A deep neural network (DNN) is a multi-layered computational model composed of stacked processing units that learn hierarchical feature representations directly from raw or preprocessed input data such as CSI amplitude or phase measurements. In WiFi sensing, DNNs are valued for their ability to automatically extract complex spatial and temporal patterns relevant to tasks like crowd counting, localization, and activity recognition without relying on extensive hand-crafted features. Key variants employed in this domain include convolutional neural networks (CNNs) for capturing spatial structure in CSI matrices, recurrent architectures such as LSTMs for modeling temporal dynamics, and hybrid or multi-task networks that jointly optimize for related objectives like simultaneous regression and classification, as seen in systems such as Wi-CaL.
Source Papers
- CrossSense: Towards Cross-Site and Large-Scale WiFi Sensing ↗ — CrossSense: Towards Cross-Site and Large-Scale WiFi Sensing
- Time matters: Empirical insights into the limits and challenges of temporal generalization in CSI-based Wi-Fi sensing ↗ — Time matters: Empirical insights into the limits and challen
- Towards Energy Efficient Wireless Sensing by Leveraging Ambient Wi-Fi Traffic ↗ — Towards Energy Efficient Wireless Sensing by Leveraging Ambi
- Wi-CaL: WiFi Sensing and Machine Learning Based Device-Free Crowd Counting and Localization ↗ — Wi-CaL: WiFi Sensing and Machine Learning Based Device-Free