Voltage Ref. shopping question

Well there are some REF195 and 192's in DIP 8. (which I like.) GH

ADR421 (jfet sort of bandgap) is about the best around.

but not cheap.

Some of the LM4040 shunt parts are pretty good.

Yikes, two bucks in reels.

I use quite a few LP2951s (380 mV dropout, 100 mA, 40 cents in 100s). Their untrimmed accuracy isn't stellar, but they're pretty stable.

Cheers

Phil Hobbs

Thanks, I don't really need the best. I do want low dropout... Run the ref close to the rail. I did think of an LM4040, which is in stock.

Say I used a 5V LM4040 as a 'zener' in the negative 5 volt level shifter from a week ago. (negative logic) It works great... a lot faster than I expected.. I hardly needed the 'speed up' cap in parallel. so thanks for that too. :^)

George H.

It's tricky to run a shunt reg close to the supply rail. Just a resistor can result in a huge spread of currents. A more complex current limiter is hardly worth the effort.

A beta-limited PNP is cool, if only because it freaks so many people out.

Electronics is fun.

Price is no object :^) I think I've been using them since the stone age.

OK thanks, for this I don't need accuracy.

George H.

I've had real good results with the LM4140. We use the 2.5 V version, but found that the 1/f noise gets much worse at higher input voltage, so we run it at about 4 V input to get the best performance. They also have

Jon

Metal film / thin film resistors plus an adjustable reference. Thick films are horrible.

Cheers

Phil Hobbs

I've been using the LM4128, because it comes in a sot-23-5 package, has 0.1% accuracy and costs $2. Only 200mV dropout.

Thanks, excess noise at higher V, maybe some avalanche? an easy guess.

George H.

No, thank you. :^) There are way too many refs on DK, I end up ordering the list by number in stock, figuring the herd mind must know something.

I'm a man still attracted to a DIP pac. (at least for proto's)

George H.

Yes, likely. We were going for absolute minimum 1/f noise, and used an OP706 and the LM4140 after some other parts didn't do well enough. The last fix was to drop the supply voltage to the LM4140, and we've been using this circuit for over a decade with excellent results.

Jon

Maybe more power dissipation, so more sensitivity to air currents.

Sometimes putting a little cover over an opamp or something greatly reduces LF noise.

Crystal oscillators, too. A potting-shell cover can make a $2 XO behave like a $20 XO.

John Larkin wrote

Yes

but do not stop there

and not even there

I've convinced myself that the best insulation to put in a small box is air. There's not much convection in a small tight box. Without convection, most anything conducts heat more than air.

John Larkin wrote

Yes, exactly what is in the small black box, some 'tronics and air. Logs over 6 years show the temperature always stayed within +- 1 degree C. Maybe better, but that is the resolution of the detector display. The temperature control loop software was done experimentally in an afternoon..

For low noise and good stability, though, a stirred oil bath (fluorinert or silicone oil if you can afford it) and high convection rate is good, too. As long as temperatures are steady and unchanging, either convection maximum or minimum can be beneficial. You just don't want temperature gradients to change, usually, but if thermocouple effects and low DC matters, the stirred bath wins.

black_box_IMG_3868.GIF

You need to work out the Rayleigh number.

If it is less than 600 there isn't much convection, and if it is less than

It rises as the cube of the characteristic length of the box. I worked it o ut for the (small) reaction vessel I used in my Ph.D, project and it came o ut at 7,500, so your small box can't be all that big - my reaction vessel w as about 30mm in diameter about 100mm long (a bit over an inch in diameter and about four inches long).

Aerogel comes very close to non-convecting air - it is 99.8% air.

It's what you ought to be putting into all but your smallest boxes.

That's not a reasonable thing to put on a PC board.

When we were in the NMR gradient driver business (before Agilent killed off their NMR division) we built the critical parts (current shunt, DAC, error amp) on an aluminum block with closed-loop heater.

The custom-stamped, annealed manganin current shunt was embedded in the block, under the PC board. There were some interesting eddy-current effects.

We did a couple of special OCXOs that way too.

but nowadays incredibly good OCXOs and synthesizer chips are cheap.