AI-Native Wireless Systems: The Future of Connectivity

The wireless industry is experiencing an unprecedented surge in demand, with more than 7.1 billion human mobile users and a growing number of wireless machine-to-machine (M2M) connections.

Wireless is undergoing a major surge in demand, with more than 7.1 billion human mobile users of wireless machine-to-machine (M2M) connections, and for engineers, designing wireless systems and networks is becoming increasingly complex. Traditional predefined designs lack adaptability and cannot flex when requirements and environments are altered.

Addressing this need, AI-native systems are set to become the foundation of next-generation networks, embedding AI into network frameworks to enable engineers to achieve better coverage, higher capacity, and greater reliability.

A Consensus Among Industry Leaders 

The wireless standard organization 3rd Generation Partnership Project (3GPP) has been vocal about AI’s significant role in the forthcoming 5G Advanced and 6G standards. They’ve emphasized and pushed AI’s functionality to improve wireless elements like positioning, beam management, and Channel State Information (CSI) feedback. Similarly, the Wireless Broadband Alliance (WBA) praises AI’s potential to help wireless engineers in indoor positioning and beam management. 

These endorsements from industry leaders stress the growing importance of AI in wireless systems and the importance for engineers to integrate AI-native concepts to stay ahead in the next-gen wireless system race.

Benefits of AI-native Wireless Systems Compared to Traditional Designs

AI-native wireless systems are designed to learn from and adapt to their environment, a significant departure from traditional designs that are based on more rigid, predefined models with scalability limitations. These traditional designs also often require costly, time-consuming signal-processing resources.

On the other hand, AI-native systems offer three key benefits for wireless engineers: better coverage, higher capacity, and increased reliability. These improvements are achieved by the AI-native wireless system inherently incorporating AI algorithms directly into its operational framework.

Engineers designing AI-native systems need large real-world measured data sets. Most of this data is sourced from physical prototypes or by measuring real-world signals. However, most engineers use digital twins – representative virtual models that can be simulated – to augment data to train AI-native systems. Digital twins ensure that AI-native systems have sufficient data to handle adverse situations and efficiently manage system elements.

Design, Development, and Integration of an AI-Native Wireless System

Through a four-step process, engineers can address the complex task of developing an AI-native wireless system design workflow, which includes gathering data, training and testing the model, then implementing and integrating the model into the wireless system.

1. Gathering and Generating Data

Data is the backbone of AI-native wireless systems. The first step of building the system requires data collection through acquiring over-the-air (OTA) signals or synthesizing data from a digital twin. Synthetic data is especially useful as it facilitates scalability testing, fault tolerance, and anomaly detection, while also aiding in environment modeling and system configuration optimization. To ensure maximum model efficiency, engineers must ensure that training data is representative of real-world scenarios that the system will experience. Engineers can use the collected data to perform training and validation of AI models, testing and simulation, and optimization and performance tuning. With the data gathered, the next step involves simulation and modeling.

2. Training and Testing Model

When training an AI model for a wireless system, it is essential to determine the quantity of system parameters, including bandwidth allocation, latency, signal strength, modulation, and coding. By pairing these parameters with the full dataset obtained from step one, engineers can select and optimize machine learning algorithms for key system functions like autoencoders, channel estimation, channel feedback optimization, and resource allocation. There are a few factors that affect real-time performance that engineers must consider during the training process, including computational complexity, memory usage, and parallel processing on GPUs or clusters.

Once an AI model is trained, the model is tested for reliable performance in real-world systems. At this stage, the model's performance is repeatedly tested and adapted as needed to correct for biases, errors, and inefficiencies. After the adaptation is complete, the AI network is ready to be pruned, which involves converting the model to a fixed point and removing the neural network layers that do not contribute to the system's overall behavior. At this stage, the model can be implemented in the wireless system.

3. Implementing the AI Model

The first step to implementing an AI model as part of a real-world system requires scaling and resource assessment, which involves evaluating the processing power, memory requirements, and data throughput needed for the AI models to operate efficiently.

The second step is using automatic code generation for deploying pre-trained AI models on desktop or embedded targets using low-level code. This automates the implementation process and reduces manual coding errors. 

The last step to finalize the implementation is validation. Validation compares the performance of the implemented system to that of the original AI model. After engineers have identified and addressed discrepancies or performance issues, they are ready to perform model integration.

4. Integrating the Model

Once data is gathered and the AI model is trained, tested, and implemented, engineers can integrate the implemented AI models within the overall wireless system. This phase ensures that the newly implemented AI solution works seamlessly with the full legacy system. Before full-scale integration, engineers must ensure interoperability with existing system components by analyzing the end-to-end system performance rather than individual algorithms and subsystems.

An AI-native wireless system inherently incorporates AI algorithms into its operational framework.

MATLAB® can help engineers throughout all these stages of wireless system development. In MATLAB, they can perform tasks such as data generation, algorithm optimization, code generation for implementation, and model validation.

Challenges of Using AI to Design Wireless Systems

While the benefits of using an AI-native model are strong, there are a variety of hurdles to overcome in integrating AI into wireless systems, including balancing conflicting performance metrics and ensuring superior performance relative to legacy systems. The end goal is to achieve a balance that supports operational objectives by delivering high-quality overall performance.

Balancing Performance Metrics

Optimizing one metric typically compromises another, emphasizing the importance of finding an acceptable balance that meets the system’s overall goals. For example, increasing the network's throughput may lead to higher power consumption and latency, necessitating a trade-off to maintain energy efficiency. Modeling and simulation can help engineers explore a plethora of scenarios and configurations to balance the desired metrics. This predictive approach helps in making informed decisions and identifying optimal configurations without disrupting the actual system. 

Ensuring Superior Performance

Transitioning from legacy wireless systems to AI-enhanced systems without disruption is challenging, but essential for achieving superior performance. AI models that continuously learn are essential to ensuring a transition without disruption, as they teach the system to adapt to dynamic network conditions among wireless networks. Achieving superior performance requires training models using diverse, representative datasets. 

By simulating the integrated system before full-scale deployment, engineers can ensure AI components are positioned to interoperate properly with legacy systems. Engineers use tools like MATLAB to facilitate interoperability testing and identify potential compatibility issues and performance bottlenecks.

Conclusion

The wireless industry is at a critical juncture between traditional and AI-native models. With the upcoming rollout of 5G Advanced and 6G standards, the next generation of wireless systems will deploy more AI-native technologies to keep up with network demands. Engineers with roles dedicated to designing modern wireless systems have concluded that the integration of AI is no longer optional, but vital to the future of connectivity. By incorporating AI-native design principles, wireless engineers can develop systems and networks that meet today's needs and are equipped to evolve with tomorrow's wireless requirements and advancements.