My version of the G = 1000 low noise measurement amp (for Ikoflexer)

- LMH6702 won't drive a 20ohm load, it has only 80mA output current. - Why do you need a servo if you have C1 DC block? The servo would only mess the input stage, the DC loop is broken.

80 mA would be enough to drive the feedback to +-1.6V or the sources to +-16 mV. With 80 dB of total amplifier gain that should be enough for input voltages that drive the amplifier into saturation.

The servo fixes the node act_load at 4Vdc, so I can set the drain current with VCC & the load resistor or a current source. A depletion FET should make a nice low dc-drop CCS load resistance. I'd like to use max. 4 Lithium cells in total.

Also, there is a window comparator that checks the 4V OP and if far off it reduces the input bias resistor, so power up does not take forever.

In the log file, there are the operating points, but no noise contributions.
 
Last edited:
Why don't you use a 300mA output op amp, plenty around, the extra dynamic range won't harm. An 1.7 GHz 3000V/uS op amp in that position doesn't make much sense here, from an AC perspective.

Sounds very complicated, without an obvious benefit. The act_load is well defined by Vcc-Idss*R12 disregarding the cascode mosfet parameters. All you have to take care is that the mosfet Vgs exceeds the jfet saturation Vds. For noise reasons, I was always very nervous of controlling the jfet input from a servo, too many chances to inject noise or to create not well controlled LF zeros in the loop gain (with the associated LF gain bumps).

If you want to move to an active load (CCS) that's a different stroy, you need the servo, but then there are simpler solutions to control the bias, see above.
 
Yes, I poked around in the bias loop. Nothing interesting. Yes, it is possible to induce motorboating with the wrong parameters.

I have opened the bias loop and forced the correct bias from the previous run with a voltage source. That made no difference. +-7V was more than the abs.max. ratings of the LMH6702. That has been corrected.

I don't care much about about an additional bias op amp. It is an OPA2134 in real life, and the second half is on the board available for free. If desparate, I still could use a DeBoo integrator.

And we use LMH6702 by the pound. It did never disappoint or surprise us. This is a RF shop here. I wanted really a short fb loop delay. I have written a few pages earlier that I think that the JFET in this circuit feels like a capacitively loaded follower, given the cascode and the hard feedback.

The ADA4898 in a similar circuit had severe convergence problems. At least its model from spring 2018 gets the noise right. The older model was off in every respect.

And I don't want anything that can drive +-300mA on my tiny board. It buys me nothing. Clipping is, yes, exactly that. There is no more headroom to be gained.
 

Attachments

  • bias_open_loop.png
    bias_open_loop.png
    369.1 KB · Views: 433
Last edited:
Yes, I poked around in the bias loop. Nothing interesting.

This makes things clearer. I see four slopes in different places (1/f to the 0,1,2,3). At the lowest frequencies it looks more like 30dB/dec rise like a 1/f effect meeting a 1/f*f gain. R16 looked at as a current noise has a 1/f*f rise for instance.

When all else fails sometimes I research the various induced gate current noises to see if they are modeled. Just thinking out loud, because the usual problem is that the sim is OK but the circuit is not as well as usually being high impedance circuits. OTOH the Cgs here is huge compared to what I am used to.

The BF862 has a 1/f component of current noise due to induced gate current as well a some microphone capsules due to the sheet rho of the diaphragm coating. I have posted a very nice IEEE paper on this in the past.
 
Last edited:
If you want to move to an active load (CCS) that's a different stroy, you need the servo, but then there are simpler solutions to control the bias, see above.

Controlling the current source looks elegant, but I cannot (and would not like to) run an IF3601 at/near Idss, let alone 4 of them. That could easily be 1A, for one.

Large source resistors are impossible because of their thermal noise, so controlling the gate voltage is the only solution.
 
Last edited:
Oh, thanks, I think you've got it. When I move the input to the LMH6702 input, I see 2.3 nV/rt Hz but with a 3 KHz corner. Since there is only 40 dB gain in front of it, it could not be better than 2.3 nV at 30 Hz, definitely a performance impact just from the 2nd/3rd stage.

We use the LMHs as buffers for precision oscillators, and we see the 1/f on the signal source analyzer, but we do not get the full broadside because not all of it is mixed up to 5, 10 or even 100 MHz.

happy bed time now :) Gerhard
 
Controlling the current source looks elegant, but I cannot (and would not like to) run an IF3601 at/near Idss, let alone 4 of them. That could easily be 1A, for one.

Large source resistors are impossible because of their thermal noise, so controlling the gate voltage is the only solution.

BTW, to lower the drain current, there’s no reason why you could not bias the gate(s) at a fixed negative bias, like you are doing in your beloved jfet based RF amps, and then keeping the drain CCS bias control loop.
 
That could also make some unlucky JFET pairs work together.

One might use left & right outliers for the CS and cascode with the same Vgs when I look at the picture below. :) Extra bonus if you publish that on diy without a word about the selection. Just the circuit and "It sounds absolutely eargasmic!"

What 1/f corner did you see for these ON FET/BJT cascodes?
 

Attachments

  • 8pcs_IF3602.png
    8pcs_IF3602.png
    265.4 KB · Views: 542
Last edited:
I've only built myself a version with BF861 cascoded with individual bipolars. Aluminum milled case 12mm thick on all sides renders about 1.5nV of 50Hz harmonic noise (under 1nV @150Hz) in a VERY noise polluted lab.

My conclusion was that BF861's have lower flat noise compared to CPH05 (8 x BF861 @12mA Idss was consistently 0.28nV/rtHz) but have higher corner noise (about 400Hz).
 

Attachments

  • IMG_1898.JPG
    IMG_1898.JPG
    664 KB · Views: 459
See Post #1 on this thread.


Not sure what you mean, I see only a DC bias feedback loop encompassing a photovoltaic SSR. My suggestion is to bias the gate(s) with a fixed negative voltage and use a DC bias feedback loop on the CCS load (schematic above). This arrangement would have a high enough PSRR to allow a simple 317/337 power supply (which is what I did). Using a simple resistive load (and no DC bias feedback loop) requires a very low noise power supply, or batteries as Gerhard is using.
 

Attachments

  • gridstop.png
    gridstop.png
    45.4 KB · Views: 742
  • gridstop2.png
    gridstop2.png
    16.3 KB · Views: 751
Last edited:
I am attempting to build my own (very modest) version of this amplifier.

The design is practically finished (spice only at this stage), and during the sanity checks, I discovered an unpleasant quirk when the source impedance is comprised between a few hundreds ohm and a few hundreds kilohm: a bump at the end of the bandwidth:

attachment.php


I have used a somewhat different topology, but I retained the feedback to the source of the FET because of the obvious advantages it brings, and that's where the problem comes from: initially, I was unaware of it, because the FET seems to work in common source (due to the 1 ohm source resistor), but of course, that is not the case: with the opamp, it forms a composite follower driving the 1 ohm with a much lower impedance.
Because of the large Cgs of the FET, this forms part of a Colpitts-type circuit, and the second capacitive part is virtual, provided the opamps phase-shifts.

This causes a transformation of the input impedance, insufficient to cause oscillations but sufficient to degrade the FR.

That is problematic, because I opted for a FET input precisely because of its low noise current and high input impedance, and this quirk negates all the advantages.

I have the intention of using this LNA as a GP lab amplifier, and it has to remain well behaved for all conditions, including various source impedances.

The fixes I have found are unsatisfactory, as they completely ruin the top frequency: to cover every possible case, the compensation capacitor(s) need to be quite large.
Fortunately, in the topology I used the problem has a much reduced magnitude, but it is present and I would prefer not to compromise.

I have searched the thread, and I have seen that the problem has been detected, but I have seen no magic-bullet solution to cure it.

Maybe I searched with the wrong words?

Anyway, if such a solution exists, I would be glad to know about it.
 

Attachments

  • LNA0.png
    LNA0.png
    108.4 KB · Views: 524