Paper Title (use style: paper title)

1 downloads 3 Views 599KB Size Report
developed file system [12], based on UCBs Matchbox file system [13], is being used, which makes the storage system simple and efficient. Additional sensors ...

wildCENSE: GPS based Animal Tracking System Vishwas Raj Jain, Ravi Bagree, Aman Kumar, Prabhat Ranjan# Information and Communication Technology Dhirubhai Ambani Institute of Information and Communication Technology Gandhinagar, India {vishwas_jain, ravi_bagree, aman_kumar, prabhat_ranjan}

Abstract—wildCENSE is a Wireless Sensor Network (WSN) system which attempts to monitor the behaviour and migration patterns of Barasingha (Swamp Deer). The system would collect the micro-climatic as well as positional information of the animal and communicate it to a base station through flooding of data using peer-to-peer network. The base station, using a gateway, will upload all the collected data to a database server on Internet and portray the information using browser based visualization software. Each node would monitor five parameters namely position (using GPS), temperature, humidity, head orientation and ambient light. Also, the node will have a real time clock for the synchronization of the network and to keep timing information. An external data flash memory would be used to record the data collected from sensors and other peer nodes. A radio transceiver would transmit the data to the base station by using a peer to peer communication protocol. A solar energy harvesting system for recharging node’s power supply batteries is being added to prolong the lifetime of nodes. The system would be integrated in the form of a collar that can be easily fitted on the neck of animal. Keywords- GPS Tracking; wildCENSE; micro-climate sensing; Wireless Sensor Networks; wildlife tracking

I. INTRODUCTION Wireless sensor networks (WSN) invariably employ sensing from spatially distributed autonomous nodes. With a little jugglery of sensors, micro-controllers, radio transceiver and an energy source, low-power and inexpensive sensor nodes (we’ll simply call them nodes) can be made to cooperatively monitor physical or environmental conditions, such as temperature, sound, vibration, pressure, motion etc. at different locations. The task becomes more challenging when the nodes are mobile. To further question the engineering effort is the case, where the node’s power supply should be sufficient for it to last years. Hence the acquisition, accumulation and relay of data impose a great challenge on the WSN's design viz. the strict energy constraints. The recent past has seen a wide variety of WSN applications namely Habitat monitoring, Seismic Detection, Environmental monitoring, Health monitoring systems etc., of which mobile nodes, dynamic network topology, communication failure, limited power supply, harsh environmental conditions are few of the varied challenges. To address the issues in wildlife monitoring and to understand the complex relationship of animals with their surrounding, scientists had to collect the required data

The Project is partially funded by Wildlife Institute of India, Dehradun

manually by visiting the site. In some cases the search was made easier by tagging the animal with radio transmitters to relocate them easily, but yet the seemingly bigger part of the picture remained un-addressed: the efficient data collection. There are numerous reasons why it is difficult and not advisable to visit the site frequently. Firstly, studying the species without avoiding human contact is almost impossible. Frequent human visits or disturbances affect the species in ways unknown [1]. Secondly, keeping track of animal activity in the dark after dusk become more of an adventure than an experiment or a study. Finally, it is not only time consuming but also money intensive job to keep track of animal migration as well as its feeding habits without using dedicated low cost sensor networking equipments. An automated system would thus be desired, equipping natural spaces with numerous networked sensor nodes to enable long-term data collection at times (even at night), scales and resolution which are very difficult if not impossible, to achieve by manual monitoring. It also allows collecting data without disturbing the ecology and yet represents a substantially more economical method for conducting longterm studies than traditional one. Significant proofs of concepts and previous attempts to monitor wildlife movement and habitat have been made like the ZebraNet [2] and Great Duck Island Experiment [1]. Learning from the experiences of the aforementioned, wildCENSE is an attempt on the same footprint, designed to have lower power consumption, better range, varied environment sensing features and more robust data backup system. We briefly mention here some of the previous work. Wildlife monitoring: ZebraNet [2] is a classic example of WSN, using GPS to monitor the behavior of zebra. wildCENSE is a completely new design of the nodes based on different frequency of communication and using various additional sensors to get insight into the animal migratory pattern. Another major difference is the migratory path is not restricted and can span over thousands of kms. Other applications and deployments: In 2002, an attempt was made to study the habitat and population of birds on Great Duck Island [1]. It deployed 43 motes for the same. Energy Scavenging: In the past numerous attempts have been made to scavenge energy for sensor network node from environment. Prometheus [3] is one such example of it, which uses energy storage elements, such as super capacitor, and intelligence of microcontroller to efficiently harvest energy

from solar cells. Prometheus deploys two stage storage systems with super-capacitor at primarily level and lithium battery at secondary level. wildCENSE is a WSN system which attempts to monitor the behavior and migration patterns of Barasingha (Swamp Deer). System being designed can be suitable for many more species of medium to large size. Equipped with a GPS, Radio transceiver and various other sensors, the hardware is designed to support the needs of wildlife monitoring. The captured data can be provided to the wildlife researchers for their research and study purposes. It will be helpful to them to understand the needs of the endangered species, and the relationship these species share with the surroundings. The paper fundamentally discusses the hardware and software design architecture of the wildCENSE system at the node, base and network levels. In particular, it embodies the issues and constraints, which were met during the design and testing of the system. Section 2 of the paper discusses the design parameters taken into account for wildCENSE. Qualitatively and quantitatively the system is described in Section 3. Section 3.1 and 3.2 cover the hardware and software ends of the system, while 3.3 details the energy management. Experience gained, System performance and our field testing results are covered in Section 4. Finally we conclude by enumerating the challenges and experience gained from inception to tracking. II. GPS BASED ANIMAL TRACKING SYSTEM The Barasingha is native to India and Nepal. Once it populated throughout the basins of the Indus, Ganges and Brahmaputra rivers, as well as parts of central India reaching out till the river Godavari. But in past few decades its population has declined significantly listing them as endangered species by IUCN from 1984 to 1996 and as vulnerable since 1996 [4]. Wildlife researchers while surveying Jhilmil Tall (Uttaranchal) area came across some 30 heads of the Barasingha on February 3, 2005 [5]. Trails indicate that they might have migrated across the Nepal border. But yet their exact migratory path is unknown; hence it is important to monitor their movement and protect them. wildCENSE is an attempt towards discovering this path along with helping the researchers know more about their habitat. The input for the design of the system came from wildlife researchers. In the system proposed, specially designed light-weight collar, with sensor node attached, would be put on the animal. These collars would collect data about the desired parameters from the vicinity of the animal. The data may then be sent to other nodes or the base station, depending on the availability of either (base station getting a preference in case of both being available). The prime requirement is to track the migratory movement of the animal which is done using a Global Positioning System (GPS). Besides the location, the animal's habitat and its ambient environment parameters are of interest. Also a study of the animal’s activities viz. the grazing patterns of the animal needs to be recorded. Since the area under surveillance is large, the acquired data needs to be propagated on a node to node basis until it is transferred to a base station. Lastly the system

needs to run continuously for a minimum of 12 months, so the power supply design and its usage need optimization. Positional Logs: The GPS reading needs to be accurate and precise, in view with the migration pattern of the animal. As researchers specify, a location reading every 3 hours would be enough to draw a close enough movement track of the animal over a year. Ambient Environment: With the animal covering a lot of ground during its migration over the year, the researchers also need to monitor the environment in which the animal dwells and grazes. Sensors for measuring the temperature, humidity, and light as well as animal activity are embodied in the system. Data Transmission and Recovery: To collect the dispersed data for analysis by the researchers, it needs to be transmitted to the base station(s). Since the Barasingha has a fairly large movement track it is not possible to equip the entire region with numerous base stations. To address this issue, the data needs to be moved through the network, employing node to node communications as was attempted in Zebranet[2]. In order to compensate for high latency, the node has a large external flash to accommodate data generated on the node as well as acquired through peer interaction. Section 3.1 discusses the communication in more detail. Energy harvesting: The nodes need to be alive for a minimum of a year, tracking the migration path, avoiding any human intervention. Their only contact is the wireless link with other nodes or the base station as the case may be. Also, since there is a limitation on the weight of the node, a bulky power supply is forbidden. Hence, the node needs to have lightweight power back up system. Given that the animal will mostly be in large fields under open skies, the required power supply could be equipped with solar energy harvesting features. With careful energy management policy, supplemented by harvesting, the energy requirements can be easily met. The power supply is discussed in more detail in Section 3.3 III. System Overview Broadly the wildCENSE system is divided as in Figure 1, namely the hardware, related system software and drivers, middleware servers with data logging and web hosting services and finally the browser based visualization software. 1) Hardware Architecture The complete sensor node along with the battery recharging system is in the form of a collar to be worn by the animal. Hardware system architecture of wildCENSE node is as depicted in Figure 2. The design issues as discussed in Section 2 have been carefully met. Each component has been carefully selected based on earlier prototypes to meet accuracy, power, voltage compatibility and cost considerations [6]. The components that make up a single node are as follows: Micro-controller – ATMega1281V [7], with 128K bytes program memory, is the core processing unit of our design. It has 4K bytes of EEPROM and 8K bytes of SRAM. The availability of 2 USART ports enables independent

Browser based Visualization System Data logging onto servers and backend hosting of the data on the internet.

Base station and Gateway to transmit data to servers. System software and component drivers.

Hardware including sensors, power supply, solar cells, GPS, radio and other circuitry. Figure 1. wildCENSE System Overview

communication of GPS and Radio transceiver with the core processing unit simultaneously. This allows us to remove the multiplexing overhead as described in the software section 3.2. The internal resonator is not accurate enough for serial communication, so an external crystal of 1.83728 MHz is used. (limiting baud error to zero percent [7]). Real Time Clock - DS3231 [8] - For node discovery, all the nodes need to wake up at the same time requiring them to be synchronized. External RTC is required to accurately synchronize these nodes. It also generates periodic interrupts to wake up the micro-controller from its “Power Down” sleep mode. Features like extreme accuracy, integrated temperature compensated crystal oscillator (TCXO), I2C interfacing at different baud rates make the device ideal for this application. Any mismatch in the time in the system (between two interacting nodes) can cost a lot of power in network synchronization. RTC running on different nodes are skewed due to the environment. To maintain the accuracy of node to node communications, the RTC is synchronized by the GPS device every five days, keeping the clock skew within 1 sec. GPS – Lassen iQ GPS Receiver with antenna [9] – It has a small footprint, low energy consumption (89mW at 3.3V) in active mode. To achieve high accuracy, it uses twelve processing channels to track the GPS satellite signals. Lassen iQ GPS supports the required NMEA protocol with GPRMC message format, which contains all the required information namely date, latitude, longitude and time. It serially communicates with the microcontroller at 4800 bps. Our GPS is used in On/Off mode since readings are taken every 3 hours. To utilize “Warm start” feature of GPS, we use a battery backup mechanism. Radio Transceiver – XBee-Pro [10] - This Digi-Key communication module is based on ZigBee/IEEE 802.15.4 standard. It operates at 2.4GHz (only freely available ISM band in India), providing a range of more than a kilometer. While using this frequency results in higher power consumption for same range compared to 900 MHz, we gain in terms of much higher data rate and smaller compact antenna. Low cost, low power and ease of use are among the other advantages. It also provides five sleep modes to meet various needs of different applications. We use lowest power sleep mode as it is not a time but power critical system. Delay of few

Figure 2. wildCENSE Hardware setup depicting various components, their interfacing and power supply.

milliseconds of wake up are well within the system’s tolerance Memory – ATMEL AT45DB16B Data flash [11] A high memory storage is required to complement the long latency of communication between the base and the node. For our WSN, a node needs to collect data from its peers, asking for a higher memory capacity. AT45DB16B, with SPI interfacing looks quite promising for the scenario. An operational in-house developed file system [12], based on UCBs Matchbox file system [13], is being used, which makes the storage system simple and efficient. Additional sensors - To collect the ambient environment parameters of the animal, the node mostly incorporates digital sensors, as from the experiences of Great Duck Island Experiment [1]. The humidity sensor has an inbuilt heater to evaporate absorbed water. Among the set of sensors are the Sensirion’s SHT11 [14] which is a digital temperature and humidity sensor (Resolution: 0.01°C and 0.05% RH). This sensor is shielded by a cap (IP67 standard) which lets it sense the environment but at the same time protects it from the same. We use a high sensitivity digital light sensor from TAOS, TSL2561t [15]. To monitor the activity of the animal an analog accelerometer from Freescale Semiconductor MMA6270QT [16] is used. This data along with the position logs provides more insight into the migration pattern of the animal as well as its micro-climatic preferences. The node has been designed employing numerous noise reduction techniques. To reduce the ADC noise, a LC filter (L=10mH and C=0.1uF) has been added to the ADC pins of the micro-controller. Also, the AVCC is connected to the main power supply without any in between fan out lines, to reduce noise [6]. The whole PCB has copper pouring to keep the noise at a minimum level as also to dissipate any heat generated by the node. Figure 3 depict the node. The size of the node is 5 x 6cm2, weighing only 34gms. Including the power supply (Liion battery with solar charging mechanism), the total weight of the system, excluding the collar is less than 300gms (using 4 Li-Ion cells weighing 148 gms).

(b) (a) Figure 3. wildCENSE node, (a)top view (b)bottom view

2)Software Architecture Addressing the main design constraint, the energy; wildCENSE software implements effective scheduling and synchronizing of events. The node is kept in sleep/inactive mode for most of the time. Desired data is collected from sensors and GPS on the basis of periodic interrupts generated by a real time clock (RTC). The accuracy of the RTC helps the node in synchronization during node to node interactions as also with the base. GPS samples are taken every 3 hours and the wake up of respective sensors is scheduled every ten minute. Since the Radio transceiver, GPS and external flash memory are on separate ports namely USART0, USART1 and SPI, we can afford to use them simultaneously. While the GPS is switched on and tries to get a fix, a radio transmission could be achieved if another node/base is discovered in the vicinity. The pointers to current read/write data segment on the flash are stored in the EEPROM of the micro-controller. This is advantageous in restart situations (e.g. a watchdog reset), the pointers to data segment can be retrieved from EEPROM. All of the nodes wake up at the same time and the data exchange process is initiated, once the node/base discovery is done. Data transferred to the base station is successively deleted from the nodes. The communication baud rate between the micro-controller and radio transceiver is set to be 57600, though a higher baud rate of 115200 is supported by this radio transceiver. Our test results show better data reliability at the lower baud (57600) and hence the trade-off. 3)System Energy Management WSN energy requirements are by far the most critical and important design considerations. In case of wildlife monitoring the scenario becomes more challenging, where a trade-off is needed between the weight of the node and its energy requirement. In this section we discuss our techniques for energy management, at the software level and the hardware level as well for wildCENSE. An instance of the node’s power consumption is shown in Figure 4. The details of energy consumption calculation are given in part C of this section. A. Hardware Level Energy Management The desired node is populated by number of sensors along with other peripheral including RTC, external flash memory and radio transceiver. Designing a simple power supply for such complex system was a challenge. All components and sensors were carefully selected to have low energy profile and almost similar input supply range with 3.3V as the common voltage. The system is powered by a re-chargeable Li-ion battery, which can safely have voltages from 2.7 V to 4.2 V. Solar power is being added to further enhance the node life.

Figure 4. Power Consumption breakdown of a wildCENSE node

Also the unused pins and digital input buffers are configured as output pins and disabled respectively, to minimize their energy leakage. The decision of using a common voltage (3.3V) not only made the power supply for the node simple but also saved lots of energy, which otherwise, would have been wasted in regulating it for different voltages. To utilize the battery energy to the maximum, a DC/DC converter, TPS63001 buck booster from Texas Instruments [17], is used. It provides a constant 3.3V output with a maximum of 1.8A of current; being rated up to 96 percent efficient. B. Software Level Energy Management wildCENSE employs “Power Down” sleep mode of the micro-controller for the time the node is inactive, thereby saving substantial amount of energy. Another option could be to use a dual clock scheme instead of sleep mode, but while using an external Real Timer Clock, we are able to save the required power in our “Power Down” sleep mode itself. The Power Down mode features putting everything to a complete shutdown, including the clock source [7] and typically consumes less than 10uA of current with “Watchdog” enabled at 3.3V. Brown-out detector remains the only analog module in terms of power consumption during a sleep mode state [18]. But since even this module is not required in our design we have turned it permanently off ensuring further lower powered “Power Down” sleep mode. Another power saving mode, the “Power Save” sleep mode is employed when smaller delays are required between processing and reading sensors. Along with microcontroller other peripheral are also put to sleep mode to minimize energy usage. The GPS is switched off while the node is in sleep mode. This is implemented by the use of power switch, TPS2092 [19]. The application being not very time critical gives us the opportunity to put the radio transceiver in lowest power mode which consume about 5 times less power than the other modes. C. Node lifetime estimation Table 1 explains the power requirement of various components of the node. The following assumptions have been taken in estimating the energy requirements of the node for it to survive minimum of one year:

The node takes measurements every 3 hrs.

It is assumed that most of the time, the Barasingha, would be under open skies, getting a clear GPS lock.

The node tries to discover other nodes/base every hour, synchronized by the RTC.

Only 70 percent of the total rated Li-ion battery energy has been assumed usable [20].

Lastly, we assume that a node per day will receive a maximum of 7 pages from its peers and transfer one page to the other. In a month’s time the node will transfer all its content to the base

With the above assumptions and using the stated hardware, the node requires a total of 13.5mAh energy per day or 7040 mAh in a year. To meet the above requirements, Li-ion battery pack of 8Ah capacity is sufficient. Solar charging would further improve the life time of nodes. In the Table 1 we will represents Transmit mode by “Tx”, Receive mode by “Rx” and Power Down by “PD”. The average current (in milli Ampere) requirements of components in different modes are given in Column 4. Column 5 “T” gives the typical time (in sec) taken by sensors and other peripheral to take single reading. Last column i.e. “C” is the per day current requirements (mAh) of components in different mode. IV.FIELD TRIALS In order to test our system in a real environment we deployed one of our sensor nodes on a cow in the nearby forest. The woods simulated the final deployment area. Testing was done on the domestic animal instead on wild animal at this stage (before deploying on final species) since at this stage we are testing our device and it’s important to be able to deploy and retrieve the node easily. This is not possible once we deploy it on wildlife. Verification of data obtained equally crucial at this stage, which can be easily done on domestic cattle.


For testing purposes, we programmed the node to take reading every 3 minutes and try to communicate with the base station after every 30 minutes. Among the major problems faced, we found that battery drained quickly due to sleep mode not being activated. The beltdesigned was loose making the node shift from side to side of the neck, resulting in skewed neck movement readings and also increasing the fixing time of GPS.The ambient environment sensors were put on one end of the node; hence with the node shifting sideways, the sensing became more localized and got affected due to the body of the animal. Trial 2 (Sept 2-4, 2008) - In the second field trial (shown in Figure 5), we rectified all the shortcomings of the previous trial. An improvisation in the belt was made. The battery pack was hung at the lower end and the node at the top, resulting in the node being always at the top. Also a provision to attach the node with the horns of the animal was made to further make its location static from the cow’s reference frame. This trial continued for 46 hours without any human intervention and

Trial 1(Aug 22, 2008) - A collar belt along with the node and power supply was prepared and care was taken to ensure the radio transceiver and GPS were facing perpendicular to the node towards the sky.

Figure 5. Field trial on Sept 2, 2008

Figure 6. 24 Hours animal track of Sept 3, 2008 on Google map. Sensor reading at one of the instances is expanded, depicting respective temperature (°C), humidity (%RH), light (lux) and accelerometer readings (deg

data was collected and uploaded to our servers via the base station. Figure 6 shows the data collected with the visualization software developed using Google Map APIs. We have expanded an instance to get readings of various sensors at that point. The readings were counter checked by manual readings which were taken from a distance. Also our GPS readings were within a precision of 15 m, which is well within the tolerable range of 50 m. Readings from various sensors are depicted in Figure 7. The sensor data closely co-relates to the ambient environment readings which were taken manually for verification. Temperature readings varied around 30 to 45 °C as local to Gujarat state. Light sensor data is scaled on the secondary axis varying from 0 to 25000 lux.

ACKNOWLEDGMENT We would like to thank Bharat Jethwa of GEER foundation (Gandhinagar) for always being available when needed. Discussion with WII researchers P R Sinha (Director), S P Goyal, K Sankar, B Pandav, Q Qureshi and others have been very helpful. We gratefully acknowledge S Chandola (Chief Wildlife Warden, Uttarakhand) for explaining the various issues related to the protection of Barasingha. We would also like to thank Chris Sadler for sharing his experiences with Zebranet project. We would also like to acknowledge tremendous contribution made by earlier team members of wildCENSE, specially Prabhat Saraswat and Obulpathi Naidu.. REFERENCES [1]

V. CONCLUSION This paper presents an operational prototype for wildlife monitoring using WSN. wildCENSE is compact, accurate and does energy efficient sensing. Besides being energy efficient, it provides detailed position logs with a very high accuracy. The software protocols and the hardware implementation have all been carefully crafted to optimize the systems energy requirement. Though units like the GPS and Radio transceiver consume considerable energy, utilizing the solar recharging mechanism, node lifetime would be enhanced. We would now proceed to deploy nodes on Cheetal (Spotted Deer) and test node to node communication as well as reliability of nodes to survive in harsh conditions. This is being done since our target animal Barasingha is in endangered species list and permission to deploy nodes on them are very rare and we cannot afford to experiment on them. This trial would also give wildlife researchers a glimpse into social behaviour of Cheetal. Once we are satisfied with this trial, we would proceed to deploy nodes on Barasingha. This system would be useful on many other wildlife species and would be tried out by WII researchers, whenever opportunity comes along.



[4] [5] [6]

[7] [8] [9] [10]

[11] [12] [13] [14]

[15] [16]

[17] [18] [19] [20] Figure 7. Sensor data readings spectrum over 46 hours.

Alan Mainwaring , David Culler , Joseph Polastre , Robert Szewczyk , John Anderson, Wireless sensor networks for habitat monitoring, Proceedings of the 1st ACM international workshop on Wireless sensor networks and applications, September 28-28, 2002, Atlanta, Georgia, USA. (references) Pei Zhang , Christopher M. Sadler , Stephen A. Lyon , Margaret Martonosi, Hardware design experiences in ZebraNet, Proceedings of the 2nd international conference on Embedded networked sensor systems, November 03-05, 2004, Baltimore, MD, USA X. Jiang, J. Polastre, and D. Culler, "Perpetual Environmentally Powered Sensor Networks," in 4th International Conference Information Processing in Sensor Networks (IPSN'05), Los Angeles, CA, 2005 wildCENSE Project Report [HTTP online Document] T.pdf Atmel ATMega1281 Datasheet [HTTP Online Document] DS3231 Timer module [HTTP Online Document] http://www.maxim¬ Trimble Lassen iQ ® GPS Module [HTTP Online Document] XBee-PRO OEM RF Modules Product manual [HTTP Online Document] AT45DB16B Data Sheet [HTTP Online Document] Wildcense V 2.0 – [HTTPOnline Document] The Matchbox File system [HTTP online Document] SHT11 [HTTP online Document] TSL2561t [HTTP online Document] MMA6270QT [HTTP online Document] f TPS63001 [HTTP online Document] TPS2092 [HTTP online Document] Carter B. , Matsumoto J. , Prater A. , Smith D. , Lithium-Ion battery performance and charge control, Proceedings of the 31st Intersociety Energy Conversion Engineering Conference, 1996. IECEC 96. J. Clerk Maxwell, A Treatise on Electricity and Magnetism, 3rd ed., vol. 2. Oxford: Clarendon, 1892, pp.68–73.