In my intro to logical design class, we were shown this simple question as one of the first examples for the exercises we will have. The question was:

An analog voltage is in the range of 0–5 V. If it can be measured with an accuracy of ±50 mV, at most how many bits of information does it convey?

The answer we were provided stated that:

Accuracy of ±50 mV means that the analog signal is divided every 100mV, so we have 50 discrete measurements from 0 to 5V. Thus $$log_250= 5.64$$ bits. My questions are:

What does accuracy of 50mV really mean?

How did we come up with a division of every 100mV?

If out analog voltage was in range 0 to 6V, then would we say that 6V = 6000mv, so $$\frac{6000}{100} = 60 \text{ discrete measurements, thus } log_260 = ... \text{ ?}$$

And lastly, what am I really being asked here? What do 5.64 bits really show us?

I would appreciate any help given!!


2 Answers 2


What does accuracy of 50mV really mean?

It's an accuracy of +/- 50 mV and that means there is a spread of 100 mV. Put another way, you can't rely on a measurement being absolutely accurate so, if you measured a value of (say) 1 volt, the real voltage will be somewhere between 0.95 volts and 1.05 volts.

How did we come up with a division of every 100mV?

The +/- 50 mV inevitably sub-divides measurements into blocks having a range of 100 mV. That defines the error-free accuracy when converting the analogue voltage to a digital value.

What do 5.64 bits really show us?

It's a number that you can use to make comparisons. Many ADC specifications do this; an ADC may have 16 bit resolution but, at full scale and taking into account quantization and noise, the effective number of bits might only be 15.5. It's not a real number - it's a comparative guide number that tells you how good one ADC is versus another.


Dover Books sells a fine introduction (various probability examples, and philosophy of information theory) paper back.

On the other hand, in the game of 20_questions, people are challenged to rapidly zero in on the correct answer, using binary_chopping questions if possible.

Such as "thinking of a number between 1 and 100." How many questions are needed to detect that number?

  • < 50 YES

  • < 25 NO

  • < 37 NO

  • < 45 NO

  • < 48 YES

  • Is the number 47 NO

  • Is the number 46 YES

Except for the last question (used to verify), these are binary_chopping style of numbers.

In a pure binary_interval, 1----128 for example, the # of questions is log2(size) exactly.

Now --- what if the interval sizes, in our example being exactly ONE, are not uniform? In ADC analog_digital_converters, that non_uniformity is called Differential Nonlinearity.

All these imperfections go into ENOB effective number of bits.

  • \$\begingroup\$ I can't see what the last two paragraphs have to do with the question(s) asked. \$\endgroup\$
    – Simon B
    Commented Oct 17, 2020 at 10:51
  • \$\begingroup\$ I see the connection, but it is only half made and doesn't really answer the question. If the rest of it were there it would be OK, but there's too big a gap between "assume a spherical chicken" and "X=1" for this to be useful. \$\endgroup\$
    – JRE
    Commented Oct 17, 2020 at 11:32

Your Answer

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

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