GDM enhanced CNN-based first arrival path delay estimation framework

We propose a method that utilizes Generative Diffusion Model (GDM), a sequence encoding algorithm, and an image decoding algorithm to jointly assist Convolutional Neural Network (CNN) in estimating the First Arrival Path (FAP) delay. In the task of utilizing CNN for FAP delay estimation, there is a challenge of insufficient dataset. To mitigate this limitation, we augment the dataset for CNN training using GDM. Specifically, we employ the Cross-Correlation (CC) algorithm to extract delay features from wireless signals, thereby generating Cross-Correlation Function (CCF) sequences that serve as the input data. To capitalize on the capabilities of GDM in processing image data, we use the sequence encoding algorithm to encode CCF sequences into images, which are then used as dataset for training the GDM. After training, GDM generates new image data. The image decoding algorithm can decode the generated image data back into sequences, which are used as augmented data for CNN training. The results of the simulation experiments demonstrate that the sequence encoding algorithm and the image decoding algorithm can effectively assist GDM in augmenting the data for CNN training in the task FAP delay estimation.