This project develops and presents a cutting-edge prototype designed to rigorously assess the effectiveness of federated learning (FL) over bandwidth-constrained wireless channels. The mission is to explore how well FL performs in real-world wireless environments where bandwidth limitations and channel variability present significant challenges.

The core of this prototype revolves around a novel configuration that leverages ZigBee and National Instruments Universal Software Radio Peripheral (NI-USRP) devices. This setup enables client devices to broadcast their locally trained machine learning model updates to a centralized server, where these updates are aggregated to form a generalized global model. By enabling this distributed learning across multiple wireless nodes, the prototype demonstrates the feasibility of decentralized training in low-bandwidth scenarios while maintaining communication efficiency and model accuracy.

To validate the effectiveness of this prototype, performance is systematically evaluated using key metrics. These include global model accuracy, which measures the precision of the aggregated model, and time complexity, which assesses the speed of the FL process under different operational conditions. Specifically, the project examines the impact of various factors such as transmission power, data heterogeneity, and the configuration of local learning processes on overall system performance.

The mission is not only to propose this innovative FL setup but also to critically analyze its performance in constrained environments, thereby advancing the understanding of how federated learning can be effectively applied in bandwidth-limited wireless networks. This work aims to pave the way for future improvements in wireless FL, particularly in real-time, resource-constrained applications like the Internet of Things (IoT) and smart devices.

The key objectives of this project are as follows:

• Privacy-Preserving Machine Learning (ML):
  This project aims to establish a federated learning (FL) protocol that is inherently designed to safeguard user privacy. Instead of transmitting raw data, the protocol focuses on sending model parameter updates, ensuring that sensitive information remains secure. These updates are transmitted over wireless channels, maintaining data confidentiality and adhering to privacy standards throughout the learning process.

• Real-time Machine Learning (ML):
  The project emphasizes the seamless integration of machine learning with cutting-edge hardware, specifically National Instruments' Universal Software Radio Peripheral (NI-USRP) and ZigBee technology. This integration enables real-time communication of updates during the FL process, facilitating instantaneous feedback and adjustments in the learning model. By leveraging these hardware platforms, the project demonstrates how real-time learning can be efficiently implemented in practical, wireless environments.

• Decentralized Model Training:
  One of the key advancements of this project is showcasing the efficiency gains achieved by decentralizing the training process across multiple hardware devices. By distributing the learning workload among the devices, the project aims to reduce bottlenecks and improve the overall performance of the system. This approach highlights the potential of federated learning to scale across diverse and distributed hardware, leading to faster and more efficient model training without centralizing data.

This project serves as a comprehensive, hands-on guide to establishing a federated learning (FL) environment utilizing Universal Software Radio Peripheral (USRP) devices and ZigBee technology. By combining FL principles with practical hardware platforms, the project facilitates real-world implementation of privacy-preserving machine learning across wireless networks.

The convergence of FL and hardware devices like USRPs and ZigBee opens up new possibilities for collaborative machine learning, where multiple devices contribute to the training of a model without sharing raw data. This privacy-conscious approach to ML is particularly significant in wireless communication, as it enables secure learning across distributed networks, such as IoT systems, without compromising data security.

By setting up this environment, the project illustrates how theoretical advancements in FL can be translated into practical applications. It bridges the gap between conceptual research in federated learning and its real-time deployment on wireless hardware platforms, making it possible to explore the full potential of FL in scenarios involving resource-constrained and distributed devices. This integration paves the way for future innovations in wireless federated learning, ensuring that these systems are not only conceptually viable but practically implementable.