AbstractThis work addresses the problem of digital terrain modelling for estimating radio path propagation within a mobile communication system. The ideal requirements are for a data structure which is storage efficient and computationally efficient for calculating profiles, whilst elevation errors should be constrained and radio path loss errors should be minimised. For a digital terrain model (DTM) to be considered viable as an alternative to the regular grid, it should:
(i) produce a storage saving of at least 75% over the regular grid;
(ii) be error constrained to a maximum absolute error of 10 metres;
(iii) produce only a small overall average elevation error;
(iv) preserve critical terrain characteristics such as ridges, peaks and slopes;
(v) produce 95% of profiles to within a radio path loss error of ± 6 decibels; and
(vi) be as computationally efficient as the regular grid.
This research focuses on the implementation of a number of prototype DTMs, including a regular grid, sub-sampled grids, variable density grids, elevation difference grids, polynomial models of fixed and variable degree, surface patch quadtrees, and triangulated irregular networks (TINs). Each of these DTMs are examined in terms of the criteria outlined above. No DTM fulfils all of these requirements. The user should identify the relative importance of each requirement before selecting a specific model. For this study, computational efficiency is identified as the criterion which can be considered the least important.
With this in mind, two original DTMs are developed. These are optimised with respect to storage and error constraints. The proposed Huffman-encoded DTM represents the deviations of a regular grid of heights from linearly predicted values as variable-length codes, whilst the Implicit TIN is a storage-efficient triangulated irregular network which reconstructs the original topology of the triangulation at the application stage. Both methods produce storage savings approaching 90% over the regular grid for the data sets tested and are suitable for parallel implementations.
|Date of Award
|Derek Smith (Supervisor) & David Knight (Supervisor)