Bio-signal processing algorithms for the removal of artefacts in photopletysmography signals collected at the wrist and the lower extremities of the body

UoM administered thesis: Phd

  • Authors:
  • Arturo Vazquez Galvez


In recent years there has been a substantial interest in wearable devices that measure heart rate (HR) via photoplethysmography (PPG) sensors placed at the wrist. To date, algorithms for motion artefact removal from the PPG have focused on the use of a co-located accelerometer to record the motion. In this work, a switch feature is introduced between co-located accelerometer, gyroscope and magnetometer sensors to allow three, six and nine and a “mixed” approach to the degrees of freedom estimates of the motion present. An analysis on PPG sensors placed at the lower extremities, on the ankles and on the feet, is also presented. These locations reflect the locations on the body where the most amount of kinetic energy is available for collecting by an energy harvester. There are two different sets of data used for which the sensors brand is different but the exercises performed by the subjects are similar. For the upper body analysis, eight records of wrist PPG recorded during bike riding were collected, with simultaneous accelerometer, gyroscope and magnetometer recordings, are used. For the lower extremities analysis, data from a total of 25 subjects is used, where they completed four stages of exercise: one stage of walking, one stage of running, and two stages riding a bicycle, all of which involve different motion and impact performances. In both data collections cases a reference chest electrocardiogram (ECG) was also recorded to give a gold-standard comparison of heart rate. These are processed using an adaptive filtering based signal processing algorithm. For sensors placed at the wrist, this research quantifies for the first time the trade-off between battery energy consumption and heart rate estimation accuracy when using different motion sensors for estimating the motion artefact present. Combining the gyroscope and accelerometer sensor data gave the most accurate average estimates for heart rate. For the lower body analysis, during walking the mean error across all records was 11 bpm, and during running 19.5 bpm. These figures are in-line with practical performances obtained from current wrist based PPG units, despite the significantly larger motion artefacts present at the feet and the ankles. In about half of the records the heart rate estimation accuracy is better at the ankle/foot than at the wrist. If the potential energy that can be harvested at the foot is added, a very desirable power vs accuracy trade can be integrated with the implementation of PPG sensors placed around the ankles and on the feet.


Original languageEnglish
Awarding Institution
Award date1 Aug 2020