Open RAN Networks Will Upend Telecom as We Know It

The Telecom Industry’s Secret Is Out: Your Carrier’s Tower Is a Locked Box

For decades, the radio access network (RAN) — the hardware that connects your phone to the internet — has been a black box. One company, say Ericsson or Nokia, sold the whole thing: the baseband unit, the radio unit, the software that runs it all. If you wanted to upgrade, you bought from the same vendor. If you wanted to tweak performance, you called their engineers. The system was proprietary, vertical, and deliberately opaque.

That is about to shatter.

In a 2023 tutorial published in IEEE Communications Surveys & Tutorials, researchers Michele Polese, Leonardo Bonati, Salvatore D’Oro, and Stefano Basagni laid out the blueprint for a different kind of network: Open RAN, or O-RAN. The paper, which has already accrued 855 citations, is not just a technical specification. It is a declaration that the telecom industry’s old guard is no longer necessary. The authors show how O-RAN replaces closed, single-vendor hardware with disaggregated, interoperable components connected by open interfaces and optimized by artificial intelligence. The result is a network that can be programmed, tuned, and even built by anyone with the right software.

This is not a slow evolution. It is a structural shift in who controls the infrastructure that connects five billion people.

What O-RAN Actually Does (And Why Your Carrier Hasn’t Told You)

The core idea is simple, but its implications are radical. Traditional RANs are monolithic. A single vendor provides the radio unit that sits on the tower, the baseband unit that processes signals, and the software that manages it all. The interfaces between these components are proprietary. You cannot swap a Nokia radio unit into an Ericsson baseband unit. You cannot run third party optimization software on a Huawei RAN.

O-RAN breaks this apart. The architecture, as Polese and his colleagues describe it, splits the RAN into three main building blocks: the radio unit (RU), the distributed unit (DU), and the central unit (CU). These components communicate through standardized, open interfaces defined by the O-RAN Alliance. Crucially, the software that controls the network is separated from the hardware. A central abstraction layer, the RAN Intelligent Controller (RIC), sits above the physical components and uses data driven, closed loop control to optimize performance in real time.

What does that mean in practice? Imagine a stadium during a concert. Thousands of people are trying to stream video, post photos, and call friends. A traditional RAN would struggle. The vendor’s proprietary algorithms might try to balance load, but they are limited by the hardware and software they came with. An O-RAN network can do something different. The RIC, running machine learning models, can detect the congestion and dynamically adjust radio parameters, reroute traffic, or even allocate more spectrum to the busiest sector. And because the RIC is a software platform, not a piece of hardware, it can be updated, improved, and even swapped out for a better one.

The authors write that this enables “programmatic optimization through a centralized abstraction layer and data driven closed loop control” (Polese et al., 2023). In plain language: the network can learn and adapt, just like a software application.

The Architecture That Makes It Possible

The paper provides the most detailed tutorial on O-RAN architecture to date. It is worth walking through the key pieces, because the design choices reveal what the authors believe is possible.

The RAN Intelligent Controllers: Two Types, One Goal

O-RAN defines two RICs: the non real time RIC and the near real time RIC. The non real time RIC operates on a timescale of seconds to minutes. It handles tasks like policy management, network slicing, and training machine learning models. It sits in the cloud, far from the tower. The near real time RIC operates on a timescale of 10 to 100 milliseconds. It runs the trained models and makes split second decisions about radio resource allocation, interference management, and handovers. It sits closer to the network edge.

The authors emphasize that this separation is deliberate. “The architecture and interfaces enable AI/ML workflows that can be used to effectively control and manage 3GPP defined RANs” (Polese et al., 2023). This is not theoretical. The paper reviews early experimental results where researchers used the near real time RIC to optimize scheduling in a testbed, achieving measurable gains in throughput and latency.

Open Interfaces: The End of Vendor Lock In

The key to O-RAN is the open interfaces. The O RAN Alliance has defined specifications for how the RU, DU, and CU communicate. These are not abstract suggestions. They are detailed protocols that any vendor can implement. If you build a radio unit that conforms to the O RAN fronthaul specification, it will work with any O RAN compliant distributed unit. The same is true for the midhaul and backhaul interfaces.

This is the part that terrifies incumbent vendors. In a traditional RAN, once you buy from a company, you are locked in for years. Upgrades, expansions, and maintenance all go through that vendor. With O RAN, you can mix and match. You could buy radios from one company, baseband processing from another, and run optimization software from a startup. The authors note that this “multivendor, interoperable” approach is a foundational design principle (Polese et al., 2023).

Why This Matters: The Economics and the Politics

The technical details are important, but the real story is what O RAN does to the telecom industry’s power structure.

Lower Costs, More Competition

Today, building a mobile network is expensive. A single macro cell site can cost tens of thousands of dollars, and a large carrier might deploy hundreds of thousands of them. Most of that cost goes to a handful of vendors: Ericsson, Nokia, Huawei, Samsung. O RAN changes the equation. By opening up the interfaces, it allows smaller companies to compete. A startup that writes a better scheduling algorithm can sell it to any carrier running O RAN. A hardware manufacturer that builds a cheaper radio unit can plug it into any network.

The authors do not provide cost estimates, but the logic is clear. When you can buy components from multiple vendors, you can shop for the best price. When you can swap software without replacing hardware, you can upgrade more cheaply. The paper frames this as a “new paradigm for RAN design, deployment, and operations” (Polese et al., 2023). It is also a direct threat to the oligopoly that has dominated wireless for thirty years.

National Security and Supply Chain Resilience

There is another dimension. Governments around the world are nervous about having a single vendor, especially Huawei, dominate their telecom infrastructure. The United States, the United Kingdom, and others have restricted or banned Huawei equipment. But banning one vendor does not solve the problem. You still need to buy from someone else, and that someone else might be just as locked in.

O RAN offers an alternative. If a network is built with open interfaces, no single vendor has a chokehold. You can swap out a radio unit from a company that becomes a security risk and replace it with one from a trusted supplier. The paper does not dwell on geopolitics, but the implications are clear. O RAN is not just a technical standard. It is a tool for national security and supply chain resilience.

How O RAN Handles AI and Machine Learning

This is where the paper gets genuinely exciting. The authors spend significant time on how O RAN enables AI and ML workflows. This is not an afterthought. It is a core feature.

Training and Inference at Different Timescales

The non real time RIC is where machine learning models are trained. It has access to historical data, network logs, and performance metrics. It can train a model to predict traffic patterns, detect anomalies, or optimize beamforming. Once the model is trained, it is deployed to the near real time RIC, which runs inference in milliseconds.

The authors describe this as a “closed loop control” system (Polese et al., 2023). The network measures its own performance, feeds that data to the non real time RIC, which updates the model, which is then deployed to the near real time RIC, which changes how the network behaves. The loop runs continuously. The network gets smarter over time.

What This Means for Your Phone

This is not abstract. If you are in a crowded area and your video call starts to stutter, an O RAN network can detect the problem and adjust. It might hand you off to a less congested frequency band. It might allocate more resources to your session. It might even predict that congestion is about to happen and preemptively balance the load. The authors note that the architecture enables “datadriven closed loop control” (Polese et al., 2023). The network does not just react. It anticipates.

What the Research Does Not Prove

The paper is a tutorial, not an experimental study. It does not claim to have built a production scale O RAN network and measured its performance against a traditional one. The authors are careful to call this a “review of early research results” and an outline of “future directions” (Polese et al., 2023). They do not provide benchmarks. They do not claim that O RAN is faster or cheaper today.

This is an important caveat. O RAN is not yet deployed at scale. The standards are still being finalized. The open interfaces are not all mature. Security is a major concern: with more components and more software, the attack surface grows. The authors dedicate a section to security challenges, noting that “the disaggregated nature of O RAN introduces new vulnerabilities” (Polese et al., 2023). A traditional RAN is a single appliance. An O RAN is a distributed system. Distributed systems are harder to secure.

There is also the question of performance. Can a virtualized, software defined RAN match the efficiency of a purpose built, hardware optimized one? The early results are promising, but the paper does not settle the debate. The authors call for more research on latency, throughput, and reliability at scale.

The Open Question That Keeps Engineers Up at Night

The biggest unknown is whether O RAN can handle the extreme demands of a 5G network. 5G requires ultra low latency, massive bandwidth, and high reliability. A virtualized RAN running on general purpose hardware might struggle to meet these requirements. The authors acknowledge this, writing that “research challenges include achieving the required performance with virtualized and disaggregated components” (Polese et al., 2023).

The answer may depend on how much of the processing is done in hardware versus software. The O RAN architecture allows for different splits between the radio unit and the distributed unit. Some splits keep more processing in hardware, which is faster but less flexible. Others push more into software, which is flexible but slower. The trade off is not yet resolved.

What This Actually Means

• ▸Your carrier will soon be able to buy network components from different vendors, breaking the lock in that has kept prices high and innovation slow. This means more competition, lower costs, and potentially faster upgrades. The paper’s description of “multivendor, interoperable components” is not a footnote. It is the whole point.

• ▸The network will become programmable, like a smartphone. The RIC is essentially an operating system for the radio access network. Just as you can install apps on your phone, carriers will be able to install optimization algorithms on the RIC. The authors’ emphasis on “programmatic optimization” is a direct invitation to software developers.

• ▸AI and machine learning will move from the cloud to the network edge. The two tier RIC architecture (non real time and near real time) is designed to run ML models where they are needed most. This is not about adding AI as a gimmick. It is about making the network adaptive in real time.

• ▸Security will become a bigger concern, not a smaller one. With more components and more software, there are more ways to attack. The paper’s discussion of security is not a weakness. It is an honest acknowledgment that open systems require open security practices.

• ▸The biggest winners may not be the carriers or the vendors. They may be the startups. O RAN lowers the barrier to entry. A small team with a good algorithm can sell to a global carrier. The paper’s review of “experimental research platforms” shows that testing is already possible. The race is on.

The telecom industry has been a fortress for decades. O RAN is not just opening the gates. It is tearing down the walls. The paper by Polese, Bonati, D’Oro, and Basagni is the blueprint. Now someone has to build it.