The positioning accuracy of microphone arrays in acoustic measurement rooms directly affects the interpretation of the spatial properties of the sound field (mode shapes, diffuseness, isotropy). This thesis treats a cable-driven measurement system, the Cable Robot Microphone Array (CRMA), which uses a 6×6 microphone grid and determines the position of the microphone field through acoustic localisation with respect to six loudspeakers at known coordinates. We develop a geometric model linking the position of the microphone field to the times of flight (TOF) extracted from impulse responses, and describe the actual reconstruction procedure: threshold-based leading-edge time-of-flight picking, multi-resolution grid search over translation, rotation and a scale factor, and iterative rejection of outlying microphones. The central diagnostic output is a residual control matrix between measured and modelled distances, whose statistics we formally connect to the statistics of weighted combination of independent estimates. Positioning accuracy was independently evaluated with a laser scanner and amounts to approximately 3.5 mm in the position of the centre and 0.004 rad in orientation. As an additional, independent check of the system's accuracy alongside the laser-based validation, we also perform a numerical sensitivity analysis (Monte Carlo): based on a single real measurement we simulate repetitions with added noise at the level of the quantisation limit (≈5 mm) and observe the stability of the parameters and cable lengths. In analysing the actual source code of the system we further identify and document three implementation issues — two affecting the robustness of the outlier-rejection mechanism, and one in the computation of the control matrix itself (a missing scale-factor correction, which masked the true level of fit under a systematic offset) — and propose a fix for each. The results provide a quantitative estimate of positional uncertainty and concrete guidelines for further improvement of the system.
|
