"Low" Noise Amp

Well, the 51 ohm series resistor isn't too helpful there, and it looks like your current mirror on the input probably needs more degeneration.

Cheers

Phil Hobbs

Among other things, Q8 is pouring noise directly into the input.

R8 adds a nanovolt/rthz on its own.

Seemingly not much. The alternative is a simple voltage divider, right? I tried that (replace Q8 collector with 150R to GND): same noise.

So at least, the mirror does not contribute excess noise beyond what the resistors do.

I quickly noticed the first iteration was noisy, with a fixed current mirror (R from +V to Q7) and no degeneration (R11/13). Besides transferring 100% supply noise into the summing junction (not that the supply is very noisy -- actually, it seems to be pretty quiet, I should measure it*), the lack of degeneration put full amplified transistor noise into it.

Well, no, the resistance adds with the source, for 1.4 nV/rtHz total. Is Q7-Q8 or Q4 responsible for the remaining 1.95, or something else entirely?

Also, so, now I have a different question for you: if 100R is 1.4 nV/rtHz, then /HOW/ is the wideband amp doing _1.3 nV/rtHz_ with that at the input already? Surely I must have measured it wrong.

I should add, I tried a noise simulation, but the models don't seem to have parameters. Also, I suspect Altium's noise simulation option is broken anyway, grumble...

Tim

Yeah, I'd like to run the bias from, like, -5V, but then I need a charge pump, and have to keep /that/ quiet, which seems like a losing proposition (I mean, it's already at /eight transistors/ ;-) ). What degeneration is there, definitely helps.

Alternately, I could cap-couple R5, and bleed in a little current (PNP CCS) to bias the base. I would like to avoid /four poles/ in this darned circuit though. (If I had a -5V supply, I'd just pull some bias through R8 so it can be DC coupled...)

Tim

How are you measuring it?

-- john, KE5FX

LT Spice seems to do a fair job with noise sims. A little weird, but apparently realistic.

Base current (shot) noise into the base circuit impedance is another nasty. Biasing the first transistor is tricky. A high value bias resistor and beta-biasing would be better than a low-resistance (high current noise) bias network; close a loop on the beta bias for DC stability.

Really, jfets are a lot easier to work with for 1 nV/rthz sort of gain.

Then your problem would be to make a very quiet current source!

I'd go with a big resistor from a big quiet positive voltage to bias the first transistor. Servo that.

If you put in all the noise sources. I recently had a problem with the series resistance of a photodiode being the dominant noise source.

Except for the 1/f noise, which starts at around 1 kHz for a JFET and 10 Hz or less for a good BJT. (And 30 MHz for a SKY65050 pHEMT.)

Cheers

Phil Hobbs

There are some 50-ohm wideband MMICS with astounding noise figures, under 1 dB, but their 1/f noise corner is probably high.

Maybe one could diplex a MMIC with an opamp or something.

Scope: 1mV/div, 5us/div, sample acq., 350MHz BW* Amp: 12V supply, input terminated, output to scope Gain is slightly over 10 (21dB), as measured separately. Assuming flat noise density, input referred noise is: e_n = sqrt(Vrms^2 - (noise floor)^2) / sqrt(BW) / Vgain

I measure 290uV under this condition (for the wideband amp).

Noise floor is 140uV.

For the 20MHz amp, scope is set to 2ms/div and 20MHz BW. I measure 36uV noise floor and 190uV noise.

Tim

I'm tempted to knock in a PN4392, or even J105, just to see what happens. But again, compensation, and the probable need of a negative supply to close the bias loop.

Tempted to change the CCS load to a gyrator. Puts a pole in the loop, but it's the same as the bias servo pole, so that's fine. Super easy for FETs, where I can just run at Idss and be done, but BJTs still have the problems of base current and hFE variation.

Tim

Well, it would. Can easily burn 5V on the emitter degeneration, no problemo. And it'd be a smaller current (~0.5mA vs. ~4mA), too.

Trouble is I've got that big small feedback resistor from output to base, and I don't want to add another fat ass electrolytic in the loop.

Tim

I've spent a fair amount of time trying to make a JFET/pHEMT diplexed front end, but it's hard to do since inductors really only work at low-Z. It might be possible to do something with a modified JFET diff pair sensing the input and output of a pHEMT stage--any lowish frequency stuff that didn't track would be pHEMT noise.

Dunno. Still working on it off and on.

Cheers

Phil Hobbs

If you replace Q8 and all that other stuff with a single resistor, base of Q4 to ground, the whole thing collapses to a Vbe multiplier sort of amp. The resistor might be, say, a few hundred ohms, so its Johnson noise won't be awful.

I was thinking you could AC couple into the MMIC and hang a fet diffamp across the coupling cap to capture the missing DC part and then mumble mumble.

Or just AC couple into the MMIC and DC amplify the input with a good opamp, ditto mumble. I don't think you'd need inductors.

Incidentally, few 50 ohm mmics are actually 50 ohms in. Usually lower. You can tune Zin by fiddling the output load. Some can be fooled into actually being a 50 ohm input.

That could be a niche product, a super low noise DC to 5 GHz amp in a box.

The lamented BF862 worked fine at Idss in a cascode. Noise below 1 nv and no bias worries at all.

Resistors have this nasty habit of being 300K noise sources. ;)

If you want to fix that, as John says you have to run much higher resistances from much higher voltages and then apply feedback to make it work the way you want.

Alternatively you can use all capacitive feedback and fix it up in the second stage.

Cheers

Phil Hobbs

You can still get them, so everyone should stock up. I have about 40k of them, split between the BF862 and CPH3910. I'm unlikely to need that many myself, but it would be good to be able to supply my consulting customers.

The BF862 is about a decibel noisier than the CPH3910 but runs much more happily at I_DSS, as you point out.

Cheers

Phil Hobbs

If current noise is the issue, bigger is better. For voltage noise, bigger is badder. This looks like a low impedance amp, so a big bias resistor should be best.

Use a jfet with a 100M resistor, gate to ground.