# Understanding relay isolation specifications

I'm building a board with some relays to switch a bank of HF filters. I want to be sure I'm understanding the isolation I can expect from a given relay so as to not degrade a filter's performance by allowing unselected relays to couple into the output.

Let's say I have this circuit, built of relays with a specified isolation: [full datasheet]

simulate this circuit – Schematic created using CircuitLab

Does the fact that the unselected input sees a short to ground change the isolation I can expect to something more than the 37dB in the datasheet, since the unselected input will be able to deliver very little power into the short? Or does it not matter since there can still be current through the short which can couple the signal inductively across the relay?

• Why do you want to short the unused input? If you're going to abuse your sources that way, why not use just one SPDT relay turned around the other way and let the unused source see an open rather than a short? Or if you're going to use two SPDT's why not terminate the other outputs properly? – The Photon Apr 7 '17 at 0:52
• Also, can you link the relay datasheet? Because I'm more used to "isolation" referring to isolation between the coil and the switch (and having a value like "6000 V") than to two switched ports. – The Photon Apr 7 '17 at 0:54
• I guess I was thinking connecting unused inputs to ground would reduce the power from the unselected input making it to the output, but now that you ask I guess putting power into a short isn't any better than putting it into an open. Regardless, I'm having some difficulty understanding what this isolation figure means exactly. – Phil Frost Apr 7 '17 at 1:04
• @ThePhoton Datasheet linked. – Phil Frost Apr 7 '17 at 1:05
• I'd GUESS: You still get nominal 37 dB BUT it's at the signal level when shorted. So effectivelyt much better than 37 dB as reqards unwanted signal transfer to load. | The isolation spec effectively reflects an impedance across the open contact - so you will reduce output very slightly due to the ~~~ 37dB coupling to ground. WHEREAS - using a reversed SPDT and O/C for unused signal gives you Vunwanted behind the isolation impedance so more extraneous signal. – Russell McMahon Apr 7 '17 at 1:35

1. I agree with Simon, you should be better off terminating the "unused" output properly (with a resistor equal to the system impedance). This reduces the likelihood of causing the source driving that path to misbehave.

If you're going to abuse your sources by not terminating them properly when the output load is disconnected (or you know they don't mind high reflected signals), I'd probably just save a relay and use the SPDT with inputs on the two-contact side and output on the one-contact side.

2. Whether the isolation gets better or worse by terminating with a short depends whether the crosstalk mechanism is inductive or capacitive. Since TE doesn't say, you'd best assume worst case. That would probably be 3 dB worse cross-talk than specified, since you'd have equal power in the forward and reverse travelling waves when terminating with a short (or open).

3. For that matter, since these are 2 Form C relays, the isolation might be meant to be between the two switch paths, not between the NC and NO pins of one switch. If your RLY1 and RLY2 are actually the two switches within a single 2-Form-C unit, I guess this would have about the same effect in your circuit.

4. Since you say you are actually operating at 21 MHz, you should expect cross-talk to be at least slightly better than the 37 dB specified for 100 MHz signals, though you'd have to measure to know how much better.

5. If whatever's driving the SELECT line is capable of driving two coils in parallel, you can save yourself a NOT gate by just switching which output you connect to on one of the relays (so OUT is connected to the NO pin of one relay and to the NC pin of the other relay). Or maybe you already had this in mind since your device actually has just one coil driving two SPDT switches, and I'm just going down a rabbit hole here.

Intuitively: combining the two RF lines is an impedance mismatch. I'd use a third relay or a solid-state switch here.

If you connect the unselected input to GND, you get another impedance mismatch, this should be properly terminated so the power is sunk into a resistor rather than radiated or reflected.

It will not block perfectly in the relay, but adding a second stage adds more attenuation, and also improves impedance control.

• Can you add a schematic so it's clear what you are suggesting? And why does an unused input need to be terminated in a matched load? Radiating or reflecting the power is just fine, if that keeps it out of the output. – Phil Frost Apr 7 '17 at 0:21
• Notice the SWR is ~1.5 (assuming we're talking about 900 MHz). I don't know if adding more relays is really better than just keeping the traces between the two relays and the trace junction short. – The Photon Apr 7 '17 at 1:14
• Highest I intend to go is 21 MHz. Though besides the connection from the selected input to the output, SWR isn't a concern at all. Isolation is. – Phil Frost Apr 7 '17 at 1:17
• So then you should be expecting isolation somewhat better than 37 dB (the 100 MHz spec) rather than ~20 dB (the 900 MHz spec). – The Photon Apr 7 '17 at 1:18
• Yes you're right. I'll edit the question. – Phil Frost Apr 7 '17 at 1:23