blast from past

Why the high price? Can't you just add a DC offset that gets adjusted according to the duty cycle?

There are some things where integration just doesn't make any sense. Either because the performance ain't there, the heat is too hellacious or the market is too small.

That is not a looney idea. Not at all. We used to do this all the time in the days of the alumina hybrid. Days long gone, but hybrids are making a great comeback in California. Except that now they have four wheels.

We've got just that: PNP inside a loop, ferrite bead in the collector. It looks like a second bead or L would be good, possibly with a parallel damping resistor. The spiral inductors are a last resort: expensive and hard to handle.

(Hmmm, I should investigate the opamp loop a bit closer. Maybe it's doing something.)

The classic wideband bias tee has a really good small inductor first, often a Piconics, then progressively bigger L's, last one maybe a mH toroid, with resistors across the Ls to kill resonances. Finding mid-value Ls with high SRFs is the challenge. Ironically, they turn out to be old, axial-lead parts, like my low-C transistors.

John

Considering the effects of voltage, you really would need a UHF device like this to keep Co low. 2907 more than doubles Co below 10V.

It's one of the reasons HV parts won't go, in spite of low book ratings (..at >20V).

A choke might be in order here.....?

RL

A split-path thing is what one would probably have to do to make a generalized, say, true-DC-to-18 GHz amplifier. One or more of the horrific Hittite or TQ distributed amps would be the AC gain path. Alongside that would be an opamp-based DC path, split at the input and merged at the output. The output merging is tricky, keeping the crossover flat and the output impedance constant; visualize trimpots. We'll try that some day. But it's big, expensive, and a huge power hog.

Wideband coupling caps turn out to not be a big issue. The regular Mouser parts work fine.

Yup, so parts-on-boards circuit designers aren't quite obsolete. Yet.

John

I have a friend who has a hybrid (circuit, not car) business. All he does is some custom military things and a little electro-optical stuff; surface mount has pretty much killed off hybrids. A lot of the Apex amps are now just pc boards, when their stuff used to be mostly hybrids.

John

How many channels?

Well, something doesn't make sense. On a good day, you might get 5nS rise and fall from a 2N2905/2N2907. And they certainly won't do 1GHz square waves. Plus the pnp is isolated from the fast edges with some inductance.

So you must have something else in the circuit that gives the fast edges, and holds the amplitude until the pnp can catch up.

If this is true, why do you need low output capacitance in the pnp? Or why do you need the the pnp at all?

If the source impedance is 50 ohms, that means a 10V swing. With reference to the questions above, how do you get that?

Also, does anyone do plain ECL anymore, or is it all PECL?

The Hittite parts were mentioned before you described the requirements. But they might be interesting if you wanted to go to 2 or 3GHz or even higher.

I did a split thing last year. Ok, only into the hundreds of MHz but it sure is not big, expensive or a power hog. Four per card, a whole rack full of those and it barely gets warm to the touch. The DC to LF path is servoed.

If they could only figure out how to build a decent search engine. That's what keeps pulling me back to Digikey although prices there are often higher. Not having a good search engine is costing Mouser beaucoup bucks.

According to one of my professors I and all the other analog board-level guys should be obsolete since 1990 :-)

There may be some small problems with compounding. The MMIC's have an internal 50 ohm termination. If the input is AC-coupled, the termination would start to kick in around the lower 3dB frequency and start to disappear below that.

So the DC path has to supply a 50 ohm load that starts to disappear above the 3dB frequency. A series inductor might do the trick, but I'd be concerned about damping and ringing, and amplitude and phase bumps at the transfer.

Another small problem is the prop delay of the MMIC should be vastly lower than the op amp supplying the DC path. If there were several MMIC's in series to supply the gain, each would have its own AC coupling caps, which would give a funny overall rolloff.

It might be advantageous to use the DC-coupled MMIC version so only one cap is needed between stages. This might help flatten the high frequency response and simplify the problems areound the transition frequency a bit.

As you mentioned elsewhere, the DC path would have to be the exact inverse to keep the overall response flat in frequency and phase. But the DC path has longer prop delay, so it might be tricky to get everything working correctly. Do you think it could be done?

AFAICT the prop delay of a reasonably zippy MMIC is miniscule compared to the required DC/LF path.

You can use feedback on the DC/LF path.

I am partially guilty here as well. Later in life I designed what was then called "poor man's hybrids". Essentially just glorified PCB modules with some laser trim areas on there.

Yes, there are still things to be built from resistors and transistors and stuff. And you can move up the abstraction stack, putting processors and FPGAs and synthesizers and switching power supplies, things that used to fill a rack, on one board too.

The thing I wish somebody would integrate are good continuous-time lowpass filters, preferable with SPI control over pole/zero locations. The LC filter after a DDS chip can be five times the size of the chip.

John

The switch device is a gaasfet. The PNP is just a current source so the fet has something to switch.

No, 5 volts behind 50 ohms, 2.5 volts loaded. But that's still 5 volts into a high-z load, like a fet or a cmos gate. We did one output stage that swings 12 volts, 6 loaded, but it won't work at a GHz.

For most parts, ECL and PECL are the same thing; you just connect the part to different supply rails. We do both, sometimes on the same board.

John

John Larkin a écrit :

Have a look at the nxp BFQ252 (PNP) and BFQ262 (NPN).

That will only happen if a market situation arises where for example every iPhone absolutely needs one. It's just like adjustable delay lines. No market, so I always have to roll my own. I wish the SD5400 was still around. I mean for a reasonable price.

The BFQ252's fast, with only 2.5pF Ccb @ Vcb = -10v.

Alas, NXP says it's discontinued.

Best, James Arthur

And Digikey doesn't have it. Not a good sign. Now you could have waited until after the holiday weekend with such news, couldn't ya?

Not to worry, Fred's basic idea of NXP is still good.

This SOT-223 is marginal at 1w, 15v, 40ºC/W, but 5GHz and 1.8pF are attractive. Maybe John can figure out how to cut the heat.

Digikey lists both that and the comparable BFQ149 as non-stock, but orderable, roughly $0.40 @ qty 1k.

Happy holiday!

James Arthur

How do you do the delay lines? I've seen lumped-element lines done with inductors and (lots of) varicaps (HP 8133A) but that looks tedious, and I don't quite understand the impedance situation.

John