Optimizing Multiplexing for High Frequency Analog Signals

[Topher925]

I have an application that requires me to multiplex between >7 sinusoidal signals (15v ~1-100kHz) with as little distortion as possible. I'm not really sure what would be the best way to go about doing this though. My first thought would be to simply use a low capacitance multiplexer, like http://www.analog.com/static/imported-files/data_sheets/ADG1206_1207.pdf" But I'm concerned that using such a small IC and combining all the signals into such a small area is going to generate some crosstalk between the channels. I would like to avoid designing my own circuit as IC's designed by actual electrical engineers will always do a better job than me but I'm wondering if there is a better way? If I do end up using a ADG1207 are there any "tricks" to the PCB design that I could do to minimize crosstalk and distortion?

 

[sophiecentaur]

Is there a good reason why you don't want to digitise the signals then multiplex them? Your bandwidth requirement seems well within the capabilities of readily available devices and it would take care of crosstalk. Are the signals going to be used in analogue form at the destination?

 

[Topher925]

Are the signals going to be used in analogue form at the destination?

Yes. The signals will be fed into a lock-in amplifier which does not accept a digitized input.

 

[berkeman]

I have an application that requires me to multiplex between >7 sinusoidal signals (15v ~1-100kHz) with as little distortion as possible. I'm not really sure what would be the best way to go about doing this though. My first thought would be to simply use a low capacitance multiplexer, like http://www.analog.com/static/imported-files/data_sheets/ADG1206_1207.pdf" But I'm concerned that using such a small IC and combining all the signals into such a small area is going to generate some crosstalk between the channels. I would like to avoid designing my own circuit as IC's designed by actual electrical engineers will always do a better job than me but I'm wondering if there is a better way? If I do end up using a ADG1207 are there any "tricks" to the PCB design that I could do to minimize crosstalk and distortion?

To minimize crosstalk, use individual analog switches, and be careful in the layout to have ground guarding between them and their inputs. You generally can't do as good a job of ground guarding the signals when you use a single large analog MUX.

 

[Topher925]

Thanks for the reply berkeman, but what exactly do you mean by "ground guarding". Do you mean isolating the ground for each signal input and switch?

 

[berkeman]

By ground guarding, I mean surrounding the sensitive traces with ground pours. On the same layer, you will pour ground in the open spaces between the sensitive traces (and keep those traces apart from each other as much as practical). Keep the sensitive traces on the top layer with the switch ICs, and make your inner layer closest to the top layer your ground layer. That's usually your ground inner layer anyway, when you are running high-speed signals on the top layer between ICs, because you want to control your Zo geometry.

In your case, you are making that uppermost inner layer ground to help with the ground guarding. The top layer interstitial ground pours and the inner layer ground plane help to minimize the capacitance between the sensitive traces, which cuts the crosstalk down. The capacitance from a sensitive trace to ground dominates, compared to the capacitance between the sensitive traces.

It might also help to be sure that the sensitive signals are buffered individually before they are routed to the part of the PCB where the multiple analog switch ICs are. But that would only help if the signals were coming from a higher-impedance source, not if they are already coming from low-impedance buffers or amps.

Also be careful not to share any power or ground impedance between the separate signal sources or in the routing. The signals should approach the MUX area of the PCB in kind of a fan-in pattern, with the power and ground being routed from near the MUX area out to the sources of the signals. So for example, you would not put the power supply section in the upper left of the PCB, with the signal buffer circuits lined up along the left part of the PCB, since the bottom amp will share ground impedance with all of the amps above it and vise-versa. It's better to put the power supply circuit near the middle of the PCB, between the analog input and analog output areas.

Maybe post a floorplan of the PCB when you get farther along, and we can comment.

 

[berkeman]

Oh, and stitch the top layer ground pours with vias to the inner ground plane, with enough stitches to ensure good ground integrity at whatever frequencies you're working at.

 

[berkeman]

Also, I've had good luck with the MAX4644 analog switches. I used them in a recent emulator board for a mixed signal ASIC, and they performed well. They're a little pricey though.

 

[Topher925]

Thanks for the advice. I'll post a PDF of the basic layout as soon as I get it figured out.

The MAX4644 does look like a nice chip but I need +15v operation.

 

[Mike_In_Plano]

The two weirdnesses I've seen with this are:
1. What leaks through the "off" switches
and
2. little glitches that occur when the switches transition.

As for distortion, make sure that your signal level stays away from the switche's power supply rails and that you've got your supplies and drive levels correct ( I once had everything else right, but the drive level was wrong and had trouble ). Also, don't pass current through your switch (i.e. follow with a non-inverting op amp circuit instead of a summing node).

Getting back to the cross talk, I found a method that serves me pretty well. Place a 2:1 switch on each incoming line before multiplexing. On each channel, run the normally open positon of the switch to the incoming channel and the normaly closed position to the signal ground. This will load the output line when you have that channel turned off and keep you from having to worry about stray pick up.

Then, follow with 8:1 mux, or if you're running differential signals, (always best for fast and easy success) use two 2:1 for each channel and two 8:1 muxes.

As for the glitch issue, it only happens during switching and the easiest way to mitigate it is for use parts with a 5 pico-Coulomb (or less) charge injection and go with a differential signal.

Now, if you like most of use, you're using a lock in amp because your signal is small and is close to totally buried in the noise. The muxes and their associated digital circuits will do wonders to make this problem worse. So, you probably want some low noise gain before you go into the mux circuits.

Best luck with it, I wish you well

- Mike

 

[berkeman]

Getting back to the cross talk, I found a method that serves me pretty well. Place a 2:1 switch on each incoming line before multiplexing. On each channel, run the normally open positon of the switch to the incoming channel and the normaly closed position to the signal ground. This will load the output line when you have that channel turned off and keep you from having to worry about stray pick up.

That's a great idea, Mike. Thanks.

 

[dlgoff]

I found a method that serves me pretty well. Place a 2:1 switch on each incoming line before multiplexing. On each channel, run the normally open positon of the switch to the incoming channel and the normaly closed position to the signal ground. This will load the output line when you have that channel turned off and keep you from having to worry about stray pick up.

That's a great idea, Mike. Thanks.

That is a great idea. I wish I would have thought of that back in the '70s when I was designing this music synthesizer that mixed 8 audio signal to 1. Then filtering this output and mixing those components again with another 8 to 1 analog switch. However the cross talk issue wasn't too noticeable for this application.

 

[mheslep]

These analog muxes always specify crosstalk. Here, on page 3 of the OP's ADG1207, the spec is 85 db. In most cases that come to mind that is ample, so I wouldn't second guess the mux itself, and instead concentrate on preparing the incoming signals (terminate when not selected, and/or add gain) as Mike-I-P suggests.

 

[Topher925]

That's a great idea, Mike. Thanks.

That is a great idea. I'm working it into my design right now.

 

[Topher925]

Now, if you like most of use, you're using a lock in amp because your signal is small and is close to totally buried in the noise. The muxes and their associated digital circuits will do wonders to make this problem worse. So, you probably want some low noise gain before you go into the mux circuits.

I probably should have mentioned that I have low noise gain coming out the wazoo before the muxes. The low-noise gain circuit I had designed for this project uses a common ground but I'm thinking about redesigning things to make everything completely differential or perhaps just differential after last op-amp. In this case I want fast and guaranteed results, don't really care so much about how elegant or cheap things are.