SpO2 estimation from wearable PPG measurements

14 Jul 2026

Methods

First just get the data:

Code

import numpy as np
import matplotlib.pyplot as plt
import pandas as pd

DATA_URL = "https://raw.githubusercontent.com/sonjababac/Signal_Processing_For_Remote_Patient_Monitoring/refs/heads/main/Datasets/PPG_for_SpO2/PPG_for_SpO2_data.csv"

data = pd.read_csv(DATA_URL, header=0, sep=",")
data.head()

	t [s]	Red [bit]	IR [bit]
0	0.000	50757	55256
1	0.008	50756	55254
2	0.016	50755	55252
3	0.024	50754	55251
4	0.032	50753	55251

Extract relevant signals

Code

time = data["t [s]"].to_numpy()
sampling_frequency = 1.0 / np.mean(np.diff(time))

ppg_red = data["Red [bit]"].to_numpy().astype(float)
ppg_ir = data["IR [bit]"].to_numpy().astype(float)

_, ax = plt.subplots(figsize=(8, 3))
ax.plot(time, ppg_red, label="red")
ax.plot(time, ppg_ir, label="IR")
ax.set_xlabel("time (s)")
plt.tight_layout()
plt.show()

Preprocess the signals

Preprocessing focuses on reducing high-frequency (powerline) noise and adjust for baseline wander (e.g. trend or offset removal)

Code

import scipy.signal
nyq = sampling_frequency / 2.0
lowcut = 0.8
highcut = 2.0
low = lowcut / nyq
high = highcut / nyq
order = 4
sos = scipy.signal.butter(order, [low, high], btype="band", output="sos")
filtered_red = scipy.signal.sosfiltfilt(sos, ppg_red)
filtered_ir = scipy.signal.sosfiltfilt(sos, ppg_ir)

_, ax = plt.subplots(figsize=(8, 3))
ax.plot(time, filtered_red, color="tab:blue", label="red (filtered)")
ax.plot(time, filtered_ir, color="tab:orange", label="IR (filtered)")
ax.set_xlabel("time (s)")
plt.tight_layout()
plt.show()

Idenfity peaks

Now idenfity peaks to derive the heart rate

Code

thr_min_peak_distance = int(0.5 * sampling_frequency)

peaks_ir, pkprops_ir = scipy.signal.find_peaks(filtered_ir, distance=thr_min_peak_distance, height=np.std(filtered_ir))
peaks_red, pkprops_red = scipy.signal.find_peaks(filtered_red, distance=thr_min_peak_distance, height=np.std(filtered_red))

_, ax = plt.subplots(figsize=(8, 3))
ax.plot(time, filtered_ir, color="tab:orange", label="IR (filtered)")
ax.plot(time[peaks_ir], filtered_ir[peaks_ir], "o", mec="r", mfc="r")
ax.set_xlabel("time (s)")
plt.tight_layout()
plt.show()

Calculcate peak-to-peak intervals

Code

# Calculate inter-beat intervals in seconds
peak2peak_ir = np.diff(time[peaks_ir])
peak2peak_red = np.diff(time[peaks_red])

# Calculate instantaneous heart rate (beats per minute, bpm)
heart_rate_ir = 60 / peak2peak_ir
heart_rate_red = 60 / peak2peak_red

# Plot the heart rate 
_, ax = plt.subplots(figsize=(8, 3))
ax.plot(time[peaks_red][1:], heart_rate_red, "o-", label="red")
ax.plot(time[peaks_ir][1:], heart_rate_ir, "o-", label="IR")
ax.set_xlabel("time (s)")
ax.set_ylabel("heart rate (bpm)")
plt.tight_layout()
plt.show()