Multiplexing high impedance signals - too much crosstalk from driven signals

Thread Starter

novacircuit

Joined Jan 27, 2018
11
I've been trying to create a 3-channel ultrasonic transceiver (schematic attached), but I'm running into issues with crosstalk between the channels when I transmit. Each 40khz transducer is attached to X1/X2/X3. A transmitter channel is selected on IC2 DG409DJ (TRN_SEL_0, TRN_SEL_1), and the transmitter signal (TRN_SIG_IN) is a 4-pulse 40khz sine wave @ +13/-13V. The receiver channel is selected (RECV_SEL_0, RECV_SEL_1) on IC1 DG409DJ. I've attempted to additionally isolate the signals by using an ADG419 SPDT on each channel, which selectively allows X1/X2/X3 to route towards IC1 or IC2 (transmit or receive).

I tried out the design on a trace-milled PCB, and it does work, but I'm seeing unacceptable crosstalk between the channel that's transmitting and the amplified receiving channel (signal plot attached). The received pulse always contains an attenuated copy of the transmitted pulse. The transducer cables are shielded and moving them around or holding them both in my hands have no effect, and the specs on the DG409+ADG419 should give me something like 75+85dB of isolation, so the only thing I can figure is that the board traces are perhaps coupling?

I am using DIL sockets with DIP8/16 multiplexer ICs inserted into the sockets. Would foregoing the sockets or switching to SMD variants of the ICs significantly reduce the coupling? I've tried to keep the traces as short as possible, but I'm really not very good at board (or circuit!) design. Does anyone have any suggestions about how to better-layout the board? Should I look at different approaches entirely?

Thanks!
 

Attachments

danadak

Joined Mar 10, 2018
4,057
When transmitting pulse do you need to have the receiver
sampled/enabled ? Thats one way of ignoring the noise.

Another way use open drain output on a UP as an analog
switch to ground to translate the high Z line into low Z when
transmitting.

Google "pcb board crosstalk", a number of hits.

Regards, Dana.
 

Sensacell

Joined Jun 19, 2012
3,432
Transmitting and receiving at the same time implies an enormous difference in signal levels.

The scope traces looks pretty clean to me, based on my attempts at similar designs?
Typically a 'BLANKING' pulse is used during TX events, to keep the RX circuits from saturating when the TX fires.
 

crutschow

Joined Mar 14, 2008
34,281
To reduce the signal further, two analog switches can be used, one in series, and one at the output to ground.
They are switched out-of-phase so that when the series one is off, the one to ground is on and vice-versa.

Alternately, add as low resistance resistor as possible on the receive trace to help absorb any crosstalk
 

Thread Starter

novacircuit

Joined Jan 27, 2018
11
When transmitting pulse do you need to have the receiver
sampled/enabled ? Thats one way of ignoring the noise.

Another way use open drain output on a UP as an analog
switch to ground to translate the high Z line into low Z when
transmitting.
I'm looking for the exact start of the signal so I really need as little noise as possible to catch the very first rise :/

"Muting" the other transmitter (or receiver) outputs is something I've thought about, but that means running long traces out to the UP which are connected to the sensitive receive lines .. could end up picking up a bunch of extra interference.

Transmitting and receiving at the same time implies an enormous difference in signal levels.

The scope traces looks pretty clean to me, based on my attempts at similar designs?
Typically a 'BLANKING' pulse is used during TX events, to keep the RX circuits from saturating when the TX fires.
There's only one RX amplifier (channel-switched by IC1) and isolated by IC3/IC4/IC5 so the amp stays nicely out of saturation. The SNR isn't awful on the scope traces but I'd really like to have it completely quiet to get the most accurate results.

To reduce the signal further, two analog switches can be used, one in series, and one at the output to ground.
They are switched out-of-phase so that when the series one is off, the one to ground is on and vice-versa.

Alternately, add as low resistance resistor as possible on the receive trace to help absorb any crosstalk
This sounds like it would work quite well, but I'd end up with two switches per channel + 2 multiplexers .. might just be easier to have a separate transmit/receive amplifier for each channel, and then multiplex the low-impedance signals on either side. I was really hoping to avoid that to save board space though.

The receiver is directly hooked to the transducer through DJ409/ADG419 when configured to do so, and should have an rON of less than 100 ohms.

I think the noise is coming from N$4/N$6/N$11 coupling together from the transmitted signal, and then those lines are coupled to N$1/N$2/N$3. I was hoping there was something in the layout I could do to help minimize it, or switching to a different package that might help ..
 

ebp

Joined Feb 8, 2018
2,332
I see no evidence of decoupling capacitors.

If the board has been milled only to remove as much copper as required to isolated the signal paths, there is probably a lot of scrap copper left unconnected to anything and therefore serving as capacitors to couple all the signals together. Copper that is not grounded is generally a menace.
 
Last edited:

Thread Starter

novacircuit

Joined Jan 27, 2018
11
I see no evidence of decoupling capacitors.

If the board has be milled only to remove as much copper as required to isolated the signal paths, there is probably a lot of scrap copper left unconnected to anything and therefore serving as capacitors to couple all the signals together. Copper that is not grounded is generally a menace.
There are decoupling caps on the xmit/recv amp (which is sitting on a breadboard). I didn't put any on the muxes because AFAIK they're not buffers and shouldn't be drawing much current directly. I put some blobs of solder between the ground traces and the leftover copper on both sides of the board. That does make me wonder if having the amps closer to the muxes would help though, as they are relatively far away (a few inches). The capacitance of the transducers is 2nF, and at 40khz it would make the reactance about 2kOhm? So about 6.5mA driving the transmit side? Is that enough current to cause inductance problems in nearby traces? But actually, looking at the scope output, the interference is in phase with the transmitted signal, so that would make it more likely to be capacitive coupling instead of inductive coupling?
 

ebp

Joined Feb 8, 2018
2,332
Decoupling caps near the analog switches would only serve to reduce noise that might be on the supply rails from coupling to the signal path via the capacitances associated with the switches. They also help tie the supply traces to the same potential as ground for AC, which may be more beneficial.

The transducers are resonant piezoelectric "crystals" so impedances not a simple thing. I've never looked at them closely enough to understand them well - I don't know to what extent the cone typically attached to the crystal influences the whole affair. You can see from your received signal that the Q is not terribly high.

In high impedance circuitry, capacitive coupling is usually a much greater concern than inductive coupling. Grounded "guard" traces can sometimes be helpful. Don't forget that analog switch isolation specifications are always conditional on some specific impedance, and usually something moderately low. The sockets probably add a couple of picofarads between the pins - about 2 megohms of reactance at 40 kHz. I suggest checking specs for the sockets.

An aside - using an instrumentation amp just to offset a signal seems a rather expensive way to do it. Be sure the reference pin of the amp is driven with a low impedance. I'm assuming the output of the amp is going to something running on 3.3 V. Your supply rails for the amp are certainly high enough to get you into trouble with high current if you have higher output amplitude than expected.

Also, if you place the signal name on or immediately beside a net on the schematic, the netlist generator should associate the name with the net, which is often useful when doing layout - consult the manual.
 

Thread Starter

novacircuit

Joined Jan 27, 2018
11
Decoupling caps near the analog switches would only serve to reduce noise that might be on the supply rails from coupling to the signal path via the capacitances associated with the switches. They also help tie the supply traces to the same potential as ground for AC, which may be more beneficial.
Ahh I see, that makes sense. The supply rails are fairly quiet from what I can measure. I tried piggybacking a few decoupling caps on top of the DIL sockets, but it didn't seem to make much difference.

The transducers are resonant piezoelectric "crystals" so impedances not a simple thing. I've never looked at them closely enough to understand them well - I don't know to what extent the cone typically attached to the crystal influences the whole affair. You can see from your received signal that the Q is not terribly high.
Ah I vaguely recall reading that "if you treat them like a capacitor, it kinda sorta works out" - in any case, the current must be quite tiny because I barely see any a difference in current consumption on my power supply when it's transmitting.

In high impedance circuitry, capacitive coupling is usually a much greater concern than inductive coupling. Grounded "guard" traces can sometimes be helpful. Don't forget that analog switch isolation specifications are always conditional on some specific impedance, and usually something moderately low. The sockets probably add a couple of picofarads between the pins - about 2 megohms of reactance at 40 kHz. I suggest checking specs for the sockets.

An aside - using an instrumentation amp just to offset a signal seems a rather expensive way to do it. Be sure the reference pin of the amp is driven with a low impedance. I'm assuming the output of the amp is going to something running on 3.3 V. Your supply rails for the amp are certainly high enough to get you into trouble with high current if you have higher output amplitude than expected.
The reference pin is driven by a 1k voltage divider and a 47uF cap. It's not ideal but it seems to be working decently well in my tests. There's a 1k resistor inline on the AD620 output and protection diodes to GND/Vcc before going to the ADC pin. The signals are clipped +/- 0.7V to the rails and are .. well, mostly in spec with what the STM32L4 can handle.

I have an older design (attached) that used an LM358. There were 3 of these transceivers on the board (one for each channel), and the extra passives took up a ton of board space. The bandwidth of the LM358 was very close to the edge of it's rating (though it made a nice filter), and it would saturate when operating in transmit mode. Recovery took a while and the update rate of the whole system was slower than I had hoped for. OTOH, the transmitters did not interfere with each other and the signal was very clean. I actually prefer this solution, but it does use 3 expensive OPA454 opamps. If I can eliminate some of the passives and prevent saturation without increasing the part count too much, I might reconsider using the old design. Do you have any suggestions for alternate transceiver designs?

Also, if you place the signal name on or immediately beside a net on the schematic, the netlist generator should associate the name with the net, which is often useful when doing layout - consult the manual.
Good to know! I have been pretty lazy in naming most of my nets and will make a point to give at least the signals a name. Thank you for all the help.
 

Attachments

Thread Starter

novacircuit

Joined Jan 27, 2018
11
Actually something that would help tremendously (no idea if it exists) would be a 4:1 multiplexer that grounds the unselected lines, rather than going into a high impedance state. Has anyone heard of something like this?
 

ebp

Joined Feb 8, 2018
2,332
The highest isolation switches use a "T" configuration - the off channels have an open switch to the input, an open switch to the output and a closed switch to ground at the node connecting the input and output switches. I haven't looked at such things in a very very long time, so I don't have a clue what might be available. I don't recall if there were muxes or just switches - of course you can make a mux with switches, but you would probably want a digital decoder for the enable signals. In auld lang syne they were used for high frequencies, including video, where the capacitance between input and output of an OFF switch was a problem. I'd probably start looking at Maxim and Analog Devices. Be prepared to run into parts that are fabulous, except they won't operate on supply rails of more than a few volts.
 

Thread Starter

novacircuit

Joined Jan 27, 2018
11
The highest isolation switches use a "T" configuration - the off channels have an open switch to the input, an open switch to the output and a closed switch to ground at the node connecting the input and output switches. I haven't looked at such things in a very very long time, so I don't have a clue what might be available. I don't recall if there were muxes or just switches - of course you can make a mux with switches, but you would probably want a digital decoder for the enable signals. In auld lang syne they were used for high frequencies, including video, where the capacitance between input and output of an OFF switch was a problem. I'd probably start looking at Maxim and Analog Devices. Be prepared to run into parts that are fabulous, except they won't operate on supply rails of more than a few volts.
Thanks for the advice. The DG333 seems like it might fit the bill. It's a 40V 4x SPDT analog switch, which could switch each channel between ground and the transmitter signal. The enable signals aren't digitally encoded, but since I'm only using 3 channels, it'll just cost one extra I/O over the two that I'm using now (really not worth it to add a decoder imo). They're not cheap, but they're cheaper than an OPA454.
 
Top