Latent Diffusion-Based Image Generation from WiFi Channel State Information WiFi Kanal Durum Bilgisinden Latent Difüzyon Tabanli Görüntü Üretimi
34th Signal Processing and Communications Applications Conference, SIU 2026, İstanbul, Türkiye, 7 - 10 Temmuz 2026, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Doi Numarası: 10.1109/siu71813.2026.11636547
- Basıldığı Şehir: İstanbul
- Basıldığı Ülke: Türkiye
- Anahtar Kelimeler: BiLSTM, image generation, latent diffusion model, multi-task learning, transfer learning, WiFi channel state information
- Maltepe Üniversitesi Adresli: Evet
Özet
Reconstructing images from WiFi channel state information (CSI) is a privacy-preserving sensing paradigm that eliminates cameras. LatentCSI maps CSI into the latent space of a pretrained diffusion model via a CNN, but this single-frame approach ignores temporal dependencies. This work improves LatentCSI in three directions: a BiLSTM encoder capturing temporal dependencies across consecutive CSI frames, a multi-task activity head structuring the latent representation, and transfer learning accelerating convergence. On the MM-Fi dataset, the proposed multi-task BiLSTM achieves a 50.1% improvement in FID over the original LatentCSI. Controlled experiments show that the BiLSTM learns genuine temporal patterns rather than acting as a simple moving average.