5 How voltage reference noise affects noise-free resolution

While ENOB valuably represents the resolution of your ADC output, it does not account for DC performance. To understand the resolution implications of noise from a DC input to your ADC, consider finding the noise-free resolution of your circuit. Using Equation 4 you can calculate the noise-free resolution by observing the code spread in number of least-significant bits (LSB) of an ADC’s digital output while measuring a DC signal:

To highlight the impact of reference noise on the system precision performance, my colleagues and I conducted DC code spread tests for a given signal chain using the REF70 (with 0.23ppm_p-p flicker noise) and the REF50 (with 3ppm_p-p flicker noise). Both the REF50 and REF70 are high-precision voltage references used with high-precision ADCs, and have differing DC characteristics. However, in this exercise, the goal was solely to compare the noise performance of these devices in a signal-chain circuit.

The design uses batteries for a stable DC source with a voltage level close to the full-scale range of the ADS8900B 20-bit SAR ADC, which captures data at 20kSPS. OPA2320 is used with a gain = 1 to drive the ADS8900B inputs. This ADC integrates the reference buffer driver; therefore, an optional reference buffer is not required. Placing a simple resistor-capacitor low-pass filter on the output of the voltage reference further lowers the noise from the voltage reference. Figure 3 shows the setup used for these tests.

The signal-chain components beside the voltage reference also have flicker noise, which will be part of the final code spread. Because the signal chain remains the same with different references only, the impact on performance numbers must be from the voltage reference noise only.

High-precision systems employ data-processing techniques to improve the precision and increase the overall resolution. In this experiment, we converted the 20-bit raw data from the ADS8900B to a 24-bit length by multiplying the output by 16. Different finite impulse response (FIR) filters processed the converted 24-bit data. FIR filters are easy to implement and settle faster if there’s a change in the input values. The output data rate remains at 20kSPS, but with latency as defined by the filter characteristics.

At a 24-bit level, the noise (and thus the precision) of REF50 and REF70 are almost similar, with the overall noise dominated by the signal chain and its wide bandwidth noise. The difference in the average code value is because of the reference voltage difference – an accuracy specification that you can eliminate through calibration. These results can be seen in Figure 4 and Figure 5.

We used the Octave tool to conduct post-processing of the raw data with three different digital filters:

• 1,024-tap moving average filter.
• 801-tap 17Hz low-pass filter.
• 455-tap 36Hz low-pass filter.

Figure 6 shows the filter response for these filters.

Figure 7, Figure 8 and Figure 9 illustrate the impact of the digital filter on the code spreads.

Using Equation 4,you can easily compare the impact of the REF50 and REF70 with each filter profile on the ADC resolution. The results from these tests are summarized in Table 1.

This comparison shows that in the highest precision applications, the REF70 performs better than the REF50 when calculating noise-free resolution, mostly because of the devices’ difference in flicker noise levels. The reduced code spread when using the REF70 shows that its ultra-low noise can offer nearly a 2-bit resolution advantage in high-precision applications. Additionally, we can see that using a low noise reference allows using a fast 455 tap filter while still being able to maintain a high noise free resolution. Low voltage reference flicker noise will lead to lower code spread, thus enabling higher noise-free resolution. Like the ENOB, noise is an important consideration when designing your signal chain for low noise-free resolution.