Spectrum: A Multidisciplinary Overview

• Spectrum is a structured description of distributions across indices (e.g., frequency, space, or prime objects) that defines how power, intensity, or statistical likelihood is organized.
• It underpins varied applications, from wireless spectrum sensing and real-time prediction to astrophysical measurement and advanced integrated photonics, each employing distinct formalisms.
• The concept enables actionable insights through calibrated representations, semantic segmentation, and economic frameworks that govern allocation and regulatory compliance.

Across contemporary research, spectrum denotes several formally distinct objects. In wireless systems it is a scarce and congested resource whose occupancy must be sensed, quantified, forecast, allocated, and protected (Uvaydov et al., 2024). In astronomical and photonic instrumentation it is an intensity-, phase-, or likelihood-bearing representation of radiation as a function of wavelength or frequency (Bolton et al., 2011). In cosmology it is the sky-averaged extragalactic specific intensity across the electromagnetic spectrum (Hill et al., 2018). In solar physics it is a velocity amplitude per spherical-harmonic degree (Hathaway et al., 2015). In ordered algebra and real algebraic geometry it is a spectral space built from prime cones, prime $\ell$-ideals, or related prime objects (Wehrung, 2017). Taken together, these uses suggest a common role for spectrum as a structured description of distribution, compatibility, or localization, even though the indexing variable differs radically across fields.

1. Domain-specific meanings

The surveyed literature uses the term in several precise senses [(Bolton et al., 2011); (Hill et al., 2018); (Hathaway et al., 2015); (Fortuna et al., 2018); (Wehrung, 2017)].

Domain	Object called spectrum	Representative formalism
Astronomical spectroscopy	Compressed representation of the likelihood of the underlying source spectrum	$(\mathbf{f}, \mathbf{C}, \mathbf{R})$
Cosmology	Sky-averaged extragalactic specific intensity	$I_\nu$, $\nu I_\nu$
Solar physics	Velocity amplitude per spherical-harmonic degree	$$V(\ell) = \sqrt{\ell \sum_{m=-\ell}^{\ell} |A_\ell^m|^2}$$
Wireless sensing	Streaming measurements, occupancy maps, or event abstractions over time and frequency	$S_i=[(f_1,e_1),\dots,(f_N,e_N)]$, $<E_i,t_{start},t_{stop},f_{start},f_{stop}>$
Ordered algebra	Spectral space associated with prime objects	$\operatorname{Spec}_{\mathrm r}A$, $\operatorname{Spec}_{\ell}G$

A recurring misconception is that spectrum must be a one-dimensional plot of amplitude against frequency. The cited literature rejects that restriction. A spectrum may be a blackbody intensity curve, a multiscale velocity distribution, a compressed likelihood functional, a discretized RF resource over space-time-frequency, or a topological space. What remains common is that the object organizes information by a mathematically specified axis or class of indices.

2. Spectrum as intensity, phase, and likelihood

In modern astronomical spectroscopy, a spectrum is explicitly not “just a 1D extracted flux vector.” The spectro-perfectionism formulation defines a practical spectrum as the triplet $(\mathbf{f}, \mathbf{C}, \mathbf{R})$, where $(\mathbf{f}, \mathbf{C}, \mathbf{R})$0 is the extracted flux-like vector, $(\mathbf{f}, \mathbf{C}, \mathbf{R})$1 is the covariance matrix, and $(\mathbf{f}, \mathbf{C}, \mathbf{R})$2 is the resolution matrix or line-spread-function operator (Bolton et al., 2011). Its defining property is inferential: for any trial model spectrum $(\mathbf{f}, \mathbf{C}, \mathbf{R})$3, the spectrum-domain likelihood

$(\mathbf{f}, \mathbf{C}, \mathbf{R})$4

encodes the same model discrimination power as the raw CCD likelihood

$(\mathbf{f}, \mathbf{C}, \mathbf{R})$5

up to a constant offset. In that view, calibration is likelihood functional determination, extraction is likelihood functional compression, and measurement is likelihood functional projection. This makes spectrum a statistical object as much as a radiometric one.

Integrated photonics extends the term further by demanding simultaneous access to spectral amplitude and phase. A fiber-based vector spectrum analyzer can characterize passive devices and active laser sources in one setup, measuring loss, phase response, and dispersion properties of passive devices while also coherently mapping a broadband laser spectrum into the RF domain (Luo et al., 2023). The reported performance is a bandwidth of $(\mathbf{f}, \mathbf{C}, \mathbf{R})$6 from $(\mathbf{f}, \mathbf{C}, \mathbf{R})$7 to $(\mathbf{f}, \mathbf{C}, \mathbf{R})$8, frequency resolution of $(\mathbf{f}, \mathbf{C}, \mathbf{R})$9, and dynamic range of $I_\nu$0. The same instrument is used for integrated dispersive waveguides, frequency-comb spectra, and coherent LiDAR. Here spectrum is intrinsically complex-valued: intensity alone is insufficient because phase and dispersion are part of the measurable object.

These formulations make clear that, in spectroscopy, spectrum is often best understood as a calibrated representation of what models remain compatible with data. A plausible implication is that “flux versus wavelength” is only the simplest visible projection of a richer statistical or vector-valued structure.

3. Spectra of radiation and scale in astrophysics

In observational cosmology, “the spectrum of the Universe” is the spectrum of the cosmic background, defined as the summed electromagnetic radiation from all sources outside the Milky Way after subtracting atmospheric, Solar-System, and Galactic foregrounds (Hill et al., 2018). The principal quantity is the specific intensity $I_\nu$1, often displayed as $I_\nu$2. The paper gathers measurements from radio to $I_\nu$3-rays and identifies the dominant CMB blackbody at

$I_\nu$4

together with the infrared, optical, X-ray, and $I_\nu$5-ray backgrounds. The same work also treats angular structure through the spherical-harmonic expansion

$I_\nu$6

so that the monopole spectrum and the spectra of multipole moments become distinct but related objects.

Solar physics uses spectrum in a different but equally formal way. The photospheric convection spectrum is a velocity amplitude per spherical-harmonic degree $I_\nu$7, inferred from full-disk HMI/SDO Doppler observations and simulation-constrained decomposition into radial, poloidal, and toroidal components (Hathaway et al., 2015). The reported total velocity spectrum rises to a first peak at $I_\nu$8, characteristic of supergranules, and then to a second peak at $I_\nu$9, characteristic of granules. The toroidal component dominates below $\nu I_\nu$0, the poloidal component dominates above about $\nu I_\nu$1, and the radial component grows from about $\nu I_\nu$2 of the total flow velocity at the lowest wavenumbers to about $\nu I_\nu$3 near $\nu I_\nu$4. In this setting, spectrum does not refer to electromagnetic frequency at all; it refers to distribution across spatial scales on a sphere.

These two examples show that the physical meaning of spectrum is broader than frequency decomposition. It may instead quantify how power, intensity, or velocity is distributed across a hierarchy of scales, harmonics, or photon energies.

4. Spectrum as a scarce wireless resource

Wireless research treats spectrum first as a regulated and contested resource. One line of work states that spectrum has become an “extremely scarce and congested resource,” motivating real-time awareness of who is using what part of a wide RF band (Uvaydov et al., 2024). Another emphasizes that effective spectrum analysis supports regulatory compliance, interference detection and mitigation, and spectrum resource planning and optimization, while multi-site sensing systems may generate between $\nu I_\nu$5 and $\nu I_\nu$6 of raw data per day per site (Gandotra et al., 16 Jan 2026). In radio astronomy the resource interpretation is even sharper: below $\nu I_\nu$7 only $\nu I_\nu$8 of the spectrum is reserved for the Radio Astronomy Service, and only $\nu I_\nu$9 on a primary basis, while IMT identifications cover about one third of the spectrum below $$V(\ell) = \sqrt{\ell \sum_{m=-\ell}^{\ell} |A_\ell^m|^2}$$0 (Winkel et al., 28 Feb 2025).

A second line of work formalizes spectrum as a quantifiable space over power, geography, time, and frequency. MUSE characterizes the use of spectrum by a transmitter at a point in terms of the RF power occupied by the transmitter, and by a receiver at a point in terms of the constraints on RF power that can be occupied by any transmitter in order to ensure successful reception (Khambekar et al., 2015). A closely related discretized-spectrum-space framework defines total spectrum space, utilized spectrum, forbidden spectrum, and available spectrum, with the conservation relation

$$V(\ell) = \sqrt{\ell \sum_{m=-\ell}^{\ell} |A_\ell^m|^2}$$1

over unit regions, unit time quanta, and unit frequency bands (Khambekar et al., 2014). This replaces a binary occupied/unoccupied view with explicit accounting of transmitter occupancy, receiver liability, and residual opportunity.

A third line treats spectrum as an allocable economic good. Licensed spectrum sharing has been modeled as a two-sided matching problem between spectrum providers and spectrum users, with deferred acceptance used to establish stable, preference-aware pairings (Butt et al., 2018). Revenue-maximizing dynamic spectrum access has been formulated through Bayesian mechanism design, with allocations driven by virtual valuations in both frequency-division and spread-spectrum settings (Kakhbod et al., 2011). Secure multi-tier auctions add another layer: MTSSA allocates under-utilized spectrum to commercial wireless service providers while leveraging the Paillier cryptosystem to avoid possible fraud and bid-rigging (Abdelhadi et al., 2015). In this literature, spectrum is not merely measured; it is priced, leased, matched, and protected.

5. Spectrum as sensing, segmentation, and prediction data

Streaming-spectrum analytics begins from low-level measurements. One formulation defines each incoming spectrum sample at sensing instant $$V(\ell) = \sqrt{\ell \sum_{m=-\ell}^{\ell} |A_\ell^m|^2}$$2 as

$$V(\ell) = \sqrt{\ell \sum_{m=-\ell}^{\ell} |A_\ell^m|^2}$$3

with event extraction producing tuples

$$V(\ell) = \sqrt{\ell \sum_{m=-\ell}^{\ell} |A_\ell^m|^2}$$4

through per-bin recent-versus-historical histogram comparison and chi-squared testing (Fortuna et al., 2018). The key shift is from raw PSD vectors to actionable “wireless spectrum events” suitable for real-time notification, reporting, and indexed querying.

Deep learning pushes this further by treating wideband sensing as semantic segmentation rather than object detection. Semantic spectrum segmentation directly consumes frequency-domain complex samples represented by real and imaginary parts in an input tensor of size $$V(\ell) = \sqrt{\ell \sum_{m=-\ell}^{\ell} |A_\ell^m|^2}$$5, and outputs a $$V(\ell) = \sqrt{\ell \sum_{m=-\ell}^{\ell} |A_\ell^m|^2}$$6 multi-label occupancy map over protocol classes and frequency bins (Uvaydov et al., 2024). The same work introduces wideband signal stitching: isolated over-the-air captures are stitched together in frequency to create large-scale labeled data while retaining realistic channel effects. Across five wireless protocols, the reported mean Intersection over Union is $$V(\ell) = \sqrt{\ell \sum_{m=-\ell}^{\ell} |A_\ell^m|^2}$$7, and latency is $$V(\ell) = \sqrt{\ell \sum_{m=-\ell}^{\ell} |A_\ell^m|^2}$$8; the non-local block improves segmentation accuracy on the most challenging signals by about $$V(\ell) = \sqrt{\ell \sum_{m=-\ell}^{\ell} |A_\ell^m|^2}$$9 percentage points relative to vanilla U-Net.

Prediction extends spectrum analysis from present occupancy to short-horizon futures. Using $S_i=[(f_1,e_1),\dots,(f_N,e_N)]$0 days of real-world mid-band measurements aggregated into $S_i=[(f_1,e_1),\dots,(f_N,e_N)]$1 frequency bins at one-minute resolution, next-minute occupancy prediction has been formulated as

$S_i=[(f_1,e_1),\dots,(f_N,e_N)]$2

with Random Forest, XGBoost, and LSTM outperforming a first-order Markov baseline, especially under fixed false-alarm constraints (Mao et al., 16 Jan 2026). A more recent line recasts RF forecasting as autoregressive token prediction. Large Spectrum Models tokenize PSD values together with gain, frequency, FFT bin, and timestamp information, and train decoder-only transformers on over $S_i=[(f_1,e_1),\dots,(f_N,e_N)]$3 of raw spectrum data across $S_i=[(f_1,e_1),\dots,(f_N,e_N)]$4 sub-GHz bands, producing $S_i=[(f_1,e_1),\dots,(f_N,e_N)]$5 billion tokens; the best reported model achieves RMSE $S_i=[(f_1,e_1),\dots,(f_N,e_N)]$6, and $S_i=[(f_1,e_1),\dots,(f_N,e_N)]$7 of predictions have mean absolute error below $S_i=[(f_1,e_1),\dots,(f_N,e_N)]$8 (Lunar et al., 11 May 2026). This suggests a further semantic broadening of spectrum: from physical measurement, to labeled scene, to token sequence.

6. Spectrum in ordered algebra and real algebraic geometry

In ordered algebra and real algebraic geometry, spectrum is a topological construction rather than a physical distribution. The real spectrum $S_i=[(f_1,e_1),\dots,(f_N,e_N)]$9 of a commutative unital ring $<E_i,t_{start},t_{stop},f_{start},f_{stop}>$0 is the set of prime cones of $<E_i,t_{start},t_{stop},f_{start},f_{stop}>$1; the $<E_i,t_{start},t_{stop},f_{start},f_{stop}>$2-spectrum $<E_i,t_{start},t_{stop},f_{start},f_{stop}>$3 of an Abelian lattice-ordered group $<E_i,t_{start},t_{stop},f_{start},f_{stop}>$4 is the set of prime $<E_i,t_{start},t_{stop},f_{start},f_{stop}>$5-ideals; the Brumfiel spectrum $<E_i,t_{start},t_{stop},f_{start},f_{stop}>$6 of a commutative $<E_i,t_{start},t_{stop},f_{start},f_{stop}>$7-ring is built from prime $<E_i,t_{start},t_{stop},f_{start},f_{stop}>$8-ideals that are also prime ring ideals (Wehrung, 2017). These are studied inside the class of generalized spectral spaces, which are sober spaces whose compact open subsets form a basis and are closed under finite intersections. A spectral space is a compact generalized spectral space, and a spectral subspace is one whose inclusion map is spectral.

The comparison results are sharp. Every real spectrum can be embedded, as a spectral subspace, into some $<E_i,t_{start},t_{stop},f_{start},f_{stop}>$9-spectrum. Not every real spectrum is an $\operatorname{Spec}_{\mathrm r}A$0-spectrum. A spectral subspace of a real spectrum may not be a real spectrum. Not every $\operatorname{Spec}_{\mathrm r}A$1-spectrum can be embedded, as a spectral subspace, into a real spectrum. There also exists a completely normal spectral space which cannot be embedded, as a spectral subspace, into any $\operatorname{Spec}_{\mathrm r}A$2-spectrum (Wehrung, 2017). Here the term spectrum has no frequency axis at all. Its content is categorical and topological: it identifies a space of prime-like objects together with a compact-open lattice and specialization order.

This mathematical usage is a reminder that spectrum is not inherently tied to waves or signals. The word instead marks a recurrent construction in which structure is made visible by passing to an indexed space of local or prime constituents. Across the literatures considered here, that construction appears in measurably different forms, but it retains a common aspiration: to replace a coarse global description by a systematically organized field of local distinctions.