few V-grade FET's. I guess it's best to switch for beefier parts for the cascode, with my 11mA IDSS parts I have to go down to 2.5mA for the input pairs to get best waveforms.

Rüdiger

[Onvinyl]

Hi,
now I build the attached circuit. Other than indicated, the input pair are 2sk389/2sj109 with an IDSS around 7mA running at 6mA. The cascode fet's take 11mA, the bjt's take 9.

The ouput sits at 300mV DC at the moment, gain is 100. The square looks perfect till 70Khz and slowly rounds till it loses gain at perhaps 300Khz.
It is perfectly stable, power on and off are very smooth.

Can't wait to actually hear it.
Rüdiger

[Onvinyl]

sorry, bjts sit at 12mA

[CBS240]

Hi

It appears to me that the input capacitance of the j-fet is much greater than that of the BJT. The input impeadance of the BJT may be lower at 0Hz because of base current but it is more resistive, so when viewing higher frequencies like in that 70KHz SW, the input capacitance of the j-fet used as the gain stage loads the input differentials more, reducing dV/dT. Try using the BJT's as the gain stage and the j-fet's as followers. R3 could be eliminated since the j-fet follower is self bias, with source resistors. This way the BJT's, with a lower impeadance, are driving the j-fet input capacitance instead of the input stage. Also you could try BJT's for the gain stage and follower. Might have to adjust R1,2,8,9 a bit though but I would be interested in the simulation results to see if it is faster.

[GK]

Originally posted by Onvinyl
sorry, bjts sit at 12mA

You really need some emitter degeneration here to define the quiescent current through the BJT's. I'm actually surprised that you haven't smoked them yet. Consider ~10 ohms on each emitter and perhaps even replacing R3 with a series pair of 1N4148’s or similar.

Cheers,
Glen

[Onvinyl]

Originally posted by G.Kleinschmidt



You really need some emitter degeneration here to define the quiescent current through the BJT's. I'm actually surprised that you haven't smoked them yet. Consider ~10 ohms on each emitter and perhaps even replacing R3 with a series pair of 1N4148’s or similar.

Cheers,
Glen

Well, I *did* smoke them, when I made R3 too large...

I will try all suggestions as soon I have the time!

Rüdiger

edit: yes, I did try it with two bjt's, it *is* faster but it gave me some problems that looked like ringing. I will investigate further.

[CBS240]

It looks like R1/R8 degenerate J4/J8, the question is, to what degree? Some degeneration on Q1/Q2 is still a good idea.