The Promise and Challenges of AI-Native 6G Networks

The integration of AI and ML into mobile networks is key to the new era of wireless communications.

With 6G capabilities set to be released in 2028, we are moving into a future where AI is intrinsic to network infrastructure and services. Unlike in 5G, where artificial intelligence and machine learning (AI/ML) are applied in specific use cases, AI will be fundamental to the 6G network. In other words, there will be a shift from AI as a performance enhancement to AI as a key technology component: from AI-assisted to AI-native. AI will be an integral component of network orchestration and management, signal processing, and network optimization, structurally changing how we think about and design our communication systems.

THE VISION OF AN AI-NATIVE AIR INTERFACE

Imagine a world where traditional signal processing tasks are seamlessly handled by advanced ML models. An example of such a concept is the neural receiver, which aims to improve the handling of dynamic environments, improving channel estimation and equalization by significantly reducing estimation errors.

Neural receivers are already under development. For example, Rohde & Schwarz and Nvidia have collaborated to design and train a neural receiver based on the current 5G New Radio standard. This neural receiver resides in the base station and is specifically trained for Multiuser MIMO (MU- MIMO) scenarios. For demonstration purposes, two users are simulated facing different wireless channel conditions (mobile and stationary). As the neural network incorporates information about various modulation schemes and subcarrier spacings into its weights during the offline training process, different models handle the different modulation schemes and subcarrier spacings. The performance of these different models can be tested with the Rohde & Schwarz 6G neural receiver testbed. Visitors to the Mobile World Congress 2025 can see the latest version of the setup in action at the Rohde & Schwarz booth. This version demonstrates the latest milestone of the ongoing collaboration: the integration of ray tracing for wireless channel behavior emulation.

COLLABORATION IS KEY

The ongoing work within 3GPP on standardizing an AI/ML framework for 5G-Advanced highlights the importance of industry-wide cooperation.

The organization is currently exploring ML use cases, such as enhancing channel-state information (CSI) feedback using autoencoders, which is a specific type of neural network initially designed for image compression tasks. CSI-feedback enhancement is a so-called two-sided model use case, implying that one part of the ML model is executed within the device (UE; the encoder) while the other is executed on the infrastructure side (gNB; the decoder). Studies by 3GPP members indicate that ML-based CSI feedback outperforms conventional methods, making it a valuable differentiator for industry leaders. However, while companies may develop and train their models independently, cross-vendor interoperability is crucial for seamless integration.

This is where test and measurement play a key role in validating performance and enabling vendor interoperability. Rohde & Schwarz, a leading manufacturer of test and measurement equipment, is showcasing an industry-first demonstration of CSI-feedback enhancement at Mobile World Congress 2025, supported by Qualcomm Technologies.

The two companies recently achieved cross-vendor interoperability for ML-based CSI feedback. Rohde & Schwarz developed a decoder for its network emulator, while Qualcomm Technologies implemented a proprietary device encoder on a mobile form factor reference design. Despite employing distinct training approaches for their AI models, interoperability was achieved through specified reference models, marking a significant step toward standardized ML-based communication in 5G-Advanced and a future 6G standard.

GENERATIVE AI AND THE BRIGHT FUTURE AHEAD

Generative AI has potential applications in many sectors, and wireless communications is not an exception. Within the 6G AI-native interface, generative AI can greatly improve various aspects of wireless systems.

Traditional channel models, for example, are developed through extensive, long-running measurement campaigns that collect data in specific environments and frequency bands. Generative AI has the potential to create models that provide a more detailed and adaptable representation of channel environments. Generative Adversarial Networks (GAN), for example, can create realistic and dynamic channel models that capture the complex behaviors of wireless radio channels. These models can then be used in the training and inference processes to design and optimize neural receiver architectures that further increase performance.

Of course, generative AI comes with its own set of challenges, such as the high computational complexity of training and the need for large, high-quality datasets.

However, it still shows great promise for 6G communications and creating more robust and efficient wireless networks.

The journey toward AI-native mobile networks is filled with challenges, but the potential rewards are immense. With the increased integration of AI/ML, we are setting the stage for a new era of advanced communications.

Experience the role of AI/ML in wireless communications firsthand at the Rohde & Schwarz booth: Hall 5, Stand 5A80.