Predicting the complexity of signal processing algorithms

Predicting the complexity of a signal processing task is a very difficult subject, mainly because the notion of complexity itself is not well defined. We propose to link the complexity to the predicted execution of an algorithm on a given platform or processor. To do so branches, arithmetic and memory operations are classified into different types of category. Then the number of time operations of each category executed is computed for the studied algorithm. Finally weights depending on the underlying architecture are used to sum up the contribution of each category to get the final complexity. Typical weights for two Pentium processors are presented and the validity of the prediction is verified for the case of a simple algorithm