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

This is my take on Ikoflexor's low noise measurement preamp.

I used a Vishay photovoltaic MOSFET driver ( in an unusual way to make a negative bias on the input JFET. This eliminates the huge capacitor in series with the 1 Ohm gain resistor. In fact there are no electrolytics in the signal path. The back to back PV arrays make a nice bipolar bias reference and you can simply flip the diodes to reverse polarity. On dual supplies this voltage can go outside both rails (when referenced to ground). R3 is selected for ~1/2 the supply at the drain off J1 at the desired drain current. Then when R2 is ~2XR3 the drain voltage will track roughly ½ the supply voltage over a wide range with no further adjustment. The MOSFET driver can be looked at as an isolated current in current out device (gain = .001). This times the 10Meg resistor gives a gain in this loop (FET drain current to gate voltage) of 10,000, or 10,000*gm if looked at as a voltage gain. For these FETs it's around 100 which gives decent enough operating point stability from FET to FET without trim.

The nominal gain is 60dB and BW from .1Hz to >200kHz. the dual op-amp arrangement has DC gain of one and acts at AC like a differential out amplifier. So the input stage finds Vcc/2 at the drain of the FET with optical feedback and the output stage sits at the same voltage.

This op-amp can be an ordinary bi-FET. An additional requirement was that the circuit could be used with 48V phantom power and make for a very simple one box set up. Of course the multi-paralleled FETs at the input would work but phantom power is limited to 10mA.

As built with one 2SK170 it came right out at 1.1nV as simulated and with these values and AD823, .1Hz to 400kHz. No comp-cap is needed at G = 1000, but it will be necessary at lower gains. Even a TL072 will work with Fmax becoming 200kHz or so. If C6 gets too large one can induce motor-boating, I’m sure there is a optimum set of low frequency time constants.

At 6mA current draw this whole circuit could be put under a 48V phantom supply with differential outputs on the two op-amps (gain will be +6dB more). This would require a larger resistance level in the feedback network.

Big advantage, 1Hz BW with no caps in the feedback network and no electrolytics needed anywhere (except bypassing).

(Pictures in Pass Forum)

BTW Ikoflexor my motherboard overheated and failed last week. I lost my entire address book, and it took me a few days to get the energy to assemble this.


  • lna2.JPG
    100 KB · Views: 7,259
Good question.

I only sent him one 372, and it had a high Idss.
And he needed the circuit first to measure the 372. So I guess he had to use something else known and available.

But now that we know how the 372 is in comparison, the question is a valid one.

And I also wonder what is better -- a single 372 or 2x 170GR in parallel ??

Last edited:
would adopting a 372 in lieu of the 170 be of any noise benefit in Scott's circuit?

The resistor reference swamps the FET noise, so my cal line is very close to exact. To use this as an instrument front end I would go with the 372 below 20k, but I must confirm that the high frequency anomolies will not give problems. Right now I'm leaning toward trying some paralleled BF862's also.

I presume you applied this in the JFET noise measurement setup.

Could you kindly also post details of the rest ?
Both schematics and photos would of course be of great interest.

Many thanks,

Sorry if I wasn't clear this is the FET noise measurement. With input shorted you see nothing but the FET noise (and 1 Ohm). With a 10k source resistor the noise is sqrt(Noise(FET)**2+(12.38nV**2)). The FET contributes <1% error to this so I divide the waveform by 12.38 to give me a 1nV reference line. This way I don't need a gain cal.

To be totally general you would want to be able to vary Vds and Id over a large range, rev 2.

This setup compares apples to apples, all FETs are measured at ~1mA. Remenber the noise is a weak function of current. At 6ma there is only a factor of .64 improvement.

Bit busy this week, more later.
So I short the input of the circuit as posted (before the cap C5?), replace J1 with the DUT (device under test), and measure the output with a spectrum analyser, correct ?

You mentioned in the other thread :

that you need to go to double shielding. I presume it is the amplifier that you are shielding. How about power supply -- batteries ?

Sorry for being such a pain,
Stupid question :

What is the value for R6 ?
What does 2.7me means ?


2.7 Meg Ohm (me is our simulators shorthand). R6 was 3K and had about 3.8V across it. I was a little imprecise the first time just seeing if the idea worked out. Every DUT I tried biased at about the same point. I used a 9V battery with the circuit in a box and them put a grounded box all around this.
So if I were to duplicate the circuit, I should use 2.7M for R6, correct ?
What about R3 & R2 (just some typical values) ?

You mentioned a reference resistor of 20k in post #5, which one is that ?
Any particular requirements for the resistors (low tempco, 1%) ?

0.1uF decoupling cao close to U1, I suppose ?

Thx again,
4 years on, I finally got round to building something.
Only half way through, not yet tested, but simulated on Space beforehand.

Supply voltage is changed to 9V, as I want to use a single battery.




  • 150906 Wurcer LNA 1.JPG
    150906 Wurcer LNA 1.JPG
    159.1 KB · Views: 2,206
  • 150906 Wurcer LNA 2.JPG
    150906 Wurcer LNA 2.JPG
    154 KB · Views: 2,083
  • 150906 Wurcer LNA 3c.jpg
    150906 Wurcer LNA 3c.jpg
    153.6 KB · Views: 2,064