Localizing Persons using Body Area Sensor Network

Context awareness is an important aspect in many ICT applications. For example, in an intelligent home network location of the user enables session transfer, lighting and temperature control etc. In fact in a body area sensor network (BASN) location estimation of a user helps in realizing realtime monitoring of the person (especially those who require help) for better health supervision. In this chapter we first introduce many localization methods and algorithms from the literature in BASNs. We also present classification of these methods. Amongst them location estimation using signal strength is one of the foremost. In indoor environments, we found that the signal strength based localization methods are usually not accurate, since signal strength fluctuates. The fluctuation in signal strength is due to deficient antenna coverage and multi-path interference. Thus localization algorithms usually fail to achieve good accuracy. We propose to solve this problem by combining multiple receivers in a body area sensor network to estimate the location with a higher accuracy. This method mitigates the errors caused by antenna orientations and beam forming properties. We evaluate the performance of the solution with experiments. It is tested with both range-based and range-free localization algorithm that we developed. We show that with spatial diversity the localization accuracy is improved compared to using single receiver alone. Moreover, we observe that range-based algorithm has a better performance. BODY AREA SENSOR NETWORK AND LOCALIZATIONS With a number of devices equipped with sensors, micro-controllers and radio components, important parameters, such as body temperature, momentum, glucose levels, blood pressure or heart rate of the person can be monitored. In the meanwhile ambient intelligence (Ducatel, 2010; Aarts, 2003) is seen to be penetrating our daily lives. Being a way to realize ambient intelligence, the rapid development of these wireless sensor networks also extend to body area networks. Wireless sensors can be placed on persons, on cloths, etc. to monitor diverse vital health parameters of the persons. These devices form a Body Area Sensor Network (BASN). Such networks provide many novel applications in healthcare, fitness, and entertainment, which enable better quality of life of persons. A large amount of BASN applications are in healthcare domain. A BASN can help patients in hospitals gain more freedom. Wireless sensors can replace wired sensors monitoring patients. Thus they can move freely instead of being bonded to beds. More than in the hospitals, chronically ill patients or those are in recovery may return to their normal life at home with their physical conditions closely monitored by the doctors remotely. Another application of such body area sensor networks is helping senior citizens to manage their life in houses or even in public places without any support. In the above applications, context information has to be generated to provide control systems some form of context awareness. In case of emergency, data collected from sensors have to be collected and then processed in a control center to make a decision for actuation. Among the context 1 Corresponding author information, location is an important one. No matter in hospital, home or in public area, locations of the patience have to be known so that first-aid can be provided timely. Moreover, the location information also provides doctors with the movement patterns of patients so that their living habits can be analyzed and used for disease prevention and diagnosis. Especially, people with Alzheimer disease should be monitored with wearable sensors and their house is equipped with other sensors for monitoring temperature, humidity, smoke or other hazardous gas and cameras for security surveillance. Some areas are dangerous such as kitchen where there are some sharp tools, gas stoves and electrical appliances. When the patient is moving in the house, sand moves into the kitchen his location should be known to the monitoring systems so that a video-camera in the kitchen is activated. An alarm and the video are sent to caretakers so that help can be provided timely in case required. APPLYING WIRELESS SENSOR NETWORKS FOR LOCALIZATIONS One of the most well-known localization systems is the Global Positioning System (GPS). However, due to poor penetration of the radio signals this system is not suitable for indoor localization (Savvides, 2001). A revolving technique in indoor environment is to use already deployed wireless sensor devices (in hospitals, buildings and other indoor area), together with wearable BASN to estimate the locations of the persons wearing them. This technique is highly practical and cost effective since all the sensors are already deployed and connected wirelessly for their usual tasks. Hence, localization does not add any additional investment on hardware infrastructure. There are a few requirements for reliable localization system. The localization system must have the features such as: (1) Accuracy: although many applications do not require localization to be accurate to centimeter, in many applications in home, buildings or hospitals, a monitoring system has to know where a user is, e.g. a living room or bed room. (2) Low complexity: in an indoor environment, location information is likely to be provided without dedicated hardware to avoid extra cost. Localization is expected to be an additional service on other devices, such as access points and environmental sensors. These devices work with embedded devices which can even be worn on the body. Computational and memory resource are limited on these devices. The localization algorithm or method must be as simple as possible and work inconspicuously to the users (Ducatel, 2010). (3) Real-time operation: location information has to be estimated in real time following the movement of the person. Therefore it has to be updated periodically at appropriate intervals. These intervals have to be decided by the speed of the users. Since we have the indoor environment as our user scenario, walking speed of 1 to 2 m/s is considered as a moderate mobility in these scenarios. Therefore, locations shall be updated every a few seconds. In the literature, there are many approaches designed for different applications. Taxonomy for identifying location techniques in general was developed in (Hightower, 2001). The techniques are classified into Angle-of-Arrival (AoA), Time-of-Arrival (ToA), and the radio signal strength, which is frequently represented by Received Signal Strength Indicator (RSSI). Different techniques focus on different sensing data that is collected (Mutukrishnan, 2005). It has been shown that AoA and ToA methods are very accurate in ranging thus they estimate location with high accuracy. Especially when Ultra Wide Band (UWB) are combined with ToA method (Yu, 2004; Savvides, 2001) or both ToA and AoA (Deng, 2000), the accuracy can be improved even higher. A survey on UWB based localization methods is provided in (Gezici, 2005). Nevertheless, UWB requires extra hardware as reference points and all the devices have to have higher computational frequency so that the differences between ToA can be identified. Therefore it is not suitable for a BASN since the devices required by UWB methods are too large and consumes substantial power in wearable devices. Furthermore, the cost of a UWB system is much higher than that of a BASN. It is well known from many studies that the radio propagation is essentially related to the transmission distance. Given the transmitted signal power, RSSI is highly influenced by the distance. However, distance is not the only factor deciding RSSI. Both (Cox, 1984) and (Bernhardt, 1987) suggested that the distribution of receiving signal strength is a random and log normally distributed random variable with a distance dependent mean value. It is decided by a path loss exponent and a zero mean Gaussian distributed random variable accounting the effect of shadowing. Most of RSSI-based localization systems take advantage of existing indoor network devices such as WLAN or RFIDs (Jin, 2006). Along with the development of WSNs, RSSI-based methods use such a network more and more. The WSNs in most scenarios are implemented for monitoring environmental parameters such as temperature, humidity, presence, illumination etc. Together with a BASN, the two networks can cooperate to locate a person. A review of existing RSSI-based methods is provided in the next section. LOCALIZATION TECHNIQUES USING RSSI RSSI-based positioning algorithms are generally divided into two categories according to the classification in (Liu, 2004): range-based and range-free. The former makes use of the absolute distance or angle calibrated from the pre-measured RSSI map, which can be a set of “signatures” or RSSI to distance/angle relation. The latter, rather than using the information concerning the absolute distance, utilizes the geographic relationship between target motes and anchor motes. RADAR (Bahl, 2000) and MoteTrack (Lorincz, 2005) are the examples of using the “signatures” for localization. Both protocols utilize a large set of pre-measured reference points on a map. Firstly the localization systems in the two protocols have a set of anchors. Then a mobile node is put at different points, which location is know, in the deployed area to measure the RSSIs to different anchors. The RSSIs are used latter for deciding a target mote’s location. Afterwards the target node’s RSSI to anchors are compared the neighboring reference points’ RSSI to the same anchors. The closet reference points are selected. Location of these reference points form an area of which the gravity center is considered as the estimated location of the mobile node. An enhanced RSSI-based localization system is proposed in (Lau, 2008). Although it is also range-based, it does not use “signatures”. Instead, triangulation with Minimum Mean Square Error (MMSE) estimation is used to calculate the target node’s location. The authors proposed a better ranging algorithm which assumes the target’s current location is not far away from a previous one. Thus the two RSSIs should be highly correlated. So the current RSSI is averaged with the EWMA method. APIT (He, 2003) is a good example of range-free localization protocol. The authors assume a lot of nodes are available in the area where the localization is needed. Nodes can exchange their RSSIs observations of three anchors. The three anchors form a triangle. If a node in the target node’s neighborhood has RSSIs to the anchors larger than the target node’s, then the target node must be outside the triangle formed by the anchors. Otherwise it is inside. Changing another three anchors, another triangle can be formed. The whole localization area is divided into small areas; once a triangle is formed the possible areas where the target node may be are given more weight. After all the combination of anchors is tried, the highest weighted areas’ gravity center is estimated as the target node’s location. ROCRSSI (Liu, 2004) is similar to APIT. However, the rings are used to replace triangles. The RSSIp from the target to an anchor is compared to the RSSIs from the anchor to other anchors. We can always find a pair RSSIs between anchors that p is larger than one is smaller than the other one. Then a ring can be drawn that the target node is possibly inside. Again several rings add weights to the possible small areas the target node may reside. The estimation location is the gravity center of the highest weight area. Received Signal Strength Indicator (RSSI) is widely exploited for localization in the literature. However, because of the small size, wearable characteristics and cost restrictions, health care wireless sensor devices normally use an on-chip antenna which in turn has no perfect omnidirectional beam properties. As a consequence, the antenna orientation of the devices has a large influence on the received signal strength. Factors such as complex indoor radio propagation and the movement of people usually cause RSSI-based localization algorithms to fail to achieve a good accuracy. In this chapter, we propose a method, which takes advantage of multiple wireless devices in a BASN to improve the accuracy of estimates of location of a person in an indoor environment. The highlights of this method are: (a) nullifying the effects of antenna orientation by using multiple devices; (b) using the spatial redundancy in the estimations; and (c) implementation of the system to show our method performs better under similar circumstances. LOCALIZATION WITH MULTIPLE DEVICES Range-based localization algorithms usually fail to achieve good accuracy due to deficient antenna embedded, fading, shadowing, and people movements. In this chapter, we try to solve the first two problems by combining multiple wireless devices in a BASN. Their diverse antenna orientations and small distance can be exploited to increase the localization accuracy. Moreover, it was shown in the literature that the relation between RSSI and range is not one to one. Transceivers apart of different distances, sometimes up to 10s of meters, can receive packets with the same RSSI (Aguayo, 2004; Zhao, 2003). Therefore, absolute distance estimation is tricky. Although range-free algorithms avoid using absolute distance estimation between anchor and target nodes, it has to use relative distances between anchors. If the antennas on the anchors also have the problem on orientation, the relative positions of anchors are not precise anymore thus the location estimated is not accurate. We will combine the multiple-devices-solution with representatives from both categories to test how much this solution can improve performance of the protocols. In the following, we firstly introduce the advantages of using multiple devices for localization. For completeness, we introduce the two representatives from range-based and range-free algorithms that we proposed in (An, 2006; Wang, 2006). Localization with Multiple Devices in a BASN Small BASN devices normally transmit with on-chip antennas. These antennas’ transmission pattern is far from perfect omnidirectional (Tmote sky datasheet, 2006). We verified the deficient coverage of the antenna by putting the transmitting and receiving motes 4m apart within the line-of-sight (An, 2006). We measured the RSSI values at 8 different antenna directions in steps of 45°. In Fig. 1, we show that the antenna has the strongest strength of -50 dBm at 0° and the smallest signal strength of about -65 dBm at 90°. We observed that RSSI varies in a range of around 15dBm for the static case. This problem can be solved by continuously rotating the antenna direction and taking the average of the RSSI values received from different directions. However, this method is neither agile nor feasible. Since there may be many wireless sensor devices in a BASN, we can actually use the RSSI from different devices to calculate the location with the average RSSI. Or we can let the devices calculate the location independently and take the average of the calculated locations. Both ways may mitigate the estimation error induced by the deficient antenna radiation pattern. We will show the performance of both methods in Section V. Using multiple devices can also reduce the estimation error caused by fading, since the signal is transferred in different paths. We can use the so-called micro diversity, which refers to that the antennas are at a distance of an order of a wavelength, to combat fading. Figure 1. Antenna Orientation Effect (An, 2006) Range-based Localization Algorithm We proposed an enhanced triangulation algorithm in (An, 2006). The optimization is that we give anchors which have higher RSSI to a target higher weight so that they are trusted more and used firstly to calculated a location. The next question is how large weight we shall assign to an anchor. Since we do not know the reliability of any measure RSSI it is hard to assign the weight. We proposed to use the following way. We firstly have to plot out an empirical curve showing relation between RSSI and distance, as shown in Fig. 2. All the distances were estimated by this relation with received RSSIs. We can see from Fig. 2 that the higher the RSSI the steep of the curve is. We use the slopes of the curve as the weight of an anchor. To be specific, we record all the RSSIs received from anchor i and mapped them to the empirical relation curve. For each of them we have a slope of the mapped segment, i.e. if we have a RSSI of -68 dBm then we can get the slope of segment 4 meter to 6 meter. We get the weight of the anchor when it receives the j RSSI as