Question for the RF guys here.
We want a wideband RF switch IC that switches fast. Bandwidth lf to 10 GHz or better, switching in well under 10 ns.
ADRF5024 is 100 MHz to 44 GHz, but switches in about 17 ns. The 100 MHz lower signal bw makes no sense.
HMC347 looked pretty good, DC-14 GHz and 10 ns typ, but it's gone obsolete.
Who else makes parts like this?
Well the HMC347B seems to be "in production"
347B is a bare die. Mouser seems to have some of the only surviving packaged part, HMC347ALP. We could scoop up a lifetime supply, for $70 each.
Thanks.
We've used Skyworks SKY13453 (2-way) and SKY13317 (3-way) but they're only 6GHz. Check their other parts?
And Qorvo of course:
Clifford Heath
The data sheet does say: "All of the RF ports (RFC, RF1, and RF2) are dc-coupled to 0 V, and no dc blocking is required at the RF ports when the RF line potential is equal to 0 V." There is a similar device, the ADRF5025 that goes down to 9kHz, but that has much slower switching. There must be resistive filtering on the gate drive and a resistive path from the FET sources to ground to bias the FETs. The time constant must be such that there is not enough time for the gate charge to leak away during each half-cycle of the switched signal. If so, the tradeoff between minimum operating frequency and switching time would make complete sense, as would the power derating at low frequencies. The data sheet also mentions that there are no low-frequency spurious signals, so they are not using a charge pump for FET bias.
John
Maybe. I suspect the gate drivers are slow to save power.
The RF people seem to pick some arbitrary low frequency limit on parts, possibly based on the blocking caps on their eval board.
Some people apparently work with spectrum analyzers that commonly have a 9 KHz low end, so spec their parts to 9 KHz.
I did once use a Hittite 8-to-1 RF mux that was spec'd "DC-to-12 GHz." It didn't work right below about 50 MHz. I eventually got the chip designer on the phone but he wouldn't explain it because it was proprietary.
The RF world is weird.
Wow, a lot of near-misses.
I'm amplifying an arbitrary waveform as the seed of a largish laser. And I want to inject a 100 ps fiducial pulse. We can make a nice programmable 100 ps gaussian pulse, but can't passively mix it with the arb without wrecking both. So I was thinking that a switch could select the impulse and then immediately switch over to the arb. But the customer wants at most 5 ns between the fid and the arb.
Maybe we'll tell them they have to wait longer. The 5 ns requirement might not be absolute.
I could delay the fiducial some, to start it after we've told the pokey analog switch to go. Maybe.
Probably why I mostly see PIN diodes used for such things.
Joe Gwinn
PINs are cheap and can switch lots of power. They are terrible for time domain signals. At really high frequencies, their off capacitance can be a problem. 0.2 pF is about 50 ohms at 14 GHz.
There are two almost identical parts (ADRF5024 and ADRF5025) with different tradeoffs between switching speed and low-frequency cutoff. That isn't going to be explained by limitations of the test gear or blocking caps on the eval board - especially when the eval board doesn't even have any dc blocking caps on it! The data sheet is quite specific that there is a dc path to ground on all the rf pins. It is a real tradeoff and there is nothing arbitrary about it.
John
Why? Does each one produce interfering crud when it should produce nothing, or are you just not able to tolerate the amplitude loss of a passive resistive combiner?
Amplitide. Neither the arb nor the fiducual generator can make over about 0.75 volts peak, and the distributed amp needs all of that to drive the modulator. DA's invert, so adding another $300 chip in the signal chain creates new tangles.
The many available RF switches are various flavors of weird and all are poorly spec'd. I've never seen any mention of the capacitance of the mux control pins, for instance. Or of switching glitches. OIr, heaven forbid, any actual waveforms. The really fast ones need a pair of maybe-big poorly-specified negative switch control voltages.
Analog Devices acquired Hittite, which had the fastest RF switches. "Support" is mostly now forums. Some issues that I care about are asked about on the forum; some haven't been answered in four years.
TI had the same problem when they acquired Burr-Brown. Nobody knew much about the parts.
Am 27.02.22 um 16:35 schrieb snipped-for-privacy@highlandsniptechnology.com:
But this is not only for aquired products. My questions about the bleeding controls in the ADF5356 synthesizer are unanswered, too, after 2 years. Some 500 reads.
About as helpful as the Altium forum.
Gerhard
We had a lot of hassle using the LMX2571 synth, that TI made necessary by being coy. Makes no sense.
It must be difficult to find and keep good application and support engineers. They have a life that is, basically, continuous job interviews. It's a great job for a recent grad.
I guess that the chip designers, if they are even still around, need to be defended against customer questions, or they wouldn't get work done.
We have to test a lot of parts ourselves to understand them.
We need a web site where we share insights, measurements, code, bugs.
And the 1/f corner of the parts.
HP 8566B, for instance. That isn't a coupling cap, it's just the wings of the huge peak at DC. The HP 8568B has a coupling cap, which makes it harder to blow up the first mixer.
Cheers
Phil Hobbs
Could you increase the output of the fiducial generator somehow? Then a combiner with unequal resistors could have low loss for the arb.
If you don't care too much about the shape of the fiducial, there are probably also some less-broadband-on-one-port combiners that could help with lower loss for the arb.
If you can't increase the output voltage of the fiducial generator, it may be sufficient to provide multiple outputs with the same voltage, or a single, lower-impedance output driving a combiner designed for different impedances.
Could you yank the arb output high just during the fiducial pulse, with the emitter of a fast NPN? The impedance would not stay constant, but that mightn't matter.
We're using a Micrel laser driver chip, Sy88022. It makes 25 ps edges and has beautiful linear amplitude control, but is intended to drive a low impedance laser and won't swing even one volt.
Inspired by Leo Bodnar's pulse generator, which uses a similar Maxim chip.
The shape really matters. It will be used to test the response of laser bits downstream. The customer wants a 100 ps gaussian.
It's hard enough to make a 100 ps programmable gaussian pulse. The analog switch looks like the thing we need.
There are some amazing RF switches around. ADRF5025 is rated for 9 KHz to 44 GHz. It switches slow so we'd have to deal with that.
The fast switches all look like nightmares. Some need 7 volt (or 40 volt!) gate drives into unspecified capacitances.
Am 26.02.22 um 16:07 schrieb snipped-for-privacy@highlandsniptechnology.com:
The 9 KHz come from some EMC norms. If a lab wants to certify compliance to these norms, they must measure to 9 KHz. Everything better might cost one more synthesizer loop aka money or lose dynamic range because of 1/f, filter bandwidth or over all synthesizer quality.
The 8720A network analyzer goes from 20 GHz down to 50 MHz. They won't specify anything they cannot measure. And 50 MHz is just nervous DC.
Many s-parameter files end at either 3 or 6 GHz because the 8753 A/B/SE/WHATEVER ends at 3 or 6 GHz without/with the option -6.
No, just costly.
And PINs will switch even slower. They start to distort below a few MHz.
Maybe you can build something discrete with CEL 20 GHz GaAsFets.
<Cheers, Gerhard
The Hittite mux thing was real. It is supposed to terminate the unselected inputs with 50 ohms, but there is apparently a cap in series with each terminator resistor. So much for "DC."
We use a lot of MiniCircuits enhancement phemts, the SAV series. I considered making my own RF switch from a couple of them, but it would be risky. Gate drive would be interesting.
We did make a nice switchable attenuator with a SAV541.
PE42525, maybe. 9 kHz-60 GHz, SPDT, 8-12 ns switching time to
10%/90%, 48-60 ns settling time to 0.05 dB.I'd be surprised if you find anything faster than that. PSemi makes fantastic stuff (at least, when you can buy it.)
-- john, KE5FX
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.