Understanding the real behavior of Mote and 802.11 ad hoc networks: an experimental approach

Abstract

IEEE 802.11 and Mote devices are today two of the most interesting wireless technologies for ad hoc and sensor networks respectively, and many efforts are currently devoted to understanding their potentialities. Unfortunately, few works adopt an experimental approach, though several papers highlight that popular simulation and analytical approximations may lead to very inaccurate results. In this paper we discuss outcomes from an extensive measurement study focused on these technologies. We analyze the dependence of the communication range on several parameters, such as node distance from the ground, transmission data rate, environment humidity. Then, we study the extent of the physical carrier sensing zone around a sending node. On the basis of these elements, we provide a unified wireless link model for both technologies. Finally, by using this model we analyze well-known scenarios (such as the hidden node problem), and we modify the traditional formulations according to our experimental results.

Introduction

Self-organizing wireless networks are nowadays one of the hottest topics in the area of pervasive computing. The research community is devoting a lot of effort to designing protocols to support Mark Weiser’s pervasive networking vision. Mostly, 802.11-based ad hoc networks and sensor networks are being investigated. While 802.11-based devices have emerged as the de-facto standard technology for investigating ad hoc networks, no counterparts exist on the sensor-networks side. The most widespread devices are Berkeley Motes [15] that are thus assumed as the reference sensor-network technology also in this paper.

The vast majority of works on wireless networks rely on simulation models for evaluations, the main reason being the ease of development and reproducibility with respect to real experiments. However, relying just on simulations may be misleading. Specifically, it is well known that accurately modeling the signal propagation on a wireless medium is a hard task. Unfortunately, an accurate model is often required to correctly evaluate the effectiveness of higher-layer protocols. For example, [10], [18] show that the performances of routing protocols (e.g., AODV, DSR) highly depend on the physical-layer model used in simulations. In some cases, simulation results are extremely different from experimental measurements. Furthermore, the relative comparison among couples of protocols can be completely swapped by changing the physical-layer model. These observations remind us that simulation models and outcomes should be validated against experimental measurements.

These remarks are the main motivation for the work presented in this paper. Specifically, we report the main results from a wide measurement study focused on 802.11 and Mote wireless networks. The emphasis of the paper is on characterizing key networking features such as the maximum communication distance¹ between a couple of nodes, and the interactions between concurrent transmitting nodes. We study the effect on the communication distance of several environmental parameters (e.g., humidity, distance from the ground), providing quantitative evidence of their impact. For example, the communication distance of Mote nodes on foggy days can drop to one fifth of the communication distance on dry days. We also account for the effect of technology-dependent parameters, such as the bit rate, and the antenna directionality. We find that the communication distance of 802.11 nodes significantly varies with the data rate, and we sketch possible side effects on routing protocols. We also find that Mote antennas are strongly directional, and nodes need to be very carefully placed in order to communicate.

Then, we study the effect of concurrent transmitters on each other. Since both technologies adopt a CSMA/CA MAC protocol, the Physical Carrier Sensing mechanism determines the interaction between concurrent senders. Our measures show that for both technologies, Physical Carrier Sensing–and, thus, the dependence among couples of transmitters–extends far beyond the maximum communication distance. Roughly, the maximum Physical Carrier Sensing distance is (at least) twice as large as the maximum communication distance.

On the basis of these measurements we provide a channel model for both 802.11 and Mote devices. It is worth noting that, while the numerical values (e.g., the maximum communication distance) depend on the particular technology, the structures of the channels are very similar in the two cases. Therefore, the two channel models can be seen as particular instances of a unified channel model. This is very important, since the same channel model can be used to analyze both 802.11 and Mote networks, just by tuning the model parameters.

Finally, we exploit the channel model definition to elaborate on the well-known hidden and exposed node problems. The formulations currently reported in computer networking handbooks do not take into consideration the effect of Physical Carrier Sensing beyond the communication distance. Due to the large extension of Physical Carrier Sensing, we find that these formulations should be significantly revised. Hence, we provide novel formulations, which comply with the measurement outcomes.

The rest of the paper is organized as follows. Section 2 describes the experimental methodology we have adopted, and provides some background on the 802.11 and Mote technologies. Section 3 analyzes the bandwidth available at the application level for both technologies. This analysis is exploited to understand the networking features under investigation. Section 4 is devoted to characterizing the maximum distance at which a couple of nodes can correctly communicate. Section 5 presents the results related to Physical Carrier Sensing, and defines the channel model for 802.11 and Mote nodes. Finally, Section 6 surveys works related to ours, while Section 7 highlights the lessons learned from our measurement study.