OVERVIEW

To avoid the computational problems associated with the direct calculation of a set of adaptive weights for an array processor, both the LMS and maximum SNR algorithms can be used. It is shown in Chapter 9 that random search algorithms also provide a means of circumventing computational problems. These algorithms all have the advantage that the required calculations are usually much simpler than the corresponding direct calculation, they are less susceptable to hardware inaccuracy, and they are continually updated to compensate for a time-varying signal environment.

Another class of processors based on recursive methods can also be used to circumvent computational problems [1] [4]. The basic approach taken by these methods is to recursively perform the matrix inversion required by the direct calculation approach so that at no time is a direct matrix inversion computation required. The recursive algorithms should therefore exhibit the same kind of steady-state sensitivity to eigenvalue spread in the signal covariance matrix as that already found for DMI algorithms. Furthermore, since the principal difference between the recursive methods and the DMI algorithms lies in the manner in which the matrix inversion is computed, their rates of convergence are comparable. The recursive algorithms are primarily based on least-square estimation techniques and are closely related to Kalman filtering methods [5]. These methods assume that the sensor signals are available in sampled data form, and they define digital processors for updating the adaptive weights in the array processor. For stationary environments these recursive procedures compute the best possible selection...