Design of Frequency Domain Metric for Evaluation of Mode Decomposition Algorithms

Mode decomposition algorithms have been used extensively to extract and separate signal components from composite signals which constitute the semantic information in it. The individual signal components also called modes should be independent of each other and have some sparsity properties such as having a limited bandwidth and should be compact around a center frequency. However, not all algorithms effectively segregate the constituents truthfully. Different algorithms perform differently for the same input signal. This work proposes a metric to quantify the performance of mode decomposition techniques by leveraging the analogy of having a compact central frequency in the decomposed modes. The proposed method is tested on the widely used mode decomposition algorithms. The results on simulated and real signals show the usefulness of the proposed metric in evaluating mode decomposition algorithms.