Data Signal & Carrier Classifier
Overview
This model is designed for high-frequency telecommunications monitoring. It analyzes the Data Signal (তথ্য সংকেত) integrity and identifies the underlying Carrier (বাহক) modulation type in real-time. By processing raw I/Q samples, it can distinguish between various modulation schemes and detect signal degradation caused by environmental interference.
Model Architecture
The model utilizes a Temporal-Frequency Transformer architecture:
- Feature Extraction: A 1D-Convolutional front-end processes the raw waveform to extract local spectral features.
- Contextual Encoder: 8 layers of Multi-Head Self-Attention capture long-range dependencies across the Carrier wave.
- Classification Head: A linear layer that maps the hidden state of the
[CLS] token to the specific modulation class.
- Loss Function: Weighted Cross-Entropy to handle class imbalance in noisy environments:
$$L = -\sum_{c=1}^{M} y_{o,c} \log(p_{o,c})$$
Intended Use
- Spectrum Sensing: Automated identification of occupied bands in cognitive radio networks.
- Quality of Service (QoS): Monitoring the health of a Data Signal to trigger automated re-routing.
- Anomaly Detection: Identifying unauthorized Carrier frequencies in restricted zones.
Limitations
- Signal-to-Noise Ratio (SNR): Performance degrades significantly when the SNR drops below 3dB.
- Frequency Offset: Requires external synchronization; high frequency-offset values may lead to misclassification.
- Hardware constraints: Optimized for FPGA or GPU inference; may require quantization for edge deployment on low-power microcontrollers.
