Capacitors on reels--any tendency for values to cluster?

The parts are bulk (funneled) before test and taping.

Don't buy 5% from anyone offering 2% and expect a normal distribution.

RL

Interesting. Lemme guess: a bimodal distribution? -5..-2, +2..+5?

Clifford Heath.

Or all 2%.

Do you put these boards through some sort of test jig in production?

If so, could you make it electronically adjust the resistors at test and store the settings in EEPROM? You wouldn't need a proper digital pot, just a few steps of coarse adjustment (with some analog switches or even tristated microcontroller pins) to get from 5% to 2%, though it would be tempting to make it much better. For low volumes you could do one of those schemes where an operator cuts off a resistor on some units.

You can get capacitor arrays with NP0 dielectric - I wonder whether the matching within each unit is better than between separate ordinary MLCCs

- though the datasheet doesn't say and it wouldn't solve your problem anyway.

I'd expect that to be more likely. You can't afford much test time for a part that sells for a fraction of a cent.

I'll have to get a few reels and see, I suppose.

Cheers

Phil Hobbs

Sharp cutoff? That's going to be tricky at 7 poles. I like passive LC filters when feasible; they are less twitchy. Might you make your filter trimmable somehow? That could be fun.

Pity that switched-cap filters are/were so awful. Seems like someone could make a thinfilm all-analog programmable filter, but I guess it's not worth the trouble. The monolithic ceramic filters are all up at outrageous frequencies, not suited for DDS filtering or most of the real-world signals we work with. All the effort is going into narrowband wireless where the money is.

You could assign one of your lab lackeys to measure some caps here and there on a reel or two, and report back here.

We did order a reel of truly custom-brewed caps from Capax, 3.3 pF N4700, to temperature compensate our instant-start LC oscillators. That worked great. Maybe they would make you a reel of close-matched parts.

Turns out that in the hundreds of picofarads we can get 1% caps for reasonably cheap.

The filters are 12-dB transition Gaussians, made with LT1260 triple CFAs and a THS4631 output stage. I want really blameless performance--less than 0.5% overshoot, 1% bandwidth accuracy, 4:1 shape factor, ideally <

10 nV noise, good linearity.

The amps all have 1 kV/us slew rates, which helps a lot when you have no control over what ugly thing somebody's going to connect to it.

The idea is to be able to characterize photoreceivers for noise using a true-RMS meter rather than a scope FFT or spectrum analyzer, and to have a choice of bandwidths.

Simon is doing a nice box with four of them: 10 kHz, 100 kHz, 1 MHz, 5 MHz.

We really don't want to do any tweaking if we can possibly help it. Going to 10 MHz or higher would probably need LCs, at least for the higher-Q sections. The current design is three Sallen-Key sections with equal or nearly equal resistors, and one RC pole on the first stage.

Cheers

Phil Hobbs

To do optical measurement to that sort of precision, what do you use for the light source?

Here's my photodiode calibrator for the ill-fated LAM project.

One of two models of LED source, based on either a 7-ns LED ($2) or a

2-ns LED ($50). For simple stuff I usually hang a LED barefoot on a Highland P400 DDG, then dork the pulse height to get the right brightness.

Our LED source boxes are coming out Real Soon Now, with a BNC, a wall wart, mounting flanges, laser-inscribed case, perspulex running boards, ion drive, et cetera et cetera. ;)

Inside, they're a 74AC14 driving one of the magic fast LEDs, with an AP2205 adjustable LDO providing a stable VDD. Subnanosecond edges, lots of drive, cheap like borscht. That way we get everything out of the LEDs.

The AP2205 is a nice part--an adjustable small LDO with 36V max input,

2% accuracy, and _built-in polarity protection_, all for 15 cents. I may switch my allegiance from the venerable LP2951.

Nice pastel green color!

Cheers

Phil Hobbs

Back in the 1970s we had several hundreds 10% 1K TH carbon resistors. I measured a couple hundreds and crudely plotted the bins (years before "spreadsheet" software was all the rage.) Sure enough, the distribution had peaks at roughly +- 10%, and steps at ~5%. Just as everyone had been saying since the ?1950s?

So, buy a too-high value, and then discard all the +5% capacitors, to create a single gaussian peak? Heh!

OOoo! Build a highspeed machine which rapidly measures a whole spool, then mechanically extracts all the off-value components. Then next, SELL THE MACHINES. (For a much higher price, also offer the machine which re-spools SMT components, where the originating spool has many positions unfilled. Hmmm, maybe such products already exist?)

That looks like a nice part (the adjustable startup time might be useful), but I don't see anything about reverse polarity protection in the datasheet, only reverse current protection.

There is a new product announcement for the part which touts reverse battery protection, but the datsheet is lacking such information.

cu Michael

Yeah, I see that. I don't have any more info either--when they come in I'll try it out.

Cheers

Phil Hobbs