Science and engineering
Machine learning for wireless communications
Wireless engineers use machine learning to classify modulation types, to estimate channels and decode symbols with learned receivers, and to predict beams and allocate resources in 5G and 6G networks.
How machine learning is used in wireless communications
Classifying radio signals
Models identify the modulation type or emitter of a received signal from raw IQ samples, which helps spectrum monitoring and cognitive radio.
Real exampleOver-the-air radio signal classification (O'Shea, Roy and Clancy, 2018)Compared deep learning with a baseline of higher-order moments and boosted trees for radio signal classification under impairments such as carrier frequency offset and multipath fading, including over-the-air measurements with software radios.
Learned receivers
Neural networks replace or assist receiver blocks such as channel estimation, equalisation and demapping, and can adapt to impairments that hand-designed blocks model poorly.
Real exampleDeepRx (Honkala et al., 2020)A fully convolutional neural network receiver that runs the whole pipeline from the frequency-domain signal to uncoded bits in a 5G-compliant fashion.
Channel estimation and detection
In OFDM systems, the receiver must estimate how the channel distorts each subcarrier. Networks trained on simulated channels can estimate and detect symbols jointly.
Real exampleDeep learning for OFDM channel estimation (Ye, Li and Juang, 2017)Used deep learning for joint channel estimation and signal detection in OFDM and found it more robust than conventional methods with fewer pilots, no cyclic prefix or nonlinear clipping noise.
Beam prediction from sensing data
High-frequency links need narrow beams pointed at the user. Cameras, radar, lidar and GPS at the base station can help predict the best beam without a full search.
Real exampleDeepSense 6G (Alkhateeb et al., 2022)A large-scale real-world dataset of co-existing multi-modal sensing and communication data, built to support deep learning research across sensing, communication and positioning.
Can you learn machine learning for wireless communications?
Yes. Wireless engineers can add machine learning to their physical-layer knowledge, and machine learning engineers can learn communications basics. Start with Python, linear algebra and probability, then classical machine learning and deep learning on IQ samples and channel data, alongside wireless fundamentals.
- PythonSimulate links and process IQ samples with NumPy.Learn it in Python
- Linear algebra and probabilityMIMO, channel models and detection theory are built on matrices and probability.Learn it in Maths
- Machine learning basicsClassifiers and regressors on signal features, with careful separation of training and test channels.Learn it in Machine Learning
- Deep learning for signalsConvolutional networks and transformers on IQ samples and channel matrices, with differentiable link simulators such as Sionna.Learn it in Deep Learning
- Wireless fundamentalsOFDM, MIMO, channel coding and the 5G NR physical layer.
Moving into machine learning for wireless
If you already work in the field
Wireless engineers know the physical layer, the standards and link-level simulation. Add PyTorch or TensorFlow and try a learned channel estimator or receiver in a simulator such as Sionna, compared with your standard algorithm. Equipment vendors, chipmakers and operators hire for AI in the radio access network.
If you come from data science or ML
Machine learning practitioners need digital communications basics: modulation, OFDM, MIMO, channel models and how 3GPP standards work. A wireless communications textbook and open simulators make it hands-on.
Roles that do this workWireless systems engineerPhysical layer engineerMachine learning engineer, RANResearch engineer, 6G
Tools you will use
| Tool | What you use it for |
|---|---|
| Sionna | Simulate link-level wireless systems and train learned receivers end to end. |
| TorchSig | Generate and load radio-frequency signal datasets for deep learning. |
| GNU Radio | Build software-defined radio pipelines to capture and test signals. |
| PyTorch | Train neural receivers and modulation classifiers on IQ samples. |
Public data to practise on
- DeepSig RadioML: Synthetic IQ samples of many modulation types across SNRs for modulation classification.
- DeepMIMO: Ray-traced MIMO channel data for beam prediction and channel estimation research.
Your first project
Download a DeepSig RadioML dataset, classify modulation type with a small convolutional network, and compare it with cumulant features and gradient boosting using accuracy at each SNR.
Common questions
How is AI used in 5G and 6G?
For channel state information feedback, beam management, positioning, scheduling and network optimisation. 3GPP has studied AI and machine learning for the 5G air interface, and AI is a design theme for 6G research.
What is a learned receiver?
A receiver in which neural networks replace or assist blocks such as channel estimation, equalisation and demapping, trained on simulated or measured signals.
Can I practise wireless machine learning without hardware?
Yes. Link-level simulators and public datasets such as RadioML and DeepSense 6G let you train and test models on a laptop or a single GPU.
Do wireless engineers need deep learning skills?
Increasingly, for research and system design roles at vendors and chipmakers, although strong communications fundamentals remain the core requirement.
Courses for these skills
- Deep Learning with PyTorchExplore sequence representations after a baseline.Syllabus available · lessons open as produced
- Maths for Machine LearningConnect signal vectors and objectives.26 published lessons · all planned lessons open
- Statistics and Probability for Machine LearningUnderstand noise and error-rate estimates.41 published lessons · all planned lessons open
- Data Tools: NumPy, pandas, Matplotlib and scikit-learnPrepare arrays and preserve sampling metadata.23 published lessons · all planned lessons open
- Practical Machine Learning in PythonBuild and evaluate a compact predictive baseline.51 published lessons · more lessons in production
- Python for AI and Machine LearningUnderstand input validation and reproducible transformations.33 published lessons · all planned lessons open
New to programming or maths? Start with Python, or follow the machine learning path in order. Published lessons are free to read; compare Free and Pro.