Demonstrations

These guides start from reusable system templates, train external replacements, return portable artifacts, and run paired benchmarks. Physical-layer examples capture selected tensors at a block boundary. End-to-end image examples instead use file-backed data contracts and train both communication endpoints together.

For the shared UI concepts and reproducibility rules, read the physical-layer learned-method workflow.

Flagship Interactive Demo

Break the comparison is the shortest route into Noema’s core idea. It loads the canonical launch evidence, starts with a valid paired comparison, and lets you introduce condition, aggregation, metric, and comparator-role faults one at a time. Its static companions and canonical result table are collected in the F0–F5 launch visual system. Watch the complete learned QPSK I/Q calibration workflow for the clean installation, contract export, external training, returned model, and UI comparison.

Resource Allocation

• Learned OFDM subcarrier allocation

• Reliability-aware OFDM allocation with delayed CSI

• Learned joint ISAC OFDM allocation

Channel Estimation and Feedback

• Learned 2×2 MIMO-OFDM channel estimation

• Learned CSI compression and feedback

Receivers and Signal Classification

• Learned QPSK receiver calibration

• Learned QPSK carrier tracking

• Automatic modulation recognition

Localization and Sensing

• Learned two-dimensional range localization

• Learned narrowband AoA estimation

• Learned near-field XL-MIMO range-angle focusing

Beamforming

• Learned MISO beam selection

• Learned LEO-NTN Doppler prediction and beam handover

End-to-End Communication

• DeepJSCC vs. capacity-matched JPEG over AWGN

• Blind DeepJSCC over slow Rayleigh fading