Quick general summary: In analog time division multiplexing, after combining several continuous signals into a single line, do you generally need to use some form of sample and hold to reconstruct the continuous signals on the demux output side? Or do mux IC's do this automatically?

My specific application: I'm collecting 256 voltage signals from tiny wires embedded in brain tissue, eventually collecting them on a DAQ card and doing online analysis. The multiplexer (perhaps ADG1606 on both mux and demux side) is intended to reduce the wire-count in the long, expensive cable bundle that connects the on-head signal buffers to the rest of the amplification and digitizing hardware (please see figure - I hope the size is appropriate). From the answers I've gotten so far, I drew what seems to be the necessary general idea, and I'm wondering about specifications and unforseen limitations.

For the sample-and-hold, I'm considering this setup, except that the droop rate (2mV/ms) and sample time (3ms) seem bad for my application. Maxim seems to offer inexpensive SH packages with better specs (DS1843 DS), would you recommend using those instead of making SH stages from opamps and switches as the tutorial outlines?

Am I right in imagining that independent SH circuits synced with the mux timing will reconstruct the original independent pre-multiplexed signal (at least the lower-frequency parts that I care about - 0.1Hz to 9kHz)? Or does the SH switching introduce periods of corrupt output data?

May I skip my late 5000x amplification step (256 channels) by 5000x amplifying the 8 mux lines just before demux? This would save me some space and money.

And, do I need to lowpass filter the brain signal BEFORE it gets to the mux? I have not done this in my current setup (which doesn't have any mux), but with mux I think I may need to worry about aliasing?

You may be able to tell that I'm new to this, so clues that may seem obvious to you are still very much appreciated!

Recording setup diagram here and below

enter image description here

  • \$\begingroup\$ This qustion is too general to allow a good answer to be give. Rather than having a multi session to and fro q&a exchange why not tell us all the details available?| Knowing what originates and "reads" these signals will help in answering. | You need to advise the maximum acceptable time between reading each channel and the maximum frequency needing to be transferred end to end. What is the input amplitude and what is the amplified output level? What is the length of the wiring loom and what is the general environment? |Re "for amplification" - why? then what? ... \$\endgroup\$
    – Russell McMahon
    Nov 20, 2011 at 22:41
  • \$\begingroup\$ For the output of an analog multiplexer to be available when the channel is not "on" then you need to provide an analog memory = sample and hold. This can be reasonably low cost and simple. eg a unity gain opamp buffer using a modern low input bias current opamp will require extremely small capacitance value by normal standards to "hold up" and S&H input well enough during off periods at typical sampling rates and multiplex ratios.|ALL of the "weasel word" qualifiers in the prior sentence are intentional. Once you know what you need to do you can translate all these into actual specs. \$\endgroup\$
    – Russell McMahon
    Nov 20, 2011 at 22:49
  • \$\begingroup\$ @RussellMcMahon Thanks for the guidance. I wasn't sure how much detail people would want to read here, so your clue was helpful. \$\endgroup\$
    – ImAlsoGreg
    Nov 21, 2011 at 5:34
  • \$\begingroup\$ More tomorrow - rushing now. But - what sort of brain? Very demanding requirements for human interface under these conditions so I rather suspect that this is "something else". | Re level of detail - need to get level right is appreciated. but what you have mow provided is utterly crucial for a good solution. Consider also joining www.piclist.com Both communities will have value on many questions. Their less format allows you to wander though more peripheral aspects more readily. [I'm at both places as are other people here]. \$\endgroup\$
    – Russell McMahon
    Nov 21, 2011 at 7:04

2 Answers 2


If you are going to continue to use 256 amplifiers, ADCs, etc. you will need a sample and hold that is triggered each time you activate the mux channel. One thing you'll have to pay attention to is the droop rate of the SH.

Alternatively you could now get away with a single amplifier, ADC, etc. for each group of 4 or 16 channels. You would know which channel was being sampled because you are driving the mux to select the proper channel. This would greatly reduce the system complexity if you are able to do it.

Note for any of these configurations you will now have to send several signals to your mux to select the proper channel. This means you'll need some device to handle the sequencing, and this device could also be used to handle the triggering of the SH amplifier, or the per block ADC.

  • \$\begingroup\$ Thanks a million for getting me started. I rephrased the question and added a figure to see if I could get a few more bits of info on some specific things before I mark the question answered. Will upvote as soon as I'm allowed to by the site! I like the idea of cutting down the number of amps & ADC's by keeping the signal multiplexed. Except that I also need to bandpass filter (I imagine that can only be done when the signals are broken back out to independent channels). I think I can use this idea to save an amp step though. \$\endgroup\$
    – ImAlsoGreg
    Nov 21, 2011 at 5:55

A few comments:

  • You really want to do your amplification before going through the MUXes. Each mux will introduce some series resistance, as well as causing small spikes when it switches (see the datasheet, it's described as "Charge Injection"). Furthermore, the mux has some distortion, as well as non-linearities.
    Basically, you want as big a signal as possible, to reduce the contribution of the multiplexers.

  • Why are you demuxing at all? Most NI DAQ systems have their sample rate described as "Aggregate", which means that internally, they have a single channel ADC, and a bunch of muxes. As such, the DAQ switches the ADC between channels very rapidly, to sample multiple channels.
    If you have a data acquisition system where the sample rate decreases as you use more channels, it's probably internally multiplexed already.
    The only advantage I can see to demuxing before entering the DAQ is that it lets you put the filters there, rather then at the preamp.
    On the other hand, a simple 3 or 6 dB per octave filter is very compact, so it should be pretty simple to put it at the preamp end.

If you have long wires (and you do), you should really consider using differential signaling. With the signal levels you are working with, you will have EM pickup issues is you use single ended signaling, particularly if you do not amplify the signals before sending them through the cable.

Also, shielding the 3m cable is a must.

  • \$\begingroup\$ Re: amplification & filtering before MUX, I'm reluctant to add even the necessary R's and C's on the pre-amp PCB. The brain I am recording is a brain inside a rat running around in a maze, so everything is as small and light there as possible. But if these steps are necessary to get things working, I will put some time into trying to pack them into the design. \$\endgroup\$
    – ImAlsoGreg
    Nov 21, 2011 at 18:14
  • \$\begingroup\$ Re: Allowing the NIDaq card to demux for me, it's a really interesting idea. I don't think I will stay with the NIDaq cards forever though, so I'm worried that designing the rest of my front-end around the specifics of those cards will leave me with something unportable. There's also the issue of getting the right gate sequence out from the NIDaq cards - I believe users can get the master clock signal, but I haven't seen a way of getting the 5 MUX address lines. Do you know anyone who has done this trick? I will ask them on their forums now, thanks for the idea. \$\endgroup\$
    – ImAlsoGreg
    Nov 21, 2011 at 18:19
  • \$\begingroup\$ @ImAlsoGreg - allowing the DAQ card to do your demuxing will not render the system unportable. Many (I'd say most) multi-channel DAQ systems use a single, switched, fast ADC to sample multiple channels sequentially. Anyways, really, any DAQ which has a single ADC channel that can take data at your samplerate * number of channels will work. It's not NI specific at all. \$\endgroup\$ Nov 22, 2011 at 1:17

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.