0
\$\begingroup\$

Would there be any benefit to splitting analogue sensor signals and sending them in parallel through a low res & high throughput rate ADC and high res & low throughput rate ADC, and analysing/comparing the separate data/sticking it together? Any systems this is used in? Maybe the signal would be somewhere imbetween the ADC specs (ie. bandwidth is double/equal the throughput rate and SQNR is high/low)

(

Previously I have used very standard sensor set ups with Arduino Mega etc. This week I have been venturing into EE territory to improve the bit between 'resistance information' and network training. WOW, analog electronics is complicated, hats off to you guys in the know! I mostly work in physics/mathematical modeling of complex systems, so some basic theory is there, however really I have only been doing any EE for 5 days now.

@Wesley Lee The application is using neural networks to predict the various strains from sets of resistance measurements of a stretchy sensor array to predict its position/state. If it is then I might set something up and see if any significant increase in information (obviously there will be more data points) arises from it. If nobody does this anywhere, that would suggest it isn't worth it.

\$\endgroup\$
  • \$\begingroup\$ Are you asking this hypothetically or do you actually have an issue in mind? \$\endgroup\$ – Wesley Lee Nov 17 '16 at 1:52
  • 1
    \$\begingroup\$ the benefits of added complexity can be combined in one solution defined by better specifications for sensor bandwidth, SNR, stability, averaging, alarm thresholds ..etc. 1st define functions, resolution, accuracy, SNR, interference and outputs with signal rate and averaging , limits, tolerances for all inputs, temp range before you even think of design!! \$\endgroup\$ – Sunnyskyguy EE75 Nov 17 '16 at 1:53
  • 1
    \$\begingroup\$ Fast and slow are really bad quantifiers! What is the bandwidth of the signal and what are the sampling rates of the two ADCs? Also what is the desired resolution and the bit rate of the ADCs? Only then we can have a conversation. \$\endgroup\$ – vini_i Nov 17 '16 at 1:53
  • \$\begingroup\$ @Tony Stewart Yes, I like your definition, and see your point. I have expanded a little, any areas you know that do this? \$\endgroup\$ – ben horseradish Nov 17 '16 at 4:33
  • \$\begingroup\$ @vini_i They certainly are! They are merely relative terms, I'm not really looking for a numerical breakdown of anything. \$\endgroup\$ – ben horseradish Nov 17 '16 at 4:33
1
\$\begingroup\$

GREAT, I have found a very recent book on this, if you are interested. It is called dual path parallel sampling / parallel sampling :

Power-Efficient High-Speed Parallel-Sampling ADCs for Broadband Multi-carrier Systems (Analog Circuits and Signal Processing)

'The authors focus on exploiting the a-priori knowledge of the system/application to develop enhancement techniques for ADCs, with particular emphasis on improving the power efficiency of high-speed and high-resolution ADCs for broadband multi-carrier systems.'

Implementations of the merits of parallel pipelined low res and SAR high res are discussed. Particularly good for increasing the dynamic range.

\$\endgroup\$
1
\$\begingroup\$

Analog sensor signals are often considered slow changing (strain gauges, photoresistors). A typical digital sampling rate for industrial-scale systems, that do logging on sensors is 1 Hz. A good audio ADC will have 20 bits of effective dynamic range. This is huge, requiring better than 120dB SNR from the rest of the system if you're gonna do bit accurate measurements per sample. Delta-Sigma converters can convert with 48kHz and up to somewhere around 2-300kHz.

Personally, I recommend going the route of an audio ADC if you need REALLY high dynamic range. For practicality IMO a 14-bit ADC (you can get these even faster) will do the job. Average some samples and you have a nice value with a low sigma. If the sensors are connected in a Wheatstone bridge, I recommend linearizing the bridge with a constant current drive, to increase the linear range of measurement.

My bottom line is, I cannot see a reasonable reason to do double ADC-ing, except for increasing both cost and complexity. Modern ADCs are so broad and cover so many niches, you should be able to find one that does everything with the needed speed and resolution.

For getting a good sensor value, maybe look into kalman filters, they do an excellent job if you give them enough variables, and they are super fast. (for a kalman filter, speed is more important than resolution, though both help)

And since you're working with a sensor array, you probably would like to use multi-channel ADCs.

I hope your project is going / went well.

( I know it's a late answer, but maybe someone else can benefit from it as well )

\$\endgroup\$

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.