• State Resolved
  • Locked Locked
  • Replies 3 replies
  • Subscribers 46 subscribers
  • Views 3682 views
  • Users 0 members are here
Support feedback
Options
Options
Related

This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

Multiple Channels from a Single Channel ADC - Is it feasible?

Alex52381
Intellectual 810 points
Other Parts Discussed in Thread: ADS7951

Hello,

I'm looking to add an analog to digital converter to my design. After looking through the various TI options, I'm starting to think that it might be more cost effective to buy a single channel ADC and somehow multiplex diffrent analog signals to it. I know this can be done, as various micro controllers use this technique. I don't know how it would be implemented outside of a micro controller however.

My design requirements are:

- Have an analog bandwidth of 1 MHz (and be able to digitize this to obtain a good representation of the original waveform)

- Interface to a standard micro controller or some programmable logic device

- Have two channels

Using the rule of thumb that you need at least 10 points over a period of a waveform to re-generate the waveform acceptably, I think that I would need at least a 10 MSPS converter. If I multiplexed this into two channels, that means that I would need a 20 MSPS converter. I would probably need to have transmission gates as the analog mux.

Does anyone have any ideas or insights on this before I purchase my hardware?

Thanks a ton,

Alex

  • 0
    Genius 4630 points

    The short answer is: Yes.

    Texas Instruments even makes ADC chips with as many as 16 inputs which are internally implemented as an analog mux following by a single-channel ADC.

    The long answer is that if you are going to build your own, then you need to be selective about the technology used in the ADC.  There are at least three kinds of ADC technologies: Successive approximation, flash conversion, and delta-sigma.  The latter, delta-sigma, is not generally appropriate for the situation where multiple signals are multiplexed into one ADC.  That's because the delta between two unrelated signals is not the same as the delta within a single channel.  I believe that there was a Texas Instruments article describing how you can get around this by running the delta-sigma converter through several samples before switching to the other channel - enough to clear out the digital filter and get a true conversion - but this would increase your 20 MSPS rate several times.  Not just delta-sigma, but any ADC with digital filtering built in will not be generally appropriate for multiplexing several signals into a single ADC, due to the impulse response of the digital filter.

    That said, any successive approximation or flash converter will work, and your analog multiplexer can support two channels, four, eight, sixteen, thirty-two, or even more.

    I have been using the ADS7951 to great effect, but it is limited to 1 MSPS.  In your application, that would result in 500 KSPS per channel.  I'm sure that Texas Instruments has a faster 2-channel ADC.

    I'm sure you know that 2 MSPS is all you need to recreate an analog signal of 1 MHz via sampling, according to Nyquist.  I'm not sure where you got the idea that 10 points are needed to regenerate a waveform.  But it's true that if you want to view an analog waveform on a digital scope, you'll need a much higher sample rate.  The real test is whether you need to recreate square waves, because it takes several harmonics of the fundamental frequency in order to regenerate a square wave on output.  If you can qualify the true bandwidth needed on your output signal, then you really only need 2 x the rate in order to regenerate the original, band-limited signal.

    Another consideration is bit depth.  As far as I know, the only technology which approaches 24 bits per sample is delta-sigma, which is challenging for multiplexed signals (needing many more times oversampling to clear the impulse response).  Most successive approximation converters are limited to 12-bit, 14-bit, or 16-bit.  Flash converters seem rare, and usually have an even smaller bit depth due to the massive silicon real estate needed when a comparator is allocated to every possible code (e.g. an 8-bit flash converter requires 256 comparator circuits!).  Again, if your application is a digital sampling oscilloscope, then you really do not need any more than 16 bits, and you could even suffice with less.  If you think that you need 24-bit accuracy, then be prepared for an extreme challenge to design a circuit which approaches 144 dB of signal-to-noise performance.  The best I've seen is 110 dB, and you'll be hard-pressed to exceed 100 dB, which is equivalent to 16-bit or 17-bit.  Your best bet for a two-channel multiplexed ADC design is to stay at 16-bit or less.

  • 0 in reply to Brian Willoughby
    Intellectual 810 points

    Thanks for the quick reply.

    I decided that I'm not intrested in the extra hassle of doing multiplexing by myself. I'm aiming towards a dual ADC device.

    The issue that I've run up against now is the choice of anti-aliasing filter. I understand that for maximal flatness, I need a Butterworth filter. With an 8 bit system, I would need to reject 48 dB in the stop band. Let's say I sample at 10 MSPS, and that I am intrested in visualizing signals up to 1 MHz in bandwidth minimum. I need to heavily oversample to make sure that I get many points so that I can get a good picture of the waveform.

    I'm thinking that I should design the filter so that its cutoff frequency is a bit past 1 MHz (because I don't want the amplitude errors associated with the -3 dB point). The stop band needs to be -48 dB by the time that the Nyquist frequency rolls around. By that measure, I would need a -48 dB attenuation to be reached at 5 MHz (assuming the 10 MSPS sample rate). Theoretically, this would require a filter that drops about 100 dB/decade. From the cutoff a bit past 1 MHz to the stopband at 5 MHz, that is half a decade. With the aforementioned filter, I would get -50 dB of attenuation at the 5 MHz point. I think that would require a 3rd to 4th order Butterworth. Am I on the right line of thought here?

    Thanks,

    Alex

  • 0 in reply to Alex52381
    Genius 4630 points

    Again, I should stress that you need to define the total frequency content of the signals that you want to analyze.  That's because, at the most basic level, you only need to sample at 2x the highest frequency in order to recreate the waveform faithfully.

    However, if you're talking about waveshapes with significant harmonics, then the fundamental frequency is not the frequency you should be concerned with.  Instead, you should determine how many harmonics you need to get a good picture of the waveform, and then double that to account for Nyquist.  In other words, instead of thinking of sampling at 10x your 1 MHz fundamental, you should determine the highest harmonic needed, and design for a sampling rate of 2x the highest harmonic.

    As you're finding, anti-aliasing is making the design problem harder.  Filters with a sharp cutoff are complex, but they relax the sampling rate requirements.  Conversely, a simple filter is easier to implement, but has a gradual slope and thus requires a higher sampling rate.

    I'll let someone else answer specifically about your 3rd/4th-order Butterworth and whether it is appropriate for your situation.  If I were you, I would simulate the filter in SPICE or something similar to double-check that the cutoff works as well as you need it to.  Some filter design tools will allow you to specify a maximum amplitude deviation within the passband, which should address your concern about amplitude errors at the -3 dB point.  Also, don't forget to look at the phase response of the filter, particularly near the "cutoff."

    In any event, I'm hoping that someone else has more time to chime in here with better advice.