Poster Abstract: Pushing a Standard Wireless Sensor Network Stack for Ultra-low Data Rates Usman Raza University of Trento and Bruno Kessler Foundation Trento, Italy raza@fbk.eu Amy L. Murphy Bruno Kessler Foundation Trento, Italy murphy@fbk.eu Gian Pietro Picco University of Trento Trento, Italy gianpietro.picco@unitn.it ABSTRACT Time series forecasting aims at improving energy efficiency in wireless sensor networks (WSNs) by reducing the amount of data traffic. One such technique has each node generate a model that predicts the sampled data. When the actual, sensed data deviates from the model, a new model is gen- erated and transmitted to the sink. Reductions in applica- tion data traffic as high as two orders of magnitude can be achieved. However, our experience in applying such fore- casting in a real world deployment shows that the actual lifetime improvement is significantly less due to networking overheads. The study reported here reveals that careful, coordinated network parameter tuning can leverage the re- duced traffic of forecasting techniques to increase lifetime without compromising application performance. Categories and Subject Descriptors C.2.2 [Network Protocols]: Applications; Routing Pro- tocols; E.m [Data]: Miscellaneous; C.4 [Performance of Systems]: Measurements General Terms Design, Experimentation, Measurement, Performance Keywords MAC, Routing, Optimization, Energy efficiency, Low power 1. INTRODUCTION Time series forecasting [3,7] can be used to reduce the data rate in WSN applications that can function with only an approximation of the data sensed by the distributed nodes. In this technique, each node locally computes a model that predicts the data trend. This model is transmitted to the sink, which uses the collected models to approximate the data sensed by each node. As long as the “forecasts” of each model remain within well-defined error bounds of the actual Permission to make digital or hard copies of part or all of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. Copyrights for third- party components of this work must be honored. For all other uses, contact the Owner/Author. Copyright is held by the owner/author(s). SenSys’13, Nov 11-15 2013, Roma, Italy ACM 978-1-4503-2027-6/13/11. http://dx.doi.org/10.1145/2517351.2517423 . Table 1: Data Reduction with DBP WSN Application Datasets Data Reduction Tunnel Lighting Light 99.74% Soil Ecology Air Temperature 91.83% Soil Temperature 98.80% Indoor Sensing Humidity 99.50% Light 97.58% Temperature 99.60% sensed values, no communication is required. Otherwise, a new model is computed and sent to the sink. Our novel, linear modeling technique, Derivative Based Prediction (DBP), has demonstrated [6] up to 99% reduc- tions in the amount of data generated in a sample network. DBP takes a small sequence of sensed data and constructs a line approximating the trend within that sequence. Future sensed data is compared to the data predicted by the linear model. If the sensed values are too far away from the line for too long, a new model is generated and sent to the sink. To evaluate the effectiveness of the model, we applied DBP to six different data sets and their applications: i) a WSN in a road tunnel used to monitor and control the light- ing [2], ii), soil data from the Life Under your Feet Project [4] used by biologists to study micro-climates, iii) indoor hu- midity, light and temperature data from a testbed at the Intel Berkeley research lab [1], applied to building climate control. When applying reasonable application-dependent error tolerances, Table 1 shows that DBP achieves reduc- tions in the transmitted data from 91 to 99%. 0 0.5 1 1.5 2 2.5 3 3.5 500 1000 1500 2000 2500 3000 Avg. Duty Cycle (%) Wakeup Interval(ms) Idle Listening CTP Beacons DBP Models Figure 1: Energy Break- down. Nevertheless, when this extremely small amount of data is sent on top of a standard WSN network stack composed of CTP and BoX-MAC, the sys- tem lifetime improves only 3-times [6]. As shown in Figure 1, this relatively low im- provement can be at- tributed to the large control overheads present in these protocols. For example, BoX-MAC controls the radio duty cycle, periodically wak- ing up the radio to check for a possible intended reception. With shorter wakeup intervals, transmission times at each hop are reduced. However, as there is very little data, the receive checks are often unnecessary, resulting lost energy. Therefore, the size of the wakeup interval must be adapted