Comparing beacon selection and handover methods

The challenge

Selecting a strong beacon and deciding when to switch connections are different decisions. A method must control outages and unnecessary switching as signal quality changes, including under conditions that differ from its training data.

My contribution

• Compared a spectral-power baseline with a small neural network for beacon subchannel selection across synthetic noise and frequency-offset conditions.
• Implemented a Python handover controller with eligibility checks, hysteresis, time-to-trigger and minimum dwell, including recovery when the current link is lost.
• Compared immediate strongest-link selection, classical hysteresis and learned signal-quality prediction on held-out synthetic episodes and a separate set with faster variation and stronger noise.
• Measured outages, switching, short-lived returns to the previous link and forecast errors. Added controller tests for stale observations, invalid candidates and loss of all usable links.
• Defined the observation data, test cases and evaluation criteria for a later comparison on labelled radio measurements.

Results and validation

The hysteresis-based controller reduced outages and repeated switching relative to immediate strongest-link selection in both the held-out and altered-condition synthetic tests. The benchmark compares classical and learned methods using the same acquisition costs, eligibility checks and evaluation splits. Controller tests cover invalid observations, link loss and recovery.

Technical details

The beacon study uses synthetic signal bursts; the handover study uses synthetic signal-quality traces. The reported improvement belongs to the classical hysteresis controller. Beacon decoding, confirmation of a new connection and hardware execution sit outside this offline benchmark.

Python · NumPy · Signal processing · Machine learning · Handover · Simulation · Regression testing