A novel software algorithm to maximize the accuracy of high-resolution analog-to-digital (A/D) converters is presented. Implementing this approach, the data acquisition system is required to use only one hardware voltage reference as an input of the A/D converter. The value of this reference used by the software algorithm is optimized for narrow input ranges. The transfer function (TF) of the A/D converter is linearized locally, and the result is an overall error of approximately 1 least significant bit (LSB), greatly improving existing systems. When wide input voltage ranges are desired, the algorithm is used to derive a piecewise, software-compensated approximation of the transfer function of the A/D converter.

High-end data acquisition systems generally use at least one stable, high-accuracy voltage reference to compensate for temperature drifts on the TF of the A/D converter and associated circuitry (a general block diagram of a data acquisition system is shown in figure 1). The accuracy of the A/D converter over the temperature range is dominated by the drift performance of the voltage reference. Among the other sources of errors that affect A/D converter accuracy are noise of the voltage reference, integral linearity error, differential linearity error, transition noise, full-scale error, and full-scale error drift. Because of all these errors, it is extremely difficult to obtain accuracies better than a few LSB’s when using existing approaches.

As an example, in a 16-bit A/D converter, it is acceptable to have 12 bits of accuracy. Normally, the device could have a resolution of 1 to 2 LSB’s, but when issues like noise, drift, and nonlinearity are considered, the accuracy of the A/D converter is not much better than 12 of 16 bits. When 12-bit accuracy is not sufficient and speed requirements do not allow the use of higher-resolution A/D converters, some algorithms may be used to increase the effective accuracy by 1 or 2 LSB’s. Yet, accuracies of 14 bits or better are very hard to achieve with existing 16-bit A/D converters and existing software algorithms.

When narrow input voltage variations are expected, the TF of the A/D converter may be locally linearized to achieve the highest accuracy within that narrow voltage range. Figure 2 better describes the concept.

Figure 1. General Data Acquision Block Diagram

The locally linearized, software-compensated TF of an A/D converter using a single hardware voltage reference and ground for compensation is given by (1):

(1) where is the number of counts (quantization levels) of its respective voltage, V_ref is the value of the voltage reference used in software (variable we want to optimize), and V_out is the software-calculated representation of V_in. Note from figure 2 that the error between the linearized TF and the actual TF [e₂] is greater than the error between the locally linearized TF and the actual TF [e₁] for a given input voltage: .

Following the same approach, we may define an array of voltage references to provide maximum accuracy through a piecewise linearization over a wide input voltage range (see figure 3). The value of the voltage reference used in software will be determine by the input voltage and by the simple software rule given by (2):

(2)

Key accomplishment:

• Successfully implemented algorithm on Shuttle Tire and Strut Pressure Monitor (TPM) with greatly improved measurement accuracy.

Key milestone:

• Formalize software algorithm for implementation in several existing projects.

Contact: J.M. Perotti (Jose.Perotti-1@ksc.nasa.gov), YA-D5-E, (321) 867-6746
Participating Organization: Dynacs Inc. (Dr. P.J. Medelius, Dr. C.T. Mata, and B.M. Burns)