Doppler Shift

The observed frequency f' differs from the transmitted frequency f by a factor proportional to the relative velocity v and the speed of light c: f' = f * (c / (c ± v)). A positive shift (higher frequency) occurs when the transmitter and receiver move toward each other; a negative shift occurs when they move apart. This is the same principle behind the changing pitch of a passing siren.

In multipath environments, Doppler shift does not produce a single frequency offset but a Doppler spread—a range of frequency shifts caused by different path angles. The inverse of the maximum Doppler spread defines the coherence time of the channel, the interval over which the channel impulse response remains approximately constant. For a 2.4 GHz signal with a vehicle moving at 30 m/s, the maximum Doppler shift is approximately 240 Hz.

Doppler-induced spectral broadening distorts the fine-grained frequency-domain features that fingerprinting models rely on, such as carrier frequency offset (CFO) and IQ imbalance signatures. A model trained on stationary device signatures may fail when deployed on a mobile platform because the Doppler spread smears the very hardware impairments it was trained to recognize. This is a core domain shift problem.

Channel-robust feature learning addresses Doppler shift through several techniques:

• Domain adversarial training forces the feature extractor to ignore Doppler-induced variations.
• Data augmentation with synthetic Doppler profiles during training improves generalization.
• Pilot-based channel estimation can estimate and remove the bulk Doppler shift before feature extraction.
• Time-frequency representations like the Wigner-Ville distribution can separate Doppler effects from stationary hardware signatures.

These two phenomena are often confused but are fundamentally distinct. Carrier frequency offset (CFO) is a hardware impairment caused by oscillator mismatch between transmitter and receiver—it is a static or slowly varying device-specific signature. Doppler shift is an environmental effect caused by motion. A robust fingerprinting system must disentangle the two: CFO is a valuable identifying feature, while Doppler is a nuisance variable to be suppressed.

Doppler compensation is critical in several operational contexts:

• Vehicular networks: V2X communication at highway speeds produces significant Doppler spreads.
• Drone identification: UAVs exhibit rapid velocity changes and rotor-induced micro-Doppler modulation.
• Satellite ground links: LEO satellites have relative velocities exceeding 7 km/s, causing Doppler shifts of hundreds of kHz.
• Railway communications: High-speed trains create predictable but severe Doppler profiles that must be normalized.