Recently I am designing a Data Acquisition system which utilizes FPGAs and ADCs. As my project needs to have a good balance between price, performance and quality, I have looked into the market to find available medium-cost ADCs. My current selection is the single-channel AD977 ADC (100/200 Ksps, 16-bits where I need 12 Effective Bits (ENOB)) from Analog Device. Also there is the AD624 instrumentation amplifier which is again, single-channel but is great if you want to programmatically amplify your signal in the analog zone, not losing your dynamic range.

But my design requires 4 or 8 analog inputs. So what I thought was to place the ADG508 analog mux in the first stage of the DAQ, just before the near Hi-Z inputs of AD624 instr. amp.

However my instructor believes that this is a bad workaround, suggesting to use multi-channel ADCs which have the Analog MUX On-Chip which, reduce noise. Surely that would be a better choice, but if I am going to use a multiple channel ADC with the same parameters and quality of ADG508 (sample rate, SNR, SFDR, ...), I will be have to heavily increase the budgets!

I had some research on the internet for comparison of MUXes. Everyone just seem to compare various MUX ICs with respect to bandwidth and RDS(ON) (On-Resistance). From the noise aspect in Communications Circuits Engineering, the equivalent thermal noise introduced on the signal could be equivalent to sqrt(4 K T R B). But is that all the noisy source? Other people have also had the same question but in other senses, such as: How much is the noise of a multiplexer for analog inputs?

The question is: Is my discrete analog MUX solution a big deal from the signal integrity and noise aspect? Did anyone have these problems with the same approach? If it is really a bad idea, what would be other ways of solving this issue?

I have important design rules of analog systems such as grounding (Analog GND, Signal GND, Power GND, Digital GND,...) and shielding in my mind.

Please note that the input signal of each of the channels is thought to have these properties: about 10 KHz maximum frequency, maybe as weak as 1mV or even less, impedance in range of 10KOhm~300KOhm

Extra Note: Does inserting a buffer before the MUXes make things better?

Thanks to All

  • \$\begingroup\$ The AD977 is quite expensive. Given that you only need 12 ENOBs this is a poor choice. Go to digikey.com and do parametric search. \$\endgroup\$
    – Mario
    Sep 1 '16 at 14:12
  • \$\begingroup\$ That's right Mario. I have to correct my earlier saying. I need "at-least" 12-Bit ENOB. I know this ADC is expensive but I don't want to easily put a low-cost low-quality ADC for the matter. I could simply place an AVR microcontroller and use its ADC. \$\endgroup\$ Sep 1 '16 at 19:02
  • \$\begingroup\$ (For the record, in my place I don't have access to digikey.com, I wish I had) \$\endgroup\$ Sep 1 '16 at 19:02
  • \$\begingroup\$ Alright, I understand. I have one or two AD977 in my lab and almost always I find something cheaper that does the job equally well if not better. And since you care about the price, I just wanted to point that out. What about an AD7606-4 or similar? \$\endgroup\$
    – Mario
    Sep 1 '16 at 19:25
  • \$\begingroup\$ You know I must admit that when looking in digikey or directly manufacturer website, Analog Device, many other choices are possible. But in my case I must also make sure that I can find those in my country's market. Otherwise I have to import. \$\endgroup\$ Sep 1 '16 at 20:59

Is my discrete analog MUX solution a big deal from the signal integrity and noise aspect? Did anyone have these problems with the same approach? If it is really a bad idea, what would be other ways of solving this issue?

It's not a bad idea because for instance you only need one instrumentation amplifier with high gain to amplify your small signal of 1 mV but the problem comes with the output impedance of your small signal and the input capacitance of the instrumentation amplifer.

Just as a guessed example let's say the input capacitance is 10 pF; for this to reach the full potential of the input you need to "sample" for significantly longer than 5 x CR i.e. 5 x 300k x 10pF = 15 us. Like I said you need significantly better than this to charge the capacitor to get an accurate reading.

So maybe you have to "dwell" on each channel for 100 us. If you had two channels you have effectively 200 us sampling time and that means the highest frequency you can resolve (due to nyquist et al) is 2500 Hz and probably it'll be closer to 2 kHz.

So immediately you hit a problem with the multiplexer and I would advise you either buffer each signal (to give it a low impedance) then feed the multiplexer. In fact the same problem will arise with most common ADCs that contain a multiplexer so my advice is do some more reasearch.

I might add that I have designed a similar system for measuring thermocouples i.e. a DG309 (if I remember correctly) multiplexer feeding an InAmp then to an ADC BUT the scan rate is 100 Hz maximum! However, it does prove the point that noise needn't be an issue using an external multiplexer.

  • \$\begingroup\$ I also need to remark the mentioned "input capacitance". You're right about it. \$\endgroup\$ Sep 1 '16 at 18:59

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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