Diffusion Model-based RSSI Fingerprint Generation for Indoor Localization in Dynamic Environments

The demand for indoor localization technology is increasing in various fields, including smart cities, security monitoring, and smart homes. However, frequent changes in indoor environments often render existing RSSI fingerprint data obsolete. Traditional methods require frequent data recollection, which is time-consuming and labor-intensive. While existing approaches such as multi-sensor fusion and transfer learning partially alleviate this issue, they still suffer from high costs or the need for partial data recollection. This paper introduces an innovative approach that treats RSSI data as images and utilizes a denoising diffusion probabilistic model (DDPM) to learn their distribution characteristics across different environments, offering a new perspective on applying diffusion models to indoor localization. By generating RSSI data distributions for new environments, our method eliminates the need for onsite data recollection. Simulation experiment results demonstrate that under a 10% error tolerance range (ETR), the proposed method achieves a localization accuracy of 91.01%, confirming its effectiveness. However, the model's performance degrades in more complex environments, highlighting the need for further improvements to enhance adaptability in such scenarios.