SoK: Acoustic Side Channels

Ping Wang, Shishir Nagaraja, Aurélien Bourquard, Haichang Gao, Jeff YanarXiv 2023 · DOI · arXiv

This Systematization of Knowledge (SoK) paper provides a comprehensive look at the world of Acoustic Side Channels (ASCs)—a field where the subtle sounds of technology, from the click of a keyboard to the hum of a 3D printer, are used to steal sensitive data. The authors move beyond just listing attacks; they provide a rigorous framework to define what an ASC actually is and how it differs from related security threats.

Clearing the Conceptual Fog

For years, the literature on acoustic security has been described as "confusing and sometimes chaotic." Researchers often used the term "side channel" loosely, sometimes conflating it with covert channels (where two parties intentionally hide a message in audio) or signal injection (where an attacker sends sounds to manipulate a device, like "DolphinAttack" on voice assistants).

The authors address this lack of clarity by providing formal definitions and a structured taxonomy. As shown in Figure 4, they categorize ASC attacks by their leaking source, the type of information leaked, and the nature of the attack (active vs. passive, offensive vs. defensive).

The proposed taxonomy of Acoustic Side Channel (ASC) attacks, illustrating the relationships between leaking sources, information types, and attack characteristics.

Methods and the "Inverse Problem" Bridge

The paper contributes several novel analytical frameworks to the field:

  • Formal Definition: They define an ASC as a functional mapping $S : I \times M \to O$, where an attacker tries to infer the input ($I$) from observations ($O$) given some adversarial influence ($M$).
  • Active vs. Passive: They distinguish between Passive ASCs (PASCA), where the attacker just listens, and Active ASCs (AASCA), where the attacker actively probes the environment (e.g., using ultrasound to "sonar" a user's finger movements).
  • The Inverse Problem Bridge: The authors make a novel attempt to bridge computer security with the mathematical field of inverse problems. They argue that since every side-channel attack is essentially an attempt to infer a cause from an observed effect, security researchers can use established mathematical concepts like "stability" and "uniqueness" to evaluate how much information a channel can actually leak. This allows defenders to focus on making the problem "ill-posed"—for example, by adding noise so that a tiny change in sound leads to a massive error for the attacker.

The Landscape of Acoustic Leakage

The systematic review highlights that acoustic leakage is far more pervasive than just recording keyboard clicks. Modern "Active" attacks turn smartphones into sonar systems, while the "hum" from LCD monitors, 3D printer motors, and even cooling fans can leak proprietary IP or cryptographic keys. Notably, the VoIP "Hitchhiker" effect allows CPU computation sounds to be captured by a laptop's own microphone and transmitted over standard calls (e.g., Zoom or Skype), enabling remote cryptanalysis.

A comprehensive summary of acoustic side channel research, detailing categories, sources, leaked information, and attack parameters across various studies.

As summarized in Figure 5 (Table 1 in the paper), the study evaluates quantitative performance across various categories:

  • Keyboard Recovery: Modern machine learning approaches can recover keystrokes with up to 94% accuracy.
  • 3D Printing: Attackers can reconstruct G-code (the "blueprints" of a 3D-printed object) from motor sounds with 78% to 100% accuracy.
  • Device Fingerprinting: Variations in microphone manufacturing allow attackers to identify specific devices with 98% accuracy based on their acoustic signatures.

Limitations and Future Work

A key finding is a significant "imbalance": while researchers are incredibly creative at finding new ways to eavesdrop, research into countermeasures is often "crude," theoretical, and lacks empirical validation. Many proposed defenses (like adding loud masking noise) are impractical because they annoy the user.

The authors call for more research into "naturalistic" settings where background noise and distance might render lab-perfect attacks useless. They urge the community to adopt common metrics and shared experimental protocols to move the field from "ad-hoc" discoveries to a rigorous scientific discipline.

Summary updated

BibTeX
@misc{wang2023sok,
  author = {Ping Wang and Shishir Nagaraja and Aurélien Bourquard and Haichang Gao and Jeff Yan},
  title = {{SoK: Acoustic Side Channels}},
  year = {2023},
  eprint = {2308.03806},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2308.03806},
}
Figure