Mobile technology has profoundly transformed our society. Now, a new transformative wave is emerging – the AI revolution.

As artificial intelligence becomes increasingly integrated into our daily lives, it's changing not just how we interact with our devices but also reshaping the very infrastructure that powers our digital world, the radio access networks (RAN).

Noriyuki Fujino, Senior Director Radio Network Development Div., SoftBank said: “At SoftBank Corp., we have already begun taking action in transforming our network architecture to support the AI age and get ready for the massive increase in data traffic that AI-enhanced services will generate. We are collaborating closely with Nokia on this journey.”

The AI-RAN Alliance has introduced a framework for AI-RAN that considers three dimensions: AI on RAN, AI for RAN and AI and RAN. I hope you had a chance to attend Nokia’s Tech Winter Horizon 2024 event last week, where we explored all these different aspects of AI-RAN.

Now, I’m diving a bit deeper into AI's impact on 5G networks and why the uplink will be the cornerstone of emerging AI-powered services.

The rapid advancement of AI-enhanced applications

Imagine a world where your smartphone isn't just smart – it's downright genius. A world where your device can edit photos with the skill of a professional artist, engage in witty banter that rivals the sharpest minds, and even predict your needs before you're aware of them. This isn't science fiction; it's the reality that's unfolding before our eyes, thanks to the rapid advancements in AI technology.

But here's the kicker: all this intelligence comes at a price. And that price is paid in data – lots and lots of data, which needs a robust uplink connection.

The chart below depicts a moderate prediction for global consumer mobile traffic. As we can see, both direct and indirect AI traffic are expected to grow at a CAGR higher than non-AI-related traffic. When end users create prompts toward AI services and tools, they generate direct AI traffic. Indirect AI traffic is generated by AI-based algorithms. For instance, a social media platform could use AI algorithms to maximize user engagement by suggesting ad-hoc video content that matches the user’s profile. This will typically lead to the user consuming more of the AI-suggested video content, which increases the traffic volume.

The Global Network Traffic Report from Nokia Bell Labs provides more insights into the projected network traffic growth, including AI traffic.

Traditionally, mobile networks have been optimized for downlink traffic. After all, we've spent years consuming content – streaming videos, downloading files and browsing websites. But AI is flipping the script. Our devices aren't just passive receivers of information; they're active participants in a complex dance of data exchange.

Let's break it down with some eye-opening statistics from our research.

• Instantaneous data speeds in uplink are reaching tens of megabits per second.

• Even multi-modal tasks like AI-assisted writing and generating AI-enhanced images can spike uplink usage to 25Mbps when sending the content for processing.

• Voice conversations with AI assistants, while less data-intensive, are latency-sensitive, requiring consistent uplink speeds of 1-3Mbps.

• Data volumes in uplink are increasing, amounting to tens of megabytes more data per day.

These numbers might not mean much to the average user, but for network engineers, they're setting off alarm bells. We're looking at a potential surge in uplink data traffic that could overwhelm our current network infrastructure if we're not prepared. For comparison, the typical uplink in today’s 5G networks is 10-15Mbps on average, and in many networks, it’s even below 5Mbps.

The AI traffic conundrum

As we explore the world of AI-generated traffic in more detail, we uncover a new mix of data flows. It's not just the volume of data that matters; it's the unique characteristics of AI traffic that can pose challenges for our networks.

Bursty and unpredictable: One of the defining features of AI traffic is its burstiness. Unlike streaming a video, where data flows at a relatively constant rate, AI applications tend to generate short, intense bursts of data. Picture this: you're using an AI photo editing tool. Every time you make a change, there's a spike in uplink traffic as your device sends the updated image to the cloud for processing. These spikes can reach up to tens of Mbps, but they only last for 1-2 seconds, making it essential for networks to allocate resources efficiently.

Real-time responsiveness: Another crucial factor is latency. Low latency is critical for many AI applications, especially those involving real-time interactions like voice conversations. This puts additional pressure on networks to handle high volumes of data with lightning-fast responsiveness.

On-device processing vs. cloud: Interestingly, different device manufacturers are taking varied approaches to AI processing, which directly impacts network traffic. Based on testing the latest commercial software of Apple, Google Pixel and Samsung devices*, we discovered three approaches. Some devices favor on-device processing, which reduces network load but still requires significant uplink for cloud synchronization. Conversely, other devices heavily rely on cloud processing, generating more consistent network traffic. And finally, some take the middle ground, with some tasks processed on-device and others in the cloud. This diversity in approaches adds another layer of complexity to network planning and optimization.

AI impact on uplink and downlink traffic

Preparing for the AI-augmented future

As we stand on the brink of this AI revolution, the question isn't if mobile networks will be impacted, but how we can prepare for the future where these high-demand AI applications become the norm rather than the exception, causing an inevitable surge in uplink traffic.

Mobile network operations should consider:

• Conducting extensive traffic modeling to forecast future AI usage patterns.

• Investing in network infrastructure upgrades to handle the anticipated increase in uplink traffic.

• Collaborating with AI developers and device manufacturers to optimize applications for network efficiency.

Traditional network optimization strategies have focused primarily on downlink capacity. Now, mobile network operators should start preparing their networks to handle increased uplink demands.

While 5G networks are already capable of handling significant data loads, the unique demands of AI traffic are pushing the technology to evolve further. Nokia’s products and solutions are designed to handle the demands of the future.

In discussions with Taiwan Mobile, they stated that for telecom operators, it is equally important to leverage AI to enhance network operational efficiency and optimize network performance ensuring customers enjoy the best AI-powered service experience. They also believe that the combined power of RAN, cloud and AI is set to unleash growth and new opportunities. Taiwan Mobile is developing innovative AI-enhanced applications and use cases and accelerating their service upgrade to support the future evolution in data traffic.

Nokia’s 5G-Advanced solutions help tackle the uplink traffic surge

Nokia’s 5G-Advanced portfolio offers a range of powerful capabilities to optimize uplink performance and deliver a premium user experience, helping our customers tackle the uplink traffic surge. Let’s explore some of these solutions in more detail.

2x2 Multiple Input Multiple Output (MIMO): This feature can double the uplink speed, as the device sends data using two antennas instead of just one. It works only in 5G Standalone architecture. In addition, you can combine the power of 2x2 MIMO with carrier aggregation by using the 5G uplink transmit feature.

Uplink and downlink carrier aggregation: By combining uplink and downlink carrier aggregation, we can maximize the use of available spectrum assets and dedicate more resources to uplink traffic for superior throughput. This ensures that even the most demanding applications run flawlessly.

Zero-forcing Multi-User MIMO (MU-MIMO): This advanced beamforming technology mitigates interference, allowing for simultaneous transmission to and from multiple users, maximizing uplink capacity and ensuring smooth, uninterrupted service.

Spectrum refarming: Refarming the FDD spectrum from 4G to 5G allows operators to allocate more bandwidth to the 5G uplink, further enhancing its capacity and performance.

Scheduling optimization: Our advanced scheduling algorithms ensure that uplink resources are allocated efficiently, minimizing delays and maximizing throughput.

Network slicing for quality of service (QoS) differentiation: With advanced slicing capabilities including edge slicing, we can prioritize different applications and users, ensuring that critical enterprise applications and time-sensitive services such as video conferencing and online gaming receive the bandwidth they need.

AI/ML-driven network management and optimization: By leveraging the power of artificial intelligence and machine learning, we can predict traffic patterns and adapt network resources in real time, ensuring optimal performance in all traffic conditions.

AI-RAN is powering the transformation

The AI revolution is not just coming; it's already here. To lead this transformation, mobile network operators and technology suppliers must recognize and adapt to the unique demands of AI traffic. This will allow operators to capture the emerging monetization opportunities, powered by the synergy of AI and RAN.

If you want to know more about how our industry is approaching AI-RAN, read this blog by Nokia’s Mobile Networks President, Tommi Uitto.

So, the next time you ask your AI assistant a question or edit a photo with an AI-enhanced, cloud-based tool, take a moment to appreciate the complex dance of data happening behind the scenes.

*Apple is a trademark of Apple Inc., registered in the U.S. and other countries and regions. Google Pixel is a trademark of Google. Samsung is a trademark of Samsung Electronics Co., Ltd.

Resources:

AI-RAN Whitepaper