PARALLEL KALMAN FILTER NETWORKS FOR KINETIC METHODS OF ANALYSIS

An algorithm for the correction of errors arising from between-sample variations in the pseudo-first-order rate constant for kinetic methods is described. The algorithm is based on the linear Kalman filter and uses a set of models (typically 40 or more) in parallel. The best model for a given set of data is selected by examination of the innovation sequences. Limitations of this approach with respect to data range, number of data points and measurement noise were investigated with simulated data. Experimental data from the molybdenum blue method for the determination of phosphate were used to demonstrate the insensitivity of the algorithm to variations in the rate constant. Principal advantages of the new method over other error-compensation methods are its simplicity, adaptability, stability, fixed cycle time and memory requirements, and its ability to provide results in only one pass through the data. The method can be used in real time and is well suited for digital signal processors and parallel computing architectures.