In this edition of Voices from the Lab, we meet Dr Abdelaziz Salama from the University of Leeds.

Dr Abdelaziz Salama’s work spans AI, cloud computing, and next-generation wireless networks. As a Research Software Engineer, he is building intelligent systems that can predict and optimise network performance in real time, paving the way for more autonomous and energy-efficient 6G networks.

Can you describe your research in simple terms? What problem are you trying to solve, and why does it matter?

Our research focuses on making mobile networks, such as 5G and future systems, much more energy efficient. Today’s networks consume a significant amount of power, with radio units (RUs) being one of the main contributors.

We are developing an intelligent agentic AI system that can automatically predict network demand and manage network resources, including deciding when to switch radio units on or off. This helps reduce energy usage while still maintaining reliable service for users.

This is important because energy consumption in telecom networks is rapidly increasing, leading to higher operational costs and greater environmental impact.

What is the key challenge your work addresses, and why has it been difficult to solve until now?

The main challenge is deciding when it is safe to turn network components off without harming user experience.

This is difficult because network traffic is highly dynamic, it changes based on time, location, user movement, and even unexpected events like concerts or weather changes. Traditional methods rely on static rules or thresholds, which cannot adapt to these complex and unpredictable patterns.

Additionally, decisions in one part of the network affect others due to interference and load redistribution, making the problem highly interconnected and hard to optimise.

What is novel about your approach compared to existing solutions?

Our approach is novel because it combines multiple AI techniques within a coordinated “agentic” framework.

We develop an end-to-end 5G O-RAN optimisation framework based on three specialised AI agents that work collaboratively and are validated using a digital twin emulator:

• One creates realistic network scenarios
• One predicts future traffic demand
• One makes energy-saving decisions

We integrate advanced models as tools for the deployed agents, such as Graph Neural Networks (to capture spatial relationships between cells) and Transformers (to model traffic trends over time), all guided by an AI reasoning layer.

The key innovation lies not in the individual models themselves, but in how they are orchestrated into a unified decision-making system that is predictive, context-aware, and explainable.

How does your research contribute to the evolution towards 6G or “connected intelligence”?

6G is expected to move towards fully autonomous, intelligent networks, often described as “connected intelligence.”

Our work contributes to this vision by demonstrating how networks can:

• Sense their environment (through data and context).
• Forecasting future conditions of the network.
• Act autonomously using AI-driven decisions.

By embedding intelligence directly into the network control layer, our framework shows how future networks can self-optimise in real time, which is a key requirement for 6G.

How could this work translate into real-world applications where we might see it used?

Telecom operators could deploy this technology in real network environments to manage energy consumption automatically.

For example:

• Urban networks could reduce energy usage during low-traffic hours (e.g., late night)
• Networks could adapt to sudden events like sports matches or festivals
• Rural or low-demand areas could operate more efficiently without over-provisioning

It could also be integrated into cloud-based network management platforms used in Open RAN deployments.

Who stands to benefit most from this innovation (e.g. industry, healthcare, infrastructure, society)?

Several groups benefit from this work:

• Telecom operators: lower energy costs and improved network efficiency
• Infrastructure providers: more sustainable and scalable network deployments
• Society: reduced carbon footprint from digital infrastructure
• Public services (e.g. healthcare, transport): more reliable connectivity with efficient resource use

Overall, it supports greener and more sustainable digital infrastructure.

How does your work align with CHEDDAR’s vision or its pillars (Emergent, Sustainable, Human-Centric Systems)?

• Emergent Systems: The framework enables autonomous, intelligent behaviour through interacting AI agents that adapt to changing conditions in ORAN networks.
• Sustainable Systems: It directly reduces energy consumption in large-scale network infrastructure, addressing environmental sustainability.
• Human-Centric Systems: The use of explainable AI allows operators to understand and trust the system’s decisions, ensuring human oversight and control.

What has been the most challenging or unexpected part of your research journey so far?

One of the biggest challenges has been integrating multiple AI agents and ensuring they work together reliably in a real-time system.

Another unexpected difficulty is handling unpredictable real-world events, such as sudden traffic spikes in the target area. These situations require the system to continuously adapt, stay up to date, and dynamically switch between network configurations or policies without compromising stability or performance.

What are the next steps? What still needs to be explored before this can be deployed?

Key next steps include:

• Testing the framework in larger-scale and more diverse network scenarios
• Validating performance in real-world deployments beyond simulation (digital twins)
• Improving robustness to rare or extreme events
• Enhancing coordination and utilisation of the integrated agentic AI system to support multiple network tasks simultaneously based on real-time needs

In one sentence: what impact could your research have on the future of connectivity?

This research enables future networks to become self-optimising and energy-aware, significantly reducing power consumption while maintaining reliable, high-quality connectivity.