Rate, Leakage, and Distortion for Stationary Gaussian Sources: Water-Filling on a Tilted Spectrum

Manuscript prepared for the IEEE Transactions on Information Theory (version tit-stationary-1.4), with its sources, check scripts and their recorded outputs, the notebook that reruns the checks, the Lean 4 module, and a short companion note on Observation Theory. An encoder describes part of a stationary Gaussian vector process to a decoder under a weighted quadratic distortion. A second party holds a noisy filtered observation of the process and never decodes. The description has two costs per symbol, its rate and its conditional leakage to the second party. For memoryless sources the frontier is reverse water-filling on the decoder's weight tilted by the second party's information. We show that for sources with memory the same holds frequency by frequency, under one water level. For a periodic source the region splits exactly over the Fourier frequencies, and it converges to a spectral region as the period grows. For a scalar process with a scalar context the minimum leakage is an integral of a closed form, and the two optimal descriptions can occupy disjoint bands. A delay in the second party's observation changes nothing when it keeps its whole observation. For the stationary source we prove that the spectral region is the operational region, with deterministic encoders, for every continuous positive definite spectral density matrix and a second party that observes the process through a filter with a continuous frequency response in noise of continuous positive definite spectrum. The region is unchanged when that party has its whole observation process, not only the part during the coded block. When that party's map is memoryless and its noise white, finite-window costs converge to the spectral cost from above, with a gap at most a constant over the window length for Lipschitz spectra.

Authors

Institutions

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-18
DOI
https://doi.org/10.5281/zenodo.24329940
Primary Topic
Wireless Communication Security Techniques
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Rate, Leakage, and Distortion for Stationary Gaussian Sources: Water-Filling on a Tilted Spectrum

Andrew H. Bond
Zenodo (CERN European Organization for Nuclear Research)
Wireless Communication Security Techniques
preprint

Rate, Leakage, and Distortion for Stationary Gaussian Sources: Water-Filling on a Tilted Spectrum

Andrew H. Bond
preprint en

Abstract

Manuscript prepared for the IEEE Transactions on Information Theory (version tit-stationary-1.4), with its sources, check scripts and their recorded outputs, the notebook that reruns the checks, the Lean 4 module, and a short companion note on Observation Theory. An encoder describes part of a stationary Gaussian vector process to a decoder under a weighted quadratic distortion. A second party holds a noisy filtered observation of the process and never decodes. The description has two costs per symbol, its rate and its conditional leakage to the second party. For memoryless sources the frontier is reverse water-filling on the decoder's weight tilted by the second party's information. We show that for sources with memory the same holds frequency by frequency, under one water level. For a periodic source the region splits exactly over the Fourier frequencies, and it converges to a spectral region as the period grows. For a scalar process with a scalar context the minimum leakage is an integral of a closed form, and the two optimal descriptions can occupy disjoint bands. A delay in the second party's observation changes nothing when it keeps its whole observation. For the stationary source we prove that the spectral region is the operational region, with deterministic encoders, for every continuous positive definite spectral density matrix and a second party that observes the process through a filter with a continuous frequency response in noise of continuous positive definite spectrum. The region is unchanged when that party has its whole observation process, not only the part during the coded block. When that party's map is memoryless and its noise white, finite-window costs converge to the spectral cost from above, with a gap at most a constant over the window length for Lipschitz spectra.

Zenodo (CERN European Organization for Nuclear Research)
San Jose State University (US)
Wireless Communication Security Techniques
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon