Update on: High gain current amplifier has excess noise

Well, two or more cascaded opamps have a lot more gain-bandwidth than one.

John

Isn't that a voltage amplifier with a 10GOhm feedback resistor? That is almost a wide open amp, no wonder noise is high. What is your required bandwidth of your design? Put a filter on it to cut the bandwidth down, noise is proportional to BW. What is the noise spec of your amp? Get a better amp.

Be careful doing that, Alan, what was the parasitic capacitance of your "another large resistor"? For example, a 10G resistor with 0.1pF of self capacitance just looks like a capacitor above 160Hz.

Have you remembered to include the effects of e_n-Cin noise? With 8nV and 80pF this noise swamps the Johnson noise of a 10G resistor above f = sqrt(4kT/R) / 2pi en Cin = 315Hz. You can improve this by changing to an opamp with lower e_n, such as an AD743, assuming you don't mind 150pA rather than 3pA of bias current. Or you can make a discrete JFET input stage. For example, I'm using Toshiba's 2sk170, which has under 1nV of noise, 8x better than your LT1793. Its gate leakage is less than 1pA for drain voltages under 8V.

Keep in mind that the e_n-Cin noise density increases as f, which means it looks flat above a single-pole rolloff (this compares to white noise, which only increases as sqrt f). With a 10G resistor your e-n-Cin noise dominates above 300Hz, and 0.1pF of uncorrected parasitic capacitance means your 10G would rolloff above 160Hz, so your signal would go down but your noise would be flat above 300Hz until the opamp's 4MHz bandwidth limit hits at 900Hz (see below).

Your modest f_T = 4MHz opamp has enough gain-bandwidth to extend the frequency response of an 80pF input node with 10G feedback resistor to about f = sqrt(f_T / 2pi Cin Rf) = 892Hz, assuming you can keep the 10G resistor's parasitic capacitance to under 0.018 pF, that is. Hah! Or you can use the R-C-R trick.

Alan, did you tell us your bandwidth and current-noise specs? What are you working on?

I used a 100Meg, and tested it against a Keithley 427 to make sure the freq response was flat.

Yes. And I also found out that for 10G, the LT1793 gave the lowest overall noise in the hundreds of Hz range.

Or you can

How does one make such a stage?

While testing the noise only, I didn't add any input capacitance, so the e_nCin noise wouldn't rise above the Johnson noise until at much higher frequencies. Therefore the noise was flat, but the freq response was still dropping with 1 pole, even with the RCR trick.

I tried and failed. See above. The freq response still falls off with

STM amp.

Never mind, I botched up my understanding of the composite amp. What resistor and cap values are you supposed to use inside the composite amp anyway?

BTW, I built a simple composite amp by inserting a 10x stage inside the feedback loop. It didn't prevent the roll-off, so I don't think the problem was with f_T