Low noise current source

ish

s there

ething

ce

ctor of 2

ut the

there

ise of

-)

=A0 =A0 =A0 =A0 ...Jim Thompson

=A0 =A0 =A0 ...Jim Thompson

ves

=A0 =A0 =A0 ...Jim Thompson

=A0 =A0 ...Jim Thompson

=A0 | =A0 =A0mens =A0 =A0 |

=A0 | =A0 =A0 et =A0 =A0 =A0|

=A0|

=A0 =A0 =A0 |

My recollection is one of the gain stages of the LT1028 isn't very high gain, so the design is only semi-insane.

re

Basically, are you suggesting a BJT with a base voltage and an emitter resistor?

I have the reference very seriously filtered. Since it is a constant voltage, the time constant here can be very long.

I am using the LT1021 with a two pole filter. Unfortunately temperature drift looks a lot like low frequency noise to my application. I need it to hold still for minutes.

OK.

Yep. BJTs are really good, so above audio frequencies it's usually best to leave them alone to get on with the job.

If the emitter resistor drops N times kT/e and the base voltage is constant, ideally your output current noise is 20 log N dB below shot noise. This performance is limited mainly by the Johnson noise of the extrinsic base resistance Rbb' and the shot noise of the base current, so you want a high beta or else some scheme to return the base current to the collector. (Don't use a Darlington, or you'll be limited by the noise of the driver stage.) Parallelling smaller devices can help both beta and Rbb', and that will also distribute the dissipation among several emitter resistors.

Thermal drift will be a serious worry, but if you use a cascode you can reduce that a lot by reducing the active device's dissipation and keeping V_CE nearly constant. The cascode contributes its own base current shot noise but is otherwise completely benign.

There will still be a thermal settling time after you change the current. For small changes, biasing at the maximum power point helps, but it's hard to make it work over 0-500 mA!

In a laser noise canceller design, I once tried using a fairly clever feedback loop that adjusted V_CE to keep the device dissipation exactly constant. The dissipation was constant, all right, but unfortunately the current error caused by the Early effect was almost as big as the drift I was trying to get rid of!

Since you have an op amp looking after the DC performance, you might be able to make something like that work in this case. (Of course if you really need to go down to zero current, you'll need a fair amount of voltage headroom.) ;) You'll probably at least want to make sure that the total dissipation on the heat sink is constant--besides speeding up the thermal settling a lot, that'll help keep the temperature drifts of everything else in the box under control. Thermocouple offsets are a bear at this level.

Getting the current to be sub-Poissonian at the bottom of the range might need range switching, or else degeneration by diodes instead of a resistor. The good news is that diode-connected transistors work over the full output range, and have a lower noise temperature (Tj/2). The bad news is that you effectively only get kT/e worth of degeneration per Vbe of drop, so (taking the lower T_noise into account) they wind up taking about 15 times more voltage drop for the same noise reduction.

It's usually the low, low baseband where the trouble lies. Also watch out for the capacitor tempcos (see below).

The LT1021 is a nice part but I'm a bit worried by the low frequency noise spec--they use 0.1 Hz as the LF cutoff, vs 0.01 Hz for the LM329, and the noise is a bit worse even so, at least going by the plots. There are also the Intersil floating gate parts (they can source and sink current, so you can stack them to get higher SNR), and the expensive but very good LTZ1000. It's a pity that the LM329's TC is 50 ppm/K. Dirty old National Semi discontinued the LM399, which was the bee's knees for that sort of thing. (*)

If holding still during a measurement is more important than absolute accuracy, you can just charge up a huge metallized plastic capacitor periodically from your reference via an analog switch, and hang a good quality op amp buffer on it. Since the charge on the cap is fixed, that'll be limited by the tempco of the dielectric constant, so it isn't a complete solution either, but the thermal diffusivity of plastic is so very low that it takes a physically big cap a long time to drift much. Polyester and polypropylene have opposite TCs around room temperature, so you could get a first-order temperature compensation by putting two in parallel--probably good for a factor of 10 without trimming, once you've characterized them.

Temperature control will be your friend here. The one major second-order imponderable in the use of BJTs for this is the temperature coefficient of beta, and at some point it's just easier to keep T from varying than to keep dialling in more tweaks.

Cheers

Phil Hobbs

(*)I wish someone would build little temperature controlled modules with a few LM329s and a bunch of 0603 resistors and SOT23 transistors sitting on a 2-oz copper ground plane, and one of Jan's little PICs adjusting all the dissipations to keep the temperature fixed and the first and second order gradient terms zero. That would make an amazing reference and still be a lot cheaper than an LTZ1000.

Sounds like a good place for a transformer.

John

So perhaps: ------Load !/ C--! !\\e -+----B ! Vref RF bead !/ from battery ----()-----! ? !\\e A----+---/\\/\\--- GND

D A-----!+\\ ! ! >---+-+-/\\/\\---C --!-/ ! ! =3D=3D=3D Vref-/\\/\\-+---------+--/\\/\\---B

Holding the collector several EB drops above the emitter to reduce the power.

A circuit then uses the drop between (D) and (C) to measure the base drive of the upper transistor. Two times this can be added to the load current with by sinking it from the emitter of the upper transistor.

The measured base current can also be fed back to the Vref so that the battery lasts longer.

Maybe I need to make several of these. I can switch in the number needed for the current I am working at.

Off, 100mA to 500mA will also work for me.

Maybe I can use a big chunk of copper and servo its temperature to always be 50C.

Fortunately I have room so great big film capacitors are an option.

Notice that I am now suggesting a battery. Unfortunately, batteries have a bad tempco too. They are very low noise however so it may be worth the trouble they cause.

I am also thinking of a thermoelectric cooler on the reference. If I keep its temperature at lets say 10C, it will be at a fixed temperature without the increase in noise and shorter life that 40 or

I have already built a circuit like this once. It used a great big slab of aluminum on standoffs inside a box. The thermal time constant low pass filtered the temperature changes. I didn't switch electronically. I used a mechanical switch. Once the voltage was set, the switch was thrown. The capacitors weren't open circuited. A circuit with many meg held them to the reference voltage on the average. Basically:

Vref--/\\/\\---+---/\\/\\---BufRef ! ! -----+---/\\/\\----+----! >---BufRef -!-/ ! ! !/ ! ! =3D=3D=3D BufRef-/\\/\\---+---\\/\\/--- ! GND

=A0Since the charge on the cap is fixed,

This could be interesting:

regards, Gerhard

ote:

Yes, very worth reading. It is likely that the low frequency rise in the noise is the effect of things like temperature.

I don't see any reason that a battery that will hold a charge for 6 months wouldn't have a noise level well below:

because the fall off is very slow at the fully charged end of the charge vs voltage curve.

That's an interesting paper as far as voltage noise is concerned, but it's completely cracked when it comes to current noise. You can't measure the current noise of a low impedance source by putting an 825 ohm resistor in series with it, because the resistor itself will suppress the current noise enormously.

Cheers

Phil Hobbs

Yes, the current part of the paper makes little sense at all. It is like they needed to fill a few more pages so they added it.

I am still casting about for a better current source topology that somehow magically ends up with very little noise. Ideas involving super conductors etc have a little too much complexity.

I think it will end up with some very large film capacitors in it.