A non-parametric test for detecting the complex-valued nature of time series

Although the emergence of multivariate signals in natural sciences and engineering has emphasised the problem of signal representation, that is, whether signals are by their nature a set of independent observations or multidimensional vectorial quantities, little research has been conducted on detecting the true nature of such signals. It remains unclear, therefore, when the complex-valued approach is to be preferred over the bivariate one, thus, clearly indicating the need for a criterion that addresses this issue. To this cause, we propose a nonparametric statistical test, based on the local predictability in the complex-valued phase space, which discriminates between the bivariate and complex-valued nature of time series. This is achieved in the well-established surrogate data framework. Results on both benchmark and real-world IPIX radar data support the approach.