On-Device Neural Receiver
Deep Learning Neural Baseband Receiver on UE
An on-device deep neural network replacing conventional MIMO detection, channel estimation, and demapping modules directly inside the user equipment modem.
Technical Explanation
In 6G, user terminals encounter severe non-linear hardware distortions, high mobility Doppler shifts, and complex near-field channel conditions. The on-device neural receiver utilizes compact quantized neural network architectures (such as lightweight CNNs or Transformers) implemented on dedicated handset neural processing units (NPUs). It jointly performs channel estimation, equalization, and soft-bit LLR calculation end-to-end, delivering 2-4 dB block error rate gains over classical linear minimum mean square error (LMMSE) algorithms while fitting within thermal and battery envelopes.
Key Functions
- Replaces cascaded estimation and detection with learned end-to-end inference
- Compensates for handset power amplifier non-linearities and I/Q imbalance
- Operates in microsecond latency on mobile NPU accelerators
- Adapts dynamically to rapid terminal velocity and Doppler spreads