Updates on Indoor Localization System Kaifei Chen, Siyuan Jack He, Randy Katz, David Culler Old System Design Problems • • • In old system, application collects data from sensors and sends requests to localization server • It becomes throughput bottleneck • It becomes functionality bottleneck (deal with different sensors and algorithms) A sensor/algorithm can be used by several components (multicast) System breaks if any components doesn’t conform to data structure New System Design System Dataflow Acoustic Ranging Using De-convolution WiFi Fingerprinting WiFi Signal Strengths Are Not Stable WiFi Appearance Possibility Are Stable • Use Message Queue as Overlay Network • Mobility, caching, multicast, security • Separate Sensor, Algorithm, and Users • Sensors are deployed by their owner. They publish to their own topics. • Algorithms are instantiated by and talk to our Golang wrapper via Cap’n Proto. They define their input and output data schema. • Users instantiate algorithm by publishing input and output topics to an algorithm wrapper. • Broker verifies data schema on each topic Signal Strengths Result, 69% Accuracy Appearance Possibility Result, 88% Accuracy • Send Sine Sweep Wave f(t), and denote its time reversal as f’(t) • Record acoustic response g(t) • Get impulse response y(t) = g(t) * f’(t) • Find peaks in y(t) and calculate distance