Acoustic Data Exfiltration via Fan Speed Modulation: A Replication Study

Even air-gapped systems can be vulnerable to covert acoustic techniques that exploit physical features of the machine. This paper replicates key elements of the Fansmitter technique on a commonly available Windows desktop. We modulate a single controllable GPU fan between two fixed speed settings, which represent a binary 0 and 1. We recorded the resulting acoustic noise using the “phyphox” app, varying distances between 20 and 50 cm. Then they were analyzed using a Python pipeline, which estimated transmitted bit patterns and computed both binary and confidence-weighted (“soft”) error metrics. Within each window a “closeness” score is computed as (1 − vote ratio), where vote ratio is the expected bit. Results show that fan-speed modulation can produce measurable, decodable structure at close range, but decoding degrades substantially with distance and under realistic background noise. We conclude that the channel is inconsistent beyond short distances in our setup however covert channels can remain partially intelligible.