Should I use a mechanical relay or an SSR for the input of this LNA?

[Verdefluox]

Hi folks,
I am working on a very humble (I hope this is the case) LNA, which I am willing to use to amplify down to around tens of uV (realistically, 50 uV) over a wide band (realistically, at least up to 100 kHz). I'd like to amplify both low DC and AC signals.
The question is about the input, right before the JFET differential amplifier: I am wondering which one of these configuration is better NOISEWISE, hence that leaves the signal as untouched as possible. First switch switches between input and GND (thought to connect it to use it for a rough calibration, other than physically disconnecting the system from the input), while the second is for AC/DC selection.
I don't know much about noise in switches, hence I don't know which of the two proposed configurations is better in that sense.
Mind that the SSR proposed is also used afterwards, with the signal already amplified. But for a low signal coming in, I have no clue on which one should I go for and why. I have read around that usually SSRs are usually better for noise (which confuses me, don't they have higher ON resistance?), but I am sure there are more important technicalities on the choice.
Any ideas?

P. S. In case they are useful:
SSR datasheet: https://www.vishay.com/docs/72107/dg417b.pdf
Relay datasheet: https://www.mouser.it/datasheet/2/307/K024_E1-3476493.pdf

[tooki]

SSR ≠ analog switch IC. A solid state relay galvanically isolates the control side from the switched side. (Usually by using an optocoupler internally.) An analog switch does not.

You’re proposing an analog switch. The fact that its datasheet doesn’t list any noise and distortion figures makes me suspect it’s not optimized for that.

[vk6zgo]

It all depends on what you are using the relay for---if it is to add or remove the VNA in a receiver only situation, that is quite different from needing to bypass it when RF power is being fed up the cable, in a transceiver type operation.

In the former situation, depending upon the frequency range, a discrete circuit using one or more dual-gate FETs was a common method.

Plain old analog switches like the old 4066 series have substantial series resistance, & are probably inappropriate.

Electromechanical relays for RF use have been around for along time & proper ones have been optimised for low through loss & high
isolation.

I would suggest looking at how commercial designs approach this problem.

[mtwieg]

There's a few ways these devices can introduce noise:
1. Johnson/thermal noise. White noise with spectral density equal to sqrt(4KTR), so easy to calculate from resistance and temperature. But I doubt this will contribute to overall noise unless the resistance is in the hundreds of ohms or more.
2. Flicker noise. Pink noise which will exist in analog switches or SSRs, but not in mechanical relays. I don't think datasheets will actually characterize flicker noise though.
3. Coupling from control circuits. Not really noise, but rather a path for noise from other circuits to contaminate the signal being switched. For example, parasitic capacitance from the relay coil driver to the contacts, or from the supply voltage of an analog switch to the channel. If your source and load impedance is very high, this might easily be the dominant noise source, but it's very unlikely that this information is characterized.

[Verdefluox]

SSR ≠ analog switch IC. A solid state relay galvanically isolates the control side from the switched side. (Usually by using an optocoupler internally.) An analog switch does not.

In this case for sure I wanted a galvanic isolation for the input, and obviously I didn't necessarily need it for AC/DC. I just wanted to know if there are any obvious reason to choose one of the two, as to start.

Electromechanical relays for RF use have been around for along time & proper ones have been optimised for low through loss & high isolation.

This is why I went with this series, which should be good for the task, as shown in the comparison below (see pics).

There's a few ways these devices can introduce noise:
1. Johnson/thermal noise. White noise with spectral density equal to sqrt(4KTR), so easy to calculate from resistance and temperature. But I doubt this will contribute to overall noise unless the resistance is in the hundreds of ohms or more.
2. Flicker noise. Pink noise which will exist in analog switches or SSRs, but not in mechanical relays. I don't think datasheets will actually characterize flicker noise though.
3. Coupling from control circuits. Not really noise, but rather a path for noise from other circuits to contaminate the signal being switched. For example, parasitic capacitance from the relay coil driver to the contacts, or from the supply voltage of an analog switch to the channel. If your source and load impedance is very high, this might easily be the dominant noise source, but it's very unlikely that this information is characterized.

For 1 and 2 I think the mechanical relay is better. Also, if I recall correctly in a wideband configuration flicker noise should be secondary, but since I included afterwards an optional LP filter @ 1 kHz (in case I need to test a small DC signal) it may start to become influential, and using a mechanical relay should be a better option.
What I still don't get is why I keep finding information that say that SSR have in general lower noise (sorry, I know I'm not justifying the reasons, but it's my first time approaching these things).
According to the info I attached they should be pretty similar in that regard, and right now I'm thinking that maybe I could go with the LATCHING TYPE of this mechanical relay series, so to avoid whatever noise is generated in non-latching relays by keeping the coil de-energized for most of the time. In normal operation I am not expecting to change between CAL and INPUT, so unless I get major backslash could this be viable?
Oh, and obviously the cost also made me wonder if I could go with the SSR, since the mech relay is ~5-6 euros, while the DG419 is ~2 euros per piece.
In conclusion: I'd like to avoid two mech relays because of cost and size, but I feel like I'm risking it by using an SSR given all the possible noise sources; also, coupling seems similar if I got the correct graphs, but here I'm assuming that for the mech relay, for frequencies below ~5 MHz which are not shown the crosstalk becomes simply negligible, which is my main concern right now.

[Kleinstein]

The CMOS switches are usually quite good in not adding more noise than the resistor noise from the on resistance. There may be a little thermal EMF from temperature gradients. With the low power this would be from other sources. The off state still has some leakage and one may want some protection in front, as the CMOS switches don't like voltage outside the supply range.

Mechanical relays, especially the nonlatching types can add thermal EMF from heat caused by the coil. Latching types are much better in this respect.
SSR (e.g photomos switches) can also have some thermal EMF, but at least the type I tested was surprisingly good ( < 2 µV) and not much (if any) extra noise, just the restance. The electronic switches do have a little leakage also when off, especially at higher temperature. There is also quite some capacitance that can be nonlinear and cause issues with higher impedance AC signals.


With a JFET amplifier there is likely more offset drift than the small error from thermal EMF.

[Verdefluox]

Mechanical relays, especially the nonlatching types can add thermal EMF from heat caused by the coil. Latching types are much better in this respect.

Exactely what I thought, for sure in case of a mech relay I would go for that kind.

SSR (e.g photomos switches) can also have some thermal EMF, but at least the type I tested was surprisingly good ( < 2 µV) and not much (if any) extra noise, just the restance. The electronic switches do have a little leakage also when off, especially at higher temperature. There is also quite some capacitance that can be nonlinear and cause issues with higher impedance AC signals.

With a JFET amplifier there is likely more offset drift than the small error from thermal EMF.

Talking about leakage, tell me if I'm wrong: in DG419BDY, it tells me I would have to consider ~60 nA max. This parameter is a DC value, right? In AC it shouldn't count (because of the capacitor), and in DC operation, given there is this the MOhm resistor, it would simply result in an offset (which has just to be accounted for in calibration). Did I get it right?

So, in conclusion: mech relay seems to me less risky for me, given that my lack of equipment makes me choose the most "out of the box" solution, even though both the Xtalk of the two are very similar. Also, I don't know if I'm deducing something correct, but confronting the two dB losses I'm assuming the mech relay also has way less capacitance than the SSR (if the losses are also due to capacitance, and not only to resistance). Too bad it costs me more, but I may always test another configuration if I have a functioning LNA.

Something else obvious I'm missing? Otherwise I would go with mech relay and not think too much about it.

I would suggest looking at how commercial designs approach this problem.

Found something, might have to go deeper with others but it's giving me some hints: SR560 from Stanford Research https://www.thinksrs.com/downloads/pdfs/manuals/SR560m.pdf uses mech relays for AC/DC coupling, and the instrument is doing a similar job as mine (obviously, way better).

[Kleinstein]

The maximum leakage specs for the electronic switches are often very conservative and more test limits. It takes more time and thus costs extra to check the leakage to a lower level. The actual leakage level to expect can be significantly lower. For the DG419 likely more in the 1-10 pA range. Another point is that with some of the switches the leakage is mainly near the extremes in the voltage. Some 0.1 V away from the supplies the leakage from the "diodes" to the supply can already compensate to a large part.

For AC there is coupling capacitance in the CMOS switches. So there is AC leakage too. For really good AC isolation one may need 3 switches with one to ground.

The AC/DC coupling in quite some modern DSOs is with a photomos switch. Leakage and capacitance across the switch are not critical, but low capacitance to ground and the control signal is wanted.

A latching mechanical relay is likely the safer / easier solution. 
There are small, affordable latching relays, even with gold contacts for small signals:  Zettler AZ850, Fijitsu FTR1, Kemet EC2/EE2
Reed relays can be tricky: they are essentially all non latching and have often quite some thermal EMF and surprisingly the leakage can be relatively high.

[tooki]

SSR ≠ analog switch IC. A solid state relay galvanically isolates the control side from the switched side. (Usually by using an optocoupler internally.) An analog switch does not.

In this case for sure I wanted a galvanic isolation for the input, and obviously I didn't necessarily need it for AC/DC. I just wanted to know if there are any obvious reason to choose one of the two, as to start.

But the DG419 "SSR" whose datasheet you link in your original post is NOT an SSR, and does NOT have galvanic isolation. It is an analog switch, not an SSR.

That's why I think you are still confusing SSRs and analog switches.