Stabilizing pHEMTs

The output stage is a CFB op amp with about 120 MHz BW. I could try using a bit of stripped-back coax as an antenna, but I sort of like my digital wavemeter. ;)

Yup. I've got a saved search in for an 8566B at this point. Should have bitten the bullet and got the nitrogen dewar as well. If this project goes to the next stage, I will.

Cheers

Phil Hobbs

Cool. I remember those--they're pretty similar to the late lamented Si8901 MOSFET ring mixer.

Cheers

Phil HObbs

Don't forget to connect the substrate when trying them out. As embarrassing as it is, that happened on my first copperclad lash-up.

Roight.

I discover that Vishay Siliconix has re-used the Si8901 part number for an unrelated product! I'm looking at the NMOS ring modulator datasheet from the 1989 databook and the dual PMOS datasheet from web site--same P/N.

Probably someone wanted to make sure the old part never ever came back.

Cheers

Phil Hobbs

This?

In the days of the Intenet that's next to impossible.

Right, but at r.f. it's entirely possible the cap will self-resonate. That's what stacking values avoids.

For example, 1nH parallel-resonates with 1uF at 5MHz, and with 10nF at 500MHz.

Either capacitor alone in a broadband amp runs risk of resonance, but the two in parallel avoids it.

That radiates and capacitively couples--certainly not helpful. I wonder about the gate line too. Might need fancy treatment, like a bootstrapped shield. Whew.

-- Cheers, James Arthur

lso help

er.

spec

find a

ut 12 GHz,

oved my

to valley

quency

and

from

ook

rs?

Lurking... Just wondering (out loud) if the 'small' oscillations could be pick-up? With Phil's finger providing the antenna or shield.. (depedning on the disposition of his other digits)

George H.

document:

For testing, it runs inside a nickel-plated steel can (used for 70 mm film reels). There are some nice Russian 47-nF feedthrough caps on the power supplies, so it's clean as a whistle. Open the box and it goes nuts.

Cheers

Phil Hobbs

Yup, that's the old one. Made some really amazing mixers.

"You can't bring that part back, it'll screw up our documentation system."

Cheers

Phil Hobbs

The best mixer ever was the Plessey SL6440. Before that ship went under I quickly bought a small stash. It follows the mantra "Nothing can replace a high IP3 rating, except for a higher IP3 rating".

They should do the right thing and retire old part numbers. Just like it's done with jerseys in football.

BTW, I sent you a few PM replies (totally other subject though) but just realized that it was your gmail address and the one above is different. If they didn't make it I can re-send.

Take a look at X2Y capacitors from Yageo and Johanson. I think they are promising on relatively thick boards because their via distribution makes some vertical currents cancel (for deblocking planes). I have bought a few from DK, but not yet formally tested.

regards, Gerhard

But talk to your assemblers first, don't drop those on them as a surprise.

GHz,

moved my

valley

frequency=20

The problem is that the finger is calibrated only attached to Fred.

right

from

depending

look

yours?

too

Sweet. But it chaps my hide to think about how many months income that thing costs.

?-)

A real brazen RF dude rolls his own, of course. Transistors with >50GHz ft are around a Dollar in singles these days.

My little LNA has another problem right now, its NiCd batteries are totally dead and small special size :-(

Might need fancy treatment, like a

I tried a bootstrapped shield the other day, and found it made things worse. Eventually I decided it was (probably) because the signal I was interested in measuring was around a nanoamp, and the BF862 I was using had a similar level of noise ~1nV/rt Hz - so I was adding more than I was cancelling (yeah I'm mixing units, but my point is both were small so the voltage was probably significant). Surely Phil will face the same problem with his 62-electron signal...?

Bootstrapping is really worthwhile if the bootstrap amp is quiet and accurate enough, but it still introduces a noise current equal to its input noise voltage differentiated by the capacitive reactance. This is the same problem faced by TIAs, of course, but interestingly, you win by combining them. They work sort of independently, though not exactly so--in arm-waving terms, the capacitance improvement adds linearly, and the noise only in RMS.

It's also initially pretty surprising how small the gain and phase error in a bootstrap has to be, if you want it to make the effective capacitance to go down by a factor of 10 or more. With a BF862, it's very worthwhile using a current source load, and hanging a couple of really fast emitter followers (also with current source loads) on the FET's source. Use one follower to bootstrap the BF862's drain and the other one to drive the capacitance. Spice says that one's good for a factor of 300, if the capacitance to ground from the sensitive node (a photodiode in that case) is below about 0.1 pF. That's the sort of place where a bootstrapped shield makes a lot of sense--there's no other way to get the capacitance to ground that low.

Cheers

Phil Hobbs

moved my

valley

frequency=20

and

One last thing, how big is that fingertip? In mm by mm please. And if you have it the sheet resistivity in ohms per square.

?-)

So I finally got the amplifier all measured up, with three different pHEMTs: Skyworks SKY65050, Avago ATF38143, and Avago ATF34143. Joerg's suggestion of bodging in a bypass right at the collector load made the amp stable even with the Avago parts, which are hotter than the Skyworks one. I dremelled down to the ground plane, so the inductance is only

Executive summary: horrible 1/f noise, but 0.28-0.4 nV noise in the flatband, and < 1 pF input capacitance. Second stage contribution becomes important above ~100 MHz. See

Measuring all that was interesting--getting decent accuracy for noise levels that small at high frequency is hard.

Cheers

Phil Hobbs