An algorithm for positive time-frequency distributions

The time-frequency distributions (TFDs) of Cohen's class provide a method for high-resolution representation of time-varying signals. This benefit does not come without cost; the most noticeable is the existence of negative values in most desirable fixed-kernel TFDs. We examine the relation between the negative values and several aspects of TFDs, such as cross terms and marginal properties. We suggest that it may be advantageous to obtain positive TFDs which at the same time satisfy the marginal properties. A post-processing algorithm is developed to modify any existing TFD to provide a corresponding positive TFD. It is demonstrated in computer experiments that the high resolution property is preserved quite well.