diodes are such rotten diodes

John Larkin a écrit :

BAS116, BAV199, BAW156, BAW170

^^^^^^

Bingo! Drop-in replacement, and cheap.

Wow, that's nice. 10 pA typ at 60C!

Thanks, I'll get some.

I tried a couple of high-voltage dual diodes, but they all leaked ballpark 5 nA at room temp. I got to thinking, who makes a diode the same way that people make transistors, a real planar/lithographed/passivated process? The only answer I came up with was "varicap." Sure enough, a Skyworks SMV1234 leaks about 1.8 pA.

John

But those will be really close to the cliff with your +/-15V requirement and then one sunny day ... *phut*

Also, they set you back 30 cents. For that kind of money you can get two Girl Scout cookies of the "thin mints" kind (my favorite) _and_ a BAV199 :-)

How about BAS116 or in the extreme case, SSTDPAD1 from Linear Integrated Systems.

Thanks. The 116 is nice, just about like the BAV199 that Joerg suggested. The 199 is a dual, which will drop into my BAV99 footprint.

The SSTPADs are actually jfets. They have a lot of series channel resistance, ballpark 2K, so aren't good as clamps. And they really leak around a pA. The c-b junction of an NPN transistor, BAT25 or BCX70 maybe, leaks less, costs less, and has way lower series R.

But I need a dual SOT23, and can tolerate a few nA over temperature, so the BAV199 is ideal.

John

Is there some reason you wouldn't clamp at the input of the filter rather than at the op amp? I'm assuming leakage at the input wouldn't be a problem since if it was, the Salen Key filter wouldn't work so well. That is, the filter assumes it is driven by a low impedance.

A lot of engineers get stuck in the Salen Key mode, but if you learn leap frog design, you can come up with filters that use a "stockable" cap, that is always use a 0.1uF for example in every filter stage, then use that cap for all your products. Cost you more op amps though. There are other solutions besides leap front that give this flexibility. Generally the one op amp per pole designs do this. You stock less parts and can get a price break on the caps. I examined a lot of modems back in the day that used this kind of scheme.

The input to the filter comes from an INA154 powered from +-17 volts, and it can potentially rail. So I can't clamp that. The input of the first opamp in the SK filter is conveniently about 7K ohms downstream, an ideal place to clamp. It's tricky, because the input range of the AD7699 ADC is 0 to 4.096, so I have to clamp just below 0 volts, to avoid jamming the ADC ESD diodes too hard, but I don't want to add drift or nonlinearity.

ftp://jjlarkin.lmi.net/22S490B_sh22.pdf

This is basically an anti-aliasing/noise filter. We have lots of digital filtering downstream of the ADC, and customers will usually be down-filtering considerably, so the AC response of this filter isn't critical.

One nice thing about S-K is that the DC gain is 1.000 and doesn't depend on component values. I use the TI FilterPro software, which comes up with standard value caps. If my transient or frequency response is a little off, my customers won't notice. If my DC gain drifts with temperature, they sure will. So I want the filter gain to not depend on resistor TCs.

I sometimes design software/FPGA versions of S-K and state-variable filters. They have the same advantage as the analog SK, namely unity gain per section. The coefficients and gains don't get insane like a butterfly tends to do.

John

It could be clamped after R413, though 17V to ground would be about

I'm stating the obvious here, but the power supply going to the clamping diodes needs to be able to sink current. Many supplies regulate well when sourcing current, but are just fine being yanked high when pulled by a diode. This is a common problem in latchup testing components. That is, the person doing the test forgets to load the power supply with a resistor so that it can sink the latchup test current.

On a sunny day (Thu, 10 Jun 2010 08:43:02 -0700) it happened John Larkin wrote in :

But clamping there changes the filter characteristics no? The caps will get pushed to strange values, how does it come out of a clamped situation? Would it not be better to add a separate opamp stage before this to do the clamping?

And it's an 0603! The additional current would just ride the voltage further up the diode curve and make the clamp that much worse. I want an ideal diode and I want it now.

I'm currently using an LM8261 opamp to make the clamp rails.

If I add a small resistor, 50 ohms maybe, from the last opamp into the ADC, I can clamp to ground and +5, same place as now, and not stuff too much current into the ADC esd diodes. Using BAV199s, that should work.

John

Only if the customer has railed the signal, past my 5% headroom. At that point, my obligations as regards signal quality cease.

situation?

clamping?

The board is packed with parts as is! Any clamping is going to do weird things to a signal; you just get to decide what sort of weird.

Actually, clamping at this point in the filter is good, because it passively attenuates spikes before things get nonlinear.

John

What about a current limiter? Signal goes into the AC side of a FWB. Bridge is biased with CCS's from + to -. Signal goes out the other AC pin, into a load resistor. When current through the load resistor exceeds CCS, the output clips.

Downsides include lots of diodes in the path (potentially requiring matched offsets and tempcos) and doing nothing about capacitive coupling (though for a filter, that's probably not a problem).

Do something with op-amps. Op-amps and diodes make great ideal diodes.

Tim

Yeah, lots of parts and potentially lots of DC error. One LSB of the ADC is 78 microvolts, which is why the diode leakage was so annoying.

Lots of parts again. The board is currently rev B, about to be C, and it's very dense, so I can't add much stuff.

Rev A had fatal pinout errors on the ADCs, impossible to kluge, so I didn't learn much from it. Rev B works, and we have customers playing with them and writing drivers and such, but the clamps are scary as regards offset and linearity over temperature. So, rev C.

John

aks

ke

. I

Just make sure the 5V supply can't be "lifted" by the diode.

A >$1 opamp for clamping a rail? ... ... That's like using Hennessy Cognac as rubbing alcohol. Us regular folk will grudgingly spring for a TL431, at the most, and then agonize over the 6-7 cents it costs along with the two resistors.

BTW, it wasn't me who suggested the BAV199. I would have but Monsieur Fred beat me by three hours :-)

No problem: it's an entire VME crate!

John

Hennesey? Martell is better, in my humble opinion. But I recently discovered Ron Zacapa 23, and the Cordon Bleu is gathering dust.

There's only one LM8261 on the board now, to make the 2.048 volt Vcm supply for the ADCs. I love these opamps. We just figured out that the parts cost on this board will be about $750. 16 16-bit ADCs, a mess of LT1124s (*three* dollar opamps!), a $50 FPGA, 33 relays, a bazillion Susumu thinfilm resistors... The bare board alone will cost about $42 each at qty 100.

Just how many beers am I supposed to buy?

John

Most of my clients would have me flogged if I ran up a tab like that :-)

I suppose Fred would prefer du vin rouge, I'd go with a Russian Ale at Zeitgeist :-)