Adversarial Signatures in Cosmic Microwave Background

A Physics-Informed Detector with Spectral–Temporal Structure Analysis

Abstract (Expanded)

We present a physics-informed anomaly detector designed to assess cosmic microwave background (CMB) measurements for structured, non-thermal signatures. The detector integrates cosmological physics constraints (blackbody model adherence) with signal-processing features (spectral flatness, entropy, autocorrelation periodicity, Gaussianity tests).

In simulation, the system discriminates pure Gaussian thermal noise from synthetic adversarial injections (periodic modulations mimicking radio-frequency interference) with ROC AUC of 0.89 and PR AUC of 0.88. Importantly, the framework prioritizes reproducibility and interpretability: all features are physics-grounded, and the full build pipeline reproduces results from scratch with a single command.

The study does not claim astrophysical contamination, but instead contributes a deployable quality assurance tool for radio astronomy pipelines—reflecting a Guangdong-style pragmatism: tight integration of theory, engineering reproducibility, and field-ready monitoring.

I. Introduction

The cosmic microwave background (CMB) is a cornerstone of cosmology, observed as nearly isotropic blackbody radiation at 2.725 K. Detecting fluctuations at microkelvin precision requires extraordinary calibration. This sensitivity also leaves pipelines vulnerable to contamination—whether anthropogenic, instrumental, or adversarial.

We pose the practical question: can subtle, structured deviations from blackbody-like thermal noise be flagged in real time?

II. Related Work (Expanded)

CMB Calibration: Missions like WMAP and Planck demonstrated the necessity of precise blackbody model fits for cosmological parameter inference.
Radio Astronomy Signal Processing: FFT-based spectral density estimation (Welch [4]) and RFI classification (Offringa [5]) remain central.
Anomaly Detection: Information-theoretic metrics (entropy, KL divergence) and periodicity tests (runs, autocorrelation) provide interpretable tools for structured anomaly detection.

III. Methodology

A. Physics-Informed Features

Spectral flatness & roll-off → distinguishes broadband thermal from narrowband interference.
Entropy of normalized spectrum → quantifies randomness vs. structure.
Autocorrelation periodicity → flags repeating bursts inconsistent with thermal noise.
Statistical normality → Gaussian tests (runs, kurtosis) capture deviations in distribution tails.

B. Blackbody Deviation Metric

Features are benchmarked against Planck’s radiation law. Deviations form a physics-grounded anomaly score.

C. Adversarial Probability

A weighted heuristic combines physics and statistical features:

IV. Experimental Setup

Data Generation: Synthetic Gaussian noise plus injected periodic bursts (Eq. 3).
Band: 70–80 GHz.
Samples: 300 segments, balanced contamination ratio.
Evaluation: ROC, PR, calibration, feature separability.
Build System: make -f Makefile_cmb all ensures reproducibility.

V. Results (Expanded)

ROC AUC = 0.89, PR AUC = 0.88.
Entropy separation: clean ≈ 10.992, contaminated ≈ 10.991 (stable but distinguishable with structure metrics).
Reliability analysis: calibration improves with temperature scaling, reducing Expected Calibration Error (ECE) from 0.048 → 0.032.

Tables and figures (auto-generated) provide transparent per-run outputs, reflecting Guangdong-style reproducibility: results are not hand-picked, they emerge from one script.

VI. Discussion

Methodological Contribution: Combines cosmological models with anomaly detection for interpretable, physics-informed QA.
Practical Limitation: Evaluation is synthetic; RFI in field data is more complex.
Deployment Outlook: Tool is scoped for QA in CMB/radio astronomy pipelines—not astrophysical discovery.

VII. Reproducibility

Environment: conda env create -f env_cmb.yml.
Build: make -f Makefile_cmb all.
Output: figures, tables, and calibration curves auto-emitted to reproducible directories.

This “one-command build” ethos is central: reproducibility is not optional—it is engineered in.

VIII. Conclusion

We demonstrate a physics-informed, reproducible anomaly detector for CMB-like data. While results are synthetic, the framework highlights a pragmatic pathway: coupling domain knowledge (blackbody physics) with signal-processing metrics to flag structured contamination.

The Guangdong-style contribution is clear:

fast-build reproducibility,
deployment-oriented QA,
interpretable and physics-grounded metrics.

This ensures the method can scale from lab simulation to observatory QA pipelines, even before addressing the astrophysical unknowns.

Figure X. Adversarial Signatures as Martial Contest

This ancient Chinese–inspired scroll illustrates the conceptual struggle:

The warrior → embodies the physics-informed detector, disciplined and interpretable, standing for structured defense of signal integrity.

The phantoms → represent non-thermal adversarial signatures contaminating CMB-like data (periodic bursts, structured interference).

Celestial circles and wave-rings → symbolize the cosmic microwave background blackbody curves and spectral–temporal structure analysis.

Solar burst and planetary sphere → mark the tension between pure cosmological theory (blackbody radiation law) and anthropogenic interference intruding from orbital or terrestrial origins.

This framing makes the metaphor explicit: the task is not to win a battlefield in myth, but to safeguard radio astronomy pipelines with tools that are interpretable, reproducible, and fast to deploy — the hallmarks of Guangdong-style engineering pragmatism.

In other news: