Few researchers have performed measurements of a Wi-Fi channel at the frame level in order to understand performance issues by relating frame errors and signal stre ngth. This paper presents results relative to ad hoc measurements in a rural environment. We show that a simple double regression propagation model like the one used in the ns-2 simulator can miss important transmission impairments that are apparent even at short transmitter-receiver distances. We propose atworay propagation model which improves on those commonly used for simulation purposes. Frames were transmitted and received by using two cheap laptops with standard Linux drivers. Measurements also show that packet loss at the frame level is a Bernoullian process for time spans of few seconds, and that l onger time spans exhibit a more complex behaviour, meaning that th e 2-state Markov-modulated process often used in the literat ure is not a good match for rural areas. I. I NTRODUCTION: CONTEXT AND OBJECTIVES THE combination of decreasing prices of wireless local area networks (WLANs) and increasing wireless link capacity has significantly encouraged the deployment of WLANs in homes, entire cities, corporate enterprise and academic ca pus networks. Initially, much of the WLAN research was conducted primarily through the use of analytic models [1], and simulation techniques [2]. Only a few researchers have tack led the expensive task of measuring WLANs [3] to understand performance anomalies and the implications of installatio n choices. However, accurate WLAN measurements have proven to be more elusive than those in wired LANs due to the characteristics of the wireless medium. For instance, measurem ents over a single wireless hop, such as in an 802.11 infrastructu re network, can provide different results depending upon the h op distance, cross and contending traffic, the building struct ure and even human motion within a measurement test-bed. In general, capturing aspects of WLAN performance requires more than collecting measurement data at any one layer in the protocol stack, and proper investigation is needed at all la yers. As far as the MAC layer is considered, a complete packet loss model needs to consider a frame error model, an ARQ (Automatic Repeat reQuest) model and a multi-rate switchin g model that implements an dynamic rate switching algorithm. In this paper we examine how ad hoc point-to-point WiFi behaves at the frame level, with both ARQ and dynamic This work was supported by the CNR/MIUR program ”Legge 449/9 7 (project IS-Manet) and by the European Commission under the European Satellite Communications NoE (SatNEx II, IST-27393) withi n the 6th Research Framework Programme. rate switching disabled. As far as we know no results have been published of analogous measurement campaigns. In fact, measurement campaigns have usually been conducted on complex network setups [4], or in simple scenarios where ARQ algorithm was always used, hiding the underlying frame error process details [5], [6]. An additional peculiarity o f our measurement process is that transmission is not greedy, but instead individual frames are sent at precisely controlled time intervals, thus allowing a precise timing characterisatio n of the frame error process. One aim of this paper is to explore a relationship between transmission range, transmission rate and height of transm itter and receiver from ground. In order to do that, we relied on measurements of the received power level as seen by the network card. This procedure has been adopted by few experimenters, such as [6], but to our knowledge no publishe d results are available at the frame level. We found that a tworay model is adequate to describe the relationship we intend ed to study; we also observed that using the power level meter built in the network cards is a reliable method for evaluatin g the proximity of the critical distance where the frame error rate becomes significant in rural area environments. Howeve r, in contrast with thetwo-ray CMU Monarch model used in ns2 [7], in our measures we observed that the received power does not monotonically decrease with distance, but has a significant “hole” where the direct signal and the groundreflected signal interfere destructively. The improved two -ray model we propose reflects these findings. In order to collect detailed information about frame transmission on wireless channel, researchers need to use tested procedures: descri ption and validation of such a procedure with associated software is an additional contribution of this paper. We present two main results, the improved two-ray model and the finding that the frame error process is Bernoullian at time scales of few seconds. These results can be useful for simulations of mobile ad hoc rural networks, particularly for evaluating t he effects of mobility. Additionally, our findings are a starti ng point for future studies on how the frame error rate depends on the received power level after the critical distance wher e frame loss becomes significant. Once such a relationship is discovered, it will be possible to evaluate the performance of loss-based dynamic rate switching algorithms in rural area s. II. M EASUREMENT ENVIRONMENT We performed our outdoor rural measurement campaign using two IBM Thinkpad R40e laptops (Celeron 2 GHz with 256 MB ram running Debian Linux with a 2.6.8 kernel), equipped with CNet CNWLC-811 IEEE 802.11b wireless cards and standard drivers. The cards were put in ad hoc mode, so that it was not necessary to depend on an access point, and no management overhead was present except for the periodic beacon [8]. Important settings for IEEE 802.11 b network cards are the fragmentation threshold, which we disabled in our measurements, the RTS/CTS threshold, which we also disabled, and the transmission rate. We were interes ted in frame-level measurements, so we disabled retransmissio ns (ARQ), and we disabled the dynamic rate switching. We used different fixed speeds of 1, 2, 5.5 and 11 Mb/s, with a fixed frame length of 1000 bytes, for different transmitter-rece iver distances.By disabling ARQ, the MAC layer transmits each packet only once, rather than trying to retransmit a frame up to 8 times after a loss. With this arrangement, it is possible to sample the channel at a constant rate, in order to accurately measure the frame error process in the time domain. Frames are transmitted every 5 ms for bit rates of 11, 5.5 and 2 Mb/s, and every 10 ms for the bit rate of 1 Mb/s. We transmitted 200 000 frames for each measure. The tools used in order to collect statistics about frame errors and power levels are released with a free software copyright license and are available for download at http://wnlab.isti.cnr.it/paolo/measurements/Softwar e.html. III. T HE TWO-RAY PROPAGATION MODEL Previous studies found that path loss characteristics in LO S (line of sight) environment are dominated by interference between the direct path and the ground-reflected path, as in t he two-ray model, in the following referred to as 2RM (see Fig. 1). This model is characterised by a break point that separates the different properties of propagation in near and far regi ons relative to the transmitter; before the break point, the mea n attenuation is close to the free-space path loss 1/d, while after that point it decreases as 1/d. A good approximation of this behaviour is the double regression model suggested by [9]. The authors propose a model with two slopes for approximating the 2RM. In particular, th ey described the existence of a transition region where the bre ak point b can be placed:
[1]
Isabelle Guérin Lassous,et al.
Experiments with 802.11b in ad hoc configurations
,
2003,
14th IEEE Proceedings on Personal, Indoor and Mobile Radio Communications, 2003. PIMRC 2003..
[2]
Giuseppe Anastasi,et al.
Wi-fi in ad hoc mode: a measurement study
,
2004,
Second IEEE Annual Conference on Pervasive Computing and Communications, 2004. Proceedings of the.
[3]
Prasun Sinha,et al.
Understanding TCP fairness over wireless LAN
,
2003,
IEEE INFOCOM 2003. Twenty-second Annual Joint Conference of the IEEE Computer and Communications Societies (IEEE Cat. No.03CH37428).
[4]
E. Green,et al.
Microcellular Propagation Measurements In An Urban Environment
,
1991,
IEEE International Symposium on Personal, Indoor and Mobile Radio Communications..
[5]
Abbas Jamalipour,et al.
Wireless communications
,
2005,
GLOBECOM '05. IEEE Global Telecommunications Conference, 2005..
[6]
Hayder Radha,et al.
Performance analysis and modeling of errors and losses over 802.11b LANs for high-bit-rate real-time multimedia
,
2003,
Signal Process. Image Commun..
[7]
A. Girotra,et al.
Performance Analysis of the IEEE 802 . 11 Distributed Coordination Function
,
2005
.
[8]
David A. Maltz,et al.
A performance comparison of multi-hop wireless ad hoc network routing protocols
,
1998,
MobiCom '98.
[9]
Robert Tappan Morris,et al.
Link-level measurements from an 802.11b mesh network
,
2004,
SIGCOMM '04.
[10]
Chris Heegard.
Range versus Rate in IEEE 802.11g Wireless Local Area Networks by
,
2001
.
[11]
Chris Heegard.
Range versus Rate in IEEE 802 . 11 g Wireless Local Area Networks
,
2001
.
[12]
K. Papagiannaki,et al.
Characterization of 802 . 11 Wireless Networks in the Home
,
2005
.
[13]
Joao Angeja,et al.
A new packet loss model of the IEEE 802.11g wireless network for multimedia communications
,
2005,
IEEE Transactions on Consumer Electronics.
[14]
C. Soras,et al.
Analysis and design of an inverted-F antenna printed on a PCMCIA card for the 2.4 GHz ISM band
,
2002
.