Abstract

With the growing reliance on cloud-based services, latency in wireless transmission of cloud data remains a critical challenge in environments where cloud connectivity is intermittent or restricted, such as cloud avoidance networks (CANs). These networks operate under the paradigm of minimizing dependence on centralized cloud infrastructure, leveraging edge or fog computing for timely data processing and delivery. Traditional transmission frameworks often fail to effectively manage latency in dynamic and decentralized networks. In CANs, unpredictable node mobility, variable signal strength, and heterogeneous device capabilities exacerbate delays. There is a lack of robust machine learning-based solutions that adaptively optimize routing and data transmission strategies in real time to reduce latency. This study introduces a Dual-Head Ensemble Transformer (DHET) model tailored for latency-aware wireless transmission in CANs. The first head of the transformer predicts short-term transmission latency across multiple hops based on real-time network conditions, while the second head assesses the reliability of the path by evaluating historical trends and signal consistency. Ensemble learning is used to fuse predictions from diverse transformer sub-models trained on varied wireless scenarios, ensuring generalized performance. A latency-prioritized routing algorithm then utilizes these predictions to dynamically select optimal paths for cloud data transmission. Simulation results demonstrate that the DHET-based approach achieves an average 18–25% reduction in end-to-end latency compared to baseline protocols such as AODV and DSR. The dual-head design allows for a balance between latency minimization and transmission stability, making it well-suited for CAN deployments in smart cities, autonomous fleets, and remote monitoring systems.

Authors

J. Jasmine¹, Hariharankumar²
Karpagam College of Engineering, India¹, Tata Consultancy Services Private Limited, Kakkanad, India²

Keywords

Latency Management, Cloud Avoidance Network, Wireless Transmission, Dual-Head Ensemble Transformer, Edge Computing