Transferring Features to Ultra-Wideband Radar: A Study on Heart Rate Measurement
In the quest for accurate heart rate measurements using innovative technologies, our latest study explores the effectiveness of ultra-wideband (UWB) radar. By leveraging learned features from existing datasets, we aimed to enhance the precision of heart rate detection using UWB radar sensors.
Data Collection Process
We conducted a study involving UWB radar data alongside electrocardiogram (ECG) and photoplethysmogram (PPG) data, which served as our reliable ground truth for heart rate measurement. The UWB radar sensor was strategically placed in typical positions where users hold their phones—either on a table in front of them or on their lap—to gather data in realistic scenarios.
Comparison with FMCW Datasets
The dataset collected from UWB radar consisted of 37.3 hours of data, significantly smaller than the 980 hours of frequency-modulated continuous wave (FMCW) data we previously analyzed. It is essential to note that due to the lower bandwidth of UWB radar, its range resolution was also inferior compared to the FMCW dataset.
Model Optimization and Transfer Learning
To optimize our measurement model for the UWB dataset, we retrained it after performing preprocessing steps. This involved adjusting the mm-wave FMCW radar data to align more closely with the characteristics of the UWB dataset, leading to a reduction in range resolution. Fine-tuning on the IR-UWB dataset allowed us to achieve a mean absolute error (MAE) of 4.1 beats per minute (bpm) and a mean absolute percentage error (MAPE) of 6.3%, representing a 25% decrease over the baseline error rate.
Results and Compliance with Standards
The original baseline performance for heart rate measurement using UWB radar was 5.4 bpm MAE and 8.4% MAPE. Through the application of transfer learning, we enabled the UWB radar to meet the Consumer Technology Association’s standards—aiming for an accuracy of up to 5 bpm MAE and 10% MAPE for consumer devices.
Conclusion
Our study highlights the potential of utilizing transfer learning to enhance heart rate measurement accuracy in UWB radar systems. By effectively modifying and retraining existing models, we can achieve significant improvements, paving the way for more reliable consumer health devices.
Related Keywords: UWB radar technology, heart rate measurement, transfer learning, ECG and PPG data, FMCW radar comparison, consumer health devices, radar data processing.
