A Dual Distance Measurement Scheme for Indoor IEEE Wireless Local Area Networks*

A Dual Distance Measurement Scheme for Indoor IEEE 80.11 Wireless Local Area Networks* Murad Abusubaih, Berthold Rathke, and Adam Wolisz Telecommunication Networks Group Technical University Berlin Email: abusubaih@tkn.tu-berlin.de Abstract In this paper we address the problem of distance measurement in indoor IEEE 80.11 WLANs. The Received Signal Strength Indicator (RSSI) and the Signal Propagation Time (SPT) have been widely used for distance estimation in indoor WLAN deployments. However, neither approach by itself is accurate. Moreover, there is no mean by which WLAN nodes know whether the estimated distance is accurate or not. In this work we propose a multi-measurement approach. Specifically, we combine an RSSI and a propagation time - based schemes and utilize both results for mutual cross-validation to improve the estimation accuracy and reduce the number of false decisions based on errored estimations. The performance of our combined mechanism has been evaluated through real experiments. I. INTRODUCTION Over the last years wireless local area networks (WLANs) [1] have become quite popular. Due to decreasing costs of the equipments (wireless access points APs and wireless network cards) and fixed broadband connections (digital subscriber lines), WLANs have become the preferred technology of access in homes, offices, and hot-spot areas (like airports and meeting rooms). Although originally several standards for WLAN have been competing, today virtually all WLANs are based on the IEEE 80.11 standard. The knowledge of WLAN users locations is becoming increasingly more important recently for both location based applications and network performance improvement. Therefore, determining the location of a wireless network client has considerably attracted the attention of many researchers and manufacturers. From the network point of view, the ability to measure the distance between mobile users and APs would help in many issues like: Handover decisions, AP selection, Locating Rogue APs, Locating sources of interference, Locating specific users, and balancing the load among WLAN APs. A fundamental issue in positioning and location systems is the determination of the distance between wireless nodes. Two common approaches have been extensively used so far, namely are: The RSSI- and the Signal Propagation Time SPT-based approaches. The RADAR system in [] is probably the first positioning system using an IEEE 80.11 for indoor WLANs deployments. The system is based on RSSI maps constructed in an offline phase. Other several studies that make use of the RSSI to infer STAs locations can be found in [3], [4], [5], [6]. 0 * This Work was Partially Supported by Proximetry Inc. A propagation time based approach is being used in both outdoors and indoors positioning systems like GPS [7]. Hoene et al. [8] have proposed an improved propagation time based distance measurement approach for IEEE 80.11 WLANs. The authors have used a packet latency based approach for outdoor distance measurements. They utilized the immediate acknowledgment feature of IEEE 80.11 and measure the time difference between sending a data packet and receiving the corresponding acknowledgment. Recently, another Round trip time based approach that uses RTS/CTS frames has been proposed in [9]. However, the RTS/CTS mechanism is not commonly used due to the overhead it incurs. Unfortunately, probably due to the multipath fading effect and obstructions, the accuracy of each previously cited approaches is low (-5m) when utilized alone. Moreover, there is no mean by which WLAN nodes could know whether the estimated distance value is accurate or not. This calls for new enhanced algorithms that improve the estimations accuracy. In this paper we present a multi-approach that utilizes results of more than one scheme to cross-validate the correctness and increase the degree of confidence on the estimated distance. Note that the idea presented in this paper applies for any number of independent schemes although we focus on a dual scheme based on the RSSI and Packet Propagation Delay parameters. We show that such a hybrid algorithm could reduce the time period required to obtain estimations normally specified for any approach. Therefore, one could save the wireless bandwidth required for signaling overhead during measurements. Estimations validation can reduce the false decisions that might be based on inaccurate results The remainder of this paper is organized as follows: A packet propagation delay (PPD)-based distance measurement approach is described in Section II. Section III presents an RSSI distance measurement approach, then we present a dual approach that utilizes both PPD and RSSI parameters to validate the estimated distance in Section IV. The experimental setup is described in Section V. Performance evaluation metrics are described in Section VI and then the experimental

results are presented in Section VII before we conclude our paper in Section VIII. II. PACKET PROPAGATION DELAY (PPD) - BASED DISTANCE MEASUREMENT As we utilize the PPD approach of [8], we briefly describe the basic idea of that approach. It utilizes a very important feature of IEEE 80.11 which is the immediate acknowledgment. Every unicast data packet is immediately acknowledged. The approach takes the advantage of drifting clocks to determine propagation times that are forty times smaller than the clocks quantization resolution. Basically, the PPD scheme works as follows: The time span from the moment at which a packet starts to occupy the wireless medium to the time at which the immediate acknowledgment is received is firstly measured denoted as T Remote. The time duration between the reception of a data packet and issuing the corresponding immediate acknowledgment is also measured denoted as T Local. Then, the distance could be simply computed based on the relation between the distance traveled and the speed of light as follows: d P P D = (T Remote T Local ) c where c 3 10 8 is the speed of light. As operating system interrupts add some extra delay, the WLAN card MAC time stamps have been used rather than the operating system time stamps. The resolution of WLAN cards is 1µs during which a signal would reach 300m. Therefore in order to be able to increase the resolution, time estimation should be based on a huge number of packets which makes the approach difficult to be utilized in real world applications. III. RSSI BASED DISTANCE MEASUREMENT APPROACH The power of the received signal at a WLAN node can be related to the transmitted power as: (1) P Rx = P T x P L + P I () where P T x is the power of the transmitted signal in dbm. P I 4.(dB) denotes the environment power correction factor [10]. P L is the path loss in db given as: P L = P L (d 0 ) + 10nlog 10 ( d d 0 ) + P s (3) where d 0 is the reference distance usually chosen to be 1m, P L (d 0 ) 50.4(dB) is the reference path loss [10], n = 3.1 is the path loss exponent, and P s is the loss in power due to shadowing. A typical values used for P s can be found in [11]. The relation between the received signal power at a WLAN node and the RSSI value recorded by and reported by adapters is vendor dependent. For example, with Atheros chipsets, the received power P Rx is simply computed by subtracting 95 from the reported RSSI value. Hence, P Rx = RSSI 95 (4) After substituting the constant parameter values and rearranging, the distance can be estimated as: d RSSI = 10 (P T x +48.8 RSSI P s) 31 (5) IV. OUR COMBINED DISTANCE MEASUREMENT APPROACH In this section we introduce a new combined distance measurement approach which we refer to as the dual approach in the following sections. Then we present the way this approach works. The idea behind this approach is to benefit from independent schemes. As shown in section II, the measured distance d P P D is a linear function of the round-trip time whereas in section III d RSSI is an exponential function of received power level. In fact, both methods use different parameters with different functions to estimate the distance between two WLAN nodes. Under the assumption that both measurement methods are statistically independent from each other, it seems to be natural that each method could be used to validate the other one. Even if there is some statistical dependency between the two methods (e.g. due to fading effects of the wireless channel) and both methods might fail to estimate the distance, we expect that both methods will not converge to the same value, because they use different approaches to estimate the distance. Let us denote the measurement time of the RSSI and PPD approaches by T RSSI and T P P D respectively. We assume that the PPD is the primary approach and use the RSSI approach to cross-validate the PPD s estimations. Let ɛ be a design parameter that denotes the acceptable absolute difference between the estimations of the two approaches. Also, denote the estimated distance by D. The Dual approach works as follows: 1) Run the PPD algorithm. ) Estimate d P P D as described in section II. 3) In parallel, utilize the RSSI value of the received frames and estimate the distance d RSSI as described in section III. 4) if = Average( d P P D d RSSI ) ɛ for a continuous time period of T w seconds, Then: { // Both approaches converge // Assume Correct Estimation D = d P P D+d RSSI. Stop the measurements}. 5) else if elapsed measurement time < T P P D Then: Go to step ) else: First Option:{ D = d P P D+d RSSI Stop the measurements}. Second Option: // measurement time has elapsed Go to Step 1). // Restart the measurement procedure

The following comments regarding this algorithm are important: If convergence occurs before T P DD elapses, the measurement time could be decreased. In fact, this time reduction depends on how often the values converge and how early do they converge. The second option of step (5) offers more confidence to the correctness of the measured value although it is not a sufficient condition for its correctness. In spite of the fact that restarting a measurement requires additional time, the possibility of convergence has already saved some time in other measurements. The beauty of this dual approach is that the RSSI value is available anyway and therefore it introduces no extra cost. In the next subsection we derive the average measurement time with the dual approach. A. Measurement Time Analysis Denote the average time after which the estimations from both RSSI and PPD converges to an accurate value as T, T T P P D (i.e the default specified required measurement time for the PPD approach). Also denote the probability of convergence as α, 0 α 1. The objective is to find the average measurement time T avg with the dual approach. Basically, T avg can be computed as: T avg = αt + (1 α)t N (6) where T N represents the time required in case of divergence. First: If the measurement process is stopped at some time instant in case of early convergence that lasts at least T w and just continued up to T P P D in case of divergence, then simply T N = T P P D and the average measurement time would be: T avg = αt + (1 α)t P P D. (7) Second: If the measurement process is repeated in case of divergence, then T N is recursive and given by: T N = T P P D + αt + (1 α)t N (8) In fact, T N can be written as a series expansion by: T N = (1 α) n T P P D + α(1 α) n T (9) Substituting in (6), we have: T avg = αt + (1 α)(t P P D + αt ) (1 α) n (10) The series in (10) is a power series that converges to 1 α. Therefore, the average measurement time in (10) can be computed as: T avg = T + (1 α)t P P D (11) α Third: If one only assumes that accurate estimations should converge up to T, and repeat the measurements if they diverge at time instance T, then in this case T N is also recursive and given by: T N = T + αt + (1 α)t N (1) Following the approach above, it can be easily shown that the average measurement time is given by: T avg = T (13) α Therefore, by utilizing the PPD and RSSI approaches and if the probability of convergence is high, we are not only able to validate the correctness of our estimations but also we might be able to reduce the measurement time and consequently save some of the wireless bandwidth used by the signaling packets used in measurements. To investigate the performance of the dual approach, a systematic measurement in an office environment will be performed. Different characteristics of the wireless channel and different distances between two WLAN nodes will be selected. The following major questions are to be answered: First, what is the probability that both methods converge? This probability gives insight about the time that could be saved in case of convergence and the additional time required to repeat measurements in case of divergence. Second, what is the residual probability that both methods converge to an inaccurate value? This probability captures the efficiency of the proposed criterion. The lower is this probability, the more efficient is the dual criterion as it would filter out just inaccurate estimations. Third, what is the probability that both methods do not converge, but one estimation of the two methods (PPD estimation as assumed) is accurate? This probability could help in deciding whether a measurement should be repeated or not in case of divergence. If this probability is low, then most likely the results are not accurate and measurement repetition is a better option. V. EXPERIMENTAL SETUP In this section we describe the setup of the distance measurement experiment we have performed. The experiment is comprised of an Access Point (AP) and two WLAN Laptops. The AP and the WLAN cards are all based on the Atheros chipset. Also, they all support the three physical operating modes 80.11a/b/g. The used WLAN adapter driver is a modified version of the open source MADWIFI device driver. The objective of the experiments is to determine the distance between the AP and a WLAN client. As the WLAN chipsets record only the time stamps of the incoming packets, the second Laptop is used in monitor mode supported by the MADWIFI driver to capture and record time stamps of incoming as well as outgoing packets. The monitoring Laptop is placed close to the measuring client in order to eliminate any extra propagation delay that might influence the results accuracy. We have positioned the AP in some place and performed the experiments for different client locations, namely: very near to the AP, between the AP and cell edge and at the cell edge. The locations have been selected so as to have

different wireless channel characteristics. Figure V depicts this setup. The experiments have been carried out under the 80.11g operational mode with two different physical rates (11Mbps and 54Mbps) so as to capture the effect of the physical rate on the estimation accuracy. To compute the propagation time, the WLAN client sends a sequence of ICMP- Ping request packets (figure ), the AP then replies with the immediate acknowledgment. Then T Remote is computed from the MAC time stamps recorded on the transmitted request and received acknowledgment packets. Also, T Local is computed by subtracting the MAC time stamps of the received reply and the issued acknowledgment packets. Afterwards, the distance is calculated following the algorithm presented in section (IV). Fig. 1. Experiment Setup - WLAN nodes Placement, A) High fading, low path loss. B) LOS, low fading,low path loss. C) Low fading, high path loss Fig.. Experiment Setup - Packet Propagation Delay Approach (PPD) VI. PERFORMANCE EVALUATION METRICS FOR DISTANCE MEASUREMENT Actually there are a couple of performance metrics for evaluating distance measurement algorithms. First, the achievable accuracy of the measured distance which is in fact the main objective of any algorithm. Second, the required measurement time. Third, the probability that a scheme fails to attain the accuracy after the required measurement time has elapsed. Four, the signaling overhead imposed by distance measurement algorithms which can be interpreted as the amount of consumed wireless bandwidth. In addition to the aforementioned metrics, we show the usefulness of presented dual approach by answering the the questions raised in section IV. Namely: We are interested to know the probability that both RSSI and PPD approaches converge P(Converge) as well as the probability that they converge to an inaccurate value P(Inaccurate Converge). Moreover, we would like to know the probability that both approaches do not converge but the result of at least one (PPD as assumed) is accurate P(One Accurate Diverge). However, we firstly outline the assumptions for the evaluation. We assume a result to be accurate if the error is within 15% of the true distance, where the error is the absolute value of the difference between the measured and true distances. We also assume that the two approaches converge if the average of the absolute value of the difference between the average measured distance from both is within 15% of the true distance over a time window of T w seconds. T P P D and T w were set to be 00 seconds and 40 seconds respectively while ɛ was set to be 0.15 True Distance. VII. RESULTS In this section we present the experimental results of the proposed dual measurement technique. We show the usefulness of such hybrid scheme for validating the estimated distance in indoor WLANs. The overall results have shown that the mean value of the error rate for all measurements around 0%. Also, we have observed that a better estimation could be achieved with low physical rates. Figures (3) and (4) depict sample measurements, where the time progress of the measured distance for both PPD and RSSI approaches is illustrated. In Figure (3), both approaches converge to the true distance, while in Figure (4) RSSI has failed to converge. The figures also show that a significant time could be saved if we detect the convergence early. Figure (5) plots the estimations accuracy versus the percentage of measurements found to fall below that accuracy for the PPD, RSSI and Dual approaches. The figure reveals that utilizing only RSSI or only PDD, around 40% of the measurements have accuracy lower than 80% (error > 0%). However, with the Dual approach only around 15% of the measurement experiments have accuracy lower than 80% (error > 0%). The accuracy improvement with the Dual approach is due to the filtering of divergent measurements, resulting in a more confident distance estimation. Figure (6) plots the percentage of measurements found to fall below a given accuracy rate utilizing the Dual approach for different distances. The results show that for long distances only about 10% of the measurements have accuracy below 90%. i.e Positioning the wireless devices closer to each other decreases the accuracy. Finally, in figure (7) a comparison between the measurement time requirements of PPD and Dual approaches is plotted. Specifically, the figure depicts the percentage of measurements where the consumed measurement time falls above some time percentage of the default measurement time. Clearly utilizing only one single approach (PPD), then every measurement requires 100% of the default measurement time. However, with the Dual approach, just about 35% of measurements have consumed 100% of the default measurement time. However, all measurements have required more than half the default measurement time. Additionally, using the two simple RSSI and PPD schemes, the experimental results have shown that, the probability that the two approaches converge: P(Converge)=0.673. However, the probability that both approaches converge to an inaccurate value: P(Inaccurate Converge)=0.13. This reveals that only 13% of the decisions based on convergence might be

errornous. However, utilizing only RSSI or PPD around 40% of the estimations were erroronous as stated earlier in this section. The following comments can be drawn from these results: The presented dual criterion could be used to validate the correctness of the distance estimations without any additional overhead. Although the approach has been tested for simple PPD and RSSI - based mechanisms that are not accurate enough, it applies for any improved versions of the algorithms or even totally different ones. Depending on the probability of convergence and the time at which it is detected, the approach might reduces the measurement time significantly which translates into a reduction in the wireless bandwidth. VIII. CONCLUSIONS This paper proposes a combined mechanism for validating distance estimations in Indoor WLANs. It presents experimental results obtained from the utilization of the Received Signal Strength (RSSI) and the Packet Propagation Delay (PPD) parameters for distance estimation between two indoor WLAN nodes. As the WLAN node has no mean to validate the measured distance, we have shown that the RSSI value could be used as a validation indicator for some acceptable threshold difference. Moreover, we have shown that such combined approach could reduce the amount of time required for distance estimation and therefore save the wireless bandwidth. REFERENCES [1] IEEE - Information Technology - Telecommunications and Information Exchange Between Systems - Local and Metropolitan Area Networks- Specific Requirements - Part 11- Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications, September 1999. [] Paramvir Bahl and Venkata N. Padmanabhan. An In-Building RF-Based User Location and Tracking System. In In INFOCOM 000, pages 775-784, Tel-Aviv, Israel, March, 000. [3] Kremenek T Muntz R. Castro P, Chiu P. A probabilistic location service for wireless network environments. In In Proceedings of Ubicomp, pages 18-4, September, 001. [4] Kogan D Smailagic A. Location sensing and privacy in a context-aware computing environment. In IEEE Wireless Communications, 9(5):10-17. [5] Youssef M. Horus: A WLAN-Based Indoor Location Determination System. In Department of Computer Science, University of Maryland,1999. [6] Kogan D Kaemarungsi K. Distribution of WLAN Received Signal Strength Indication for Indoor Location Determination. In Proceedings of the 1st International Symposium on Wireless Pervasive Computing, January 006. [7] Misra P. Enge P. Special Issue on GPS: The Global Positioning System. In IEEE, 1999. [8] Gunther A. Hoene C. Measuring Round Trip Times to Determine The Distance between WLAN Nodes. In Proceedings of Networking, Waterloo, Canada, May, 005. [9] Barcelo F. Paradells J. Zola E. Izquierdo F., Ciurana M. Performance evaluation of a TOA-based trilateration method to locate terminals in WLAN. In Proceedings of the 1st International Symposium on Wireless Pervasive Computing,, January, 006. [10] Larry G. Ivan S. Praveen G., Predrag S. A Method for Predicting the Throughput Characteristics of Rate Adaptive Wireless LANs. In Proceedings of the IEEE Vehicular Technology Conference VTC 04, Los Angelos, CA, pp. 458-453, September, 004. [11] Chris R. Joel B. Binghao L., Andrew D. Hybrid method for localization using WLAN. In Spatial Sciences Conference, Melbourne, Australia, pp. 341-350, September, 005. Measured Distance (m) 8 6 4 0 - RSSI and PPD Distance Measurement 0 50 100 150 00 50 300 Time (Sec) RSSI PPD Fig. 3. Dual distance Measurement, for a True Distance of m and 11Mbps 80.11g RSSI and PPD Distance Measurement Measured Distance (m) 14 1 10 8 6 4 0 0 0 40 60 80 100 10 140 160 Time (Sec) RSSI PPD Fig. 4. Dual distance Measurement, for a True Distance of 5m and 54Mbps 80.11g Fig. 7. Fig. 5. Fig. 6. Performance of RSSI, PPD, and the Dual Approaches Performance of Dual Approach for different Distances Time Requirement Comparison of PPD,RSSI and Dual Approaches