Hess, KristinaKristinaHessVollmer, FrederikFrederikVollmerKölpin, AlexanderAlexanderKölpin2026-06-292026-06-292026-04IEEE MTT-S International Microwave Biomedical Conference, IMBioC 2026https://hdl.handle.net/11420/63687The performance of continuous-wave (CW) radarbased non-contact cardiac monitoring strongly depends on the utilized signal representation and preprocessing. The received radar measurements can be represented as in-phase and quadrature (I/Q), phase, or magnitude signals. In addition, various preprocessing strategies have been proposed to enhance heartbeatrelated components, including bandpass filtering in the pulsewave and heart-sound frequency ranges, and the application of first- and second-order derivatives. By systematically comparing these preprocessing variants across the different signal representations, this study aims to identify which variants and combinations are most suitable for machine learning-based heartbeat detection under consistent evaluation conditions. The highest mean F1 score among individual features (94.99%) was obtained with the second derivative of I/Q signals and was slightly increased to 95.08% by combining multiple preprocessing variants.enContactless vital-sign measurementcontinuouswave (CW) radargated recurrent unit (GRU)heartbeatmachine learningpreprocessingTechnology::610: Medicine, HealthTechnology::621: Applied Physics::621.3: Electrical Engineering, Electronic Engineering::621.38: Electronics, Communications EngineeringComputer Science, Information and General Works::006: Special computer methods::006.3: Artificial Intelligence::006.31: Machine LearningConference Paper10.1109/IMBioC69142.2026.11541167