You are most welcome and thank you for sharing with us.
I have not done any contact printing for years now;
it reminds me of the time when I etched my own boards.
I myself find inductors at the gate not a good idea.
They do not dissipate energy, just store them.
Resistor does, but then as they add noise.
We used to use J111 to cascode the BF862 in follower or IV circuits.
Sometimes we get away with 220R gate stoppers, sometimes we need up to 1k.
But I thought Samuel uses multiple BF862s in parallel in his LNA.
And he managed to tame them somehow ?
Cheers,
Patrick
I have not done any contact printing for years now;
it reminds me of the time when I etched my own boards.
I myself find inductors at the gate not a good idea.
They do not dissipate energy, just store them.
Resistor does, but then as they add noise.
We used to use J111 to cascode the BF862 in follower or IV circuits.
Sometimes we get away with 220R gate stoppers, sometimes we need up to 1k.
But I thought Samuel uses multiple BF862s in parallel in his LNA.
And he managed to tame them somehow ?
Cheers,
Patrick
We had some bank holidays here, so i could spend some time on the Fet low noise amplifier. No matter what I do, it has somewhere negative input impedance and it
will oscillate with the right inductor across the input.
Circuit is nothing fancy: CS fet stage, cascode, TIA for loop gain, feedback 1000:1 or so
back to the source. The cascode is biased by a current source.
The usual remedy, a base stopper resistor defies the purpose of a low noise amplifier.
So far, I have made several versions: one with 2 pairs of IF3602 or many BF862 and
a bipolar cascode and a different one with upto 25 pcs. 2SK2394-6 or BF862 , each
with a MMBFJ310 cascode transistor of its own. That is easy because the 310 gate
can directly be attached to the 2SK2394 source without any ado about biasing.
I have used only 15 pairs because I didn't have enough MMBFJ310.
The IF3602 version was kinda hopeless already without any attempt at bootstrapping;
the 2394/310 pair has the bootstrapping for free and looks much friendlier.
I measured the input inpedance with a R&S ZVB8 vector network analyzer; sorry
that its lowest frequency of operation is 150 KHz (really specced at 300KHz).
I measure S11 of the amplifier and get a Smith diagram. That is really an impedance
plot with real values on the X-axis, inductance in the upper half and capacitance
in the lower. It is morphed to a circle so that the display goes from 0 to infinity on
a finite piece of paper. 50 Ohm without LC is in the middle of that unit circle. That
could be normalized for other impedances, but 50 is standard.
Everything in the circle has a positive real part, everything outside has a negative
real part. That can happen only with active circuits, where the analyzer gets
more power back from the circuit than it feeds into it.
The picture of the morphing is from the analyzer handbook, I think this is fair use.
The Smith chart is from the IF3602 amplifier; the greenish line leaves the unit
circle already near Marker 1, and at 2 MHz it reenters the passive region, so above
that the amplifier could not oscillate from any input impedance presented.
Sorry for the colours. The markers display the computed impedance and L/C
at their frequency.
For the 2394/310 pairs, the line also leaves the passive circle, but it stays much closer
to it, so it probably would take less effort to stabilize it completely.
I have read Samuel Gröner's amplifier article in Linear Audio. He also seems
to have problems with negative input impedance but he looses
me when he writes about negative capacitance. Negative capacitance is
inductance, nothing bad by itself. The problem is the negative real part.
No amount of additional "positive" capacitance could cure that; I tried up to
100 nF.
I have also tried Würth 600 Ohm 0805 beads as gate stoppers; they don't damp
anything, they behave like pure inductors. One bead for all 15 transistors made
the input inductive at high frequency, the trajectory went clockwise to the upper half,
following the circle.
Somehow I feel it is wrong to call the input fet common source. The drain
looks into the low impedance of the cascode and the source follows the
input voltage by the action of the feedback. So the FET does not see 0.2 Ohm
to ground but a synthesized current source. From its point of view it could
be called source follower at any time.
Any ideas about how to tame such a thing?
I don't want to take control of this thread, but I think here are the right
people for good S/N
I don't think the feedback loop is compensated properly, in particular with that difficult opamp. The easy way out is to patch a lead-lag compensation in the cascode drain to the ground (like 47nF series with 22ohm, values depend on the total transconductance of the input fet(s)) and then lower the opamp pole cap to a minimum (47pf will do). This cap will then only have a phase correction function, around the ULGF.
Gerhard:
I have been down that road a few times. Specifically it oscillates when you have the cascode in place. Fist guess is that the cascode transistor is oscillating and bringing everything with it. My guess is that its acting like the active part of a crystal oscillator (Pierce I believe but I can never keep them straight). My first step is a resistor in the gate of the cascode transistor. Despite my best efforts with the big Jfets I usually end up with a series R (10 to 100 Ohms) and 10-50 pF cap to ground.
I have been using the depletion mode Mosfets for Cascodes with success. The right ones handle the current and can handle a lot of voltage if needed. No bandwidth issues either. Cascode gate to input fet source makes it simple. However I have not tried it with so many discrete fets.
I have been down that road a few times. Specifically it oscillates when you have the cascode in place. Fist guess is that the cascode transistor is oscillating and bringing everything with it. My guess is that its acting like the active part of a crystal oscillator (Pierce I believe but I can never keep them straight). My first step is a resistor in the gate of the cascode transistor. Despite my best efforts with the big Jfets I usually end up with a series R (10 to 100 Ohms) and 10-50 pF cap to ground.
I have been using the depletion mode Mosfets for Cascodes with success. The right ones handle the current and can handle a lot of voltage if needed. No bandwidth issues either. Cascode gate to input fet source makes it simple. However I have not tried it with so many discrete fets.
I've had similar problems with very high transconductances making very tight local loop oscillators, once a snubber from the source of the cascode to ground cured it.
IIRC when you isolate a microwave transistor and aim for max frequency of oscillation it amounts to a Colpitt's oscillator.
Attached is my measurements, 1Khz-100MHz S11.
8 x BF862 in parallel, bipolar cascoded, opamp is OPA132 (not 100% sure, didn't open the case to check), compensated as described with 100nF in series with 22ohm, gate inductors are 1.2uH 0.5ohm ESR, closed loop gain is 1000, phase correction cap is 100pF, noise 0.33nV/rtHz, corner frequency around 400Hz, completely different servo.
And yes, it does oscillate with inductors lower than 500nH or even with 2.2uH with very low ESR (under 0.2ohm).
As you see, absolutely no problem.
P.S. 4195A
8 x BF862 in parallel, bipolar cascoded, opamp is OPA132 (not 100% sure, didn't open the case to check), compensated as described with 100nF in series with 22ohm, gate inductors are 1.2uH 0.5ohm ESR, closed loop gain is 1000, phase correction cap is 100pF, noise 0.33nV/rtHz, corner frequency around 400Hz, completely different servo.
And yes, it does oscillate with inductors lower than 500nH or even with 2.2uH with very low ESR (under 0.2ohm).
As you see, absolutely no problem.
P.S. 4195A
Attachments
Last edited:
Nono, if it is not unconditionally stable, that's definitely a problem for
a measurement amplifier.
I could get the IF3602 amplifier into a state where I could measure its
noise, but when it oscillates under only slightly non-optimum conditions,
then it is part of the problem, not part of the solution.
Your noise results seem reasonable; someone wrote on this site
that there are BF862 from HK and from mainland China. Those from
HK are said to have a slightly better noise corner. Mine are all marked
2AW, that would be the worse kind.
The 2SK2394 is probably slightly worse in absolute terms, but should
have 40 Hz corner. If the bootstrapping worked, one could simply make
it up with more chips.
I have bought a Locky_Z curve tracer in May, but it seems it is still
somewhere near Africa on the ship. I will resume work on the IF3602
when I can characterize them more easily.
Right now I try to understand the loop gain probes for LTspice.
Hairy.
I want 1 Mhz BW, so I cannot simply damp down everything to death.
regards, Gerhard
Last edited:
Thanks!
In Spice it seems to behave well. I also tried Genesys, that costs
me favour points,
, but I could not even get the bias loop workingwith their simplistic opamp models.
That said, using the ADA4898 model in LTspice: you don't need
any other noise source if you have that. Builds confidence.
My first Spice was 2G4 or so on a Telefunken TR4, the first commercial
microprogrammed computer. Us electronics students were allowed to
run the data center when the pro operators left in the evening. Kind of
very early personal computer. Still have some replacement boards full
of Ge-Transistors, pieces of core matrix and other trophies. Microprogram
store was some hefty volume of AA112 diodes. But it had 96 bit
double precision, 48 bit integers and instructions like "solve polynomial".
Where is my bottle with the Gerontol forte?
Forget about the Locky_Z.
All you need is a triangular wave func gen of 100Hz with variable DC, and a USB scope with enough resolution (12 bit or more).
And a reference resistor to measure current (0.1% 25ppm/K).
We now do all our JFET matching with a Handyscope HS3, and we do thousands.
Patrick
All you need is a triangular wave func gen of 100Hz with variable DC, and a USB scope with enough resolution (12 bit or more).
And a reference resistor to measure current (0.1% 25ppm/K).
We now do all our JFET matching with a Handyscope HS3, and we do thousands.
Patrick
Mine has 0.4MHz closed loop bandwidth at G=1000, that's 400MHz gain-bandwidth, not too shabby. Since mine has switchable gain, at G=100 I measured 4.5MHz bandwidth. It is not easy to compensate for both gains with the same network, but then again, the lead-lag with phase correction did the trick.
I have no experience with the IF3602 monster, but presumably it has a much larger Cgs, which may resonate with the a gate inductor with low ESR well into the gain bandwidth, potentially transforming the amp into a good Colpitts.
AOE3 talks at length about BJT noise measurements and presents measured data for quite a few low noise devices, including four BJTs with lower measured rbb than the vaunted (and discontinued) 2SB737. Here's what they say on p.509
Winfield Hill, one of the authors of the book, mentioned the noise testing here on diyAudio (link)
[Our noise test fixture] is basically the device under yest (DUT) configured as a grounded emitter stage, with settable current and settable collector-to-emitter voltage, and provision for determining its voltage gain. Knowing the latter, you measure the output noise voltage spectrum [with a spectrum analyzer -MJ], first with the input bypassed (to get voltage noise density en), then with a series resistor at the input (to get current noise density in). The fixture is described in detail in section 8.12.2.
Once you know en and the temperature of the DUT during measurement, you can calculate rbb.Winfield Hill, one of the authors of the book, mentioned the noise testing here on diyAudio (link)
Not that hard, Quantech did it 40yr. ago all discrete. I'm sure the AOE circuit is fine.
I don't have access to AoE till this weekend.
But I'll surely take a look.
But one of my question would be how to make sure the noise comes from the BJT itself,
and not from the supply (Vc) or the base (Ib).
One does need a constant and noise-free base current to bias the BJT, not ?
Patrick
But I'll surely take a look.
But one of my question would be how to make sure the noise comes from the BJT itself,
and not from the supply (Vc) or the base (Ib).
One does need a constant and noise-free base current to bias the BJT, not ?
Patrick