Agilent 8753ES A24 Microwave transfer switch substitute

[inaxeon]

Hello Community

I've got a dud transfer switch on my hands (full part number 5087-7067). It's OK from common to port 2, but from common to port 1 I'm seeing around 20 dB attenuation.

Second hand prices on these appears to be $500 to $1000. I'd rather not go spending that much on this instrument if I can avoid it!

What's to stop me from whacking a microwave switch IC i.e. from AD/Hittite on a PCB and using that in the instrument instead? Looking at the specs of some of these parts i.e. HMC1118, HMC849 surely they'd be up to the job? Power supply / control voltage problem is easy to solve by the looks of it.

Admittedly I cannot find any spec for the original switch so some parameters are unknown when comparing against off the shelf parts...

[joeqsmith]

Maybe check the salvage for something you could use. 
https://www.surplussales.com//Relays/rfcoaxialrelays/rfcoax_dcv.html

[inaxeon]

Well that is an interesting site. I'll have a look through all of that. I was looking through some bits on eBay but found that it was too expensive or mechanically impractical.

The 33311 series switches look interesting because they'd drop straight in but are mechanical. I'm concerned that a mechanical switch would be a bit annoying and would also wear out quite quickly.

[Joel_Dunsmore]

Ah yes. That was my favorite design, and 3rd patent. The switch is hybrid solid state design with sets of GaAs SPDT switches for low frequency isolation and sets of shunt pin diodes for high frequency isolation. As I recall, the insertion loss is between 2 and 4 dB from DC to 6  GHz, and isolation is more than 100 dB.  The control was designed to match the mechanical switch controls as they were back in day so 24 volts on the center. That part number was 5086-7539.  the 5087-7067 was a re-design for the 8753ES in the driver circuitry, moved to SMT boards (the 7539 was thru-hole), and I seem to remember we stretched it up to 8 GHz as a special version of the 8753 was pushed up about that far.  And I think (but I honestly forget, as it was 30 years ago) that it was changed to run off 15 V. 

You might be able to substitute the U9397A but it is DC blocked as I recall so won't go down to 10 khz like the 7067.  Oh, I remember now, I had to redesign the bias when we pushed the start freq of the 8753 below 30kHz. Turns out the 7539 had a negative regulator that had a low-level oscillation at a couple hundred kHz; never affected the first version of 8753 as it went down only to 300 kHz.

To this day, I don't think there is any commercially available switch with similar characteristics.   But there has got to more than 50 thousand of them out there in the world (see the serial number on this one) https://www.ebay.com/itm/203580892036?mkevt=1&mkcid=1&mkrid=711-53200-19255-0&campid=5338722076&customid=&toolid=10050

Check the drive voltage on the 8753ES, if it is 15 volts then you can't use the 7539 switch but I think you can find a 22V supply floating around somewhere (I think it went out one of the rear panel connectors intended to drive external switches). If you can find it, you might be able to use a 7539 switch off of that. and I don't the the 7539 had pull-up resistors so you might have to add a couple of external pull-ups. It was designed to have one input pulled down and the other pulled up, but put on a volt-meter to check the control and common voltages.  Note: I'm not -sure- it will work but I guess it would.  The 7067 was designed so it only need a drive on one pin, high-for-port 1 low for port 2, but the harness was kept the same so as to allow it to be compatible in older products. I think it has a pull-up inside.

There is no protection for the mechanical switch cycling and so I guess it will wear out quite quickly if you do a full 2-port cal and don't hold the sweep.  I don't think the second control line is active in the 8753ES so it will switch it right but not left; you could jigger up a little drive circuit to inverse the control1 line to drive the control2 input on the mechanical switch.

[inaxeon]

As I recall, the insertion loss is between 2 and 4 dB from DC to 6  GHz, and isolation is more than 100 dB.

100dB? That is ridiculous. Was that an actual requirement? I was trying to figure out what the isolation of mine was, but threw in the towel after testing to 80dB. I measured the insertion loss to be more like 1.6 dB, but yeah perhaps that wasn't a very accurate test.

But there has got to more than 50 thousand of them out there in the world (see the serial number on this one)

I have no doubt, but the challenge is finding one that isn't dead. Most if what I could find is the same deal as mine, port 1 dead, port 2 OK.

[inaxeon]

Check the drive voltage on the 8753ES, if it is 15 volts then you can't use the 7539 switch but I think you can find a 22V supply floating around somewhere

It provides a 22v supply with 16v appearing on the drive pins. I assume that the chassis was used as GND hence the big warning on it.

There is no protection for the mechanical switch cycling and so I guess it will wear out quite quickly if you do a full 2-port cal and don't hold the sweep.

Trying to avoid going mechanical if I can...

I think it has a pull-up inside.

Yes it does.

I don't think the second control line is active in the 8753ES so it will switch it right but not left; you could jigger up a little drive circuit to inverse the control1 line to drive the control2 input on the mechanical switch.

Both control lines appear to be present in mine, switching from +16V to +0.8V.

[Joel_Dunsmore]

100dB? That is ridiculous. Was that an actual requirement? I was trying to figure out what the isolation of mine was, but threw in the towel after testing to 80dB. I measured the insertion loss to be more like 1.6 dB, but yeah perhaps that wasn't a very accurate test.

I have no doubt, but the challenge is finding one that isn't dead. Most if what I could find is the same deal as mine, port 1 dead, port 2 OK.

Well, in fact the isolation is more like 110 dB with the on/off ratio spec at 100 dB. You need that kind of isolation to maintain the 100 dB dynamic range (leakage out the switch at port 2 will reflect off the stop band of a filter, for example, and limit the ability to measure stop band isolation). Some cell phone tower filters have specs that exceed 130 dB isolation requirement.

I just had an idea: the switch has an input protection diode-structure which is a very low capacitance diode-string (custom HP/Keysight part) to ground on the input line. Those work by clamping the max voltage than can get to the GaAs FET switch.  They will die if over-voltage is applied (like too much RF power) and they die one-at-a-time in the shunt-string. Eventually all the diodes short and the input becomes shorted to ground through the string.  You can open the switch up (there are 2 pozidrive - like philips- on the housing, and then the microcircuit has one screw holding the lid in place with a pressure clamp).  Look at the input line on port 1 and you will see the protection diode on the substrate (or take a picture and I'll help you identify it).  pop it off and if it was the cause of the pad port, it might solve the problem.  Removing it won't affect the normal operation, but of course now you don't want to ESD the input port or you will blow the GaAs switch as well. 

[Joel_Dunsmore]

Both control lines appear to be present in mine, switching from +16V to +0.8V.

As I said, I don't remember for sure but I think that was required for the 7539 but not for the 7067.  When you open it up you can tell if the control2 pin is connected something or nothing. 

[inaxeon]

As I said, I don't remember for sure but I think that was required for the 7539 but not for the 7067.  When you open it up you can tell if the control2 pin is connected something or nothing.

Control2 pin is NC. It's wired to the chassis but it goes nowhere internally.

[inaxeon]

or take a picture and I'll help you identify it

Attached. I see how you get the 100+ dB now. 2 switches...

[inaxeon]

I have another question too...

How come this thing has so many large tantalums on it?

[inaxeon]

The original repair option I was thinking of a custom PCB with some off the shelf switches like the one attached (I had mostly designed it before starting this post)

I'm using two HMC1118's back to back, which should get 100 dB isolation. Frequency range and insertion loss look to be OK for this too. RF resistors for the termination (whatever I can get my hands on, very little at the present time) and a fair bit of care taken to provide clean supply / control too.

I know this will impair the instrument, but I'm just a hobbyist here. So long as I can make some approximate measurements with it and not stress over a $1000 bill if I have an accident that would be brilliant.

[Joel_Dunsmore]

The back to back switch might work. I found parasitic ground inductance really wreaks havoc with isolation on the switches; as well as radiation leakage.  I'd be curious how it turns out.  I started with such a structure and could only get about 60 dB isolation at 6 GHz, but of course entirely different switches in 1986.   

But you might also try tor repair your switch. It will either be the GaAs IC switch that went bad or the protection diode. If it is the protection diode you can just reach in with a tweezers and pluck it off. I outline in blue the protection diode.

now that I see inside, I recognize that's a redesign of the PC board with new SMT parts which happened when something on the through-hole board went obsolete; done by our production engineering team. The Tantalum capacitors are needed to bypass all the control lines (6 of them) plus the voltage inverter (I think that's the LT1054).  We need a lot of bypassing to keep 100 dB isolation at 10 kHz.

Oh, I should say you can test it without the lid just to see if it is working after plucking the protection diode. When you go to put the lid back on, take the grey conductive rubber bits off the base and stick them back in the lid first.

[inaxeon]

I am going to try repair the one I have. When I get a chance I'll pluck that diode and see what happens...

[inaxeon]

While I have you on the line, would you be able to identify the numbered components for me? I'm just interested to know...

[Joel_Dunsmore]

1: DC feed likely with 50 pF discoidal bypass capacitor
2: Bonding island to ground (just a gold brick).  Ground the resistor to the left that is connected to the main transmission line
3: Single layer capacitor, maybe 100 pf, used to bypass the bias-control lines to avoid leakage in them.
4: Pin Diode mounted ont top of a very large single-layer capacitor; the top of the capacitor is biased negative to turn on the pin diode. Bonding island 2 is biased positive to provide bias and voltages are set so the top of the Pin diode is 0 volts, that way no blocking capacitors are needed.
5: bonding island to the top of the cap, as the plating on the cap doesn't bond well
6&7: A little tuning stub, but it was not in the orginal design.  I guess the production engineer added them in case then needed to use them.  We used to use tuning ribbons but no one likes to have them in production.

[inaxeon]

Brilliant. Thanks for typing that up.

I forgot to ask, either side of the GaAs switch, are those two blocks bypass capacitors?

[Joel_Dunsmore]

yes, there's a total of something like 5 bypass on each line. In fact I think later other versions added a couple more as we found some leakage still occurred along the bias lines.

[inaxeon]

I've plucked that diode and the results are... better...

Using the power-on default S11 measurement into an open connector on the front panel, the "before" was pretty ugly.

The "after" looks a lot better but still not like S22 which shows a flat 100% reflected, so perhaps the GaAs switch is damaged too.

I'll press on and make my switch and see how that goes...

[inaxeon]

I pulled the switch out again and hooked up to my spectrum analyser/TG. I now can't see any measurable difference between port1/2 on either through or blocking states. Maybe pulling that diode did fix it. Maybe something else on port 1 is faulty. Argh!

More investigation needed...

[inaxeon]

Swapped my couplers over, and the remaining problem has moved to port 2. So, your diode plucking method has fixed the switch. Thanks for that. No doubt others will find this advice useful too.

I think the story of this instrument is becoming clearer: The CPU board died (repair story here: http://www.mattmillman.com/acquisition-and-repair-of-an-agilent-8753es/ ) so whoever owned it transferred some bad bits into it and dumped it. This is evident from both the switch and the port 1 coupler having felt tip markings on them in the same handwriting.

Without even opening it I'm guessing that coupler is another exotic difficult-to-repair hybrid.

[Joel_Dunsmore]

Feel free to start another thread on the coupler, (which is actually a bridge;  I designed it at about the same time as the bridge).  Normally S11 should be flat across the band, but the fact it goes up and down indicates it is likely a low-freq issue (like a bad contact).  But for investigation (presuming port 2 is the bad one) show also S21 and S12.  and show S22 with an open and a load.  Then we'll open it up and probe a few bits to see what we find.

Very happy to hear the diode-trick worked. 

[inaxeon]

Here is the thread: https://www.eevblog.com/forum/testgear/agilent-8753es-a21-test-port-coupler-repair/

I switched the bad coupler back to port 1

[inaxeon]

Back in here again. In the throes of trying to repair my switch I bought this 5086-7539 off eBay for 20 bucks. Seller described it as "intermittent"

... which turns out to be BS. It's not intermittent as far as I can see. The actual problem with it is that when Port 1 is in the closed state, it's not connecting through to the 50Ω termination, instead reflecting the input.

I had a bit of a poke around and the voltages are up to a couple of volts different from side to side. Slim chance it's the control board, will look into it further.

Either way, it's interesting to look at. Surprised to see the hybrid is quite a bit different to the one 5087-7067. Pictures attached!

[inaxeon]

One of the GaAs switches is damaged. Bummer. Seller refunded me. I wonder how this one died...

Unlike mine both of the input diode strings are OK, because both of the "through" paths works OK with no attenuation, so can't have been an RF overload that killed it. It's just the unwanted reflecting on port1 that's the issue. Still, it is actually useable as an overkill antenna switch or something.

Sweet ;-)