The influence of Bivariate Empirical Mode Decomposition parameters on AI-based Automatic Modulation Recognition accuracy

On the one hand, the AMR (Automatic Modulation Recognition) realm has recently shown an increase of interest, particularly as an application for monitoring the physical layer
of wireless transmissions. It consists in determining the employed modulation type of a sensed Radio Frequency (RF) signal at a given time, space and frequency. Moreover, it is a key component of intelligent radio systems such as Cognitive Radios (CR) that
are key devices for Massive IoT (MIoT), autonomous cars, drones, 5G, 6G, etc. On the other hand, Bivariate Empirical Mode Decomposition (BEMD) is a signal decomposition method that can distill signals into a finite number of Intrinsic Mode Functions (IMFs) through a process known as sifting. BEMD is specifically designed to decompose bivariate (e.g. complex) signals, such as complex IQ samples of telecommunication data time series. The
IMFs in conjunction with an AI architecture permits modulation classification.
This paper specifically focuses on the influence of BEMD parameters on component extraction, namely the number of applied sifts and projections. The impact of linear interpolation method vs cubic spline interpolation method is also presented.