Transferring realized options to ultra-wideband radar
We then ran a examine that collected UWB radar knowledge, together with electrocardiogram (ECG) and photoplethysmogram (PPG) knowledge as our floor reality for coronary heart fee, utilizing a setup that positioned the UWB radar sensor in positions the place customers sometimes maintain their telephone, i.e., on a desk in entrance of them or on their lap. In comparison with the FMCW dataset, which was 980 hours of information, the UWB radar dataset was a lot smaller, with 37.3 hours. Because the UWB radar configuration was shut to what’s possible on a cell phone, with a a lot decrease bandwidth, its vary decision was far decrease than the FMCW dataset.
To make sure that our mannequin was optimized to switch to the UWB dataset, we retrained it after performing further pre-processing steps to change the mm-wave FMCW radar knowledge to raised resemble the goal IR-UWB knowledge, successfully reducing its vary decision. We then fine-tuned this mannequin on the IR-UWB dataset, attaining an MAE of 4.1 bpm and imply absolute share error (MAPE) of 6.3%, a 25% discount over the baseline error fee. Our baseline for efficiency on UWB radar was 5.4 bpm MAE and eight.4% MAPE, achieved by choosing the right mannequin skilled from scratch on our UWB dataset. With switch studying, we enabled the UWB radar to fulfill the Shopper Expertise Affiliation requirements for coronary heart fee measurement for client gadgets: an accuracy of as much as 5 bpm MAE and 10% MAPE.
