Development of a Testbed for Wireless Underground Sensor Networks

Transcription

1 1 Development of a Testbed for Wireless Underground Sensor Networks A. R. Silva and M. C. Vuran Abstract Wireless Underground Sensor Networks (WUSNs) constitute one of the promising application areas of the recently developed wireless sensor networking techniques. WUSN is a specialized kind of Wireless Sensor Network (WSN) that mainly focuses on the use of sensors underground. Some recent models for the wireless underground communication channel were proposed but few field experiments were realized to verify the accuracy of the models. The realization of field WUSN experiments proved to be extremely complex and time-consuming and also presents novel challenges in comparison with the traditional wireless environment. To the best of our knowledge, this is the first work that proposes guidelines for the development of a WUSN testbed aiming the improvement of the accuracy and the reduction of time for WUSN experiments. Although the work mainly aims WUSNs, the majority of the presented best practices can be also applied to WSNs. Index Terms Wireless Underground Sensor Networks, WUSN Testbed, Transitional Region, Standardized RF Measurements. 1 INTRODUCTION WIRELESS Underground Sensor Networks (WUSNs) are a natural extension of the wireless sensor network (WSN) phenomenon to the underground environment. WUSNs have been considered as a potential field that will enable a wide variety of novel applications in the fields of intelligent irrigation, border patrol, assisted navigation, sports field maintenance, intruder detection, and infrastructure monitoring [1]. Despite their potential, very few field experiments have been realized, delaying the proliferation of WUSN applications. One possible explanation for the lack of a significant number of field experiments for WUSNs is that such experiments proved to be extremely complex and present novel challenges compared to the traditional wireless environment. Moreover, constant changes in the outdoor environment, such as the soil moisture, can contribute to the problems related to the repeatability and comparisons between WUSN experiments. A WUSN testbed was built in University of Nebraska- Lincoln City Campus on a field provided by the UNL Landscaping Services during August-November 2008 period and the experiments [6] followed the guidelines described in this work. Based on the experiences acquired from hundreds of hours of WUSN experiments in this testbed, this work presents the details about the development of an outdoor WUSN testbed. To the best of our knowledge, this is the first work that proposes guidelines for the development of a WUSN testbed to improve the accuracy and to reduce the time for WUSN experiments. The recommended practices in this work range from radio frequency (RF) measurements using sensor nodes to the use of paper/plastic pipes in the A. Silva and M. Vuran are with the Department of Computer Science and Engineering, University of Nebraska-Lincoln, Lincoln, NE, s: asilva@cse.unl.edu and mcvuran@cse.unl.edu experiments to facilitate the installation and removal of the buried sensors. The main objective of this work is the proliferation of best practices in the area of WUSNs aiming: The improvement of the accuracy of WUSN experiments. An easier and standardized way to proceed with comparisons between WUSN experiments. The time reduction for the WUSN experiments through the use of a WUSN testbed. The future establishment of a standard methodology for WUSN measurements. The rest of this paper is organized as follows: In Section 2, an overview of a WUSN testbed and its physical layout are presented. In Section 3, diverse aspects to be controlled in a WUSN experiment, such as the digging process, the soil composition, the soil moisture, the antenna orientation, and the transitional region are discussed. In Section 4, detailed guidelines to preserve the quality and accuracy of the experiments, even when sensor nodes are used as RF measurement tools, are presented. The overall architecture of a WUSN testbed and the aspects of its software are provided in Section 5. The preparation for the experiments and the real results of an outdoor WUSN testbed are presented in Section 6. Finally, the conclusions are discussed in Section 7. 2 PHYSICAL LAYOUT OF A WUSN TESTBED The communication scenarios in WUSNs are classified into three types, depending on the direction of the communication: Underground-to-underground: the communication occurs entirely using the soil medium, as illustrated in Fig. 1(a).

2 2 Fig. 1. The three communication scenarios supported by the WUSN testbed: underground-to-underground (a), underground-to-aboveground (b), and aboveground-to-underground (c). The distance between the center of the antenna of the buried sensor and the soil surface is called burial depth (d bg ). The symbol d ag refers to the distance between the center of the antenna of the aboveground device and the soil surface. The symbol d h refers to the horizontal inter-node distance between the sender and the receiver. Underground-to-aboveground: the sender is a buried sensor node and the receiver is an aboveground device, as illustrated in Fig. 1(b). Aboveground-to-underground: the sender is an aboveground device and the receiver is a buried sensor node, as illustrated in Fig. 1(c). In addition to supporting the above types of communication, a WUSN testbed must allow an easy configuration of the physical deployment aspects. As shown in Fig. 1, these deployment parameters reflect the location of the sensor nodes. The parameter d bg, also called burial depth, is defined as the distance between the center of the antenna of the buried sensor node and the surface of the soil. The distance above the ground d ag, used in the underground-to-aboveground and abovegroundto-underground scenarios, is the distance between the center of the antenna of the aboveground device and the surface of soil. Finally, the parameter d h is the horizontal inter-node distance between the sender and the receiver nodes. From these definitions, it is clear that from the communication perspective the antenna is the element of interest. In fact, the actual locations of the sensor, processor, and transceiver modules are not essential in a WSUN testbed. In Fig 2, four different schemas for the deployment of sensor nodes are shown. However, all these sensor nodes have the same burial depth because the distance of the center of the antenna to the surface is the same in all cases. Fig. 2. The burial depth of a sensor node depends exclusively on the antenna location. The location of the sensor module itself (if existent) and the radio/processor of the node are not essential. One very important concept in wireless communication testbeds is the grid. The basic idea is to perform multiple simultaneous point-to-point tests, speeding up the overall time spent in an experiment. The Fig. 3 illustrates the grid concept applied in a WUSN testbed initially designed for underground-to-underground experiments. As shown in Fig.3(a), one of the sensors temporarily has the role of sender and it broadcasts a sequence of test messages. Only one node can be select as a sender for each experiment. The remaining nodes in Fig.3(a) are potential receivers. After the end of the test, it is possible to verify the results of the experiments consulting each receiver individually. However, the schema in Fig.3(a) hides a problem: the results from the experiment may not be accurate. In Figs. 3(b) and 3(c), it is illustrated how the grid could interfere in the results. If the grid is only formed by the 2 nodes in Fig. 3(c), no obstacle would exist between them. However, as shown in Fig. 3(b), the third element in the grid represents an obstacle in the communication path between the mentioned nodes. Therefore, it is clear that the original grid idea must be modified in order to maintain the accuracy of WUSN experiments and also to provide the flexibility of having multiple simultaneous tests. A simple solution is shown in Fig. 3(d). This new schema proposes a direct line-of-sight (without obstacles) between the hole where the sender is located and the holes where the receivers are located. The difference is more clear when the top views of Figs.3(a) and Fig.3(d) are compared. With this new design, the grid imposes two constraints in the WUSN testbed: A specific hole is designated to place the senders: Only one hole presents the direct line-of-sight feature in relation with all other holes. The senders are placed at different burial depth in this specific hole (senders hole). Only one sender can be active at a given moment. At the senders hole, no receivers are allowed: If receivers are placed at the same hole as the sender, one of them can be an obstacle for the other. However,

3 3 Fig. 3. The grid concept can be applied to speed up the experiments in a WUSN testbed (a). However, one node in the grid can represent an obstacle in the communication path, thus, interfering in the results (b). Without the grid schema, this interference would not occur (c). A simple solution is to modify the positions of the holes (d). In this new schema, instead of a line of holes, direct line-of-sight is provided between a hole where the sender is placed and each hole containing the receivers. (a) (b) (c) Fig. 4. Top view layout of 10cm-diameter holes for an outdoor WSN testbed (a). The design is aiming undergroundto-underground communication with commodity WSN sensors using transmit power level up to +10dBm. The 5 holes in the central part (vertical) are only used by sender nodes. The remaining holes at the horizontal line are reserved for receiver nodes. At the right side of the hole 0, there are the holes used for redundant receivers. A is the redundant node for A and both nodes have the same inter-node distance. Two examples of experiments are shown: inter-node distances varying from 18 to 150cm (b) and inter-node distances varying from 52 to 155cm (c). if there is not any sensor between the sender and a specific receiver at the same hole, the experiment can be realized. Considering the dimensions of the sensor nodes and the communication constraints empirically verified in [6], the physical layout for a WUSN testbed shown in Fig. 4(a) is proposed. The layout is presented in a top view, where each circumference is actually a hole. The use of multiple sensors in the same hole, shown in Fig. 3(d), is optional. In fact, for some experiments, all the tests are performed for a unique burial depth. In this case, every hole in the proposed layout would only contain one sensor. The 5 holes in the central part of Fig. 4(a), going from 0 to 50cm, are only used by sender nodes. Only one of these holes can contain an active sender for an experiment. The horizontal holes in Fig. 4(a), with the exception of the one marked with 0, are assigned to contain receivers. Multiple receivers holes can be active in an experiment. The holes at the right side of 0 are reserved for redundant receivers. The presented layout for the WUSN testbed considers the use of 10cmdiameter holes and commodity WSN sensor nodes with a maximum transmit power of +10dBm. Naturally, the distances can be modified if larger and more powerful sensor nodes are used. As shown in Fig. 4(a), the testbed is proposed to provide: Direct line-of-sight between sender and receiver without any artificial obstacle. Simultaneous experiments for different inter-node distances and, optionally, different burial depths. High accuracy in the results through the redundancy in the measurements. In Fig. 4(b), an example of an experiment is shown. The following inter-node distances between the sender and receiver are possible with this specific schema: 18, 32, 46, 61, 76, 91, 105, 120, 135, and 150 cm. In Fig.4(c), additional inter-node distances are possible only changing the location of the active sender: 52, 58, 67, 78, 90, 103, 112, 126, 141, and 155 cm. In both Figs. 4(b) and

4 4 4(c) it is observed the redundancy feature of the testbed. The same inter-node distance is used for the receivers A and A. After the end of the experiment, the results of the receiver A are expected to be very close to the measurements from the sensor A, assuming they have the same burial depth. 3 CONTROLLING THE MULTIPLE VARIABLES IN AN OUTDOOR WUSN TESTBED This section provides the reasons for the common errors in realizing WUSN experiments. The challenges of burying and unburying sensor nodes are first presented and the use of paper and plastic pipes are proposed. Next, specific issues related to the underground-toaboveground and aboveground-to-underground experiments are discussed. The analysis of the soil texture and soil moisture of the WUSN testbed is included as an essential part of the results of the experiments. The errors caused by the antenna orientation and the use of sensor nodes to make RF measurements are also discussed. Finally, the section presents the issues related to the transitional region of WUSNs. Moreover, for each variable which can potentially cause errors in the WUSN testbed, specific guidelines are provided. 3.1 The Digging Process Burying and unburying sensor nodes are very timeconsuming tasks in underground settings. For instance, in our experimental testbed, almost 2 hours were necessary to dig a single 20cm-diameter, 1m-depth hole, even with the use of an electric power auger. Therefore, an initial consideration about the dimensions of the holes is necessary. The larger a hole is, the larger is the modification of the soil density at that area and this parameter affects the signal attenuation caused by the soil [6]. A second aspect is related to the depth of the hole. The majority of the WUSN applications will not require burial depths higher than 1m [6], [2], [1]. Therefore, the proposed WSUN testbed considered in this section assumes a burial depth that varies from 60 to 100cm. The process of digging deeper holes is only feasible with special machines. On the other hand, for shallow holes, there are many simple and manual digging tools available in the market considering that the diameter of the hole is restricted to up 4cm. In the case of our testbed, the required minimum diameter was 7.5cm due to the dimensions of the sensor node. Therefore, 8cm-diameter holes were dug with power augers. The difficulty to bury a sensor node also highlights an important aspect for the success of WUSN applications: the deployment of hundreds or thousands of these devices needs to be relatively simple. In this sense, sensor nodes with cylindrical form and a tiny diameter (2.5 to 4cm) are recommended. (a) Fig. 5. The built and installation of paper pipes are very simple (a). An example of the use of a paper pipe in a 10cm-diameter, 90cm-depth hole for a temporary WUSN testbed. Besides the difficulty and the time spent in the process of burying and unburying sensor nodes, the repetition of an experiment that requires positioning the sensor node and its antenna at the same place is not an easy task. This issue is aggravated with the use of small holes, such as a 10cm-diameter hole. The use of paper and plastic (PVC) pipes proved to be a good solution. In our testbed, preliminary tests at 433MHz were realized to verify how the adoption of paper and plastic pipes would interfere in the results of WUSN experiments. The comparison between the results with and without paper and plastic pipes, showed an additional attenuation ranging from 2 to 8dB. These values correspond, respectively, to the use of paper pipes and different thicknesses of plastic pipes. These values are still considered small in comparison with typical values of the soil attenuation which range from 20 to 50dB [6]. However, for different frequencies, it is important the realization of the mentioned comparison before proceeding with the experiments and subtract the additional attenuation from all the results related to RSS (received signal strength). The Fig. 5 shows the use of a paper pipe which was made from a 55x70cm poster board. With one poster board, one 9cm-diameter pipe can be mounted and covered with plastic seal tape to protect against the soil moisture and finally installed at a 10cmdiameter hole. Depending on the soil moisture, it is not expected that the paper pipe lasts more than one month, making it an option for temporary WUSN testbeds. For a permanent WUSN testbed, it is recommended the use of plastic pipes. To obtain a smaller attenuation value due to the introduction of the plastic pipe, smaller thicknesses can be used. Considering the easy construction, facility of use, low cost, and mainly, its neutrality on the WUSN experiments (the pipes only introduce a fixed attenuation in all results), these schemas with pipes were extensively used at our WUSN experiments. The above schema is only part of the solution. The paper/plastic pipe helps to preserve the physical structure of the hole for multiple experiments. However, to perform the experiments, we also have to bury the (b)

5 5 Fig. 6. Paper pipes can be used to facilitate the process of having the sensor placed at a specific burial depth; a plastic sack filled with soil can also be used (a). More than one paper pipe can be used to produce a specific burial depth for the sensor (b). Multiple sensors can be easily installed in the hole, each one with a different burial depth (c). sensor, that is, to cover it with soil. Therefore, the re-use of the hole for multiples experiments is still a problem. A solution for this issue is to use paper pipes filled with soil. In our testbed, additional 7.5cm-diameter paper pipes were used. These new paper pipes contain the same soil which was taken out from the digging process. These pipes, with both ends sealed, can have different lengths, helping to make experiments for different burial depths. Fig. 6 shows a typical use of the paper pipes in the testbed. In Fig. 6(a), a unique paper pipe is used for the experiment. In Fig. 6(b), 2 paper pipes are used in order to provide a specific burial depth for the sensor in the experiment. Finally, in Fig. 6(c), 3 paper pipes were use to realize an experiment involving 2 distinct burial depths. In Fig. 6(a), it is observed the use of a small plastic sack containing soil to complement the final stage of the hole. 3.2 Underground-to-Aboveground Experiments Some WUSN experiments involve the communication between an aboveground device and the buried sensor node. The presented WUSN testbed can also be used in this new scenario. As shown in Fig. 4(a), the presented underground-to-underground testbed has 5 special holes for the senders nodes and 20 holes for the receivers. Extending the WUSN to aboveground experiments imply that the sender or the receivers will be located aboveground. Accordingly, the grid schema must be adapted for this new scenario. The following guidelines are necessary for extending the WUSN testbed for aboveground experiments: The surface of the paper pipe must be aligned with the soil surface, as shown in Fig. 7(a). No plastic soil sack above the soil surface is allowed, as shown in Fig. 7(b). The propagation of the antenna cannot be disturbed by the paper pipes filled with soil. The mentioned paper pipes can be used but the antenna must Fig. 7. Extending the WUSN Testbed to support aboveground experiments. The antenna must be positioned in the direction of the aboveground device and without any obstacle (a). Some aspects allowed for underground-tounderground experiments are not allowed with aboveground experiments (b). The aboveground devices can be easily installed through the use of a small PVC pipe and a wood stake (c). positioned in a way that it points into the direction of the aboveground device(s), as shown in Fig. 7(b). Use a hole with a direct line-of-sight without obstacles to the aboveground device(s). It is recommended the use of the sender hole marked with 50 cm in Fig. 4(a) for installing the buried nodes. The aboveground nodes devices can be easily installed using a PVC pipe in conjunction with a wood stake. The Fig. 7(c) shows a tower of receivers at different heights (d ag ) for an underground-toaboveground experiment. All the devices and schemas presented in this Section proved to be speed up the realization of our experiments. Without these schemas, the same experiments would last more than 3 times. At the same time, the accuracy of these experiments was not compromised. 3.3 Soil Texture and Soil Moisture The characteristics of the soil have a strong influence on the signal attenuation [3], [1], [2], [6]. As a consequence, WUSN experiments realized without the characterization of the soil are incomplete. In parallel with the preparation of the testbed, soil samples must be collected and sent to a specialized laboratory for soil analysis. The soil texture analysis provided by the laboratory presents very important parameters to be added in all results from the testbed. In Table 1, the soil analysis from our testbed performed by a specialized laboratory [15] is presented as an example. Besides the soil texture, the volumetric water content (VWC), or soil moisture, is other parameter to be included in every WUSN experiment report. However, different from the soil texture which is very stable for the same site, the VWC is dynamic and depends on the

6 6 TABLE 1 Example of a soil analysis report. These parameters from the testbed location must be included in all experimental results. Depth Organic Matter Texture %Sand %Silt %Clay 0-15cm 6.4 Loam cm 2.6 Clay Loam cm 1.5 Clay Loam environment and the weather. Moreover, the VWC also varies as a function of the burial depth [9], [8]. These facts are important because the VWC can significantly modify the results of an experiment, as suggested in [2], [6]. Although there are soil moisture sensors that can be used for the VWC measurement, a simple solution is called oven drying method [12], which consists of separating and weighing a sample of the soil used in the experiments. Then, this soil sample is completely dry in an oven and it is weighed again. The difference in the weights divided by first measurement represents the VWC in the soil sample, a number varying from 0 to 1. If no rain or artificial irrigation occurs, a unique daily measurement of the VWC may be enough because it usually does not change very quickly [8]. On the other hand, the occurrence of a sudden rain can cause real problems in the experiments. Naturally, the VWC measurements must be collected frequently to confirm the continuity of the VWC conditions. This is especially recommended when a set of experiments is partitioned into many different sessions and distinct days. The soil texture and the soil moisture must be informed together in the experiments reports. The comparisons between experiments realized in different testbeds are only feasible including with these parameters in the analysis. Moreover, significant changes in the VWC for the same set of experiments are not allowed because the results are completely compromised. 3.4 Antenna Orientation Usually, the antenna orientation is not a very critical factor for over-the-air wireless communication experiments. However, considering the extreme attenuation due to the soil propagation, small changes at the orientation of the antenna can make the difference between a high quality communication and no communication at all. Guidelines to avoid problems related to the antenna orientation in WUSN experiments are provided in this section. The antenna orientation is an additional constraint to be considered for the deployment of WUSNs, compared to traditional WSNs, especially for multi-hop underground networks, where the communication range varies based on the antenna orientation. Accordingly, the experiments in a WUSN testbed can be easily compromised if the antenna orientation is not observed. To illustrate how critical this aspect is, antenna orientation Packet Error Rate (a) Relative angles for the antenna Relative Angle of the RECEIVER relative to the SENDER (b) PER vs. relative angle for the antenna. Fig. 8. The schema used to test the effects of the antenna orientation in the wireless underground communication. The results proves that a unique antenna orientation, the best possible, must be used in all sensor nodes, during all experiments. experiments are performed by placing a sender and a receiver, both MICA2 motes [13], at different angles as shown in Fig. 8(a). The vertical polarization of the antennae was specifically adopted because preliminary tests proved that it provided the best results for our WUSN testbed environment, however the explanation in this section also applies to the case of a linear horizontal polarization of the antenna. The original antenna of a MICA2 mote is a standard one-quarter wavelength monopole antenna with 17cmlength. The radiation pattern of this sensor node and its original antenna does not exhibit a perfect sphere and matches the dipole antenna model presented in [5]. Therefore, it is expected that changes in the antenna orientation causes variations on the signal strength of the receiver node. These variations are especially significant when the underground scenario is considered. The experiments were performed at a transitional region (defined at Section 3.6), that is, at the limit of the underground communication range. In Fig. 8(b), the packet error rate (PER) is shown as a function of the node orientation. When the relative angle varies from 90 o to

7 7 340 o, the PER increases and the orientation of a node has a significant impact on the communication success. When the antenna orientation is between 120 o and 300 o, the communication between the nodes is not possible. For underground-to-aboveground and aboveground-tounderground experiments, a slight movement of the sensors or the action of the wind can easily change the antenna orientation. In these cases, all the results from the experiments must be voided. Therefore, the process of physically fixing the antenna to the case of the sensor node, for instance, using a seal tape, is highly recommended. This guideline is especially important for the aboveground sensor nodes. To avoid the interference of the antenna orientation over the experiments results, it is very important to choose a unique antenna orientation for all experiments in a WUSN testbed. In our experiments, only the 0 o orientation was used in order to eliminate the effect of antenna orientation. However, for every combination of sensor node and its antenna, different antenna polarizations and orientations can be adopted as the default configuration for all experiments. Accordingly, an experiment similar to the one shown in Fig. 8 must be performed to maintain the accuracy of the results and also to provide the recommendation of the best configuration for the sensor deployment. 3.5 Misalignment of RF Measurements In an ideal wireless testbed, the best accurate tools are selected to be used as the instrumentation for the RF measurements. However, this is not usually the case for WUSN testbeds for two reasons. First, it is a common approach in WUSNs to use the sensor nodes to cooperate and provide the most reliable and efficient communication solution. Therefore, sensor nodes are expected to be also used as network instrumentation. Second, if a special instrument, such as a spectrum analyzer, is used at the receiver side of the experiment, the grid idea cannot be applied and dozens of simultaneous tests must be performed as distinct experiments, one-by-one. The natural consequence is the increase of the time to conclude the experiments. When the same pair of sensor nodes are used, it is not expected many problems at the accuracy of the PER and RSS measurements. However, the approach of using a grid-based testbed involves the measurements from many sensor nodes. Naturally, differences between these RF measurements can cause accuracy problems. In the context of a WSUN testbed, this issue is called misalignment problem. A node is defined to be aligned with a given set of nodes, if: its PER varies, at the maximum, 10% from the average PER calculated for the set of nodes and its RSSI average varies, at maximum, +/-1 dbm from the average RSSI for the set of nodes. Usually, the nodes present different receiver sensibilities. This fact could cause the mentioned misalignment problem and the accuracy of the experiments can be definitely compromised. Considering this, a balanced approach adopted in a WUSN testbed is to continue using the sensor nodes as part of the RF instrumentation, but selecting only a subset of them. The selected nodes for an experiment are the ones previously qualified to perform the RSS measurements. Therefore, before using the sensor nodes for the WUSN experiments, they are tested in typical WSN scenarios, using over-the-air tests, in a process called qualification test. The reason for this test can be understood from the following example. Suppose that we want to test 5 receiver nodes, all placed in the same hole, but with different burial depths. The results from this experiment can only be validated if the 5 nodes previously presented almost the same answer for an over-the-air test, thus allowing distinct underground measurements for distinct burial depths, as expected. As an example of a qualification test, the following is a description of the test used in our WUSN testbed. One sensor node is assigned with the role of broadcasting (over the air) a total of 200 packets, 30 bytes each, to a set of nodes located in the same physical position and exactly with the same antenna orientation. After the test, the results are collected from each node and only the subset of nodes that have a very similar PER and average RSSI, as previously defined, are selected to participate in the experiment. However, as expected, this kind of approach has at least two drawbacks. First, the process is very time-consuming and must be repeated every new day/session of experiments. Second, usually it is not possible to use all the available nodes for the experiment, which means that the grid is constrained by the number of qualified nodes. For instance, in our experiments, using MICA2 motes, usually only 50% of the nodes used were qualified for each day of experiments. Surprisingly, the qualified nodes were not always the same group. The use of sensor nodes as instrumentation for RF measurements requires a huge effort in order to maintain the accuracy of the results. Also, the total number of nodes to be available for a WUSN testbed is significantly higher than the actual number of nodes used in the experiments. 3.6 Transitional Region of WUSNs It is well known that in traditional wireless communication (air channel) there is a region where the reliability of the signal varies, until the point where the communication ceases. It was reported that this issue is highly accentuated in WSNs and this critical region is called transitional region [11]. However, results from preliminary WUSN experiments show that the underground transitional region is significantly smaller than its air channel counterpart [6]. As already commented, the main problem with wireless underground communication is the very high signal attenuation caused by the soil [2], [6]. At the same time, usually sensor nodes present low power

8 8 RF transceivers. The combination of these factors results in a very small width of the transitional region. This fact causes problems in realizing WUSN experiments and probably is one of the main reasons for the small number of experiments in this area. The identification of the transitional region in a WUSN environment defining the limits of the communication range is tied to the burial depth of the nodes, the soil texture, and the soil moisture. For instance, in some of our underground-to-underground experiments, the transitional region presented a width of less than 15% of the maximum inter-node distance. More specifically, with a maximum inter-node distance of 100cm and a transmit power of +5dBm, the transitional region is located between 85cm and 95cm. As expected, such small distance is very critical: an imperceptible slight movement in one direction, when burying the node, causes the change from a good communication region to a transitional region, implying in results that are not stable or predictable. Therefore, if the tests are being realized very close to the transitional region, a careless manipulation of the sensors can cause significant interferences in the results. Considering all the presented facts, the recommendation is to limit all the experiments to a secure region, before the transitional region. For instance, experiments realized at the 50% of the maximum inter-node distance present very stable results and the repeatability and comparisons between experiments are really feasible in this secure region. Naturally, the exception for this guideline is when the maximum inter-node distance and the transitional region itself are the aspects under investigation in the experiments. Restricting the experiments in a secure region or, at least, knowing the limits where the results become unstable due the transitional region, is a way to preserve the quality and accuracy of the WUSN experiments. For instance, if the VWC is the variable being investigated in a specific experiment, the results from the tests in a secure region allow the researcher to have a clear idea of the VWC effects over communication. In the transitional region, the VWC effects would be mixed with other variables and the same clear analysis would not be possible. 4 STANDARDIZED RF MEASUREMENTS The guidelines presented in this section can be applied to any WSN. In fact, their relevance with this work is specifically related to the air path of the undergroundto-aboveground and aboveground-to-underground experiments. A WSN/WUSN testbed is generally used to provide the infrastructure necessary for the realization of comparisons between experimental results and predictions made by theoretical models. However, it has been reported that sensor nodes are being used to make RF measurements, usually the received signal strength (RSS) [11], [7]. This is usually necessary and desirable because many communication protocols take advantage Fig. 9. Typical receiver circuitry of a sensor node. The input section usually has a limiter circuit and it imposes a maximum RSS to be informed as RSSI. A practical methodology is proposed to avoid distortions in RSS predictions caused by the maximum RSSI. of the use of the sensor node as a RF measurement tool to make decisions related to multi-hop schemas, topology, localization, etc. However, it is possible to identify some issues related to the use of sensor nodes for such measurements. The objective of this section is to present a methodology that can be used to avoid these problems. 4.1 Clipping Effect Theoretical wireless communication models usually uses empirically determined parameters. In the WUSN testbed scenario, sensor nodes are used in conjunction with some combinations of transmit power level and reference distances, to take RF measurements for the estimation for these parameters. However, these measurements can introduce distortions in the results. The following case involving MICA2 motes was observed in our experiments and illustrates the problem. One device is transmitting, over-the-air, a signal with 0dBm of transmit power and a second device, with an internode distance of 1m, is receiving the signal with a detected 52dBm provided by the Receiver Signal Strength Indicator (RSSI) of its receiver circuitry. However, when the transmitter is configured to transmit at +10dBm, it is observed the same 52dBm level at the receiver. Naturally, no communication model using the well known Friis free space propagation model [4] will have a good agreement with this experimental data. The above problem can also indirectly appear in a similar scenario, called path loss exponent (PLE) estimation, again causing distortions. PLE expresses the rate at which the signal power decays as a function of the distance [4] and it is an important input parameter in many WSN/WUSN communication models [11]. This

9 9 (a) Transmit power level = +10dBm. (b) Transmit power level = +5dBm. Fig. 10. The clipping effect caused by the limiter circuit of the sensor nodes. Depending on the inter-node distance and on the transmit power level, the effect can occur or not. parameter is usually calculated based on many RSS measurements and must use the final devices to be used in the communication, particularly the sensor nodes and their antennae. However, what caused the issue mentioned in the last paragraph will also influence the PLE estimation and, therefore, will cause distortions between the estimations of the communication model and the experimental data provided by the testbed. In Fig. 9, a typical RF circuitry of a sensor node is shown. If a strong signal is received above a certain limit specified by the manufacturer of the sensor, a limiter circuit will operate and a maximum RSS will be informed as the RSSI level. Accordingly, different signal levels will correspond to the same informed RSSI. This clipping effect is the reason for the mentioned example with MICA2 motes at the beginning of this section. In Fig. 10, the clipping effect is illustrated in a typical WSN/WUSN scenario. In Fig. 10(a), the stronger signal at the receivers, closer to the sender, are reported as RSSI= 52dBm, although this is the maximum RSS informed by the receiver circuitry and does not correspond to the actual value of RSS at that locations. However, as shown in Fig. 10(b), for the same scenario, just modifying the transmit power level of the sender, the clipping effect is not observed. The clipping effect is challenging because it depends specifically on the hardware of the receiver devices and even when the models of the sensor nodes are identical, the nominal value of the maximum RSS informed by the manufacturer may also vary. The consequences of the clipping effect on a WSN/WUSN testbed are as follows: Incorrect interpretation of the testbed data: The communication model can predict a RSS value and the experimental data can show a smaller result. If this smaller value is exactly the maximum nominal RSS of the receiver, probably this is not a model mismatch. Accuracy loss in the model prediction: If the communication model is using the testbed to obtain certain empirical parameters, such as PLE, the results of the model will be negatively affected by these incorrect measurements. Although the first mentioned consequence is not critical because it is only related to the way the experimental data from the testbed is analyzed, the second consequence must be avoided or solved. In the next section, a methodology to calculate PLE is presented and it essentially shows how the choice of an appropriate reference distance for the measurements can avoid the mentioned clipping effect. 4.2 Estimating Path Loss Exponent using Sensor Nodes For underground-to-aboveground and aboveground-tounderground communication in WUSNs, the path loss exponent (PLE) is an essential input parameter in the communication model. The following methodology is recommended when it is necessary to estimate the PLE using sensor nodes, which is the case for the WUSN testbed. 1) Select the reference distance d 0 : The typical approach to determine the received power from the receiver node s perspective, located a distance d from the sender node, is the use of the well known Friis equation related to the free space propagation model. However, the application of this equation assumes the availability of the detailed knowledge related to the antennae gain and also the overall losses due to transmission line attenuation, antennae losses, filter losses, etc. Another more practical approach to predict the received power at a given distance d from the sender is the use of measurements in the radio environment [4]. For this approach, a reference distance d 0 from the sender node is chosen. This distance d 0 must be determined considering two simultaneous constraints: d 0 must lie in the far-field (Fraunhofer) region: The far-field region is defined as the region beyond the far-field distance d f which is defined by [4]: d f = 2D2 λ, (1) where D is the largest physical linear dimension of the antenna and λ is the wavelength of the RF wave in meters. For instance, for the MICA2 node operating at 433MHz, D is approximately 0.17m and, therefore, d f is 8.3cm. In this case, d 0 must be greater than 8.3cm. d 0 must be smaller than any distance d used in the deployment of the nodes (d 0 <d): For instance, for the over-the-air communication using MICA2,

10 10 it is usual to consider d 0 =1m because the minimum inter-node distance between the sensors is typically higher than 1m. After selecting a start point for d 0, such as 1m, the next step is to setup the sender at its minimum transmit power and take a RSS measurement at the receiver. An additional RSS measurement is taken considering now the maximum transmit power. The difference between both measurements must be approximately the nominal difference between the maximum and minimum transmit power levels used. If this goal is not achieved, a higher value for d 0 must be chosen and the above tests must be repeated. In the experiments reported in Section 6, the distance d 0 was 10m. Naturally, this value will vary for different models of sensor nodes and their antennae. Moreover, the use of multiple receivers will improve the quality of the results in the procedures described in this section. 2) Take RSS measurements for distances d>d 0 : Configure the maximum transmit power level at the sender and take many RSS measurements for inter-node distances higher than d 0. For our experiments with MICA2 motes, using +10dBm for the transmit power, two additional distances were used for the RSS measurements: d 1 =15m and d 2 =20m. 3) Apply a linear regression technique to estimate PLE (η): Using the following equation and applying Minimum Mean Square Error (MMSE) technique [4], it is possible to estimate PLE (η) to be used by the wireless communication model. Fig. 11. A screenshot of the WUSN testbed software running in a laptop. From this software, the operator configures the parameters for the experiments, such as the number and size of messages and the transmit power to be used by the sender node. The same software is used to start the experiment and collect the results. ˆp i = p(d 0 ) 10η log 10 (d i /d 0 ), (2) where ˆp i is the measured RSS for each measurement instance i. Even if the PLE is not expected to be used, the approach observed in the presented methodology represents the set of best practices for RF measurements using sensor nodes in generic WSNs. In this way, any parameter to be used in a communication model which is based on RSS measurements of sensor nodes must follow a similar approach aiming the accuracy of the investigated model. 5 WUSN TESTBED ARCHITECTURE AND SOFTWARE Due to the complexity of the WUSN experiments, it is desirable that the WUSN testbed presents a simple architecture to balance the conduction of the experiments. The following is a description of a simple, but effective, software architecture used in our WUSN testbed. One node, called manager, sends, via radio channel, the configuration data for the experiment to a node called sender. The configuration data must include the following parameters: transmit power level, delay between the messages, size of each message, and the total number of messages for the experiment. The Fig. 11 shows a screenshot of our WUSN testbed software running in a laptop. It can be observed that the mentioned parameters can be configured for each experiment. An additional field called Description of this Test is also provided to register the details of the experiment. After receiving the configuration data from the manager, the sender broadcasts the messages. After the broadcasting period, the sender informs the manager node, via radio channel, that it finishes this phase. At this moment, the operator of the experiment can request, also via radio channel, the results from each receiver node (unicast communication). The architecture is illustrated in Fig. 12. The software in the manager node stores, in a local file, the configuration data for a given experiment, the manual annotations from the operator for that experiment, and the results from each receiver. If the receiver receives a request for the results of an experiment but it did not have anything in its buffer, it returns a message to the manager informing no results, that is, PER=100%. After sending the results to the manager, the receiver erases

11 11 Fig. 12. Basic architecture of the WUSN testbed. The manager node sends the configuration to the sender node. The sender node starts the experiment broadcasting messages according to the configuration sent by the manager. After finishing, the sender informs this fact to the manager. The manager can capture, at any moment, the results (RSSI and PER) from each receiver node. its buffer. Also, if the receiver receives messages from a new experiment, it automatically erases the previous results which were not rescued by the manager. For the realization of long-term experiments, that is, experiments that are extended for a longer period of time, such as 24 hours, some modifications in the previous architectural schema are necessary. First, the operator must configure the experiment informing its longterm feature. Then, a special message is sent from the manager node to the sender node. This special message informs the sender that it must broadcast messages more slowly, for instance, every minute. The message broadcasted by the sender to the receivers also has special information. Therefore, the receivers also know that they are part of a long-term experiment. Accordingly, the receivers will store the results into their Flash memories due the fact that the RAM memory is not usually large enough to buffer all the results. Finally, the process of capturing the results must also be modified for the longterm experiments. If the radio channel is used for the transfer of long-term results packets, the process can take hours to finish. The solution is to have each receiver directly connected to the computer acting as the manager to start the dump of the experiment results. Actually, this is the unique situation that a cable (usually USB or serial) is necessary in the WUSN testbed. Each broadcasted message in a given experiment has a sequence number. When the receiver receives that message, it saves in its buffer only a summary of the message: its sequential number and the RSSI level related to the reception of the message. The RSSI information is provided by the transceiver of the sensor node as previously discussed in Section 4.1. Therefore, no matter what is the size of the message, the summary of the message has exactly the same size in the receiver s buffer. The use of sequential numbering is also a good resource because the operator can quickly identify if the lost packets are randomly spread or not. This observation could help the operator to identify, for instance, if the experiment suffered some kind of interference during its realization, allowing the repetition of the experiment or the search for the source of interference. 6 EXPERIMENTS SETUP AND RESULTS Before starting a WUSN experiment, the following aspects must be known a priori: Soil texture: This evaluation is realized just one time, for a given testbed location, assuming that the testbed is located in a homogenous soil. The soil texture report must be done for different depths, as exemplified in Table 1. Soil moisture: This evaluation must be frequently performed as mentioned in Section 3.3. Moreover, it is very important to know the values of VWC for different burial depths of the sensors to be tested. Attenuation due the use of paper/plastic pipes: This evaluation is realized just one time, when the WUSN testbed is being built. The fixed average RSS difference between the results with and without the pipes must be recorded. If they cannot be neglected, all the RSS results from the experiments must be adjusted accordingly. Default antenna orientation: This evaluation is realized one time, for a given model of sensor node and its antenna. As mentioned in Section 3.4, the best antenna orientation must be found and fixed for all experiments with that sensor. Transitional region: The range of this value will change as a function of the soil composition, soil moisture, frequency, and transmit power. The operator of the WUSN experiment must know a priori the different values for this region and avoid experiments in this region when trying to analyze a specific parameter without any additional interference, as explained in Section 3.6. The first step in the preparation for a WUSN experiment is the qualification test, exemplified in Section 3.5. After having the set of nodes to be used, the next step is the assignment of the roles for the sensor nodes. Considering that the manager node does not interfere on the results because it only triggers the start of the experiments and captures the results, the manager node can be elected randomly from the set of available nodes and there is no need to change its role. The node presenting smaller variance in its qualifying results must be selected as the sender. It is recommended the use of the same sender node for all experiments in a single session.

12 12 However, it is not recommended the continuous use of the same sender node for different experiments sessions (e.g., different days). As expected, the remaining qualified nodes can act as receivers. After the preparation phase, the WUSN experiments can be performed. The rest of this section is composed of the presentation and the analysis of the results of our WUSN experiments. These results are being presented in this section as examples of successful use of the proposed WUSN testbed and its related guidelines. An analysis of the soil texture of the testbed environment was made by a specialized laboratory [15] and the results are shown in Table 1. The experimental results were divided into five classes: underground-to-underground, underground-toaboveground, aboveground-to-underground, effects of the volumetric water content (VWC), and long-term (24h) experiments. The results of the first 3 classes of experiments are shown in Fig. 13 and the Table 2 complements the results showing the minimum inter-node distances for a PER >70%. The results of the last 2 classes of experiments are shown in Fig. 14. TABLE 2 Minimum inter-node distance and PER >70%. -3dBm 0dBm +5dBm +10dBm UG2UG 55cm 55cm 85cm 95cm UG2AG 200cm 200cm 250cm 300cm AG2UG 15cm 25cm 55cm 85cm In Fig. 13(a), the results from underground-tounderground experiments are presented. The RSS values are shown as a function of the horizontal inter-node distance for different transmit power levels. The variance of the RSS values is also provided. It is possible to verify the clipping effect mentioned in Section 4.1 in Fig. 13(a). At the transmit power of +10dBm, the RSS reported is basically the same, and exactly the maximum RSS of the MICA2 mote. As shown in Fig. 13(a) and in Table 2, the maximum inter-node distance is found around 80 and 90cm for transmit powers of +5 and +10dBm, and 50cm for -3 and 0dBm. These results show that the transitional region, discussed in Section 3.6, is also a function of the transmit power of the sender. In Fig. 13(b), the results from underground-toaboveground experiments are presented. The RSS values are shown as a function of the horizontal inter-node distance for different transmit power levels. The sender has a burial depth of 40cm and the receiver is positioned at the soil surface. As shown in Fig. 13(b) and in Table 2, the maximum inter-node distance is found to be between 2 and 3m, depending on the transmit power level. The proposed WUSN testbed and guidelines were used to successfully obtain these results. In Fig. 13(d), the results from aboveground-tounderground experiments are presented. The RSS values are shown as a function of the horizontal inter-node distance for different transmit power levels. The receiver Received Signal Strength (dbm) Received Signal Strength (dbm) Received Signal Strength (dbm) TX Power +10dBm TX Power +5dBm TX Power 0dBm TX Power 3dBm Horizontal Inter node distance (cm) (a) Underground-to-underground experiments. TX Power +10dBm TX Power +5dBm TX Power 0dBm TX Power 3dBm 90 0 [40] 50 [64] 100 [108] 150 [155] 200 [204] 250 [253] 300 [303] Horizontal inter node distance / [actual distance] (cm) (b) Underground-to-aboveground experiments. TX Power +10dBm TX Power +5dBm TX Power 0dBm TX Power 3dBm [41] 20 [45] 30 [50] 40 [57] 50 [64] 60 [72] 70 [81] 80 [89] 90 [98]100 [108] Horizontal inter nodes distance / [actual distance] (cm) (d) Aboveground-to-aboveground experiments. Fig. 13. WUSN testbed results with commodity WSN sensor nodes. Received Signal Strength vs. horizontal inter-node distance. has a burial depth of 40cm and the sender is positioned at the soil surface. As shown in Fig. 13(d) and in Table 2, the maximum inter-node distance is found to be between 10 and 90cm, depending on the transmit power level. These results are in accordance with the extreme attenuation suffered by the signal as shown in Fig. 13(d) and due the fact that the sensitivity of the MICA2 mote is