Most consumer devices—smartphones, laptops, and smart assistants—are equipped with speakers and microphones designed primarily for human speech. The human auditory system typically tops out around 18 kHz, but standard audio hardware operates at sampling rates of 44.1 kHz or 48 kHz. According to the Nyquist theorem, these systems can physically emit and capture frequencies up to 22 kHz or 24 kHz.
This creates a hidden, software-accessible spectrum: the near-ultrasonic band. By exploiting this frequency gap, researchers have unlocked a dual-use medium capable of silent data transfer, high-precision spatial sensing, and hardware-level security manipulation—all without modifying existing hardware.
The Mechanics of the Inaudible Channel
Why does this channel exist so reliably across different devices? Standard hardware components process signals up to half their sampling rate. As noted by Si Chen et al. in their overview of short-range acoustic communication, modulating signals into the 19.5 kHz to 22 kHz range allows data transmission that remains completely imperceptible to humans while remaining fully compatible with off-the-shelf components. Similarly, Jae Choi et al. utilize the 18–21.5 kHz range to build infrastructure-free tracking systems.
Because these high-frequency sound waves attenuate rapidly in air and cannot penetrate solid walls, they possess an inherent spatial property: they are naturally bounded by the physical room in which they are emitted. Doohwan Song et al. leveraged this room-bounding characteristic to enable smart devices to form spontaneous, localized "ephemeral groups" without complex network configurations or the risk of cross-room interference.
Covert and Secure Data Transfer
Because the channel is silent, it serves as an ideal vector for both secure ad-hoc communication and covert data exfiltration. For secure short-range interactions, Bingsheng Zhang et al. demonstrated in their system PriWhisper that standard smartphones can exchange credentials securely via acoustic channels. They use a "friendly jamming" technique—where the receiver emits masking noise to prevent eavesdropping—eliminating the need for traditional cryptographic key exchanges.
Conversely, this same channel can be weaponized for data leakage from air-gapped systems. Mordechai Guri et al. showed in their SPEAKE(a)R system that software can reverse the electrical pathways of standard headphones and earphones to repurpose them as makeshift microphones. This allows an attacker to capture audio and establish a covert ultrasonic channel above 10 kHz even when dedicated microphones are disabled.
In a subsequent work on the MOSQUITO framework, Mordechai Guri et al. demonstrated speaker-to-speaker communication, transmitting data across air-gapped computers at 18–24 kHz by exploiting this same hardware reversibility. Even if a system is completely speakerless, Mordechai Guri et al. proved in their DiskFiltration study that the mechanical noise of hard disk drive actuator arms can be intentionally modulated via seek operations to exfiltrate binary data acoustically to a nearby receiver.
Acoustic Radar: Sensing the Physical World
Beyond data transfer, an inaudible acoustic signal can act as a continuous-wave radar (sonar). By emitting a constant pilot tone and recording its reflections, a device can sense its environment through the Doppler effect and multipath propagation.
[Device Speaker] ----> (Inaudible 20 kHz Wave) ----> [Moving Hand]
|
[Device Microphone] <--- (Shifted Echo: 20.1 kHz) <--------+
For example, Sidhant Gupta et al. developed SoundWave, which senses mid-air hand gestures by analyzing the Doppler shift of an ultrasonic carrier wave reflected off a user's hand. This active sonar approach can also be used maliciously; Peng Cheng et al. introduced SonarSnoop, an attack that emits inaudible 18–20 kHz waves from a smartphone speaker to track finger movements on a touchscreen, reconstructing unlock patterns from the subtle echoes.
For health and localization, the high resolution of near-ultrasound allows for fine-grained tracking. Tao Gu 0001 et al. designed DF-Sense, a system operating in the 17–19.2 kHz range that monitors the heartbeats of multiple users simultaneously up to 10 meters away. To prevent the audible "clicking" artifacts common in continuous sensing, Tianben Wang et al. developed OmniResMonitor, which utilizes sinusoidal frequency modulation in the ultrasonic spectrum to achieve non-line-of-sight respiration monitoring via multipath reflections off walls and ceilings.
Exploiting Hardware Nonlinearities for Signal Injection
The most surprising aspect of near-ultrasonic engineering is its ability to inject audible commands directly into voice-activation systems. This works because of a physical vulnerability: hardware nonlinearity.
When an ultrasonic wave hits a standard microphone diaphragm, the physical components and pre-amplifiers do not respond perfectly linearly to high-intensity, high-frequency inputs. Instead, the hardware acts as an accidental envelope detector. If an attacker modulates an audible voice command onto an ultrasonic carrier wave, the microphone's inherent physical distortion downmixes (demodulates) the signal, stripping away the high-frequency carrier and leaving behind the audible command in the digital recording.
[Ultrasonic Signal (Carrier + Voice)]
|
v
[Microphone Diaphragm / Amp] <-- Physical Nonlinearity (Acts as Demodulator)
|
v
[Audible Command Captured by System]
Junning Ze et al. exploited this physical nonlinearity to create TUNER, a backdoor attack that poisons a speaker recognition system's voiceprint during the enrollment phase without the user's knowledge. To maximize the efficiency and stealth of these injections, Yan Chen et al. developed VRIFLE, which uses Upper Sideband Amplitude Modulation (USB-AM) to inject real-time adversarial commands into voice assistants from up to 10 meters away while minimizing audible sound leakage.
Conversely, this same nonlinearity can be used for defense. Peng Huang et al. designed InfoMasker, an anti-eavesdropping system that transmits ultrasonic jamming noise using Low-Sideband Amplitude Modulation (LSB-AM) combined with frequency pre-compensation to disrupt unauthorized recordings.
However, high-intensity exposure to these frequencies is not entirely without consequence. As Matt Wixey et al. demonstrated, acoustic emissions between 17 and 21 kHz can exceed safety thresholds and induce physiological distress like headaches or nausea, even while remaining completely unnoticeable to the victim's ears.
Key Foundational Work
SoundWave: using the doppler effect to sense gestures (2012) — contributed 1 concept elements to this direction.
PriWhisper: Enabling Keyless Secure Acoustic Communication for Smartphones (2013) — contributed 1 concept elements to this direction.