Stability and Scalability Assessment of Complexity Estimation Based Work Load Balancing Approach for a Parallel Lidar Waveform Decomposition Algorithm

LIDAR is an active remote sensing technology which performs range measurements from the sensor and converts them into 3D coordinates of the Earth's surface. Recent advances in LIDAR hardware make it possible to digitize full waveforms of the returned energy. LIDAR waveform decomposition involves separating the return waveform into a mixture of Gaussians which is then used to characterize the original data. It plays an important role in LIDAR data processing because the resulting components are expected to represent reflection surfaces within waveform footprints and ultimately affect the interpretation of the data. Computational requirements in the waveform decomposition process result from two factors; (1) estimation of the number of components in a mixture and the resulting parameter estimates are inter-related and cannot be solved separately, (2) parameter optimization does not have a closed form solution, and thus needs to be solved iteratively. A current state-of-the-art airborne LIDAR system acquires more than 50,000 waveforms per second, and the number of waveforms easily exceeds tens of millions even for small area. Therefore, decomposing the enormous number of waveforms is challenging using traditional single processor architecture. Four work load balancing approaches – (1) a no weighting (NW), (2) a linear weighting based on the decomposition results (DRLW), (3) a squared weighting based on the decomposition results (DRSW), and (4) a linear weighting based on the decomposition time (DTLW) of sampled waveforms for a parallel LIDAR waveform decomposition were assessed in terms of the scalability and stability. The DTLW approach yielded the best efficiency when the number of processors is small, and the NW approach showed the most scalable and stable results as the number of processors gets larger.