This paper proposes DGFedRS, a Federated Recommender System based on Diffusion Augmentation and Guided Denoising. The system addresses the challenges of data sparsity and privacy in sequential recommender systems. Traditional methods often fail to capture the diversity in latent data distributions and may introduce noise, leading to poor recommendation quality. DGFedRS uses a diffusion model to generate interaction sequences, expanding sparse user-item interactions in a discrete space. A guided denoising strategy is employed to preserve user-specific preferences during the reverse diffusion process. Additionally, a noise control strategy is implemented to reduce the impact of noise on personalized information, and a stepwise scheduling strategy is used to input generated data into the sequential recommender model based on their challenge levels. The system also incorporates a distributed federated architecture to address user privacy concerns. Experiments on three real-world datasets demonstrate the effectiveness of DGFedRS, showing significant improvements over existing baselines. The contributions of this work include designing a federated recommender system based on diffusion augmentation and guided denoising, developing a guided denoising strategy to generate high-quality samples, and validating the effectiveness of DGFedRS on real datasets. The system is evaluated on three real-world datasets, showing substantial performance improvements compared to 13 current baselines. Ablation studies and sensitivity analysis confirm the effectiveness of the designed components.This paper proposes DGFedRS, a Federated Recommender System based on Diffusion Augmentation and Guided Denoising. The system addresses the challenges of data sparsity and privacy in sequential recommender systems. Traditional methods often fail to capture the diversity in latent data distributions and may introduce noise, leading to poor recommendation quality. DGFedRS uses a diffusion model to generate interaction sequences, expanding sparse user-item interactions in a discrete space. A guided denoising strategy is employed to preserve user-specific preferences during the reverse diffusion process. Additionally, a noise control strategy is implemented to reduce the impact of noise on personalized information, and a stepwise scheduling strategy is used to input generated data into the sequential recommender model based on their challenge levels. The system also incorporates a distributed federated architecture to address user privacy concerns. Experiments on three real-world datasets demonstrate the effectiveness of DGFedRS, showing significant improvements over existing baselines. The contributions of this work include designing a federated recommender system based on diffusion augmentation and guided denoising, developing a guided denoising strategy to generate high-quality samples, and validating the effectiveness of DGFedRS on real datasets. The system is evaluated on three real-world datasets, showing substantial performance improvements compared to 13 current baselines. Ablation studies and sensitivity analysis confirm the effectiveness of the designed components.