New Radar-Like-Algorithm (RLA) A Radial sweep-line is used to scan the terrain: = 0 ; = small fixed angle (say /45); S 1 = signal-section (T,A, ); PQ = new priorityQueue (); while ( < 360) { S2 = signal-section (T,p, + ); Pz = pizza-slice (S 1, S 2 ); PQ.add(Pz); S 1 = S 2 ; = + ; for (i=0; i < BUDGET;i++) { Pz = PQ.top(); // most distant pizza Pz 1 = Pz.leftSplit(); Pz 2 = Pz.rightSplit(); PQ.remove(Pz); PQ.add(Pz 1 ); PQ.add(Pz 2 ); Distance(S 1, S 2 ): max vertical distance average distance RMS Note: the distance is also multiplied by angle between S 1 and S 2. Basic Algorithms Given a terrain T and an antenna A: Sample T with n sample points (SP) For each point p (in SP) compute the signal-strength (A,p) Extrapolate SP using a triangulation. The Sample set (SP) can be Random, Grid Base Or an outcome of a 3D simplification of T. Experiment Parameters: Hi resolution maps: 100*100 km of different types of terrains. Radii [5-30] km, antenna height [10-50]m, sampling set [ ] Platform: Java 1.5 Win XP RF-Propagation models: knife edge (Bullington, Epstein, Deygout). Results 3-5 faster runtime. Improved accuracy. Approximating Radio Maps Boaz Ben-Moshe School of Computing Science, Simon Fraser University Main Obstacle : runtime – too many samples (cross sections) Solution : pipe-line computation along a ray using 2D terrain simplification sampling Goal given a terrain T and an antenna A, associate a signal strength to each point on T as received from A - approximate the radio map of A over T in an accurate and efficient manner. Conclusions & future work The new Radar-Like-Algorithm is: Significantly faster and more accurate. The sensitive version perform better than the fixed one. Highly robust (fits to any terrain and parameters). Current research involves an alternative extrapolation method based on the radial order of the pizza-slices. Motivation Facility location optimization algorithms, which involve locating large scale wireless networks, use approximated radio maps algorithms. In fact, computing such radio maps is often there runtime bottle-neck. Methods Implement existing methods for approximating radio maps. Develop a new Radar-like-algorithm (RLA). Conduct a large scale experiment, comparing the methods: runtime, accuracy, robustness. Basic elements Map (terrain): higher brighter Cross-section: convex-hull Diffractions Signal computation, (Generalization of visibility) Signal Strength: continuance\discrete function: Extrapolation: using triangulation, Transforming a set of samples into a continuance surface. The Brighter the stronger the signal is. runtime Random Grid TS F_Radar S_Radar Examples Sampling (1000 samples): Grid (left) Random +TS (mid) Radar (right) Extrapolation (5000 samples): Random: Uniform – hi noise, blare edge detection. S-Radar: Sensitive angle and density. None uniform – low noise, nice edge detection. Pizza-slice (middle): composed from two signal-sections Average runtime (milliseconds) for constructing a radio map.