Rohde & Schwarz has published an article that explores the evolving landscape of artificial intelligence (AI) and its growing influence on 6G wireless communications. The article begins by grounding the study in the current era of weak AI – systems that exhibit narrow intelligence in terms of capabilities such as logical reasoning, perception, and natural language processing. From game-playing algorithms like AlphaGo to virtual assistants like Siri and Alexa, these technologies highlight how AI is already embedded in daily lives. In this light, this article looks ahead to the promise of strong AI and how machine learning, particularly through multi-layer neural networks, is paving the way for revolutionary advancements.
The developments will result in a future where wireless networks feature AI-native air interfaces – radios that not only transmit and receive signals but also learn from their environment and from each other. This vision represents a significant shift in how engineers design, deploy, and experience next-generation communication systems such as 6G.
Types of Neural Networks and their use in Wireless System Design
Neural networks are in turn a subcategory of machine learning and relevant in wireless communication – as the following three examples of neural networks show:
Recurrent neural network (RNN): output from the previous step serves as the input for the current step (e.g. text processing). RNNs are useful for time series prediction (“memory effects”) and linearizing analog RF Frontends as well as antenna subsystems through digital pre- and post-distortion algorithms based on ML models.
Convolutional neural network (CNN): feed-forward neural networks with up to 30 layers. A CNN processes structured arrays of data (e.g. originally designed for image processing) and is currently one option to realize a neural receiver.
Concept of an autoencoder: a special type of artificial neural network assisting with learning efficient data coding in an unsupervised manner. It aims to train the network to ignore insignificant data. Autoencoders, for example in the form of transformers, are currently being investigated to compress channel state information feedback, which is gathered from measurements in the downlink and sent back in the uplink direction.
6G artificial intelligence and machine learning
Even though artificial intelligence is one of the ten main 6G research areas, it is not a standalone research area. It still plays into all of the other areas however, such as cell-free massive MIMO, full-duplex communication and intelligent reflecting surfaces. The performance of every single example can be enhanced by data-driven, trained systems in 6G networks, increasing energy efficiency and therefore also sustainability at the same time. Using trained machine learning models for signal processing tasks like channel estimation, equalization and demapping will further optimize the air interface compared to present-day 4G LTE and 5G NR networks.
Rohde & Schwarz supports research activities across Europe, Asia and the US and works as a partner in projects like the 6G-Access, Network of Networks, Automation & Simplification (6G-ANNA) lighthouse project. This project aims to develop a design for 6G that includes end-to-end architecture and simplifies the interaction between humans, technology, and the environment using new sensors and algorithms to detect human movements.
AI Challenge for 6G Networks
Rohde & Schwarz highlights that establishing an AI-native air interface for 6G involves replacing traditional signal processing blocks at the physical layer with machine learning (ML) models. Initially, this means substituting individual tasks like channel estimation, equalization, and demapping with a single trained ML model—known as a neural receiver. Beyond the air interface, ML also has potential applications in the RF Frontend, including the linearization of power amplifiers.
The adoption of AI/ML in 6G signal processing and hardware design can be broken into three phases:
- Phase 1: ML for RF Frontend Linearization
ML models begin to replace deterministic algorithms used in linearizing power amplifiers and other RF Frontend components. This is feasible because these components are usually designed by a single vendor, simplifying data access for training neural networks. - Phase 2: Neural Receiver Implementation
Traditional signal processing tasks in the receiver chain—like channel estimation, equalization, and demapping—are replaced with a unified ML model, enabling more adaptive and efficient processing. - Phase 3: End-to-End (E2E) Optimization
ML is used to jointly optimize the transmitter, receiver, and baseband processing. This includes adapting transmission methods to specific applications (e.g., voice, XR) and conditions. A key example is replacing traditional modulation with a custom, learned constellation that enhances system performance and enables pilotless transmission.
6G and AI or ML: Solutions and Benefits
How can test and measurement solutions provide greater insights and help improve ML-based DPD models?
Test and measurement solutions can be used to create reference models based on a classical approach using iterative methods e. g. the R&S®SMW200A vector signal generator which helps to characterize underlying hardware or the R&S®FSW signal and spectrum analyzer which allows for the sample-by-sample correction of amplitude and phase iteratively for given waveform, also known as direct DPD. Such procedures provide a good baseline.
Rohde & Schwarz also previously showcased an AI/ML-based neural receiver setup with custom constellations at the Brooklyn 6G Summit 2023. This setup uses an R&S®SMW200A vector signal generator to emulate a single user applying 2x4 MIMO transmission scheme. The signal generator is also used to add fading and noise to the transmission, emulating a real-world scenario. The signal is then captured with the R&S MSR4 multi-purpose satellite receiver by using four receive channels, digitized and streamed to a server. This server hosts the R&S server-based testing framework that includes R&S®VSE vector signal explorer micro-services. Here the synchronization to the signal is performed, along with Fast Fourier Transform (FFT) and cyclic prefix removal, before this pre-processed data is processed by a neural receiver designed by NVIDIA, using NVIDIA Sionna™, an open-source library for 6G research.
Click here to register for the tech talk on AI for wireless communications
Click here to read the original article from Rohde & Schwarz.
