Introduction
Acoustic self-positioning—the task of determining the spatial coordinates of distributed microphones and sound sources—is fundamental to technologies ranging from teleconferencing to advanced noise cancellation. Traditionally, this task relies on two distinct measurement types: Time of Arrival (TOA) and Time Difference of Arrival (TDOA). While TOA offers high accuracy, it typically requires knowledge of the source signal's waveform. Conversely, TDOA is more practical for "blind" or asynchronous environments as it only requires microphone signals, though it often leads to more complex mathematical formulations.
This paper addresses the "holy grail" of acoustic localization: achieving the mathematical richness and performance of TOA-based methods without requiring the original source signal. This is particularly challenging in asynchronous environments where microphone start times and source emission times are unknown.
Methodology
The authors introduce a novel mapping function, $f(\bullet)$, designed to transform raw timing measurements into a standardized representation. The core innovation lies in the derivation of two versions of this function—one for TOA and one for TDOA—and the subsequent mathematical proof that these versions are identical.
As shown in the timing diagram, the mapping function elegantly handles "Unknown Timing Information" (UTIm). By centering measurements around mean values and reference offsets, the unknown start times of microphones ($\delta_i$) and emission times of sources ($\eta_j$) effectively cancel out. This transformation allows properties previously exclusive to TOA, such as specific rank properties of the measurement matrix, to be applied directly to TDOA data derived solely from microphone signals.
Experimental Validation
The researchers validated the mapping function using two distinct datasets:
- Simulated Data: A MATLAB-generated environment with 20 microphones and 20 sources in a $10 \times 10 \times 3$ meter room, providing 400,000 data points under perfect yet asynchronous conditions.
- Real-Life Data: A publicly available dataset (Batstone et al.) featuring 12 fixed microphones in a $5 \times 3$ meter office, with a loudspeaker playing "chirp" signals from 65 different positions.
Results
The primary metric for validation was $\Delta f_{i,j}$, representing the difference between the mapped TOA and the mapped TDOA values.
The results demonstrate that in both simulated and real-world environments, the difference $\Delta f_{i,j}$ was approximately $10^{-16}$ seconds. This value corresponds to "machine epsilon," confirming that the mapping is a mathematical identity rather than a mere approximation. The identity held robustly even in physical office environments with real hardware, proving that microphone signals alone are sufficient to recover the "rich" information traditionally associated with TOA.
Conclusion
By collapsing the wall between TOA and TDOA sub-fields, this research proves that the source signal waveform is mathematically irrelevant for high-performance localization. Future work will focus on integrating this mapping into existing TOA algorithms to enhance "blind" localization accuracy and exploring its application in source separation and signal enhancement in reverberant environments.
Summary updated