| 摘要 |
A method and apparatus for classifying signals into a multiplicity of signal classes which employs discriminant functions of low-complexity discriminant variables that are computed directly from the passbanb signal. One possible discriminant variable is the central second-order moment, normalized with respect to signal power, of the result of applying a nonlinear operation, such as rectification, to the passband signal being classified. Other possib le discriminant variables are the lags of the autocorrelation sequence of the passband signal, normali zed with respect to the signal power. The discriminant functions can be linear or quadratic function s of the discriminant variables. The accuracy of the classification functions is maximized by empl oying discriminant functions that are selected to minimize, in a statistical sense, the probabi lity of making erroneous classification decisions. The method can be applied to the problem of classi fying voiceband data (VBD), facsimile (FAX), native binary data, and speech on a 64 Kbps digital channel. The method may be implemented using exclusively linear or exclusively quadratic discriminant functions of the discriminant variables. Hybrid two-stage systems may be emp loyed to reduce the computational requirements at the cost of an acceptable loss in classificati on accuracy with respect to a classifier that uses only purely quadratic discriminants functi ons. In such a hybrid system the first stage employs linear discriminant functions to make classif ication decisions into a smaller number of possible preliminary signal classes. The decisions of the first stage are then refined by a second stage that uses nonlinear discriminant functions such as quadratic or pseudo-quadratic functions. The second stage of a hybrid classifier then ass igns the signal into a larger number of possible classes than does the first stage of the classifier alone .
|
