For sot-23 npn, how about BFT25, 0.3pF 5GHz but beta = 80.
But I'm still curious about your pnp choice. It wasn't H81 was it? There's gotta be a better choice than that.
BFQ149, BFT92, BFT93 and the W versions ... poor beta. Maybe the NE97833, NE97733-A, etc. Nah... OK, how about Intersil's awesome HFA3135, 7GHz, beta = 57.
That's a great part; Phil turned me on to it. It makes an amazing fA leakage diode, too, orders of magnitude lower leakage, and series R, and cost, than a PAD1. But it is awfully fast, so tends to oscillate.
And it is an NPN!
We have it in stock, and it was good enough for the application. I'm reluctant to add new parts to stock; we have too many already.
Some lurker could take over and optimize this circuit.
To make a really wideband current source, you need an inductor of some sort in the collector/drain, in which case the transistor need not be hyper-fast.
--
John Larkin Highland Technology, Inc
picosecond timing laser drivers and controllers
jlarkin att highlandtechnology dott com
http://www.highlandtechnology.com
Not really, that gives you a chunky impedance null at 1/(2*pi*sqrt(L*C)). It's better above the null, but worse in the immediate vicinity.
You can use a lossy inductor (FB, etc.), but that defeats the purpose of using any inductance at all; the impedance still humps, and it's little better than using a resistor. A resistor might be preferable, because it tends to cure such parasitics, making your circuit a bit cleaner, if not necessarily sharper/straighter.
There might be a way to responsibly design a filter, but the fundamental problem is you're trying to direct high frequency energy away from a capacitor, which already has a -20dB/dec. slope of wanting to short out said energy. You can't filter it more aggressively because you don't have -40dB/dec components. It's neither a fixed impedance filter, nor a one-side-infinite design.
I wonder if a diplexer type circuit might be applicable to these situations. Divert the high frequency energy into a known termination impedance (which might be a high but fixed impedance), while allowing lower energy to pass into the collector (and reflect off it naturally as it's supposed to).
You generally use a high-Z bead, possibly with a conical coil if really needed, and pick the LC resonant frequency to be low enough that the transistor's collector is still pretty high-Z.
Cheers
Phil Hobbs
--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC
Optics, Electro-optics, Photonics, Analog Electronics
160 North State Road #203
Briarcliff Manor NY 10510
hobbs at electrooptical dot net
http://electrooptical.net
--
John Larkin Highland Technology, Inc
picosecond timing laser drivers and controllers
jlarkin att highlandtechnology dott com
http://www.highlandtechnology.com
Well, excuse me while I recall about $20 million worth of products.
--
John Larkin Highland Technology, Inc
picosecond timing laser drivers and controllers
jlarkin att highlandtechnology dott com
http://www.highlandtechnology.com
I've generally gotten by with a bead, and sometimes a string of beads and q-killed inductors for extreme situations. I've always wanted to use the Piconics conical gadgets, but I haven't found an app that justifies them, so far.
Here's a PSPL wideband bias tee:
formatting link
formatting link
--
John Larkin Highland Technology, Inc
picosecond timing laser drivers and controllers
jlarkin att highlandtechnology dott com
http://www.highlandtechnology.com
--
John Larkin Highland Technology, Inc
picosecond timing laser drivers and controllers
jlarkin att highlandtechnology dott com
http://www.highlandtechnology.com
If it were a problem, you would've noticed it on the scope. Obviously it wasn't...
A little analysis would save you the trouble; likewise, posting generalities when the analysis does not support it is, at best, irresponsible.
Here's an exaggerated example: Schematic
formatting link
Breadboard
formatting link
Waveform (680pF Ct on 2" leads, inserted around the diodes)
formatting link
You'll notice it's doing precisely what your circuit is doing (well, astable rather than monostable I suppose), with much slower transistors and a much worse layout, which makes the effect easily visible at modest frequencies (I only had a 200MHz Tek 475, back then!). Your bandwidths and dimensions will still be proportional, so the same concerns apply.
Not shown in the schematic or breadboard, the waveform was measured with some inductance in series with the current sources.
formatting link
This was 7 years ago so I don't remember what exact value I was testing with. If the waveform is around 4.5MHz, then that's 50ns/div, and the ringing is around 15ns or 66MHz. 2N4401/3 plus JFET plus breadboarding suggest on the order of 40pF loading, or 0.14uH inductance.
Eh, stray on that poor layout might've been enough (ferrite rods/beads played around the circuit would show the problem areas, as I had noted).
The maximum frequency for that thing was around 33MHz (15ns per half cycle), at something like 20mA in the source/sinks. Hysteresis is 1.4V. Suggests Co = 214pF, but a lot of that is also going to be propagation delay in the comparator, suggesting the CCSs are much faster than it, despite their relatively massive capacitance.
FWIW, minimum frequency was something like 1Hz (with the 680pF cap), placing leakage in the nA. Not bad for the relatively large junctions. Though it was also cool in that old basement.
There's no suggestion of that. Of course there's a difference between "work ing" and "working as well as it could" and a place that typically spends tw o weeks on developing a new product would seem to have left some room for i mprovement in the products that it sells.
Maybe he's just sympathetic to people who can't get work for reasons which are difficult to reverse.
One tough situation is making a 16 ns ramp that's linear to a tenth of a per cent. That is NOT visible on a scope. Every diode junction nonlinear capacitance, every possibility of ringing, every TC matters. I don't use white plastic breadboards for stuff like that! I can't even breadboard ramps like that: I have to build the entire product, on a real PC board, and hope it will work.
Another nasty is running a linear ramp into two comparators, to make programmable delay+width down to around 100 ps. That becomes dominated by layout, not so much theory. I did one recently using an opamp + PNP closed-loop current source on that one, with a BCX71 (slow high-beta PNP) with a ferrite bead in its collector. The ramp is over before the BCX71 even notices. The current source essentially forces a precise current into the bead.
In this picosecond stuff, it's amazing how much instinct and experience matter. It's not all that hard to acquire. Just do it really wrong 5 or 6 times, then think it over.
Last few ramps that I've done didn't use conventional current sources.
One is just an RC, charging from +10 volts to make a 3 volt ramp in 20 ns. Math out the curvature.
Another, where the linearity had to be real, was a bootstrap ramp. I posted pics of that. It does have an opamp in the signal path, which adds a different mechanism for nonlinearity.
I haven't tried this yet:
formatting link
Might be interesting, but probably no better than the simpler bootstraps:
formatting link
The advantage of the first one is that the ramp slope is DAC programmable. It's harder to program the bootstrap, like with a trimpot or something.
--
John Larkin Highland Technology, Inc
picosecond timing laser drivers and controllers
jlarkin att highlandtechnology dott com
http://www.highlandtechnology.com
Many, many it looks like. Thanks for the drawings!
Do you have an estimate of the output capacitance of the ne3509m04 heterojunction FET? CEL doesn't say on the datasheet. BTW, Digi-Key says that part is going non-stocking when they run out, what's up with that?
I've had good results making incredibly-fast ramps using a fast op-amp in feedback-capacitor integrator mode, the summing junction eliminates common-mode capacitance non- linearity issues, etc. Resetting the capacitor is painful, but sometimes you only need the delta-V to be accurate.
At some point, you probably have to calibrate these things, either through RLC and nonlinear balancing voodoo (oh and don't forget the thermistors), or calibrate all the damn stuff in software with a whole lot of characterization.
Doing 0.1% easily questions the propagation of signals through the AD8009 alone, not to mention everything around it, and layout, and so on.
Any reason picking an op-amp over JFET or BJT? Seems rather ponderous to put a monster in there, even if it's a fast one. A BFT25 or whatever emitter follower will get BW one or two orders of magnitude higher (at a fraction of the price). If you don't like the tempco, you can cancel that with another, diode-strapped, or a PNP follower if you have the convenience of a complement. (Same goes for the switching diode, its drop can be bootstrapped.)
ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here.
All logos and trade names are the property of their respective owners.