diodes are such rotten diodes

You mean some sort of artifical ground? Everyone would be using cheap opamps there :-)

Intermediate rails are usually bypassed, and lots of opamps don't like that. It takes a few more parts to have a regular opamp drive a big capacitive load. The LM8261 doesn't care.

Are there other c-load opamps? The 8261 is a bit of overkill sometimes.

John

Yes, needs to be isolated. If you don't want to spring for a transistor pair that means your virtual ground becomes a bit soft.

They are all expensive but audio amps can be nice here. Many live happily with 4ohms or less resistive. They are really cheap because of the masses that want to be blasted with tchk .. tchk .. *BOOM*, the louder the better.

Although, nowadays I'd consider a sync buck or class-D amp.

One section of an LM339 will make a nice sink-only rail splitter that is good for a few mA.

The outputs of rail-rail opamps are really current sources that are servoed to make the voltage. It is too bad that they don't make the internal node that sets the output current available on the singles in the SO-8. It would be a very nice way to over compensate to allow large capacitive loads to be driven.

The LM8261 compensation is effectively Miller caps on the two output transistors. That makes a dominant pole right at the output. Adding external load caps just slows down that pole. If you hang a cap on most opamps, ones where the compensation is buried further inside, it creates a 2nd order loop that is unstable.

Lots of opamps and LDOs would benefit from this trick.

John

The LM8261 looks like a good part for the fake ground. What I like is the loop gain isn't very high. In many applications, you are better off with less gain since that often (but not necessarily) means better stability. Those designers that insist on rolling their own regulators should look at this op amp.

I wonder with the LM8261 if there is a, for lack of a better phrase, "sour spot" where you have killed the internal Miller multiplication, but having rolled off the high frequency gain enough with the external cap.

The TI rail splitter (2426) has the advantage of lower Iq.

On regular opamps you can try how small of an output series resistor you can get away with. Plus a small capacitor from VIN- directly to OUT, of course. With cheap boom box amps that resistor can be surprisingly small.

I used to like the lM6361--RIP.

Cheers

Phil Hobbs

In article , " snipped-for-privacy@sushi.com" writes

I tried the TI rail splitters last year. I had +5V and -5V rails and I needed to use one IC which could only take max 5V supply total, and needed its input near mid rail. (Actually this was a poor solution which I abandoned, but that's not my point.) I created +/-2.5V rails with two rail splitters. This caused problems, firstly because you could get lockup if one powered up before the other - the other's output was pulled beyond 0V through the IC I was trying to power and latched off; and there seemed to be an oscillation as if they were fighting each other (though that may have been another problem with another IC!)

So I decided to avoid rail splitters after that, perhaps others have found similar problems and that is why they are not widely discussed. I eventually found another approach, which did not need +/-2.5V.

I think it's stable for any C. And probably for any C+ESR.

John

eap

ike

Then LM8261 is probably the way to go then unless you need low power, The Natl design certainly looks good, especially that it is push/pull.

I got some samples of the Central CMPD6001S dual SOT-23 diodes. Through a local rep, since their registration/password thing is so absurd.

I measured about 50 fA leakage at -5 volts. I say "about" because this is sorta hard to measure... 5 mV across a 100G resistor. That's a lot better than most diodes at around 5 na. Pricing is 0.099 by the reel, as compared to 0.085 for the BAV199, which I haven't measured yet.

Looks like 1.8 pF at zero bias, pretty nice part.

John

And just how does one go about measuring 5mV across 100G? Seems like an open circuit that might not collect a static charge, to me ;)

Grant.

Here's my pA parts tester:

ftp://jjlarkin.lmi.net/99A260A1.JPG

ftp://jjlarkin.lmi.net/99A260A3.JPG

ftp://jjlarkin.lmi.net/99S260A.JPG

I plugged a 100G resistor into the Z2 slot, and put the diode in as Z1.

With a 100M or 1G resistor as Z2, nanoamps or tens of pA are easy to measure. When you get to fA, everything has to be very well shielded, and you have to wait a long time for the inherent time constants to settle down. Without a solid shield, just moving around in the vicinity will induce huge offsets.

Those damned RatShack banana terminals are conductive as hell, so I had to machine the hole and add the lexan *after* it was all built. Grrrrr.

John

Oh, people would kill for those ceramic solder posts.

So how do you make the San Francisco fog disappear? Tape the windows shut and ask all your employees not to breathe out for a few minutes?

Old Tek scope things. I got a bunch at the Foothill Flea Market. But the critical node is not on a strip... all air.

I'm not seeing a lot of leakage across the lexan, or even on the dual pomona banana plugs where I mount the test resistors. My measurement uncertainty seems to be about 10 or 20 fA. My 1T resistor is an 0805, soldered across a pomona plug! 1 Tohm and 1 pF is a 1 second time constant, so measurements can get slow.

Some transistor c-b junctions, as diodes, are logarithmic from 10s of fA to 10s of mA, and leak ballpark 20 fA reverse-biased.

For serious current measurement, you could let an unknown current charge a polystyrene cap for some time (minutes, hours, days) and then measure the voltage on the cap. 1 fA would charge a 1 nF cap at 1 uv/sec, tons of signal.

John

Oh, you mean those things they sell with high grade plastics loaded with fillers of what ever was hanging around at the time or extrusion ? :)