BROADBAND FORUM AI OPTIMISATION OF NETWORKS
so AI service flows need deterministic end-to-end capabilities to let service providers dynamically tune resources based on application needs, as opposed to treating all traffic equally. For this to be achieved, flexible networks that can effortlessly allocate bandwidth, latency and loss characteristics in real-time must be a reality. Thankfully, a QoD approach delivers this, moving beyond static provisioning and centring networks around user behaviour and specific device requirements. This is especially crucial for AI-heavy environments where upstream and downstream demands will fluctuate rapidly, and where jitter or interference can break the user experience. This is a major departure from legacy broadband models, but application- oriented resource control unlocks commercial opportunities too: by exposing network capabilities through Application Programming Interfaces (APIs), service providers can offer guaranteed performance tiers to third-party developers and application providers, aligning with the broader network transition toward Network- as-a-Service (Naas). Once network functions become consumable on- demand services, establishing QoD will become both a technical requirement and a revenue-generating mechanism. FACILITATING SELF-OPTIMISING NETWORKS Enabling AI within networks also enables self-optimising capabilities, with systems capable of predictive maintenance, energy optimisation and dynamic resource allocation. This marks a major shift from today’s rule-based automation. Instead of relying on human-defined thresholds or manual troubleshooting, future networks will be self-learning, drawing on real-time conditions and historical behaviours to continuously adapt and improve performance. Self-optimisation spans the entire access and in-premises environment, covering everything from Passive Optical Network (PON) resource allocation to Wi-Fi performance management and fibre-based sensing. Using AI in this context can reduce power consumption, improve reliability and even support digital twin-based fault prediction – all essential capabilities as networks grow increasingly complex, traffic becomes more AI-driven and customer expectations for reliability rises. Ultimately, AI-driven self-optimisation is the foundation for autonomous networks, specifically Level 4 and beyond. It facilitates intent-driven operations – where service providers specify outcomes, rather than configurations – where the network determines the most optimal path for operations. This not only reduces operational costs but accelerates fault
restoration and improves the overall user experience too. At the same time, it prepares the broadband ecosystem for the demands of next-generation AI services. KEY TECHNOLOGY TRENDS FOR AI As envision by Broadband Forum’s MR- 529, AI in broadband networks will evolve beyond basic connectivity to support agentic AI services, quality-on-demand, and self-optimising capabilities. Such evolution requires enhanced network- associated computing, storage, and sensing capacities, coupled with greater network openness, resource pooling, and comprehensive visibility to enable flexible AI application deployment. Native AI builds on established network architecture, compute and edge-cloud capabilities. What is emerging now are AI functions that can use these compute resources to support both autonomous network operations and intelligent service delivery. As these AI capabilities mature, networks will require greater coordination across distributed sites, improved network resource pooling and greater visibility to enable dynamic tuning of end-to-end resources. This evolution will enable AI to run more efficiently across broadband infrastructure, while improving utilisation, reliability, and service scalability. Programmability and visibility form the operational backbone of AI-enhanced networks. with greater network programmability, service providers can make real-time adjustments across policies, tasks, functions, and algorithms. This enables closed-loop optimisation such as improved live streaming performance through flexible capability subscription. Full visibility achieved through residential gateway intelligence, network monitoring, and advanced fibre and Wi-Fi sensing, which provides the data necessary for accurate AI-driven simulation, prediction, and decision- making. Service openness extends this ecosystem by offering standardised APIs for network capability exposure, quality on demand guarantees, and sensing events, enabling third-party AI deployment and new monetisation opportunities. Well-designed data models critically underpin these capabilities by structuring diverse telemetry, OSS, and BSS data sources, enabling AI systems to derive actionable insights across performance, service quality, and customer experience. Finally, AI in security use cases presents both challenges and opportunities. As AI is added to networks, it naturally expands attack surfaces, but it also introduces new benefits such as predictive threat detection, automated incident response, and privacy-preserving techniques. In the pursuit of fully autonomous networks, service providers are
progressing through three distinct evolutionary stages. This includes establishing a software-defined foundation, automating network operations through standardised interfaces, and finally integrating AI to enable true autonomy. AN EMERGING FRAMEWORK TO MOVE FROM VISION TO REALITY With this vision and key technology trends outlined, service providers have the means to unlock new service models and new value, facilitating the evolution to services-led Broadband. However, as networks evolve, so too do the frameworks that support them. For organizations like the Broadband Forum, attention now shifts from defining what the future will look like, to how the industry gets there. This is reflected in the ongoing development of a framework that will guide service providers in implementing AI effectively and consistently within their networks. By translating MR-529’s three values into practical guidance, service providers can progress from conceptual alignment to real-world deployments. The framework is expected to define key use cases where AI delivers measurable value, outline the requirements needed to support these, and describe the characteristics needed to support AI services and AI-enabled network operations. Part of this remit will include how AI models interact with access nodes, Customer Premises Equipment (CPE), management systems and orchestration layers to ensure interoperability and consistency across vendors and technology domains. Crucially, this phase will also provide recommendations on how the roadmap can be shaped for future industry specifications, influencing how network architecture and functions evolve, and ensuring management platforms can support emerging AI-driven behaviours. Eventually, now that MR-529 has defined the destination, the framework will define the means to make AI the deployable, interoperable reality that unlocks higher-quality services, cost savings and a new generation of intelligent, tailored broadband experiences.
Manuel Paul Service Requirements Work Area Director, Broadband Forum
Tony Zeng Service Requirements Work Area Director, Broadband Forum
www.opticalconnectionsnews.com
23
ISSUE 44 | Q3 2026
Made with FlippingBook Annual report maker