Signal Classification

Signal classification - the act of taking in raw signal samples and making an action based on the `category' of the signal - becomes more and more popular in autarkic communication systems. The signal classification is related with the estimation of signal parameters - the act of recovering useful information about important parameters like carrier wave, carrier phase, modulation type and noise power. Classification and parameter estimation are often implemented by complex mathematical algorithms which are distinguished in effort and in performance. The higher the performance is, the higher the effort usually is.

In general, a specific number of samples is needed to achieve an expected performance. This number depends on the considered signal types and on the applied algorithms. One question arises consequently: is it necessary to increase the number of samples in order to step up the performance, or can we achieve the same performance by applying a more complex algorithm with the same number of samples? In both cases, the amount of extra effort is important so as to answer the question.

The amount of extra effort is usually very difficult to calculate, because the self-information of one sample depends on the corresponding signal and on its underlying probability density function. Often, mathematical bounds and stochastically methods help us to estimate the self-information of one sample and its amount of extra performance. But in general, this bounds with their approximations are very weak so that the need for sharper bounds arises. Information theoretical concepts help in finding relationships between self-information of one sample and the corresponding amount of extra performance as well as extra effort while using the ideal signal classifier or the best parameter estimator. Thus, we investigate the calculation of sharp bounds by information theoretical concepts in order to estimate the self-information reliably. Afterwards, we employ the solution so as to approximate the amount of extra effort and extra performance.


For further information contact Gholamreza Alirezaei.