Intrinsic discriminant dimension based signal representation and classification

Generally a set of signal specific features such as energy, frequency change, are used for the representation and classification of signals of interest. However, for robust representation and classification features that are not so signal specific such as a measure of information content (e.g., Renyi entropy) and measures of statistical properties of signals such as kurtosis and skewness are needed. In this paper we derive such features. Further, in this paper, an information bound based measure is developed to find the minimum dimension of the feature set that is needed for an optimum signal representation. Similarly, a decision boundary based intrinsic discriminant dimension of a feature set that can be used in optimum classification is developed. These features are verified using different signals. The minimum set of features obtained using the information bound for optimal signal representation seems to be the same - Renyi entropy, skewness and kurtosis for all signal types considered in this paper. Similarly, a subset of these features obtained for the optimum classification seems to be the same - Renyi entropy and skewness for all signal types considered here.