Overview

The techniques presented in Chapters 5 7 apply to a complete set of data that is available after the experiment is completed. Such methods are called batch processing, postflight data analysis, or off-line parameter estimation.

It is also possible to derive parameter estimation algorithms that can be used in real time, giving interim results as the experiment is being conducted. One of the methods to do this is a recursive formulation of ordinary least squares, where parameter estimates are calculated at each sample time when new measured data are available. This is called recursive least squares. The recursive nature of the calculation avoids reprocessing of old data, making the procedure efficient for real-time operation. The extended Kalman filter, which involves augmentation of recursive least squares with information about the noise processes and the dynamic system, can also be used for real-time parameter estimation. Another approach is to use batch methods on recent stretches of data to approximate time-varying parameter estimation. This is usually referred to as sequential least squares. Real-time parameter estimation can also be formulated in the frequency domain, using a recursive finite Fourier transform to provide data for a least-squares solution. All of these methods would be classified as on-line processing or real-time parameter estimation. The short list of methods just given is by no means exhaustive, but all of them have been applied to aircraft system identification problems. Each of these methods will be described in this chapter.

An important application of real-time parameter estimation is characterizing...