Simulated inductor with 100:1 current range

designing a transistor noise tester to stand in for the late lamented HP 44

George H.

switching. That sort of range isn't hard to do with a pot, and I can prog ram two BJT current sources (for the source and drain) from the same pot by using an ordinary phase splitter circuit. No problems there.

n sticking point is just how quiet the bias current sources have to be when you're testing JFETs.

f 2/(3 g_M). To look at the relative influence of the bias current source in the output noise, we can start with full shot noise and calculate the re sulting noise figure as

about 13 (11 dB), which is the pits.

But how much?

in parallel with a shot noise current source, in parallel with a resistance 1/g_M = 26mV/I_C.

noise current loops around the g_M path and half goes into the external cir cuit. Of course you get the Johnson noise of the external resistor, so a b it more algebra leads to the conclusion that the resulting current noise go es down by 3 dB when you drop 52 mV across the external resistor. It's dow n to a quarter at 200 mV, and so on. 200 mV thus gets us within 1 dB of th e Johnson noise. For a measuring instrument, one would like to be within 0 .5 dB at worst, which needs 400 mV. Not awful.

e even to get to a 3 dB NF, we need like 13 x 400 mV, or a bit over 5V. St ill not horrible.

requirement of 500V of supply headroom, and some gross amount of power dis sipation.

ng to mess up the tracking of the source and drain, and probably cause a lo t of voltage swing at the drain.

e it quieter without needing such a high voltage supply.

plier used back to front (going in the emitter and coming out the collector ). It'll need breakpoints too, but that has to be done without adding exce ss noise from the breakpoints themselves.

Go swimming? ...Jim Thompson

Yup, that's the one. Win and Paul sang its praises in their earlier editions, but even theirs died.

Cheers

Phil Hobbs

designing a transistor noise tester to stand in for the late lamented HP 44

switching. That sort of range isn't hard to do with a pot, and I can prog ram two BJT current sources (for the source and drain) from the same pot by using an ordinary phase splitter circuit. No problems there.

n sticking point is just how quiet the bias current sources have to be when you're testing JFETs.

f 2/(3 g_M). To look at the relative influence of the bias current source in the output noise, we can start with full shot noise and calculate the re sulting noise figure as

about 13 (11 dB), which is the pits.

But how much?

in parallel with a shot noise current source, in parallel with a resistance 1/g_M = 26mV/I_C.

noise current loops around the g_M path and half goes into the external cir cuit. Of course you get the Johnson noise of the external resistor, so a b it more algebra leads to the conclusion that the resulting current noise go es down by 3 dB when you drop 52 mV across the external resistor. It's dow n to a quarter at 200 mV, and so on. 200 mV thus gets us within 1 dB of th e Johnson noise. For a measuring instrument, one would like to be within 0 .5 dB at worst, which needs 400 mV. Not awful.

e even to get to a 3 dB NF, we need like 13 x 400 mV, or a bit over 5V. St ill not horrible.

requirement of 500V of supply headroom, and some gross amount of power dis sipation.

ng to mess up the tracking of the source and drain, and probably cause a lo t of voltage swing at the drain.

e it quieter without needing such a high voltage supply.

plier used back to front (going in the emitter and coming out the collector ). It'll need breakpoints too, but that has to be done without adding exce ss noise from the breakpoints themselves.

Huh, Well not obvious to me till you said it. Not your application, but you sometimes would like to have a big inductor in the feedback loop of a TIA to kill the gain at low frequency... (say when there is a lot of ambient light.) Would your backwards cap mult. work there too?

George H.

Inductors (and I mean real inductors) are wonderful; they have impedance without Johnson noise.

Yes, as long as you add emitter degeneration. If you just connect the cap from emitter to base, you get a very large effective inductance, but you al so get full shot noise. A lot of the time that's probably fine, but not us ually in what I'm doing.

Cheers

Phil Hobbs

Yup. Unfortunately at the low current end I'd need about 900 henrys.

Cheers

Phil Hobbs

Well, JL says he gets his best ideas in the shower, so maybe that would be better still.

I might try playing around with correlation techniques. Normally that does n't work well with currents, but there may be some tricks available, e.g. s plitting the drain resistor into two branches that have different dependenc es on the FET noise and the current source noise. I'd like to avoid subtra cting nearly-equal quantities, though, for obvious reasons.

I bet the HP guys did something like that.

Cheers

Phil Hobbs

Well, JL says he gets his best ideas in the shower, so maybe that would be better still.

I might try playing around with correlation techniques. Normally that does n't work well with currents, but there may be some tricks available, e.g. s plitting the drain resistor into two branches that have different dependenc es on the FET noise and the current source noise. I'd like to avoid subtra cting nearly-equal quantities, though, for obvious reasons.

I bet the HP guys did something like that.

Cheers

Phil Hobbs

These guys claim to have one.

Thanks. Their website won't accept my email address, but I'll email them.

Cheers

Phil Hobbs

Good luck.

If they don't really have stock, let me know and I'll send out a request to some T&M dealers I do work for.

Thanks, Jay.

I don't often buy gear from dealers because of the markup. My TDS 694 is about the only thing I've got from a dealer in the last 5 years or so. However, it would be nice to have one of these, so I might make an exception if the price isn't exorbitant.

I'll probably build my homebrew one anyway, because it's an unusually interesting design problem--I hadn't thought about how an un-degenerated JFET is a sub-Poissonian current source. That'll come in handy sometime.

And figuring out how to use some correlation technique on a current rather than a voltage would be very useful too.

Cheers

Phil Hobbs

I prefer the QTL units to the HP. They are easy to repair and much more common. The HP IIRC uses multiple current sources (25?) to set bias current thereby reducing in**2 by 25.

ore

I don't know about QTL. What's the model number?

Thanks

Phil Hobbs

more

Quan-Tech (also atlanta-tech?) - the 2173/2181 or 5173 are pretty much indu stry standards. The 2173 controls the transistor bias and overall gain (wi th pilot tone etc.) and outputs the noise to the 2181 which is just a filte r bank (10,100,1k,10k and 100k) with a moving coil meter for each frequency . So you could just stick the 2173 output into your 35665. The 5173 also h andles opamps with plug in adapters, but you don't have access to the raw a mplified noise. I will have the 2173 schematic on file if you want it.

So, an instrument amp feeding an integrator feeding a current source (LM13700 will do, if the current is really small). You can get the henries, but probably will lose on noise that way.

Forget the 4470A. For a while mine failed on certain ranges, and I contemplated repairing it. But then we chose a better approach, and made a relatively simple test set, that provided far better information, and worked to lower noise levels. We used it to get the excellent BJT and JFET data in our Chapter 8 tables and graphs: to well under 0.1nV/rt-Hz, and measuring 1/f breakpoints below 20Hz. A simple modification (add one resistor) would allow even lower 1/f data.

For the full circuit, with comments, see Figure 8.92. There's an adjustment to set Vce / Vds (e.g., 2.5V), and a collector resistor sets the operating current.

The feedback loop is slightly unstable at low frequencies, which can damage the 1/f data. This can likely be fixed by adding a resistor in series with C3. But we were simply too exhausted getting and processing huge masses data to try it.

What, no bling?

Why was all the aesthetically nice kit designed before the age of the "product designer"?