ENOB vs Resolution
I'm trying to clarify my understanding of the relationship between ADC resolution and ENOB (Effective Number of Bits), particularly when it comes to practical calculations. From what I understand, ENOB represents the actual performance of an ADC accounting for real-world imperfections like noise and distortion, while the resolution is the theoretical bit depth of the converter. Since ENOB is invariably less than the stated resolution, I'm confused about which value should be used in calculations.
Specifically, when calculating the LSB (Least Significant Bit) step size, should I be using: 1 LSB = V_ref / 2^Resolution (the theoretical value) or 1 LSB = V_ref / 2^ENOB (based on effective performance)?
If ENOB is the more meaningful metric for actual performance, why do datasheets prominently feature the resolution specification? Is the resolution purely a marketing/theoretical spec, while ENOB tells the real story? Or do these two parameters serve different purposes in system design?
I'd appreciate any insights into when each specification is relevant and how you approach LSB calculations in your designs.
My current understanding:
Resolution defines the theoretical quantization - a 16-bit ADC has 2^16 discrete output codes, so with a 1V reference:
1 LSB = 1V / 2^16 = 15.26 µV (the step size between codes)
ENOB represents the actual usable performance accounting for real-world noise and distortion. If the same 16-bit ADC has ENOB = 14 bits, then:
Effective noise floor ≈ 1V / 2^14 = 61 µV RMS. Signals smaller than ~61 µV will be buried in the noise (hard to distinguish from random fluctuations)
This means that while the ADC outputs 16-bit codes with 15.26 µV steps, the inherent noise causes the readings to fluctuate by approximately ±2 LSB, making the effective resolution equivalent to a cleaner 14-bit converter.
Is this correct?
1 answer
Your understanding is largely correct. You have touched on the difference between accuracy and precision.
ENOB can differ from raw bits not only due to inherent random noise, but also non-linearity. In that case, individual counts can still be useful in detecting small variations of a signal, but the absolute level of that signal is only known to the ENOB level.
ENOB can also vary over usage conditions. A/D converters degrade at high speeds. The raw bits remain the same no matter how you run the converter (within specified limits), but ENOB may be different at different frequencies, setup times, and conversion times.
A/Ds are also often specified as monotonic, even when ENOB is more than 1 bit less than the number of raw bits. That means that there will never be flipped codes, even if that were within the ENOB spec.
There is also a difference between random and systemic noise. Random noise can be reduced by filtering multiple readings, effectively giving you a slower but more accurate A/D. Systemic noise, like built-in non-linearity, will still be there no matter how clean you make the signals and how much you filter the result. Repeatable non-linearity can be address by calibration of individual units.
Both the raw number of bits and the effective number of bits over various operating points are useful in the design of the larger system.
0 comment threads
1 comment thread