blast from past

Unfortunately that is dictated by what is available. If it has to be servoed I try to get dual-FETs, else PIN diodes, sometimes a dual-gate RF FET as a controlled resistor. Usually in an L-R setup. I rarely need >500psec vernier but for really long ones there is no choice and it has to be LC with varicaps. The impedance situation is ugly but if you have to delay a clock without introducing much phase noise that's still ok.

The challenge is not so much the delay itself but to come up with a scheme that auto-calibrates. You usually can't calibrate during production because all this is rather drifty. Sometimes I can sneak in pilot tones. Other times they give me time slots for pulses, for example frame reset in imaging systems or part of the pulse trundle-out time on ultrasound systems. That's often duked out over a beer down at the saloon :-)

The only thing that I have (so far) never seen work in a noise-sensitive scenario are silicon delay lines. I designed a few out because they had the noise behavior of Niagara Falls, plus they were freaking expensive boutique parts.

I used some Maxim silicon delay lines once. Jitter was huge and linear on delay.

Micrel has some 10-bit ecl digital delay lines that look pretty good, but expensive (well, $12 or so) and vast power hogs.

John

Cool. I mean, hot. Well, good, anyhow.

Cheers, James Arthur

Designing out silicon delay lines is one of the reasons why I get along with the accounting folks at clients really well :-)

Sounds like you are all bikers. Real bikers often say that a motorcycle that is well broken in must have those blue rings around the exhaust manifolds :-)

I probably wouldn't run that thing at 1W.

Performance often requires carefully pushing parts. If I get a schottky diode that's rated at 2 volts reverse, and it doesn't leak much up til 7 or so, I may use it in a circuit that needs 4. Some gaasfets are rated at 6 volts but don't get into trouble until 24, so...

Part of taking chances like this is testing parts to destruction to know how far we are from trouble. Some parts can't make their own datasheet abs max limits, some can do a lot better.

John

... or at least some lots of them can.

Ok, I do that as well. Such as in avalanching transistors or tolerating a certain amount of electromigration on disposables that only live a few hours. But when it comes to heat I am very careful. Seen too much grief from stuff that ran borderline red hot.

The gaasfet switch pulls down. That will easily give 150pS fall times.

The problem is with the fast risetime. The 2N2905/2N2907 won't do it.

The PNP is isolated with a ferrite bead and perhaps an inductor. How can it pull up fast enough to give 150pS risetime?

So you need something else pulling up to get 150ps risetime. It also has to hold the level constant until the PNP can take over.

But if you have this already, why do you need the PNP?

That won't give 150ps risetime. The timing will depend on stray capacitance on the interconnect and the input pin. And what kind of fets work with a 5V swing?

Your 0 to 2.5V swing ignores ECL and 5V PECL. So what kind of devices do you plan on testing?

If it is a limited set, why not just pad the driver resistively to get the logic levels you want?

You mentioned the current is regulated with an op amp. With no load, the PNP current source will saturate. So you won't get 150ps risetimes.

Also, how do you control the logic one level?

That's already needed to maintain the DC level. I was talking about the problems matching the amplitude and phase through the transition frequency.

At the very least, the group delay would decrease rapidly as the input frequency increased through the transistion region.

The PNP is a nice current source, except for 8-10 pF of collector capacitance.

The L and bead are in series with the collector, isolating Cc from the load.

It works. I don't think I can explain it any more.

If I can deliver a 5-volt swing at a 50-ohm source impedance, and the customer elects to hang a lot of capacitance on it at his end, there's nothing I can do about that.

Tons of them.

It can do 5V pecl if the customer terminates to anything from +2.5 to

It's hard to program resistors. I wish it weren't.

John

You'll have to make sure the feedback work up to that transition point.

I have been using 2N2905's since late 1960's. I have never seen one with an epoxy belly. I'm sure that never existed. For example:

1. An epoxy base would not have met the power dissipation spec for the

1. The can is electrically connected to the collector. That would have been difficult to do with epoxy.

2. There is little reason to put a metal cap on an epoxy base. Make the whole thing epoxy.

1. A metal can with an epoxy base would probably not meet the JAN requirements for hermetic sealing.

The only difference between a TO5 and TO39 is the length of the leads. Here's a quote: The TO39 is easily substituted for the TO5, with the only difference being the length of the leads.

TO39 have shorter leads that TO5 packages. Since the excess lead is always cut off, either TO39 or TO5 will do the same job.

TO5 : 1.25" lead length TO39 : 0.75" lead length

Page 42,

You are getting 150ps risetime from a 2N2905?

With an inductor and bead in series with the collector?

With ~10pf collector and other stray capacitance to ground?

And 100 ohms to ground?

This I gotta see. Do you have any photos you can post?

PS - What is the value of the inductor?

You sure go to a lot of effort to be a pain in the ass.

John

Seems that you either can't understand what I'm doing, or you are calling me a liar. Either way, you're getting tedious.

John

It's easy to have the bandpass of a fast ac-coupled forward gain path, and the bandwidth of an opamp-based DC servo loop, overlap by a few decades of frequency, 10KHz to 1 MHz or something like that. So there's no transition problem.

John

Just trying to understand what you are saying.

I modeled this in LTspice. The series inductor is critical. That's what stores the energy needed to get the fast risetime. The 2N2905 collector voltage goes up much slower.

But now you have to damp the energy in the inductor, which is resonating with the stray capacitance. This is what gives the long recovery you mentioned earlier. That will make it difficult to work with different frequencies or various pulse widths.

I tried different values of damping resistance and series inductance. There doesn't seem to be any satisfactory way to get rid of the resonance and still keep the fast rise.

But I think there's a lot simpler way to do what you are trying to accomplish.