Signal Analysis Overview
Radar signal analysis can be categorized into three main types:
Time-Frequency Transform-based methods, Time-Alignment-based similarity analysis, and Deep Learning-based feature learning.
1. FFT, STFT and Wavelet Transform
These methods analyze signals by decomposing them into components using specific basis functions.
| Aspect | FT(FFT) | STFT | Wavelet Transform | |
|---|---|---|---|---|
| Basis Function | e−jωt | g(t−τ)e−jωt | ψ(at−b) | |
| Varying Parameters | Frequency ω | Time τ, Frequency ω | Time b, Scale a | |
| Meaning |
Extract frequency components over the entire signal |
Extract frequency components within a local time window |
Extract time-scale components at a specific time and scale |
2. DTW
DTW is an algorithm that measures the similarity between two signals even when their time axes are nonlinearly distorted.
| Input | Two signals (sequences). | ||
| Core Concept | Time alignment (warping) via dynamic programming. | ||
| Varying Parameter | Time index (nonlinear mapping) | ||
| Mathematical Operation | Distance minimization (finding the optimal warping path). | ||
| Meaning | Measures similarity by optimally aligning signals in the time domain. | ||
| Output | Alignment path and the “warping distance.” | ||
3. Deep Learning Model
Unlike traditional methods that require manual feature engineering,
Deep Learning models automatically learn hierarchical representations directly from raw or pre-processed radar data.
| Model Type | CNN | RNN / LSTM / Transformer | Autoencoder (AE) | |
| Primary Input |
2D Spectrograms (STFT/Wavelet) |
1D Raw Time-Series (I/Q Data) |
Normal Signal Patterns | |
| Key Mechanism | Spatial pattern recognition | Sequential dependency learning |
Data compression & reconstruction |
|
| Radar Application |
Target Classification (e.g., Drone vs. Bird) |
Modulation Recognition & Tracking |
Anomaly Detection (Clutter Rejection) |
Radar Sensor SoC & Module
WiFi IoT Module
5G/LTE/CAT-M1/NB-IoT
WiSUN/HaLow/Thread Module
