Heart Rate Detection by Discrete Fourier Transform (DFT) of a Photoplethysmographic (PPG) signal

Problems-

• Heart rate detection from motion corrupted ppg signals
• Motion Artefacts (MA) have very large frequency components in the ppg signal

​

Solutions-

• The heart rate (HR) signal is usually between 30 bpm to 200 bpm
• The MA signal is seldom found within this range
• Therefore, the two signal components (HR and MA) can be separated by filtering with an appropriate band pass filter

PROBLEMS:

​

• Heart rate detection from motion corrupted PPG signals
• The Motion Artifacts (MA) signal has significant frequency components in the PPG signal
• The MA signal peak can be greater than the Heart Rate (HR) signal

​

SOLUTIONS:

​

• The HR signal is always between 30 bpm to 200 bpm
• MA signals are seldom found in this region
• The two signals can therefore be separated by using a band pass filter with the current pass-band and stop-bands

PROJECT IN SHORT:

​

• Observe the corrupted PPG signal in frequency domain
• Create a band-pass filter that passes signals between 30 bpm to 200 bpm
• Observe the PPG signal before and after filtering
• Calculate the heart rate from the highest peak on the frequency spectrum after filtering
• Verify our results by correlating the filtered signal with the original signal and again calculating the heart rate

Dataset Info

Benchmark data for RR estimation from the photoplethysmogram used in “Multiparameter respiratory rate estimation from the photoplethysmogram”

by W. Karlen, S. Ramen.

Source:

www.capnobase.org/index.php?id=857

​

Adds demographics information such as weight, age and ventilation mode to meta structure

Free to use

Adds signal.ecg,labels.ecg, and reference.hr.ecg; allows to compare hr against ECG

Extracting Input From Dataset

Input signal

Zoomed in

Frequency Spectrum of Input

Zoomed in

Constructing the Bandpass Filter

Creating a bandpass function

Using Kaiser Window

Passband:

30-200 BPM

BPF Output

Passband matches with code

Filtering the Input Signal

Convolution in Time Domain

Multiplication in Frequency Domain

OR

Greatest Spike found at

105.5 BPM

Filtered Output Zoomed in

Reconstructed Signal

Zoomed in

Verifying the Result with Correlation

Correlation between

Input signal & recovered signal

Correlation output zoomed in

The two peaks shown in the figure has a difference of 170 (value of K1)

Converting it into BPM;

We get peak difference at

(300/170)*60 = 105.88 BPM

After Filtering, Peak was found at

105.5 BPM

Therefore, The result is verified with correlation!

THANK YOU !!

Mumtahina Islam Sukanya

​

ID: 1606039

Ishtiaque Ahmed Showmik

​

ID: 1606035

MD Obaidur Rahman

​

ID: 1606057

Mohammad Ali Kazmi

​

ID: 1606061

MD Adnan Faisal Hossain

​

ID: 1606063