Fig 1  Accuracy histogram for 9 subjects. 24Ghz radar data with sampling rate of 1000Hz.
The results of the first experiment with a high sampling rate in Table 1 show that the PAMDF is more accurate than the traditional FFT and AMDF methods. It yielded almost the same total accuracy as the ML method, with a difference of less than 0.5%. To our knowledge, there is currently no method that can provide more accurate results.
Specifically, Figure 2 shows a histogram of the accuracy of the four methods in the first experiment for data from nine subjects with different quality. The PAMDF always performs better than the AMDF, with either large or minor advantages. On the other hand, it has almost identical accuracy to the ML for all subjects, which demonstrates that the PAMDF is a close approximation of ML estimation.
In the second experiment with a lower sampling rate, the PAMDF method maintained similar accuracy with the help of rounding interpolation, and outperformed other methods. In terms of complexity, the interpolated PAMDF only slightly increased compared to AMDF. For each window with 200 complex samples in the second experiment, the PAMDF is calculated by adding only ~ 1200 real additions and ~ 110 real multiplications to the AMDF, which needs ~ 40,000 real additions but no multiplications. Comparatively, the ML estimation may also gain performance improvements with additional signal interpolation. However, applying the simplest signal rounding in would require ~ 50,000 real additions and ~ 4,000 real multiplications, and may only achieve performance similar to the PAMDF.
Conclusion: A novel PAMDF period estimation method is proposed for remote heart rate monitoring. The algorithm is implemented by advancing the classical AMDF and has low complexity. Its estimation accuracy is almost the same as the maximum likelihood estimation, which is the best of the known methods. Future studies may investigate the potential application of the PAMDF in other scenarios.
References
  1. Li C, Peng Z, Huang T, Fan T, Wang F, Horng T, et al. A review on recent progress of portable short-range noncontact microwave radar systems. IEEE Trans Microw Theory Tech. 2017 May; 65(5): 1692-706. doi: 10.1109/TMTT.2017.2650911.
  2. Aardal Ø, Paichard Y, Brovoll S, Berger T, Lande TS, Hamran SE. Physical working principles of medical radar. IEEE Trans Biomed Eng. 2013 Apr; 60(4): 1142-9. doi: 10.1109/TBME.2012.2228263.
  3. Wang G, Gu C, Inoue T, Li C. A hybrid FMCW-interferometry radar for indoor precise positioning and versatile life activity monitoring. IEEE Trans Microw Theory Tech. 2014 Nov; 62(11): 2812-22. doi: 10.1109/TMTT.2014.2358572.
  4. Carter GC. Coherence and time delay estimation. Proceedings of the IEEE. 1987 Feb; 75(2): 236-55. doi: 10.1109/PROC.1987.13723.
  5. Zhang W, Fu Y, Yin J. Micro-doppler period estimation based on concentration statistics of ambiguity function. IEEE Trans Aerosp Electron Syst. 2020 Jun; 56(3): 1722-41. doi: 10.1109/TAES.2019.2921192.
  6. Hui L, Dai B, Wei L. A pitch detection algorithm based on AMDF and ACF. Proc IEEE Int Conf Acoust Speech Signal Process. 2006 May. doi: 10.1109/ICASSP.2006.1660036.
  7. Edanami K, Sun G. Medical radar signal dataset for non-contact respiration and heart rate measurement. Data Brief. 2022 Feb; 40. doi: 10.1016/j.dib.2021.107724.