PPGs at the CoSiNe Lab

Use in Biometrics

PPGs can be used for user authentication and provide an additional layer of liveness detection. In our work, we use an existing fingerprint sensor (Synolo) to capture images that are converted into PPG signals. While this keeps overhead low and eliminates the need for external hardware, the PPG signals are monochrome (blue) and low frame rate (14 fps). 

For accurate biometric performance, we use a hybrid deep learning model for authentication that combines CNNs, LSTMs, and Transformers to capture spatial and temporal variations, and focuses on the specific features of the PPG relevant for authentication.    

Synthetic PPG Signal Generation

We are using deep generative models to augment existing PPG datasets with high-quality synthetic data to support machine learning pipelines and downstream applications such as biosignal estimation and anomaly detection. This work explores the latent structure of PPG signals by analyzing the internal representation and evaluating the quality of generated signals through metrics like MAE and RMSE, and techniques such as KL annealing, dropout regularization, and latent space visualization (e.g., t-SNE) to improve the model's performance and interpretability. The results demonstrate the model’s ability to learn physiologically meaningful patterns and synthesize diverse, high-fidelity PPG signals.

Pulze-Lock: Raspberry Pi Hardware Platform

PulzeLock is an early prototype of a device capable of estimating heart rate, breathing rate, and oxygen saturation using a user's PPG signal. The system also uses the PPG signal to authenticate enrolled users into the system.

PPG GUI for STEM Education

We have developed a Graphical User Interface (GUI)-based tool that uses the PPG to demonstrate fundamental signal processing concepts such as periodicity, noise, jitter, and peak selection. The GUI allows users to upload video recordings of their fingers and estimate their heart rate and oxygen saturation, while manipulating features such as filter order. 

After deploying our app in a Signals and Systems class, we noticed that our app:

• Helped 68% of students understand the details about their class project

• Enabled 77% of students to understand PPG concepts (PPG is the same technology used in your doctor's office to find your heart rate & oxygen saturation)

• Showed 86% of students the difference between heart rate and SpO2