What AI Brings to OpenRAN

Abstract

The telecommunications industry is undergoing a revolution, and artificial intelligence (AI) is at its core. This article explores how AI and machine learning (ML) are transforming Open Radio Access Networks (O-RAN), unlocking unprecedented network capabilities that were once science fiction.

What is AI and ML?

AI enables machines to perform tasks requiring human-like reasoning: pattern recognition, autonomous decision-making, and prediction. ML, a subset of AI, uses algorithms that adapt based on data inputs rather than explicit programming.

The core value proposition is prediction at scale. AI and ML have the potential to dramatically reduce the cost and complexity of making accurate predictions across telecommunications networks. This capability transforms network operations from reactive maintenance to proactive optimization.

In finance, they are used to detect fraud and anomalous activity, keeping their customers’ bank accounts safe. In ecommerce, they power recommendation engines, helping people find exactly what they want. For social media, algorithms are the backbone of what appears on your feed. TikTok, with its recommendations that seem to know what you want before you do, relies on algorithms. Recently, LLMs like ChatGPT, which have algorithms underneath it, have taken the world by storm. An algorithm may have sent this very article your way.

There are also more experimental technologies like autonomous vehicles, which make decisions via sensors to make split-second decisions, and robots being trained to handle complex tasks and specialized missions. There has been significant development in that regard with Waymo, a driverless taxi service, gaining popularity in geofenced areas.

Why Telecom Needs AI

AI is now being directed toward the Radio Access Network (RAN), the critical infrastructure that connects mobile devices to the internet. This convergence presents a unique opportunity: AI’s predictive power can address many of the real-time challenges of modern telecommunications.

To make this possible, industry-leading organizations like 3GPP (3rd Generation Partnership Project) and the O-RAN Alliance are laying the foundation. 3GPP develops global standards for mobile networks, including 5G, and is increasingly focused on incorporating AI and automation into future releases.

What is O-RAN?

Architecture of an Open Radio Access Network.

O-RAN represents a significant transformation in the design and operation of mobile networks. It is an open architecture that eliminates traditional vendor lock-in, enabling operators to integrate equipment from multiple vendors while ensuring seamless interoperability. In this sense, O-RAN serves as an open platform for innovation and collaboration across the telecommunications industry.

To appreciate the impact of O-RAN, it’s helpful to contrast it with the traditional system architecture. Conventional RANs are defined by several limiting characteristics:

• Proprietary, vertically integrated designs that restrict interoperability and tie operators to a single vendor’s ecosystem.
• Closed and undocumented interfaces between hardware and software, making integration and upgrades difficult.
• Strong vendor lock-in due to tightly bundled hardware and software deployments.
• High-cost, restrictive licensing models that limit flexibility and increase operational expenses.
• Obfuscated or inaccessible APIs, which hinder third-party development and innovation.
• Minimal compatibility with external frameworks or open standards, stifling extensibility and collaboration.
• Limited customization and configurability, making it hard to adapt networks to evolving needs.

In contrast, O-RAN was designed to address these very challenges:

• Emphasizes open and standardized interfaces between all RAN components, enabling true interoperability and fostering a more competitive vendor landscape.
• Software-based architecture
  • Decouples network functions from proprietary hardware, making upgrades and scaling far more flexible and cost-effective.
• Decentralized and modular
  • Breaks the network into distinct, interoperable modules such as the Radio Unit (RU), Distributed Unit (DU), and Central Unit (CU). Any component can be sourced from different vendors.
• Highly customizable
  • Leverages software-defined networking (SDN) principles, empowering operators to fine-tune and optimize their networks through software rather than rigid hardware. This greatly reduces integration complexity, even as compatibility remains important.
• Ecosystem-driven
  • Functions as a true platform, allowing operators and third parties to build and deploy new services, applications, and products, much like an app store for the network itself.

Why O-RAN Is Needed

O-RAN extends the capabilities of traditional RANs by providing several key advantages:

Cloud-Scale Economics

• Standardized interfaces: Facilitate interoperability, allowing operators to select from a broader range of hardware, including proprietary, open, and commercial off-the-shelf (COTS) solutions.
• Modular software: Supports flexible integration and replacement of network functions as needed.
• Lower scaling costs: Enables more cost-effective network expansion and resource allocation.

Network Agility

• Intelligent sensing and anticipation: Supports proactive identification and mitigation of potential network issues.
• Adaptive resource management: Enables dynamic allocation of resources based on real-time network conditions.
• Self-healing capabilities: Provides mechanisms for automated detection and correction of faults.

Future-Proof Architecture

• Evolves with technology: Reduces the need for frequent hardware replacements by supporting ongoing upgrades.
• AI-ready foundation: Designed to accommodate the integration of artificial intelligence and machine learning technologies.

O-RAN was developed to overcome the technical rigidity and commercial lock-in of traditional radio access networks which has increasingly become a barrier to the demands on modern networks. While legacy infrastructure effectively supported past demands, it was built on vertically integrated, centralized architectures optimized for stable, predictable traffic patterns. Such conditions that no longer reflect today’s dynamic, data-intensive, and latency-sensitive environments. Demand for fast, ultra low latency internet and corresponding services are rising rapidly. The infrastructure must be upgraded to match new marketplace demands and a changing landscape. Autonomous vehicles, smart factories, emergency services, and the IoT (Internet of Things) will require sub-second responsiveness, deep interconnectedness and dynamic scaling, all which legacy infrastructure will struggle to deliver.

This shift from closed, monolithic systems to open, modular, and software-driven architectures is what enables O-RAN to meet the demands of today’s dynamic, data-intensive, and rapidly evolving mobile environments.

The shift is also is gaining traction. The O-RAN Alliance reported in early 2024 that 32 mobile network operators have committed to Open RAN systems, with almost 300 companies worldwide including major players like AT&T, Bell, Ericsson, Deutsche Telekom, NTT DOCOMO, Orange, Vodafone, Telefónica, and Verizon now participating in the ecosystem. As this adoption continues, it paves the way for greater extensibility and innovation, especially through the integration of AI and machine learning.

The Value AI and ML Bring to O-RAN

AI and ML enable networks to become self-optimizing, predictive, self-correcting, more energy-efficient, and increasingly autonomous. But perhaps their most transformative advantage is high extensibility: the ability to add new features and capabilities without overhauling existing infrastructure.

Software-defined networking already streamlines upgrades. AI and ML take this agility even further, allowing operators to introduce intelligent functions incrementally and adapt models as network conditions evolve.

Integrating AI and ML into networks adds dynamism to them. It enables them to learn usage patterns, predict system load, detect threats, optimize performance, and interact with the physical environment. The potential lies far from mere communications, extending out to things like smart city infrastructure and environmental monitoring to industrial automation and immersive user experiences.

O-RAN defines an architecture for the integration and usage of AI and ML models, to be installed through the RIC (RAN Intelligent Controller). These models can be deployed within or through the RIC. However, unlike the traditional networks, it can evolve with time. Because it is modular and software based, new models and strategies can be abandoned and installed without having to alter hardware. This will take two major forms:

• Near-Real-Time RIC (Near-RT RIC): Fast decision-making (<1 second), taking the form of traffic steering, radio resource management, session management, and maintaining Quality of Service.
• Non-Real-Time RIC (Non-RT RIC): Longer time scale (>1 second to hours or days), handling network optimization, policy management, and network orchestration.

Developers can install modules and apps in the form of xApps and rApps. They carry the intelligent features driven by AI and ML, optimizing spectrum distribution. rApps are specialized microservices that run on the non-RT RIC. xApps run on the Near-RT RIC, optimizing radio spectrum efficiency.

Real-World Applications

These AI-driven applications enable real-world network optimization. For example, an xApp deployed on the Near-RT RIC can monitor and coordinate traffic flow, preventing signal loss and degradation. During high-demand events like festivals or stadium concerts, where thousands of users simultaneously compete for spectrum, this technology ensures that users stay connected to the network, and their experience is preserved under the heavy load with seamless high-throughput downloads and uploads.

Standard RIC operations achieve this with inference latencies under 50ms, which suffices for most network management tasks. However, ultra-low latency inference below 10ms, or even sub-millisecond response times, unlocks advanced capabilities that extend far beyond traditional network optimization.

Two key applications justify pursuing aggressive latency targets: intelligent RAN control and Joint Communications and Sensing (JCAS).

Ultra-Low Latency RAN Control

A 5G frame operates on a 10ms cycle. When AI inference latencies drop below this threshold, it becomes possible to complete a full capture-infer-respond cycle faster than a single frame. To achieve this, inference needs to occur on or near the radio itself, where the signal processing is happening. This breakthrough enables AI to directly control physical radio operations in real-time.

With sub-frame latency, AI systems can dynamically control beam steering, mitigate signal collisions, maximize spectrum utilization, and adapt to changing channel conditions, all with minimal processing overhead, applied directly to radio signal data. This ultra-low latency inference can transform RANs into high-performance adaptive cells that respond to conditions faster than the underlying communication protocols.

Radio inference directly on RAN hardware unlocks control and sensing.

Joint Communications and Sensing Through Radio Inference

Radio inference unlocks an additional capability: sensing. As radio waveforms propagate through the environment, they interact with objects, terrain, and atmospheric conditions. These interactions become embedded in the measured signal characteristics. When captured and processed through AI inference, these signals provide detailed sensing information about the surrounding environment.

Since this sensing occurs on base stations already providing communication infrastructure, it naturally enables Joint Communications and Sensing (JCAS). Networks gain environmental awareness, with the potential for detecting objects, measuring distances, and monitoring physical condition. The new functions occur while simultaneously delivering traditional connectivity services.

Radio inference makes JCAS applications possible, including passive radar detection, real-time spectrum monitoring, radio fingerprinting for security, and intelligent transport solutions that can track vehicle movement and traffic patterns. This dual-purpose infrastructure maximizes the value of existing radio assets while opening new capabilities for smart city applications, industrial monitoring, and autonomous systems.

The true value lies in JCAS adaptability across use cases. Different industries benefit from vastly different sensing capabilities, from detecting vibrations on roads and bridges to monitoring air quality in urban areas to tracking wildlife in conservation zones. By deploying industry-specific AI models on the same radio infrastructure, operators can transform identical hardware into specialized sensing solutions tailored to each environment’s unique requirements. This further unlocks the value of custom, open source, small scale, and private networks.

Communications and sensing from the same infrastructure.

Qoherent’s Role

The ultra-low latency RAN control and JCAS capabilities described above require more than theoretical frameworks, they demand practical deployment solutions that work reliably in real-world environments. This is where specialized tools and expertise become critical.

Enabling the value that AI brings to RAN control and JCAS is a core mission for Qoherent. Qoherent develops the software infrastructure needed to make radio inference practical on 5G networks. RIA Hub is a development platform for creating lightweight AI models optimized for radio applications, while RIA RAN integrates these models directly onto 5G base stations to achieve the sub-10ms inference latencies that enable frame-level network control and sensing capabilities.

The technical challenges are substantial: radio signals are noisy and complex, edge computing resources are constrained, and telecommunications systems demand exceptional reliability. The RIA platform enables engineers to overcome these challenges with purpose-built tools designed specifically for radio inference applications. Our toolchain produces models that address these constraints through optimized resource utilization and task-specific design, enabling high performance in lightweight packages suitable for real-time operation.

Current implementations demonstrate practical capabilities including opportunistic dynamic resource allocation, RF fingerprinting, ML-based beamforming that reduces acquisition overhead, and interference recognition and mitigation.

These developments support the broader O-RAN ecosystem by providing the missing link between AI research and operational deployment. As the industry moves from proof-of-concept demonstrations to commercial implementations, the availability of robust inference platforms becomes essential for realizing the full potential of AI-enabled wireless networks.

Looking to explore AI-powered wireless solutions? Discover implementation approaches and learn more about our work at qoherent.ai .