Invented by SO; Steven Rhejohn Barlin, NASEEM; Tasneem Khuzema, AFZAL; Muhammad Umair, AHMED; Sarim Mohammed, ZHENG; Alina, TO; Philip, BROOKS; Andrew Scott

Smart technology is changing how we monitor and fix problems in telecom networks. A new patent application describes an inventive way to use artificial intelligence (AI) and advanced language models to find and solve issues quickly in large, cloud-based mobile networks. In this article, we’ll break down the background, explain the science and prior art, and look closely at the unique ideas behind this invention.

Background and Market Context

Telecommunications networks are the backbone of our connected world. Every time you make a call, send a text, or stream a video, you rely on these networks. Behind the scenes, these systems are incredibly complex. Modern telecom networks use the cloud and virtualization, which means many different software pieces, called network functions, run on lots of servers. There can be thousands or even millions of these nodes working together, each one vital for network performance.

If something goes wrong with just one part, it can slow down the network, cause calls to drop, or even lead to outages. When that happens, customers get frustrated, and companies lose money. That’s why it’s so important to detect problems as soon as possible and fix them before they affect users. But with so many nodes and network functions, it’s very hard for human engineers to keep up or spot the first signs of trouble.

Historically, teams would use simple tools to look at logs and metrics from the network. These tools might show if memory usage is high or if there are a lot of errors. But as networks have grown more complex, these old methods have become less helpful. Networks now produce huge amounts of data every second. Sifting through this mountain of information by hand is nearly impossible, and traditional systems often miss subtle warning signs before a big problem happens.

Operators need smarter, automated ways to monitor their networks. Many have started to use machine learning and AI to help, but even these solutions have limits. They may spot clear anomalies, but they still struggle to understand the root cause or recommend how to fix it. This is where the latest ideas using AI and language models come into play, promising a new level of intelligence and automation for telecom monitoring.

Scientific Rationale and Prior Art

To appreciate this invention, it helps to know what solutions came before. In the past, telecom operators relied on rule-based systems. These systems would look for simple patterns—like a certain error message or a spike in traffic—and send alerts. But they couldn’t handle the complexity of new cloud-based networks or adapt to new kinds of problems.

As networks grew, companies tried using more advanced AI models. These could learn from past data to spot issues. For example, they might use classification models like decision trees or support vector machines to figure out if a network function was healthy or not. Some even used deep learning models, such as convolutional neural networks, to find patterns in logs or metrics. These tools helped, but they still had trouble telling engineers exactly what was wrong or how to fix it. They often needed a lot of tuning and still produced too many false alarms.

One of the big challenges is that telecom networks have many different types of network functions, each with their own logs and data formats. Some manage user sessions, others route data, and each can report errors in different ways. No single model could understand all the different data, and most AI systems were trained on just one piece of the puzzle.

Another challenge is that telecom standards, like those from the 3rd Generation Partnership Project (3GPP), are very detailed and complex. These documents contain the best practices and requirements for how networks should work. Most prior AI models couldn’t use this deep, expert knowledge to help with troubleshooting.

Recently, the world has seen breakthroughs in language models—AI that can read and understand large amounts of text, like GPT (Generative Pre-trained Transformer) models. These models can learn from huge libraries of documents and answer questions in natural language. Some companies have tried linking these models to network data, but they haven’t gone deep enough to tailor them to telecom’s special needs or combine them with other AI models that understand raw network data.

The prior art, then, includes basic log analyzers, simple machine learning models, and early uses of language models. However, none have brought together specialized AI for recognizing network function types, clever prompt creation, and telecom-trained GPT models in a single, seamless solution.

Invention Description and Key Innovations

This patent describes a new method and system for finding and fixing problems in virtualized telecom networks. The core idea is to use a combination of AI models and a special GPT language model, all working together. Here’s how it works, step by step, in simple terms.

First, the system collects data from the network. This can be test data from new features or live data from real users. The data might include logs, events, traces, and metrics—basically, everything the network produces as it runs. The system then parses and organizes this data into a format that computers can easily read.

Next, the data goes to a first AI model. This AI is trained to look at the data and figure out what kind of network function the data came from. For example, it might see that some logs are from a Mobility Management Entity (MME), User Plane Function (UPF), or other types like Session Management Function (SMF), Packet Data Network Gateway (PGW), or Serving Gateway (SGW). This step is important because each network function has its own role and reports problems differently.

This first AI model can use many different techniques. It might be a classification model like a decision tree or support vector machine, or more advanced, like a convolutional neural network. If the data is text-heavy, it could use natural language processing (NLP), such as a Named Entity Recognition (NER) model trained to spot network function names and descriptions in the logs.

Once the system knows the network function type, it uses a second AI model. This model’s job is to create a special prompt, or question, to ask the GPT model. The prompt is tailored to the network function and the context of the data. It includes not just the function type, but also the log type, the relevant data, and any extra details the AI can extract, like error messages or session IDs.

This step is key. Instead of just sending raw data to the language model, the second AI model crafts a focused, smart prompt that tells the GPT model exactly what to look for. By doing this, the system makes sure the GPT model can give more accurate and useful answers.

Now, the GPT model comes into play. This is not just any language model—it has been trained using 3GPP documentation, meaning it “knows” telecom standards and best practices. With this expert knowledge, the GPT model can read the prompt and the related data, then figure out what condition or problem is present in the network. It can also perform root cause analysis, meaning it tries to explain why the problem happened. On top of that, the GPT model can suggest ways to fix the problem or improve performance, all based on its understanding of telecom standards.

Finally, once the system knows the condition and has recommendations, it can start an action in the network. This might mean alerting an engineer, restarting a troubled network function, or even applying a fix automatically. The whole process—from collecting data to taking action—can happen with little or no human help, making the network smarter and more reliable.

The invention stands out for several reasons:

First, it brings together multiple AI models, each with a special job. The first AI model is an expert at recognizing network function types, while the second AI model is skilled at building the best prompt for the next step. The GPT model is trained with deep telecom knowledge, giving it the power to analyze problems like a seasoned network engineer.

Second, the invention uses a modular design. Each AI model can be trained and improved separately. For each network function, there can be a matching recognition model, tuned for different log types or data formats. This allows the system to grow and adapt as new network functions or data sources appear.

Third, by using clever prompt engineering, the system makes the most out of the GPT model. Instead of overwhelming it with raw logs, it delivers just the right context, boosting accuracy and reducing costs.

Fourth, the invention supports both test and live data, so it can help during development and in real-world operations. It can work on local devices or in the cloud, and it can interface with many types of user applications or dashboards.

Lastly, the system is built to be efficient. By organizing data well, compressing models, and using feedback to improve, it’s designed to handle the huge data volumes of modern networks without slowing down.

Conclusion

Telecom networks are only getting bigger and more complex, and the old ways of monitoring them are no longer enough. This patent shows a new way forward, using a blend of AI and advanced language models to spot problems early, understand their root causes, and suggest fixes—often before users even notice an issue. By combining specialized recognition models, smart prompt creation, and a GPT model trained in telecom standards, the invention brings expert-level insight to automated network management.

For operators, this means fewer outages, happier customers, and a network that can adapt to whatever comes next. For the world, it means a future where communication is more reliable and resilient, powered by smarter technology at every level.

Click here https://ppubs.uspto.gov/pubwebapp/ and search 20250217715.