Really triangular triangles

But cancellation leaves you vulnerable to mismatch, whereas air is pretty stable--only the CTE of the shield would matter, which will be in the 10 ppm/K range. (There would be a second-order effect due to the TC of the FR4 changing the fringing fields a bit, but that'll be much less than the TC of the FR4 itself.)

Cheers

Phil Hobbs

Not in New York. There people only trust air when they can see it :-)

You'd have to be able to swim in it for it to get up to 900 ppm/K!

Cheers

Phil Hobbs

From the way they drive, I didn't know they could see ;-) ...Jim Thompson

-- | James E.Thompson, CTO | mens | | Analog Innovations, Inc. | et | | Analog/Mixed-Signal ASIC's and Discrete Systems | manus | | Phoenix, Arizona 85048 Skype: Contacts Only | | | Voice:(480)460-2350 Fax: Available upon request | Brass Rat | | E-mail Icon at

The Coilcraft ones are pretty nice that way--they have a U-shaped shroud and metal alloy end caps, so they work with pick and place. I have a RFQ in--it looks like they haven't decided what to charge for them.

Cheers

Phil Hobbs

If you want really fast edge rates then you're not going to beat working wi= th a square wave (or rectangular in your case) drive of a fast buffer with = an RC-analog feedback. Your input source could be an HC Schmitt considering= the edge speed-up after the input attenuation, should be easy to get sub-n= anosecond edge rates: Please view in a fixed-width font such as Courier.

. . . . _ . /| . .--||---. . | / | . | | . | | |\ . >---+-[19R]-+----------|+\ C2 _ _ . | | > ---+--||---, _| |_| |_ . _ _ | .--|-/ | | . _| |_| |_ [R] | |/ C1 | | . | | --- [R2] . | | --- | . --- | | | . | | --- . | | . '----------+ . | . | . [R1] . | . R1C1=3DR2C2 | . --- . . .

=20

But how much mismatch would you expect? Hmm, it'll depend on geometry too, since copper is different from FR4. Could engineer some amazing hair-pullers with the conspiracy between linear and flexural CTE there. ;-)

Tim

On May 11, 4:11=A0pm, Phil Hobbs wrote:

Try this. The triangular wave appears at the inverting input to the LT1016 (labelled "out" i=3Don the schematic).

Version 4 SHEET 1 4648 1292 WIRE -224 -64 -384 -64 WIRE -16 -64 -224 -64 WIRE 208 -64 -16 -64 WIRE 352 -64 208 -64 WIRE 960 -64 352 -64 WIRE 1376 -64 960 -64 WIRE 1472 -64 1376 -64 WIRE -16 -32 -16 -64 WIRE 208 -32 208 -64 WIRE -224 16 -224 -64 WIRE 1376 48 1376 -64 WIRE -16 80 -16 48 WIRE 1136 80 -16 80 WIRE -384 128 -384 -64 WIRE 352 128 352 -64 WIRE -16 160 -16 80 WIRE 48 160 -16 160 WIRE 320 160 48 160 WIRE 576 160 400 160 WIRE 1376 160 1376 128 WIRE 1520 160 1376 160 WIRE 208 192 208 48 WIRE 320 192 208 192 WIRE 960 208 960 -64 WIRE 1136 208 1136 80 WIRE -16 224 -16 160 WIRE 1520 240 1520 160 WIRE 208 256 208 192 WIRE 480 256 208 256 WIRE 576 256 576 160 WIRE 576 256 560 256 WIRE 720 256 576 256 WIRE 896 256 800 256 WIRE 1248 256 1200 256 WIRE 1376 256 1376 160 WIRE 1376 256 1328 256 WIRE 1376 272 1376 256 WIRE 208 288 208 256 WIRE 1040 304 960 304 WIRE 1136 304 1040 304 WIRE -384 400 -384 208 WIRE -304 400 -384 400 WIRE -224 400 -224 80 WIRE -224 400 -304 400 WIRE -192 400 -224 400 WIRE -16 400 -16 288 WIRE -16 400 -192 400 WIRE 208 400 208 368 WIRE 208 400 -16 400 WIRE 352 400 352 224 WIRE 352 400 208 400 WIRE 368 400 368 224 WIRE 368 400 352 400 WIRE 1376 400 1376 352 WIRE 1376 400 368 400 WIRE 1520 400 1520 304 WIRE 1520 400 1376 400 WIRE -304 448 -304 400 WIRE 1040 448 1040 304 WIRE -384 512 -384 400 WIRE -192 512 -192 400 WIRE -384 704 -384 592 WIRE -192 704 -192 576 WIRE -192 704 -384 704 WIRE 336 704 336 224 WIRE 336 704 -192 704 WIRE 1040 704 1040 528 WIRE 1040 704 336 704 WIRE 1344 704 1040 704 FLAG -304 448 0 FLAG 48 160 out SYMBOL Comparators\\LT1016 352 112 R0 SYMATTR InstName U1 SYMBOL voltage -384 112 R0 WINDOW 123 0 0 Left 2 WINDOW 39 0 0 Left 2 SYMATTR InstName V1 SYMATTR Value 5 SYMBOL cap -32 224 R0 SYMATTR InstName C1 SYMATTR Value 4.7n SYMBOL res -32 -48 R0 SYMATTR InstName R1 SYMATTR Value 100k SYMBOL voltage -384 496 R0 WINDOW 123 0 0 Left 2 WINDOW 39 0 0 Left 2 SYMATTR InstName V2 SYMATTR Value 5 SYMBOL res 192 -48 R0 SYMATTR InstName R3 SYMATTR Value 3.3k SYMBOL res 192 272 R0 SYMATTR InstName R4 SYMATTR Value 2.2k SYMBOL res 576 240 R90 WINDOW 0 0 56 VBottom 2 WINDOW 3 32 56 VTop 2 SYMATTR InstName R5 SYMATTR Value 3.3k SYMBOL cap -240 16 R0 SYMATTR InstName C2 SYMATTR Value 100n SYMBOL cap -208 512 R0 SYMATTR InstName C3 SYMATTR Value 100n SYMBOL npn 896 208 R0 SYMATTR InstName Q1 SYMATTR Value BFR92A SYMBOL npn 1200 208 M0 SYMATTR InstName Q2 SYMATTR Value BFR92A SYMBOL res 1024 432 R0 SYMATTR InstName R2 SYMATTR Value 7.5k SYMBOL res 1344 240 R90 WINDOW 0 0 56 VBottom 2 WINDOW 3 32 56 VTop 2 SYMATTR InstName R6 SYMATTR Value 33 SYMBOL res 816 240 R90 WINDOW 0 0 56 VBottom 2 WINDOW 3 32 56 VTop 2 SYMATTR InstName R7 SYMATTR Value 33 SYMBOL res 1360 32 R0 SYMATTR InstName R8 SYMATTR Value 3.6k SYMBOL res 1360 256 R0 SYMATTR InstName R9 SYMATTR Value 1.3k SYMBOL cap 1504 240 R0 SYMATTR InstName C4 SYMATTR Value 100n TEXT -16 504 Left 2 !.tran 0 1m 1u startup TEXT 1272 504 Left 2 !.model BFR92A NPN(IS=3D0.1213E-15 VAF=3D30 BF=3D94.73 IKF=3D0.46227 XTB=3D0 BR=3D10.729 CJC=3D946.47E-15 CJE=3D10.416E-15 TR=3D1.2744E-9 TF=3D26.796E-12 ITF=3D0.0044601 VTF=3D0.32861 XTF=3D0.3817 RB=3D14.998 RC=3D0.13793 RE=3D0.29088 Vceo=3D15 Icrating=3D4m mfg=3DInfineo= n)

-- Bill Sloman, Nijmegen

I know you have your heart set on a triangle wave.... but scope and scope probes have allowed users to tweak probe padding compensation with square waves for a long time. This is a tweak that un-EE- sophisticated users do all the time as part of many procedures. Same problem? Or different?

Tim.

Air's TC of epsilon is roughly 0.2 ppm/K.

Cheers

Phil Hobbs

There are a few people posting to this thread (Fred B, Bill S and Tim Shoppa) that I can see in Google Groups on my phone but not on Supernews, which is weird.

Just so you don't think I'm ignoring you guys,

(a) (Bill S) >You can make a triangular wave with just a comparator - the comparator >output pin generates a square wave, but you generate a triangular wave >at the inverting input, and the corners will be as good as the >transition times at the output - about 250psec with the MC100EP116.

I really want a nice predictable accurate flat temperature stable high and low slope on the triangle, so that I get good rectangular pulses at the output. That puts exponentials and bipolar output stages pretty well out of court, e.g. the LT1016. Thanks though.

(b) (Tim S) >I know you have your heart set on a triangle wave.... but scope and >scope probes have allowed users to tweak probe padding compensation >with square waves for a long time. This is a tweak that un-EE- >sophisticated users do all the time as part of many procedures. Same >problem? Or different?

The idea is to make it just like tweaking up a scope probe, for exactly the reason you give. The circuit problem is how to make that happen.

It really needs to be an asymmetric triangle, because of the unusual application. This amp is something pretty special--being shot noise limited at 1 nA in a 100 MHz bandwidth is a pretty good parlour trick, if you run the numbers. The shot noise limit is SNR (dB) = 10 log(N/(2B)) where N is the number of electrons per second.

A nanoamp in 5 ns is 31 electrons, so the shot noise is about 5-1/2 electrons, which is pretty good going in a built-up circuit. (Easy in a CCD, but you can't put a wire on a CCD pixel.) So assuming it works as designed, I ought to have bragging rights for awhile.

The key to doing this is having really low input capacitance, under 1 pF and preferably more like 0.3 pF, which is very hard to do with packaged parts. Because the input capacitance is so small, I don't want to add to it, and therefore the right approach is to generate an accurate triangle wave and differentiate it with a really really small coupling capacitance. (The amp is supposed to be current-sensitive anyway.) The capacitance of an additional pad is much too large.

(c) (Fred B) >If you want really fast edge rates then you're not going to beat > working with a square wave (or rectangular in your case) drive of a > fast buffer with an RC-analog feedback. Your input source could be an > HC Schmitt considering the edge speed-up after the input attenuation, > should be easy to get sub-nanosecond edge rates:

I need really triangular triangles, for the reasons above. Square waves are a good deal easier, I agree.

I think the best candidate so far is: a 1/2-inch square Laird shield for coupling to the input node, connected to the integration capacitor of a tri wave generator. The integration cap will actually be something like 10 470 pF @ 100V 0805 NPO caps between the shield can and the ground plane, which will double as RF bypasses and keep the ringing down. I don't need the high voltage, of course, but it'll keep the nonlinearity down to a very low level and doesn't cost that much extra.

Driving the integration cap are a fixed 1 mA current source consisting of a fast op amp, a floating reference, two resistors, and a very wideband bias choke (8 uH) in series with its output, just to crispen up the corners of the ramps,

and

a steerable 1.1 mA current sink with another op amp and another choke, steered by a couple of the same pHEMTs as the front end (to keep the BOM simple) with 1k-ish gate resistors (per JL) to keep them stable. Those will be driven by an ECL comparator with a bit of positive feedback, looking at a buffered copy of the voltage on the shield can.

The result should be a 1V p-p ramp with +200 kV/s-ish rise and 20 kV/s-ish fall. With a coupling capacitance of about 0.05 pF, that's a

The idea is for the customer to be able to hook up a scope, push one button, and see a pulse response pretty enough to bring a tear to the eye. ;)

Thanks again,

Phil Hobbs

Really weird. I am on a paid server (news.individual.de) and I can see Fred's posts but not Bill's and not Tim Shoppa's. At least not in this thread.

You mean the whole Laird shield is whomping? Make sure that all this is in a metal enclosure and stuff won't leak out via wires. Else the Federales might be waltzing in some day, worst case at a client.

I sort of doubt that a 1/2 inch square shield with a 19 kHz rectangle wave on it is going to interest them.

Cheers

Phil Hobbs

ou

The circuit I posted defined the slope of the 5usec ramp with a constant current derived from a long-tailed pair of BFR92 transistors.

It's going to be very linear. The slope depends on the current through the tail. You've got 5V to play with so we could put a current source in the tail with a precision resistor, and N-FET and an op amp to make it really well defined. The current gain of the BFR92 ranges from a minimum of 65 to the typical 90 up to 135, so you could lose as much as 1.5% of your tail current through the base junction. Bipolar transistor gain rises with temperature, so this proportion would be temperature dependent.

If you want something more stable, it might be worth looking at making the long-tailed pair with two BFRT92/BFR92 complementary Darlingtons; you'd probably need to jack up the current a bit (and increase the timing capacitor in proportion) to keep the lower-current part of the complementary pair tolerably fast.

The 50usec recovery is exponential, with roughly 30uA through R1. You could make it a constant current source - in the good old days you could buy selected FETs for use as constant current diodes, but today it looks as if you'd have to buy a Fairchild J201 and put a resistor between source and gate.

About 30k would get you of the order of 30uA with a "typical" gate-to- source cut-off voltage, but you'd have to do select on test to get anything reliable or usefully predictable.

A PNP current mirror would be easier, but - as usual - you have to worry about the base current and the Early effect.

You could use a P-channel FET and a rail-to-rail op amp to set up precise current source. Farnell stock the MMBFJ177 which looks as if it could be made to work

How much you'd have to do to get the degree of predictablity and flatness that you need rather depends on what numbers you want to attach to "predictable", "flat" and "temperature independent". There are well know solutions for every level of accuracy - as you raise your standards, the components become more numerous and expensive.

E-mail me if you don't want to be explicit in public.

-- Bill Sloman, Nijmegen

Depends on how high the harmonics go and how strong they are. I've seen numerous cases where folks failed class B because a little connector can was floating. If you deliberately "light it up" it'll get even worse. Just meant as a friendly hint from a guy who has seen lots of grief when it comes to EMC :-)

Cases like a little one-line display, only six characters long or so, maybe 1/2 square-inch. They were sure it won't matter not to ground that too well. Until the day of reckoning at the EMC lab. Sometimes it's tough to bite the tongue and not say "told ya so" but as a consultant one has to keep such thoughts to oneself. The sad part was not so much the wasted $2k for the lab but it meant a re-layout.

Understood. But this one is for a research lab, and the little triangle wave will only be on for 2 minutes while someone is adjusting the TCs. If I wind up selling something of the same sort elsewhere, it'll definitely be in a metal box. (Naturally I don't sell the same circuit to more than one customer, but I've learned a lot doing this one.)

Cheers

Phil Hobbs

=20

=20

=20

=20

=20

=20

=20

=20

The square wave circuit does the differentiation for you in feedback. C2 an= d R2 represent your pad capacitance and input resistance. Then by choosing = a feedback R1 C1 with identical time constant, the square wave voltages acr= oss R2 and R1 will be equal. It's a way of putting the test waveform on yo= ur circuit input without tapping it with anything less than ultra-high Z at= your frequencies of interest. So why bother with a parts intensive fussy triangle generator, which is jus= t an approximation anyway, when you can put the end result you want there d= irectly, a rectangular waveform.=20

=20

=20

=20

=20

=20

=20

That's a very low number.

My reference says 2ppm/K for dry air, and more typically 5ppm/K (plus variations for RH and pressure, of course).

Best regards, Spehro Pefhany

Thanks--I was thinking of the optical response, but of course there are a bunch of resonances below that. Still a lot lower than the probable uncertainty of a 900 ppm/K epoxy.

Cheers

Phil Hobbs