Acoustic Demon Mode: Cross-Domain Mechanisms

• Acoustic demon mode is a cross-domain concept defined by out-of-phase dynamics that yield weak net charge and indirect cues in both sound and electronic systems.
• In underwater acoustics, DEMON techniques extract envelope modulation to reveal propeller and shaft periodicities, while in condensed matter the mode identifies neutral collective excitations.
• Applications span expert networks like DEMONet, symmetry-filtered mode conversion in sonic crystals, and active metamaterials that employ feedback for selective routing.

Searching arXiv for papers on acoustic demon modes and related usages of “DEMON”. “Acoustic demon mode” does not designate a single universally standardized object across the current literature. In underwater acoustics, DEMON means Detection of Envelope Modulation on Noise, a classical analysis of ship-radiated noise whose spectral peaks reflect shaft and blade periodicities and whose reconstructed spectra are used in DEMONet as implicit physical cues for expert routing rather than direct class prediction (Xie et al., 2024). In condensed-matter physics, a demon is an acoustic, electrically neutral collective excitation generated by out-of-phase motion of multiple electronic components, as in the multiband metal Sr$_2$RuO$_4$ and in recent theories of $d$-wave altermagnets (Husain et al., 2020, Rostami et al., 11 May 2025). A plausible implication is that the phrase functions less as a single canonical term than as a cross-domain label for hidden, weakly charged, or selectively routed modes whose observability depends on symmetry, coupling structure, or indirect control.

1. Terminological scope and principal usages

The available literature supports several distinct technical usages of “demon” or “DEMON” that can be grouped under an editorial umbrella of acoustic demon phenomena.

Domain	Sense of “demon” or DEMON	Representative source
Underwater acoustics	Detection of Envelope Modulation on Noise; physics-informed modulation descriptor	(Xie et al., 2024)
Multiband electron systems	Acoustic, gapless, electrically neutral collective mode from out-of-phase motion	(Husain et al., 2020)
Altermagnets	Acoustic spin-demon or demon mode with out-of-phase spin dynamics	(Rostami et al., 11 May 2025, Sarwar et al., 25 Jun 2026)
Classical acoustic devices	Direction-selective mode conversion or demon-like mode routing by symmetry or feedback	(Ouyang et al., 2016, Wong et al., 21 Jun 2025)

Two distinctions are essential. First, DEMON in underwater acoustics is an analysis method and feature representation, not an electronic collective mode. Second, in condensed matter the word demon denotes a collective excitation that is acoustic because its frequency tends to zero as momentum tends to zero. The overlap in terminology is therefore real but not univocal.

2. Underwater-acoustic DEMON analysis and DEMONet

In underwater acoustics, DEMON denotes Detection of Envelope Modulation on Noise, a technique for extracting periodic amplitude-modulation structure from ship-radiated noise. The physical premise is that propeller and shaft rotation modulate broadband noise, so after sub-band filtering, demodulation, and Fourier analysis of the envelope, one obtains a DEMON spectrum whose peaks reveal physically meaningful periodicities. The paper states explicitly that the 1-D DEMON spectrum peaks reflect the shaft frequency and blade frequency of the propeller (fundamental frequency and its harmonics) (Xie et al., 2024).

The extraction model is written as

$x(t) = A(1+m \sin\Omega t)\cdot \cos\omega t.$

In the reported procedure, the signal is split into sub-bands at 250 Hz intervals, demodulated using absolute-value low-pass demodulation, and transformed into a 2-D DEMON spectrum; summation across the modulation-frequency axis yields a 1-D DEMON spectrum. After absolute-value demodulation, low-pass filtering isolates the modulation component because $\omega \gg \Omega$. In this sense, DEMON is a physics-informed descriptor of target mechanics and motion.

The same paper emphasizes a central limitation: DEMON spectra are robust yet often not sufficiently class-discriminative. A three-blade propeller can be distinguished from a five-blade propeller, but DEMON may fail to separate a passenger ship from a cargo ship if both share the same blade count. The authors report that direct feature fusion with DEMON features actually degraded performance on DeepShip, and interpret this as evidence that DEMON is best used as an implicit physical cue, not as a direct semantic predictor.

That observation motivates DEMONet, a multi-expert network composed of a cross-temporal VAE, a routing layer, multiple expert layers, and a backbone network. Its logic is “clustering before classification”: reconstructed DEMON spectra infer physical similarity, inputs are dispatched to the most appropriate expert, and the backbone performs recognition. For each sample $s_i$, only the expert with maximal routing probability is activated,

$$z_i = F(\mathrm{Expert}_{I}(s_i)), \qquad I=\underset{j}{\arg\max}\, p_i[j].$$

The experts share architecture but not parameters, enabling specialization on subsets of signals with similar physical characteristics.

The denoising mechanism is a cross-temporal alignment strategy implemented with a variational autoencoder. Two 30-second segments $wav_A$ and $wav_B$ are sampled from the same recording; $d_A$ is the input and $_4$0 is the reconstruction target. Latent sampling uses

$_4$1

and the objective is

$_4$2

This cross-temporal reconstruction is designed to preserve time-invariant physical structure while suppressing noise and spurious modulation lines. In stage 2, the VAE is frozen and the reconstructed DEMON spectrum is used for routing. The paper reports that reconstructed spectra are visually cleaner and more consistent across time, and that cosine similarity between periods increases after reconstruction.

Training further incorporates load balancing loss to prevent expert collapse: $_4$3 with $_4$4 by default. Experimentally, Cross-Temporal VAE + DEMONet with 5 experts reaches $_4$5 on DeepShip and $_4$6 on DTIL, while on ShipsEar DEMONet improves over a ResNet-18 baseline but does not always exceed the best competing method. The authors therefore describe the method as data-dependent: it works best when there is enough data and sufficiently clear modulation structure for routing and expert specialization to emerge.

3. Acoustic demons in multiband metals

In condensed-matter physics, an acoustic demon mode is a collective electronic excitation in which different bands move out of phase, so their charge oscillations nearly cancel in the long-wavelength limit. Because the total density modulation is nearly absent, the mode is electrically neutral, does not couple to light in the $_4$7 limit, and is acoustic rather than plasmonic in the ordinary 3D sense (Husain et al., 2020).

The Sr$_4$8RuO$_4$9 study identifies this mode using momentum-resolved electron energy-loss spectroscopy (M-EELS) and RPA calculations. The system contains three Fermi-surface sheets, $d$0, $d$1, and $d$2, and the demon is primarily a collective oscillation of the $d$3 and $d$4 bands. The Coulomb interaction is modeled as

$d$5

with $d$6. In the RPA calculation, the demon appears as a linear branch in $d$7, with group velocity

$d$8

Experimentally, the measured mode is essentially gapless within resolution, with the gap at $d$9 bounded by less than 8 meV. The measured room-temperature velocity is

$x(t) = A(1+m \sin\Omega t)\cdot \cos\omega t.$0

and the mode persists up to a critical momentum

$x(t) = A(1+m \sin\Omega t)\cdot \cos\omega t.$1

The same work reports anisotropy and temperature dependence: at 30 K the velocity becomes $x(t) = A(1+m \sin\Omega t)\cdot \cos\omega t.$2, whereas along the $x(t) = A(1+m \sin\Omega t)\cdot \cos\omega t.$3 direction it becomes $x(t) = A(1+m \sin\Omega t)\cdot \cos\omega t.$4.

The distinction from an ordinary plasmon is explicit. The high-energy plasmon lies near $x(t) = A(1+m \sin\Omega t)\cdot \cos\omega t.$5 eV in RPA and $x(t) = A(1+m \sin\Omega t)\cdot \cos\omega t.$6 eV in M-EELS, whereas the demon is a separate low-energy branch. Partial susceptibilities $x(t) = A(1+m \sin\Omega t)\cdot \cos\omega t.$7 provide the decisive fingerprint: for the ordinary plasmon, diagonal and off-diagonal terms have the same sign, indicating in-phase motion; for the acoustic mode, $x(t) = A(1+m \sin\Omega t)\cdot \cos\omega t.$8 has the opposite sign from the diagonal terms, meaning the $x(t) = A(1+m \sin\Omega t)\cdot \cos\omega t.$9 and $\omega \gg \Omega$0 bands oscillate $\omega \gg \Omega$1 out of phase.

The neutral character is further supported by sum-rule analysis. For a charged gapless mode in surface M-EELS, the low-energy reference scaling is

$\omega \gg \Omega$2

but the measured intensity behaves approximately as

$\omega \gg \Omega$3

Because this diverges much more weakly than $\omega \gg \Omega$4, the mode carries less low-$\omega \gg \Omega$5 spectral weight than a charged plasmon should, which the paper treats as strong evidence for a neutral demon rather than a low-energy ordinary plasmon.

4. Demon modes and spin demons in $\omega \gg \Omega$6-wave altermagnets

Recent theory extends demon-mode physics to $\omega \gg \Omega$7-wave altermagnets, where two spin-split Fermi surfaces coexist with zero net magnetization. In this setting, the literature predicts both a conventional plasmon, with the two spin components oscillating in phase, and a demon mode, in which they oscillate out of phase (Rostami et al., 11 May 2025).

The Fermi-liquid formulation uses the Landau-Silin kinetic equation for $\omega \gg \Omega$8 and generalized Landau parameters $\omega \gg \Omega$9. For the lowest $s_i$0-wave components,

$s_i$1

with

$s_i$2

The demon dispersion is given as

$s_i$3

which is acoustic because $s_i$4. Its robustness is controlled especially by $s_i$5: larger $s_i$6 raises the demon velocity, increases separation from the particle-hole continuum, and sharpens the spectral peak.

A distinctive result is the strong dependence on propagation direction. In the dynamical structure factor $s_i$7, the demon can be a hidden state with zero spectral weight for propagation along the nodal diagonal,

$s_i$8

a weakly damped propagating demon near that direction, or a Fano-demon mixed state near the nematic axis, where the mode hybridizes strongly with the particle-hole continuum and acquires an asymmetric line shape. The terminology “Fano-demon mixed state” refers precisely to this interference between a collective excitation and a continuum.

A later study considers a related spin demon in a two-dimensional $s_i$9-wave altermagnet and shows how it can be made charge-visible by combining Rashba spin-orbit coupling and electrostatic gate screening (Sarwar et al., 25 Jun 2026). In the spin-conserving limit,

$$z_i = F(\mathrm{Expert}_{I}(s_i)), \qquad I=\underset{j}{\arg\max}\, p_i[j].$$0

with

$$z_i = F(\mathrm{Expert}_{I}(s_i)), \qquad I=\underset{j}{\arg\max}\, p_i[j].$$1

The spin-up and spin-down Fermi contours are ellipses rotated by $$z_i = F(\mathrm{Expert}_{I}(s_i)), \qquad I=\underset{j}{\arg\max}\, p_i[j].$$2, and the clean long-wavelength demon dispersion is

$$z_i = F(\mathrm{Expert}_{I}(s_i)), \qquad I=\underset{j}{\arg\max}\, p_i[j].$$3

or, equivalently,

$$z_i = F(\mathrm{Expert}_{I}(s_i)), \qquad I=\underset{j}{\arg\max}\, p_i[j].$$4

Rashba coupling,

$$z_i = F(\mathrm{Expert}_{I}(s_i)), \qquad I=\underset{j}{\arg\max}\, p_i[j].$$5

mixes charge and spin channels, making the mixed susceptibilities $$z_i = F(\mathrm{Expert}_{I}(s_i)), \qquad I=\underset{j}{\arg\max}\, p_i[j].$$6 and $$z_i = F(\mathrm{Expert}_{I}(s_i)), \qquad I=\underset{j}{\arg\max}\, p_i[j].$$7 finite. The RPA spin response therefore acquires the same collective denominator as the charge response,

$$z_i = F(\mathrm{Expert}_{I}(s_i)), \qquad I=\underset{j}{\arg\max}\, p_i[j].$$8

The mode remains predominantly spin-like, but it becomes accessible to charge-sensitive probes. Charge visibility is quantified by

$$z_i = F(\mathrm{Expert}_{I}(s_i)), \qquad I=\underset{j}{\arg\max}\, p_i[j].$$9

At $wav_A$0, the paper reports $wav_A$1 in the clean case and $wav_A$2 for $wav_A$3 and $wav_A$4. Gate screening modifies the Coulomb interaction through

$wav_A$5

introducing a stated trade-off between charge visibility and mode quality factor $wav_A$6. The paper’s conclusion is therefore not that the demon becomes an ordinary plasmon, but that a charge-dark spin demon can be brightened while retaining dominant $wav_A$7 character.

5. Classical-acoustic analogues: symmetry filtering and active mode selection

In classical acoustics, the strongest analogues to an acoustic demon mode arise where a structure selectively permits, converts, or amplifies one mode while suppressing another. One example is a sonic-crystal waveguide junction that achieves one-way acoustic propagation and even-to-odd mode conversion in two mutually perpendicular waveguides connected by an asymmetric resonant cavity (Ouyang et al., 2016).

The device is a 2D square-lattice sonic crystal of water cylinders in mercury with a complete band gap from approximately

$wav_A$8

The cavity consists of one elliptical water cylinder with major axis $wav_A$9 and minor axis $wav_B$0, surrounded by four shifted cylinders displaced by $wav_B$1. In the frequency interval

$wav_B$2

the waveguide supports coexisting even and odd guided modes. A cavity resonance at

$wav_B$3

is symmetry-matched to the entrance even mode and reradiates into the perpendicular exit guide as an odd mode, whereas the reverse process is blocked by symmetry mismatch. Transmission spectra show entrance-incidence peaks at $wav_B$4 and $wav_B$5, but exit incidence only at $wav_B$6. The same explanatory account notes that this is not strict nonreciprocity in the magneto-optic or nonlinear sense; it is direction-dependent mode accessibility produced by parity selection rules.

A second analogue is an active-metamaterial scheme for eigenmode-guided amplification in coupled Helmholtz resonators, described in the supplied account as a strong example of what one could call an “Acoustic Demon Mode”, while also noting that the phrase is not used in the paper itself (Wong et al., 21 Jun 2025). The dynamical model is a nonlinear, non-Hermitian coupled-mode system with cross-site gain and loss,

$wav_B$7

where the gain in one cavity depends on the amplitude in the adjacent cavity, the loss in the other depends on the local amplitude, and the total energy

$wav_B$8

remains constant. The paper interprets this as a non-Hermitian yet energy-conserving redistribution of modal energy.

The effective linearized picture predicts collapse toward the eigenstate with the largest imaginary part, which functions as a fixed-point attractor. In the dimer, exceptional points occur at

$wav_B$9

separating a PT-symmetric regime with Rabi-like oscillations from a PT-broken regime with convergent mode selection. Time-modulated gain-loss profiles,

$d_A$0

permit programmable switching between attractors, and trimer sequences such as

$d_A$1

produce sequential routing. Full-wave simulations report convergence to the predicted fixed point within about 5 ms for constant $d_A$2. A plausible interpretation is that this active framework realizes a demon-like selector not by passive symmetry alone, but by state-dependent feedback that amplifies one eigenmode while suppressing others.

6. Conceptual distinctions, misconceptions, and broader significance

Several misconceptions recur when these literatures are read together. The first is to equate all demon modes with sound waves in matter. In Sr$d_A$3RuO$d_A$4 and in altermagnets, the demon is an electronic collective excitation, not a phonon. It is called acoustic because its dispersion is gapless or approximately linear at small $d_A$5, not because it is a lattice vibration.

The second is to treat underwater DEMON features as direct semantic labels. The DEMONet study argues the opposite: DEMON spectra expose stable mechanical periodicities, but do not naturally map to semantic class labels, and forcing that mapping can introduce harmful inductive bias. In that context, the demon-related quantity is a descriptor for routing and specialization, not a stand-alone class predictor.

The third is to conflate one-way acoustic devices with genuine nonreciprocity. The sonic-crystal junction is explicitly framed as one-way mode conversion rather than a fundamental violation of reciprocity. Its asymmetry is produced by symmetry matching and mismatch between guided modes and cavity resonances.

A fourth distinction concerns terminological drift. A separate acronymic usage appears in DEMON: Diffusion Engine for Musical Orchestrated Noise, which denotes a real-time diffusion engine for music rather than a physical mode. That system is built on ACE-Step 1.5 and StreamDiffusion’s ring-buffer architecture, uses TensorRT acceleration, sustains 12.3 decoder completions per second for 60-second music on an RTX 5090, and introduces per-slot heterogeneous denoise scheduling, shared mutable per-step state, per-frame source blending, and windowed VAE decode with an 8.0x speedup (Fosdick, 27 May 2026). This usage is orthogonal to demon-mode physics, but it underscores that “DEMON” now spans signal analysis, collective-mode theory, and controllable generative systems.

Taken together, these works indicate a persistent structural theme. Whether the subject is ship-radiated noise, multiband electron fluids, altermagnetic spin-charge dynamics, sonic-crystal parity filtering, or active acoustic metamaterials, demon-related constructions repeatedly concern hidden structure revealed indirectly: out-of-phase motion with weak net charge, modulation spectra that encode mechanics without semantics, or feedback architectures that privilege one mode while suppressing competing channels. This suggests that “acoustic demon mode” is best understood not as a single settled term, but as a family of technically distinct mechanisms linked by selective visibility, indirect control, and mode discrimination.