# Non-stop ADC for full quantization of arbitrary time period signal average. Do ADCs work like this?

Lets say I have a light intensity sensor and I want to measure the exact average light intensity for one hour using an ADC sampling time that is much less than the input signal frequency. Lets also say I could setup as many ADC's as needed so that the moment one is converting, one is sampling... no overlaps or gaps between any sampling period.

Will this allow me to obtain the exact average light intensity for as long as this sampling happens?

From my understanding an ADC uses an RC circuit, charging up to the voltage of the source. Does it also deplete if the voltage drops during the sampling?

A simplified example:

The ADC has a sample time of 1ms. The sensor gives out a voltage equivalent to 20 units for 0.5ms. The next 0.5ms of sampling the sensor is at 0 units. Do I get an ADC value of 10 units for that sample period?

I am not after any particular signal frequency, just the average intensity. Is it possible to do this or do I have to focus on ensuring the sample rate is >= the Nyquist frequency to obtain this data.

• Is the light intensity really varying that much? How about putting an analog RC filter on the input which will do the averaging for you? Nov 6, 2014 at 9:42

Most types of analog to digital converters have intervals of time when they are sampling their input, and intervals of time when they are not. If one needs a time-averaged signal, there are two ways of accomplishing this:

1. Filter the signal significantly before sampling; if the cutoff frequency for the filter is well below half the sampling rate, the output of the filter will not vary much in the interval between samples. Consequently, no matter when samples are taken they will be relatively close to the average signal value.

2. A Delta-Sigma modulator, as MarkU noted, samples continuously. Not all delta-sigma ADCs do so, however. If a delta-sigma ADC counts off fixed intervals and generates a reading each time, then it may be able to guarantee that a long-term average value of the readings will equal the average value of the input signal. If the converter only generates readings when triggered, however, or if not all of its readings are captured, then such a guarantee would not hold.

Both filtering and the use of delta-sigma converters may offer useful averaging behaviors. Which is better in a particular application will depend upon one's exact requirements.

Sounds like you want a Sigma-Delta Modulator (or Delta-Sigma, I've seen it written both ways). An SD modulator doesn't really have a sample period, it just dithers its one bit output in proportion to the analog input signal. You could then count off, say, 1024 bit periods and count up how many of those periods the modulator output was 1, use that count as your ADC ouptut code. The longer you sample, the higher the counts, and the greater the resolution. (Up to some analog accuracy limitations of course.)

For further reading: How a Delta-Sigma Modulator Works

• In addition to this answer: most A/D converters do NOT work the way RobC seems to want: they charge a capacitor up to the input voltage, and then disconnect it from the input for the duration of the conversion. Nov 5, 2014 at 7:31
• I'm not worried about that conversion period because I can synchronize multiple ADCs so there is always non-overlapped sampling happening.
– RobC
Nov 6, 2014 at 3:02

You would need to supply a bit more real world data for accurate answers. What sort of resolution (bits) and speed (conversion time) you are hoping for. There are many possible options and modern systems may be fast and accurate enough for you that the technology does not matter.

In the past high resolution was achieved with sigma-delta converters because they were very accurate in radiometric circuits that made them very cheap for multi-meters and such that wanted lots of resolution (even 18 bits) but speed was not critical with conversion results taking 50 ms. Digital storage oscilloscopes would have used a flash converter that has a comparator for each voltage level and binary output logic and give the digital result at the propagation delay speeds of the device, these days they offer speeds of 500M samples per second and over 8 bits. A basic micro-controller may offer 10 bits at 10k samples/s.

A sample and hold circuit is used if the signal changes faster than the AD converter can convert the result.

For higher speeds two flash converters are sometimes used with two sample and hold circuits and they alternate to achieve twice the throughput so one is holding and converting while the other is sampling and settling.

There are also light to frequency converters that are sometimes more convenient to interface to digital systems that have wide operating ranges but limited response times.

• Ideally >= 16 bits and as fast as possible with a reasonable component price. My primary concern is if there is a high frequency component, even though I may not be able to see it, the resulting data includes that information, even though it may be smoothed out over the sample time.
– RobC
Nov 6, 2014 at 1:44
• Look at Digikey and search for ADC's that have 16 bit resolution. I suspect you could pretty easily get 20 to 50 MHz sampling frequencies, which would give you frequency resolution to at least 10 MHz. How fast does the light intensity sensor respond to changes in light? Nov 6, 2014 at 3:35