trimpot guts

What's his name from Technion has written a comprehensive treatment of bandsplitting and combining in the context of broadband video encompassing DC for instrumentation applications, but the exact reference "excapes" me just now...

In our present case, the attenuator input is a low-impedance signal from a 2 GHz opamp, and the load is the 1 pF input of another opamp. So an "RF" type attenuator isn't appropriate and just wastes signal.

Good: I have to design something, as opposed to just googling another Betty Crocker "solution" and slamming a brick on a board.

Nearly content-free, probably penned by a journalism major...

"In contrast, any MESFET- or PHEMT-based step attenuators need dc blocks, and the frequency response cannot go down to low frequencies."

Mine can.

We don't design easy stuff for a couple of reasons, including that we enjoy technical challenges, and our business model requires that we charge a lot for things. This is fun: I get to design a time-domain multi-GHz attenuator from parts. We have NE34018's in stock, so I'll see what can be done with them later today. The 3509's are just too hot here.

John

Sounds like something Lorch or Merrimac can cook up in a few minutes.

If you do manage to remember, please post the reference; I'd love to take a look.

Thanks,

---Joel

ROFL! Who wrote that? Did the author ever build something from discretes?

So did mine. Remained in production for half a decade until a new circuit didn't need them anymore.

Try the BF998. Cheap and might work.

--
Regards, Joerg

http://www.analogconsultants.com/

Not in a few minutes. And it would be in a big metal box with SMA's and hermetic feedthrus. And it would cost roughly 100x what I want to spend. And it would be characterized in the frequency domain.

John

Ummm..maybe Lorch but not Merrimac. I once got a custom high power low frequency band limited directional coupler from them, spent less than 5 minutes with the design engineer as he worked through his CAD and solidified the design, it was a lightning fast telecon, he was like a robot, three week delivery at $350 IIRC. The couplers were heirloom quality last forever type of very high quality products. Lorch was pretty good with some voltage variable octave bandwidth analog phase shifters for a non-standard IF frequency, they were impressive, and again the design engineer got right to it, had it solid in 10 minutes of telecon. If you're in a hurry, places like these will send you a parts kit you assemble yourself, did this with M/A-Com with some RF attenuators too.

OK,

The 3314G, used as a pure pot,

opamp output----------+ | | p o

What are the values of those Rs and pot?

Isn't that kind of non-linear?

have micro, will compensate?

i was kinda expecting to see a g-d resistor.....

Cheers Terry

100 or maybe 200 ohms for the pot. The output node should probably Thevenin to around 50 or so. R1 depends on the adjustment range. Since the photodiodes are reasonably repeatable, +-25% or so gain trim should be enough, so R1 is ballpark R2||R3. We'll gave to measure a few of the photodiodes before we gring out actul numbers.

The InGaAs longwave diodes run around 0.6 a/w, so I'd need a "gain" of

0.33 nom. We found a GaAs 850 nm detector, much faster than silicon, that runs about 0.4 a/w, and that needs a gain here of 0.5 nom. We cna possible come up with a single set of values to handle both diodes, maybe not, but there's latitude to play, once it's on a real pc board and we have a reasonable number of photodiodes to test.

But getting 1 GHz through a cheap trimpot is cool.

The signal path should be pretty good to at least +-200 mV swing into the second (G=10) amp. Gain control is an s-curve on gate voltage, but not too radical. The active gate control range is 0 to about -0.8 volts, but we'd need less that that to do a +-25% gain trim, probably.

So we have, to freak out PH,

+10 | | k photodiode a | | | +-----[Gain10]---[variable atten]---[Gain10]---[comparator] | | L | R, 50 ohms maybe | | gnd

where we'll use the pot for the 1 GHz, dumb version, and the phemt if we do a 2GHz, microprocessor-based box.

John

You could servo out the gate curve with a 2nd FET plus opamp. OTOH uC horsepower and DAC bits are cheap. Although we had to budget over four bucks for a 12bit octal in a similar application :-(

[...]

--
Regards, Joerg

http://www.analogconsultants.com/

It would probably mess up the frequency flatness, and would create dc errors, I think. Since my signal is always positive (light isn't bipolar!), I have some linearity advantage over a bipolar-swing attenuator.

John

Hi John,

I figured you would have a good reason. I just recalled that article by Jim Williams, in one of his books.....

Cheers Terry

FIFTY CENTS per dac? Poor baby!

John

UM6K1N "Two 2SK3018 transistors in a single UMT package." Very nice. Where do you get your Rohm mosfets?

Well, around a dime per channel:

Call me Uncle Scrooge but my first one had a servo and 8 bits. It cost a lot less because this was a multi-channel system.

--
Regards, Joerg

http://www.analogconsultants.com/

At Digikey. This one can be had for around 10c a pop in reels. Their order number would be UM6K1NTNTR-ND or UM6K1NTNCT-ND if you only want a few devices.

--
Regards, Joerg

http://www.analogconsultants.com/

The cheapest dac is a uP or fpga pin, pwm or delta-sigma. That could cost 2 cents, for the r-c output filter.

John