HEMT

signals?

In each chain, there appears to be three RF amplifier stages without any selectivity, followed by microstrip bandpass filter (apparently for only filtering out the image frequencies) feeding a simple diode mixer. Apparently a common DRO in the middle.

Just wondering, how this is going to survive any strong local signals that might be present, especially in the future, with more crowding on the RF spectrum.

signals?

I suppose this thing to be living in the focal point of a parabolic reflector. There shouldn't be all that much crowding if it has a very small field of view, and the signal strength would be quite predictable.

Jeroen Belleman

On a sunny day (Mon, 14 Nov 2011 06:40:34 +0200) it happened snipped-for-privacy@downunder.com wrote in :

signals?

I think there is selectivity, I see little stubs and stripline inductors. There may be capacitors too (PCB as insulation). Wonder what PCB material that is?

Well my LNB is over ten years old, and it has seen very strong spurious signals, for example a several watt CB antenna a few meters away, and never a problem. It is sensitive to sparks, those cause bit-errors. Never opened it, but if it ever fails I will look.

On a sunny day (Mon, 14 Nov 2011 09:29:29 +0100) it happened Jeroen Belleman wrote in :

signals?

And good screening from the back by the dish.

"John Larkin" wrote in message news: snipped-for-privacy@4ax.com...

Ferrite beads make good resistors at most frequencies above "DC". Although when you start talking GHz, you have to worry about pad capacitance and such, which the smaller and HF type chips do a fine job with.

This one time I made a 20-stage LC delay line with ferrite beads. Even though they individually aren't very lossy in the MHz, the rise time was about a microsecond -- it really piles up after a few!

I've always wanted to come up with some sort of project that uses a shitload of ferrite beads for something dramatic, like a delay line, or a switching regulator, or a shock line, but I'm afraid the region of energy storage is just too small to make much more than heat with them.

Yep, keeping in mind to suitably terminate everything. Quarter wave structures are very handy, too -- tons of them on that satellite reciever picture I found. They've got a narrow (high impedance already) trace coming off the microstrip, which is "grounded" at the bias end with a fat (low impedance) quarter wave strip, then finally the RLC bias whatever stuff the amplifiers require for DC operation.

RF structures only work at narrow band frequencies, of course. Although the resulting bandwidth may be sufficiently wide to count as "wideband". With a Q of 100 at 20GHz, you get 200MHz bandwidth, or a good 400Mbps of raw data.

It was even more true back in the day. Tubes are essentially ideal current sinks in parallel with an internal resistance (fairly small for triodes, fairly high for pentodes; occasionally negative for some tetrodes) and a vacuum capacitor. Because the transconductance is so small (~2mmho for 12AX7 up to 60mmho for the bigger sweep tubes) and the capacitance large (2~30pF), it's hard to make anything tubetronic from DC to over 10MHz. The best Tektronix did was 85MHz with massive distributed amplifiers.

But because the capacitance is constant and ideal, tuned amplifiers easily reached UHF before underlying physics (transit time, bunching, lead inductance) took over. The same can't be said for transistors, which took decades to reach fT's in the UHF. Further improvements pushed triodes into the 10s of GHz (for use only inside cavities), when physics finally forced grids to be replaced with internal cavities, fields, resonators and transmission lines.

Some totally awesome devices came from that era though.

Not to say some totally awesome devices aren't coming out of this era, you just can't see them because all the magic is written in micrometers of semiconductor. Aside from the boards they are placed on anyway.

Tim

It's a challenge to make a really hi-z wideband inductor... you need a string of inductors and beads, with shunt damping resistors. Pity the Piconics things are expensive and hard to handle.

ftp://jjlarkin.lmi.net/tee.zip

The beads we use are generally a few uH gross inductance, but it never occurred to me to make a delay line from them!

That's what they are designed for, to make heat!

Sine wave stuff! Bah! Humbug!

Yup, 585. What a beast. I think it used better toobs, frame-grid things, 6DJ8 or some such.

Phemts are amazing. Their Gm/Co figure of merit, ballpark 1 S/pF, would make an old Tek designer weep in frustration.

John

"John Larkin" wrote in message news: snipped-for-privacy@4ax.com...

Totally! On the other hand, having only a baggie of 2N3904s and -6's to breadboard with has forced me to breadboard good circuits from the start. These days I lament when I am unable to achieve under 20ns edges on the solderless breadboard. (Who needs deadbug, anyway?) :)

'Tuther day I made an ECL clock divider from a stock circuit (which is one of the most beautifully symmetric circuits, as T flip-flops tend to be!). Didn't have a fast function generator handy to test its ultimate clock rate, but it got ~5ns edges, and that as seen from a 100MHz scope.

I think I've now decided the limiting factor on 2N3904s is capacitance. There's only so much you can do with the poor buggers -- higher voltage burns 'em up; higher current doesn't get you any more slew rate due to charge storage; low voltage swings make bypassing and feedthrough more critical -- the input clock pushes the *output* up before the transistor can pull down. Interesting lessons.

On a marginally related subject, I've created a discrete MOSFET gate driver which runs faster (~4ns), and delivers more current, than any IC available on the market. So it doesn't use 2N3904s, just the next step up (MPSH10 or so).

Incidentially, you "lied" about MOSFETs -- I've pushed 2N7002s faster than they can pull. Five nanoseconds into the gate and it just sits there for the next ten. :) Fortunately, there are improved replacements available, as Zetex keeps their gate spreading resistances lower than anyone else.

Speaking^HRambling of FETs, have you played with GaN or SiC power FETs (switching, not RF)? They're tempting, but pricey (and, in the case of bare GaN dies, hard to solder and heatsink).

Tim

Things are still relatively easy at 200 ps. Below 100, it starts to get difficult.

If you're interested in fast stuff, get a Tek 11801 and a sampling head or two from ebay. Wonderful machines, for a few per cent of the original cost.

Bipolars switch slow, even the 45 GHz SiGe parts. Phemts are magic.

How much gate swing? If 6 or maybe 6.5 is enough, a TinyLogic triple buffer, NL37WZ16 maybe, is a brute, switching in around 600 ps. Parallel as many as you like!

I've put 50 volts into 50 ohms in around 1 ns with 2N7002's, through a transformer yet. Just whack the gates hard. This gidget outputs 100 volts into 50 ohms in about 1 ns, using some higher-voltage SOT-23s.

We played with a bunch of the Nytronex GaN parts. In fact, we blow up almost $1000 worth one afternoon. They claim they've fixed the problem.

I'd love to find an application for them. They behave like high-voltage phemts, which technically I guess they are. The SiC parts are more like tubes, in that they need big gate swings to turn them on and off. You can drive the GaN parts straight from ECL. Like most RF parts, the DC specs usually suck for these exotics; you're really lucky to get a transfer curve, so you have to experiment with them for pulse use.

John

"John Larkin" wrote in message news: snipped-for-privacy@4ax.com...

Double whammy of traditional nonlinear junction capacitance plus charge storage, I suppose.

Follower? That'll mainly work because feedforward is to your advantage and any sluggishness on the gate's part just forces it on more. Not working fast enough? Push more Vgs. It's like an I Love Lucy skit. I wouldn't be surprised if your transient Vgs is 30V, if only for a fraction of a nanosecond.

Trouble is you need the voltage to begin with, and to slew, you'll end up with almost as much current capacity in the gain stage. Maybe not the continuou-pulsed capacity that 50 ohms requires (1-2A), but I'm guessing it's up there (~100mA).

If you want to do the same thing common-source, you have to amp the hell out of the gate, at which point you've burned about as much power getting it to turn on as you're getting out of it. Not always a bad thing for switching, since you can still get more voltage or current than the input, and it will keep drawing current until you turn it off

No biggie, a simple mistake like not bolting down a #000 litz cable on the right tab, and *pop* goes the IGBT brick. I've seen it before. I can't believe they don't use desat protection, even on the big units. Pretty soon, I get to put my circuit in there and stop that madness.

I don't think they are pseudomorphic or high mobility (implying a 2DEG structure IIRC), but GaN is pretty slippery stuff all its own. Product flier says they do GaN on barrier on Si substrate, and the whole thing is gooped up with passivation, which is why they aren't afraid of fingerprints on the solder-balled dies I guess.

And the capacitances are teensy. Almost as low as GaN, but with all the resistivity of SiC. Still, better than plain Si, as long as you slew the gate of course.

Tim

No. Where would I get the gate drive from?

That'll mainly work because feedforward is to your advantage

No, it's 6 volts.

So don't do that. Ground the source and drive the gate hard.

John

I threw away a box 0f 723A/B reflex klystrons a few years ago. Used them as VHF parametric amplifier pumps back in the day. I guess they're collectible, now.