Gaussian Noise Augmentation is a data augmentation technique in which zero-mean Gaussian-distributed random noise is added to raw or processed CSI samples during training, simulating the stochastic interference and measurement uncertainty inherent in real-world wireless environments. In WiFi sensing research, it serves as a regularization strategy that improves model robustness and generalization by preventing overfitting to clean, idealized signals, making learned representations more resilient to domain shifts caused by environmental variability, hardware differences, or channel fluctuations. Key variants include scaling the noise by a fixed standard deviation relative to signal amplitude, applying noise selectively to specific feature dimensions or temporal segments, and combining it with other augmentations such as time masking or phase perturbation within contrastive or predictive coding frameworks to generate diverse positive pairs for self-supervised learning.
Source Papers
- Context-Aware Predictive Coding: A Representation Learning Framework for WiFi Sensing ↗ — Context-Aware Predictive Coding: A Representation Learning F
- PULSE: Physics-Aware Temporal Embedding Learning for Domain Adaptive Wireless Sensing ↗ — PULSE: Physics-Aware Temporal Embedding Learning for Domain