Back to search

Georgia Institute of Technology

Mapping the Invisible from Almost Nothing: Methods and Mechanisms for Physics-Informed Spectrum Cartography Across Propagation Model Fidelities

Abstract

dc:description.abstract

The presented research develops and validates practical frameworks for transmitter localization and radio map reconstruction that operate under very sparse sensing in realistic semi-urban environments. There are three critical gaps in spectrum cartography that, if solved, would enable the adoption of digital spectrum twins, truly automated dynamic spectrum access, and fully AI-optimized radio networks. First, a lack of evidence on when higher fidelity propagation models materially improve results over simple path loss models in sparse regimes; second, the prevailing assumption of dense, geographically distributed sensor data, which is often impractical or unavailable in operational deployments; and third, the absence of a quantitative characterization of how realistic uncertainty in the digital description of the environment propagates through ray tracing to affect the channel statistics on which geometry informed propagation models depend. These three gaps in spectrum cartography are addressed by establishing two estimation pipelines for transmitter localization and radio map reconstruction that operate under very sparse sensing in realistic semi-urban environments: a likelihood based method that marginalizes over all candidate transmitter locations to produce a received power field directly from the sensor observations, and a generalized likelihood ratio test (GLRT)-based sparse recovery pipeline that explicitly detects and localizes individual sources through sequential hypothesis testing, beam search, physics based filtering, and combinatorial model selection before proceeding with radio map reconstruction. Benchmarks are established across three propagation models, namely empirical log distance, The Terrain Integrated Rough Earth Model (TIREM), and site specific ray tracing, on The Platform for Open Wireless Data-driven Experimental Research (POWDER) testbed using all unique combinations of five fixed transmitters and mobile receivers with known ground truth. Evaluation uses average localization error, probability of detection, and precision as localization metrics and root mean square error (RMSE), mean absolute error (MAE) and bias as radio map reconstruction metric across more than six hundred transmitter and random sensor seed combinations at two sensor densities. The best propagation model depends on the estimation framework. Within the likelihood framework, log-distance with a fixed path loss exponent gives the lowest reconstruction RMSE at both densities because the marginalization requires a spatially smooth propagation matrix that geometry-informed models cannot produce. Within the GLRT framework, TIREM consistently gives the best localization at both densities thanks to its spatial discrimination capacity, but the reconstruction stage absorbs moderate localization errors and the cross-model RMSE spread stays under 1 dB at the denser regime where log distance retains a marginal reconstruction lead. At sparser sensing the localization gap widens and ray tracing becomes the lowest-RMSE GLRT configuration. For transmitters far from the sensor network or in terrain depressions, log distance fails entirely, while TIREM and ray tracing recover GLRT detection probability from near zero to 0.67. A complementary Monte Carlo sensitivity study quantifies how realistic scene-description uncertainty propagates through ray tracing, showing that 1 m building height and 0.4 m position perturbations alone cause several-decibel path gain variation comparable to the advantage geometry-informed models offer with perfect scene inputs. The findings are intended to inform the next generation autonomous spectrum management systems such as Radio Dynamic Zones (RDZs), and to clarify the tradeoffs between sensing density, propagation model sophistication, sensor error covariance models, and transmitter receiver geometry that drive operational performance.

Degree

thesis:*
Name thesis:degree_name
Electrical and Computer Engineering, PhD
Department thesis:degree_department
Electrical and Computer Engineering
Grantor
Georgia Institute of Technology
Year dc:date.issued
2026

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Tadik, Serhat
Advisor dc:contributor.advisor
  • Durgin, Gregory D.
Committee members dc:contributor.committeemember
  • Peterson, Andrew F.
  • Sundaresan, Karthikeyan
  • Patwari, Neal
  • Bhattacharjea, Rajib
ORCID person:identifier.orcid

Rights

Language dc:language.iso
English

Identifiers

dc:identifier.*
Handle dc:identifier.uri
https://hdl.handle.net/1853/82174

Chain of custody

source
Harvested from
Georgia Tech
Base URL
repository.gatech.edu/server/oai/request
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
related terms
citation

Tadik, Serhat. Mapping the Invisible from Almost Nothing: Methods and Mechanisms for Physics-Informed Spectrum Cartography Across Propagation Model Fidelities. Georgia Institute of Technology, 2026. https://hdl.handle.net/1853/82174