UPTEC IT 17 010
Examensarbete 30 hp Juni 2017
Experimental Study: Link-Quality Estimations in a City-Deployed Wireless Sensor Network
Max Wijnbladh
Teknisk- naturvetenskaplig fakultet UTH-enheten
Besöksadress:
Ångströmlaboratoriet Lägerhyddsvägen 1 Hus 4, Plan 0
Postadress:
Box 536 751 21 Uppsala
Telefon:
018 – 471 30 03
Telefax:
018 – 471 30 00
Hemsida:
http://www.teknat.uu.se/student
Abstract
Experimental Study: Link-Quality Estimations in a City-Deployed Wireless Sensor Network
Max Wijnbladh
As the capabilities and scalability of the Internet grow larger and become more advanced, new opportunities arise that enable users to make use of this enhanced manner in which to distribute data. One particular example of this is Wireless Sensor Networks - networks consisting of small sensors with wireless communication capabilities, which can monitor and record environmental phenomena with the intention of forwarding the obtained information to
individual clients or centralized servers for data mining. These types of networks are often deployed in relatively hazardous environments, which may challenge the efficiency and performance of the wireless connectivity between nodes.
A joint effort between multiple partners, called the GreenIoT project, aims to deploy a Wireless Sensor Network in Uppsala, Sweden. The partners of the project seek to learn more about the level of performance that can be expected by the network in various scenarios, such as when the distance between the
transmitting nodes is increased and when the line of sight between them is disrupted. The insights gained from this research are to be used for optimal configuration and deployment of the nodes, in order to ensure a reliable and high-performance network. Using the knowledge obtained from previous work in the area, methods and experiments were developed for assessing the effects that these demanding situations have on the performance of the wireless links.
The results showed that the observed metrics of the links, such as the strength of the received signal and the packet reception ratio, differed greatly depending on the environment in which the experiments were conducted. Based on the outcomes of the experiments, a conclusion was made that field tests on planned deployment sites before installation in the testbed will likely be necessary to ensure sufficient reliability in the network. When investigating potential areas of improvements to the existing testbed architecture, it was found in related work that techniques such as adaptive power and channel control provide higher performance in lossy and high-density networks. Since the nodes used in the project are compatible with existing solutions providing this functionality, these kinds of enhancements should prove to be beneficial for the future operation of the GreenIoT testbed.
Ämnesgranskare: Per Gunningberg Handledare: Edith Ngai
Contents
1 Introduction 1
1.1 Objectives . . . . 1
1.2 Related work . . . . 2
1.3 Limitations . . . . 5
1.4 Thesis Layout . . . . 5
2 Background 6 2.1 Wireless Sensor Networks . . . . 6
2.2 The GreenIoT Project . . . . 7
2.2.1 Project Goal . . . . 7
2.2.2 Partners . . . . 7
2.2.3 Current state . . . . 8
2.3 Sensors . . . . 8
2.4 Gateways . . . . 8
2.5 Protocols . . . . 9
2.5.1 Application Layer . . . . 9
2.5.2 Transport Layer . . . . 9
2.5.3 Network Layer . . . . 9
2.5.4 Link and Physical Layer . . . . 9
2.6 Deployment . . . . 10
2.7 Link quality in Wireless Sensor Networks . . . . 10
2.7.1 Challenges . . . . 10
2.7.2 Link Quality Estimation . . . . 11
2.7.3 Range and regions . . . . 14
3 Methods 15 3.1 Experiments . . . . 15
3.2 Locations . . . . 15
3.3 Tools . . . . 17
3.3.1 Hardware . . . . 17
3.3.2 Software . . . . 18
3.4 Parameters and analysis . . . . 19
4 Results and discussion 20 4.1 Range . . . . 20
4.2 Regions . . . . 23
4.3 Correlations . . . . 25
4.4 Recommendations . . . . 27
4.5 Method evaluation . . . . 33
5 Summary and conclusion 35
6 Future work 37
7 Appendix 38
8 References 45
1 Introduction
The rapid expansion of the Internet’s capabilities has led to an exponential in- crease in the number of Internet-connected devices [1], as well as an exploration into how emerging technologies can be applied to progress our society. One of the areas that has gained widespread attention is the Internet of Things (IoT) [2]. The idea behind IoT is that in the future, most, if not all, of our appli- ances and tools will be equipped with Internet-connectivity, which allow users to monitor phenomenons and events that have been challenging or impossible in the past.
One implementation of IoT, called a Wireless Sensor Network (WSN) [3], is a wireless network consisting of spatially distributed, autonomous devices using sensors that monitor physical or environmental conditions [4]. These sensors record various types of environmental data and forward this to a gateway, or
”sink node,” which in turn propagates the data back to a cloud for process- ing.
Devices used in a WSN create additional requirements for both the network they are connected to, as well as the devices that are a part of it, in terms of operational efficiency with regards to battery life, and reliability in the network connectivity that allows the devices to share their data. Since these devices are designed to be as energy-efficient and have as reduced complexity as possible with the intention of being deployed in various outdoor environments, their inter-connective links are often characterized by lossy and unpredictable behav- ior.
An initiative by the Uppsala Municipality, called the GreenIoT Project, aims to deploy a Wireless Sensor Network in Uppsala, Sweden, with the intention of gathering environmental data. The coordinators of the project are interested in the behavior of the wireless links in city-deployments, such as the correlation between link performance and the length of the link, the effects of obstacles shielding line of sight between the deployed nodes, and how to interpret the obtained link quality metrics for various scenarios. This information will be used for finding a good configuration and placement strategy for the nodes, as well as to provide necessary background information on the problems, and develop a foundation for future experiments to build upon.
1.1 Objectives
The purpose of this study is to evaluate the quality and behavior of wireless signals in the planned deployment locations of the GreenIoT testbed in the Uppsala City Center. In particular, this study will assess the behavior of node to node links when the distance between them increases, the maximum distance between nodes deployed that ensures a reliable link, and if this behavior differs in environments with many surrounding wireless signals and moving objects, compared with an environment of limited external disturbances. This study will also look into the effects that obstacles shielding the line of sight, such as buildings, has on the behavior and quality of the links. The results from the
experiments will be used to propose improvements to the current testbed design, in order to increase the network reliability.
1.2 Related work
Despite a lack of previous in-depth experiments in the research community on physical link quality estimations of Wireless Sensor Networks in city environ- ments, the subject of link quality as a function of distance, and the metrics that best describes it, has been a topic devoting significant amounts of research in the past. This section will summarize the results of particular studies that have influenced and inspired this paper.
Wireless Air Quality and Emission Monitoring [5]
Pushpam Joseph, Aji John
The authors describe the process they underwent to deploy a WSN in Uppsala, with the intention of demonstrating the realizability of such a system in a city environment. To analyze the reliability of their setup, they performed link qual- ity evaluations in the deployed clusters along Vaksalagatan and Kungsgatan in Uppsala. The distance between the evaluated nodes is not indicated in the pa- per, but a map illustrating the position of the nodes in the network is shown in Figure 1.
Figure 1: Deployment sites for WSN clusters. Triangles represent sensor nodes; stars represent gateways [5]
The metric chosen for evaluation was the Packet Reception Ratio, which is the ratio of packets delivered divided by the amount of packets sent. These evaluations were conducted using passive-link monitoring, where the links were observed over a months period, and the final metric was obtained using the gathered data.
Figure 2: The curves are color-coded depending on which link is measured. A packet is sent on each link every 3rd minute, which means that 19 packets need to be received per hour to achieve a 100% reception rate. A reception rate exceeding 19
packets per hour indicates duplicate packets. [5]
Their results showed that the links delivered an efficiency of 92% and 72% for the two links at Kungsgatan, and 84% for Vaksalagatan. The authors point out that although the results were in the positive territory, the network exhibited a high prevalence of duplicate and missing packets, with the cluster consist- ing of two nodes experiencing the highest amount of duplicates. The authors hypothesized that the duplicate packets were due to both nodes acting as coor- dinators, meaning that they rebroadcast packets blindly. Since the gateway was in range for both nodes, the duplicate packets could be attributed to the gate- way receiving the same packets from both of the nodes. They also attributed the missing packets to interference from nearby nodes, although this was not confirmed through further conducted experiments.
Link-Quality Measurement and Reporting in Wireless Sensor Net- works [6]
Abdellah Chehri, Gwanggil Jeon, Byoungjo Choi
The experiments performed in this study examined the effects of increased dis- tances between wireless nodes, and the environment, on the link quality of a WSN’s. The network was deployed in an underground mine using nodes utiliz- ing the ZigBee protocol operating in a 2.4 GHz frequency band. The evaluation was conducted by performing active-link measurements between the links of the network clusters, and analyzing the behavior of the connections using various link-quality metrics, such as RSSI and LQI, when the link maintained a line of sight as the distance was increased, as well as when the link was shielded by obstacles.
Figure 3: Results from experiments. Top graphs are the results when nodes maintaining line of sight, bottom graphs are when the line of sight is disrupted [6]
Their results showed that both the RSSI and LQI values decreased as distance was increased between the nodes. The obtained RSSI followed a decreasing log- arithmic curve, fluctuating around the minimum value of -90 dBm for distances exceeding 40 m for links with line of sight, and 20 m for the links with obsta- cles in the path. The LQI experienced a decrease with increased distance in all experiments, which was reflected in a linear decay of the throughput. The rate of decrease for both the RSSI and LQI metrics were increased when the line of sight between the nodes was disrupted.
SCALE: A tool for Simple Connectivity Assessment in Lossy Envi- ronments [7]
Alberto Cerpa, Naim Busek and Deborah Estrin
In this paper, a wireless network connectivity tool, known as ”Scale,” was used for collecting packet delivery statistics from a WSN testbed consisting of 55 nodes. The nodes were deployed in three different settings - an urban environ- ment of a university campus, an outdoor habitat reserve, and an office building.
The influence of the power level of the antenna was also evaluated, by varying the power level between -10 dBm, -1 dBm and +5 dBm.
Figure 4: In their experiments, the results indicated the presence of a region where the reception rate was not correlated with distance. [7]
Their results indicated that there is no clear correlation between distance and packet delivery efficiency within 50% of the maximum communication range of the nodes. In this range, a high degree of variation could be observed in the obtained metrics, and although the length of the link did not correlate with the efficiency, the temporal variations of packet delivery was reflected in the mean
reception rate. The metrics also showed that the packet size did not lead to a significant difference for the outcome of the experiments.
1.3 Limitations
The scope of this study includes measuring and analyzing the correlation be- tween distance, obstacles, and the obtained link quality metrics between nodes in the planned locations for node deployment in the GreenIoT testbed. This study will neither include experiments done with proposed enhancements to the link quality in the testbed, nor will it cover the complete set of factors that affects the performance of wireless links.
The decision to focus on the spatial characteristics of the links rather than the temporal, was established during the research stages of this paper. It became ev- ident that evaluating the reliability of a multi-cluster, meshed topology testbed with a variety of hardware becomes increasingly difficult when considering the state of the testbed at project initiation. This study will include an extensive method evaluation, where the uncertainties and limitations of the experiments will be discussed and compared to previous work in this field.
1.4 Thesis Layout
Firstly, the Background section of the thesis provides necessary information about WSN’s, the GreenIoT Project, and common reliability challenges in wire- less communication. Additionally, the means in which the link quality of wireless nodes can be evaluated and the applicable metrics that have been commonly used to evaluate link quality in similar experiments will be discussed.
Next, the Methodology section will describe the tools and methods that have been developed and designed for this study to perform the experiments. The section will also include insights towards the motivation and reasoning behind why these methods and metrics were chosen.
Further, the Results and Discussion section will highlight the obtained results, in which findings will be analyzed and compared to results found in related studies and experiments. Also, the methods used for evaluation and their shortcomings will be identified.
Lastly, the Summary and Conclusion section will summarize the work performed in this study, our findings, and propose interesting areas for further research and improvements for future work.
2 Background
2.1 Wireless Sensor Networks
A WSN is a network consisting of wireless transceivers with a communication infrastructure that supports the monitoring and recording of conditions in var- ious locations. The conditions that are recorded differ from application to ap- plication, but common phenomena include temperature, humidity, air pressure, sound, and wind, amongst several others [8].
The monitoring portion of the infrastructure that comprise a WSN consist of small lightweight sensor nodes, based upon a microcomputer, one or more sen- sors, a wireless transceiver, and a power source. These sensors collect envi- ronmental data and convert it into electronic signals which are forwarded to a microcomputer which parses the data. The environmental data collected is forwarded using the wireless transceiver, and the parsed data from the mi- crocomputer is then converted to wireless radio signals and sent through the transceiver’s antenna [3].
Since the sensor nodes are designed to be as simple as possible, they are usually not equipped with the necessary hardware to communicate using public net- works. Instead, the sensor nodes are typically grouped into clusters containing one or more sensor nodes, as well as a gateway, which acts as a bridge between a network cluster and the Internet. The sensor nodes belonging to a cluster only communicate with other sensor nodes belonging to the same cluster, as well as the gateway.
The routing and coordination schemes between sensor nodes and gateways vary between WSN’s, but the nodes in the clusters are often self-organizing and communicate their relative position and current status amongst each other.
This information is then used for creating a topological overview of the entire cluster, which is used by each node for routing purposes.
The gateway, as previously mentioned, is the bridge between the clusters and the Internet. These gateways are equipped with a transceiver able to receive radio signals transmitted on the same frequency as the sensor nodes, and are able to forward the information to their intended destinations through the In- ternet.
The end user, which can be individual clients or a single server, requests the recordings of the monitored location and receives updates from the sensor nodes continuously as they are produced. This data can be used for data mining for multiple purposes, such as creating graphs of observations over time or heat maps.
2.2 The GreenIoT Project
The GreenIoT Project is a national project funded by Vinnova, the Swedish In- novation Agency. The project goal is to deploy a WSN in Uppsala, using custom sensor nodes and gateways in a meshed topology, which are able to monitor en- vironmental phenomena and traffic data to the cloud for data processing.
Figure 5: The GreenIoT Project architecture
2.2.1 Project Goal
The objective of the GreenIoT project is to ”develop an energy-efficient Green IoT platform that enables green and sustainable growth of societies targeting for improved environment and economic growth, to investigate Internet-of-Things, cloud computing, and green networking technology to support open data for the benefits of reduced energy consumption, air pollution, and traffic congestion in major cities of Sweden. Further, GreenIoT seeks to engage companies and the general public for IoT innovations, develop products and services based on the open data, and explore opportunities for IoT businesses worldwide” [9].
2.2.2 Partners
The GreenIoT project has a number of partners involved in the development.
These include agencies such as Uppsala University, KTH, Upwis, 4Dialog AB, SICS, and Ericsson, which fullfills different roles in the development. For exam- ple, KTH, Upwis, and SICS provide hardware and software support, as well as technical expertise. 4Dialog AB provides the means to illustrate the data col- lected from the sensors, through graphical applications, such as 3D maps, while Uppsala University provides coordination amongst the partners, with Master’s thesis students developing various tools and improvements to the project, such as sensor deployment on buses, mining techniques for obtained data, etc.
2.2.3 Current state
In the Spring of 2017, a testbed was planned to be set up, with multiple clusters at designated locations in Uppsala. The deployment details are covered more extensively in Section 2.6. Two nodes, one gateway, and one sensor is mounted at Kungsgatan, which successfully collects and transmits data back to the cloud, with more nodes planned to be deployed in the near future. At the start of this project, a mobile Libellium sensor [10] was also deployed on the roof of a Uppsala bus.
2.3 Sensors
Despite the sensors that are going to be deployed in the testbed are manufac- tured and developed by multiple different partners, they follow general design re- quirements to ensure that they are cross-compatible amongst each other.
Each sensor has a programmable microprocessor development board [11]. The sensors are also equipped with RF antennas which enable them to receive and transmit data frames over the 2.4 GHz frequency band. Each sensor that is going to be deployed in the testbed include hardware for collecting data regarding temperature, humidity, light, sound, air pressure, and the amount of NO2/PM particles in the area surrounding the sensor.
The sensors run Contiki [12], an open source operating system designed to be compatible with a wide variety of hardware. The OS supports multiple Inter- net standards, such as IPv6 [13], IPv4 [14], 6LowPan [15], RPL [16] among others.
2.4 Gateways
Like the sensors, the gateways deployed in the testbed consist of micro-controller development boards and uses the Contiki OS, but are equipped with additional hardware and software that enables them to bridge the connection between the subnet of sensor nodes and the Internet. Each gateway is assigned a cluster of multiple sensors and a gateway that connects to the Internet either through a nearby WiFi hotspot [17], or through a cellular connection by using a standard SIM-card.
2.5 Protocols
The protocols follow the standardized IP-based communication protocols for IoT devices. This section will cover each protocol used in every layer of the simplified OSI-model [18].
2.5.1 Application Layer
The application layer protocol used is MQ Telemetry Transport, or MQTT, a publish/subscribe based lightweight messaging protocol designed for resource constrained devices and networks [19]. Each sensor in the testbed acts as a publisher [20] transmitting its data on a unique topic [21] to a designated broker [22]. Every client that wishes to receive the data published on that topic can subscribe to it [20], and the broker will then forward every message published on that topic to the subscriber.
2.5.2 Transport Layer
The MQTT protocol is dependent on the underlying protocol TCP [23], a connection-oriented protocol that enables multiple mechanisms for controlling the stream of data sent and received between communicating devices. It is im- portant to note that unlike UDP [24], which is a connection-less protocol, TCP enables retransmission of lost packets by an acknowledgment mechanism, mean- ing that packets which are lost during transmission will be retransmitted by the sender until an acknowledgment is made.
2.5.3 Network Layer
The network layer of the testbed is based on IPv6 for client identification, and RPL for routing. IPv6 was chosen as it is defined as the standard IP protocol for IoT devices, since it enables a much larger set of assignable addresses compared to IPv4. RPL was designed for use with low-power and lossy networks, enabling a multitude of traffic flows, such as point-to-point links, multipoint-to-point, point-to-multipoint, etc.
2.5.4 Link and Physical Layer
In the link and physical layer, the IEEE 802.15.4 [25] technical standard is used to enable the physical transmission of link-layer frames. Notable features includes the reservation of guaranteed time slots for transmissions, collision avoidance, link quality assessments and energy detection [26].
2.6 Deployment
Figure 6: The planned deployment sites for WSN clusters
The locations for the planned deployment of node clusters are shown in Fig- ure 6. There are five planned clusters in total, each covering an area around Kungsgatan, the City hall, Luthagsesplanaden, and Salagatan. The reason for choosing these locations are that they experience the highest amount of traffic and activity during the day in Uppsala, and are therefore suitable for collecting the type of data that is intended to be analyzed by the GreenIoT project cloud.
The requirements set on the physical deployment are that the nodes are placed at a height of 1.5 - 4 m and at least 25 m from large street crossings.
2.7 Link quality in Wireless Sensor Networks
2.7.1 Challenges
The performance and reliability of a WSN face several challenges that conven- tional wired and wireless networks do not. For instance, the wireless equipment used in a WSN are often deployed in relatively hostile environments, which may affect transmission quality. These hostile factors include meteorological changes, such as rain, snow, and wind - but also other factors that plague any wireless connection, such as cross-frequency interference, reflection, refraction, scatter- ing, and attenuation of the radio signals.
Constraints in the devices used in the network need to be considered when creating a WSN architecture. Since the sensors used aim to be as small and mobile and possible, they are designed to be limited in computational power and power draw. These design requirements create obstacles in the pursuit of reliability in WSN’s. By increasing the computational capabilities of the sensors, the power requirements will increase as well. This is also true for increasing the signal power of the antenna. The more power that is supplied to the signal, the stronger the signal becomes, thus resulting in better performance of a wireless link. The downside is that in order to ensure longevity, the nodes in a WSN are usually constrained in the amount of available power.
Due to these challenges, it becomes important to assess the optimal positions of the sensor nodes before deployment, in order to ensure that the connectivity does not suffer from excessive packet losses, or constant retransmissions, thus putting an energy strain on the transceivers.
2.7.2 Link Quality Estimation
N.Baccour et al. [27] performed a comprehensive survey on link quality estima- tors and their roles in assessing the performance of wireless radio links.
Figure 7: Link Quality Estimation Steps [27]
In the study, the contributors outlined the basic steps for performing a link quality estimation, which is illustrated in Figure 7. The first step is to collect network statistics, such as the amount of packet losses and signal strength, by monitoring the link that is to be evaluated over a set time estimation window.
The type of monitoring can be of three different types - active, passive, and hybrid link monitoring.
Active monitoring means that the packets sent over the link during the esti- mation window are probe packets, which are packets that are sent for the sole purpose of assessing the link and does not reflect actual real-world traffic. Pas- sive monitoring, on the other hand, exploits existing traffic for gathering data, which has the upside of using real-world behavior of the link. Hybrid link mon- itoring is a combination of the two.
The challenge of which metrics to use to accurately estimate link quality in networks consisting of wireless links, is an area devoting significant amounts of research efforts. To analyze the quality of a wireless link, there are some basic metrics commonly used. PRR is the total number of packets received divided by the number of packets sent. The PRR is measured during a time frame with a specified number of transmitted packets, and provides an estimate of the efficiency of the link for the specified time interval.
RSSI is the metric used for assessing the strength of the signal between two wireless nodes. This metric, measured in dBm, presents the relative measure of the radio frequency (RF) signal being received, where the sensitivity threshold lies at the noise floor, between -90 and -110 dBm [28].
The LQI metric is a relatively new parameter, which has been implemented in more recent radios based on the IEEE 802.15.4 standard. This metric describes how close the demodulated signal is to the ideal constellation point of the re- solved pattern. In IEE 802.15.4, the LQI is represented as an 8-bit value, and is therefore assigned a rating ranging between 0-255, with a value of 255 indicating no bit errors [28].
PRR is useful in that is gives the exact number of successful transmissions during the time window, and is widely used in routing algorithms for this reason. In a link with bad characteristics, a low or fluctuating PRR can be expected due to packets sent being lost in transmission for various reasons. The PRR value that considers a high-performance link differs between studies, but common thresholds used for the PRR are above 75% for good links and lower than 25%
for bad links. It has been argued amongst researchers that the PRR is limited in how much it can tell in regards to the quality of the link. In the following paper on link quality estimations [29], the authors summarize some of these shortcomings, and identify that the PRR can not accurately differ between stable and unstable links.
The RSSI has long been deemed an inadequate measure of link-quality in the research community due to experiments performed on early node models, where mis-calibration of the radios led to varying and unpredictable results. For ex- ample, results from experiments performed by researchers at the University of Southern California [30] indicated that while the RSSI was useful in detecting good links, the metric was inadequate in estimating intermediate links. The use- fulness of this metric has been further researched in a study made by researchers at Stanford [31], where experiments with newer radio models were conducted, correlating PRR and RSSI at differing power levels.
Figure 8: Correlation between PRR and RSSI [31]
Their experiments showed that while the RSSI was above the sensitive threshold of the nodes used, the achieved PRR exceeded 85%. The results found in the study indicates that while the RSSI might not be enough to determine the quality of imperfect links, it is a clear and easy obtainable metric for determining whether the link is of good quality or not.
While the RSSI may be sufficient in identifying good and bad links, the nature of RSSI suggests it is not competent in providing a complete overview of quality of the link. Since the RSSI is obtained by the relative measure of the RF signal being received, surrounding narrow-band interferers that are transmitting with the channels bandwidth will make the RSSI value to increase. Thus, the RSSI will wrongly indicate a strong signal in the assessed transmission, while in reality the signal between the nodes may be poor, however appearing strong due to an interferer operating in the same frequency [28].
To compensate for the shortcomings of RSSI, newer radios based on IEEE 802.15.4 implement another parameter, the LQI metric. In contrast to RSSI, the LQI is not an estimate of the strength of a signal, but rather the signal’s current quality. A study performed with ZigBee Sensor Networks [32], where researchers studied the effects of environmental factors, distance, and interfer- ence on LQI metrics, the results indicated that neither temperature, humidity, nor atmospheric pressure had any significant effect on the obtained LQI values.
On the other hand, during periods of high human activity in the vicinity of the experiment location, fluctuations in the LQI could be observed. As dis- tance was increased between the transmitting nodes, the LQI metric dropped in logarithmic fashion.
Figure 9: Results from experiments trying to correlate the obtained PRR with average LQI [31]
In ”RSSI is Under Appreciated” [31], experiments were conducted with the in- tention of estimating PRR based on recorded RSSI and LQI in received packets.
The results suggested that single LQI readings alone are not adequately accurate for a correct estimation. Rather, by averaging the LQI over a sample window of at least 120 packets, they were able to accurately estimate the obtained PRR.
The outcomes of their experiments are illustrated in Figure 10. Their results in- dicated that sampling the average LQI over a larger window of packets provides a better indication of the expected PRR of a link.
2.7.3 Range and regions
When illustrating the communication range of a wireless transceiver, it is com- mon to draw a circular disc extending around the perimeter of the antenna, where each node transmitting within this circle will achieve perfect packet re- ception.
Figure 10: A simplified range model of a wireless transceiver
This simplification of transceiver behavior, however, has been challenged in studies on the behavior of wireless signals over varying distances, such as those in [30], where the experiments identified ”gray areas” within the communication range of the sensor radios. Despite the transmitting nodes being placed within the documented maximal range of the respective radios, great fluctuations were found in the PRR, varying with as much as 20-60%. Their experiments also showed that the range, from which point this gray area becomes apparent, increased depending on the power provided to the antennas, thus indicating that the range of this region is a result of the obtained signal strength between the transceivers.
This grey area was identified in all experiments carried out, with nodes trans- mitting with various power levels and distances. The extent of this region was found to compose a significant portion of the transceivers transmission range, from at least 50% in all cases, and in some, as much as 80%.
Similar results were found by researchers performing experimental studies on Low-Power Wireless Sensor Networks [33]. By defining thresholds for categoriz- ing the PRR found within set distances from the transceiver, they found that within a certain range, they consistently managed to achieve a PRR ratio be- tween 70-100%. As distance was increased though, the PRR dropped and the standard deviation increased until no packets managed to reach their destina- tions.
In experiments performed by researchers at Berkeley [34], Woo et al. identify different regions in the communication range of wireless transceivers. Transmis- sions occurring within a certain distance threshold managed to achieve between a 90-100% PRR, with very little variation, thus indicating a stable connection.
This region are commonly called the Connective region. Once the transmitter was moved outside a certain range however, the PRR started fluctuating heav- ily, between 0-90%. As distance was increased between the nodes, the average
PRR dropped until communication ceased. The region following the Connected region, in which high fluctuations in PRR ratio became apparent, was called the Transitional Region, and the region where no packets managed to reach the receiver is called the Clear Region.
The existence of these regions have been the subject of numerous studies and experiments, both for studying its properties analytically in order to create realistic simulations [35], as well as to learn more about the attributes of the different regions and the causes of their behavior [7].
3 Methods
3.1 Experiments
To assess the impact of distance and the environment on the quality of the transmission between nodes deployed in the GreenIoT testbed, experiments were conducted in the Uppsala City Center with nodes transmitting packets over a wireless link operating in the 2.4 GHz frequency band. To isolate the physical characteristics of the link, the type of transmission was constituted by link-layer frames, thereby circumventing retransmission mechanisms used in the upper layer protocols.
In the experiments, one node acted as a receiver, and the other as a sender, and the link quality between them was assessed at incremented distances in various settings. Motivated by the research performed on link-quality estimation tech- niques and its metrics, the experiments in this study will include a combination of PRR, RSSI, and LQI in the assessments. With the PRR as the core metric for evaluation, the RSSI and LQI was used to evaluate the reasons for differ- ences in the PRR in the different scenarios. By performing correlation analysis on the obtained metrics, it was possible to evaluate which metric best describes the efficiency of the link.
By using the measured PRR for each scenario, as well the recorded RSSI and LQI, the study aims to identify the Connected, Transitional, and Disconnected region for each of the evaluated location. This provides insight as to which of the metrics best estimate the PRR during differing link conditions, and to gives an overview of the spatial behaviors of the links in the urban environment of Uppsala. Based on results from the experiments, suitable enhancements which can be implemented to increase the reliability of the testbed will be proposed.
3.2 Locations
The first experiment was conducted along the pathway in Ekonomikumparken.
This location was chosen due to an interest in evaluating a link in an envi- ronment where there are limited amounts of obstacles and activity in the sur- rounding area, which may influence the readings, such as mobile phones or other devices operating in the 2.4 GHz frequency band. Figure 11 illustrates a 3D
map of the area where the experiments were conducted. 0 m indicates the po- sition of the receiver and the blue line shows the path followed and the points from which the readings were taken.
Figure 11: The area in Ekonomikumparken where the experiments were conducted The remaining experiments was conducted in select locations in Uppsala. The locations was chosen based on three of the planned WSN clusters for the Gree- nIoT testbed deployments. Each location will describe if the nodes maintained line of sight, as well as testing when the line of sight is disrupted by objects, such as buildings. The later experiments will be performed around street corners, since the effect of these types of situations are of interest for future deployment of nodes in the GreenIoT testbed.
Three locations were chosen - the planned clusters at the north and south parts of Kungsgatan, as well as the planned cluster at Vaksalagatan. The three street corners that also were evaluated was in close proximity to the above locations.
For line of sight experiments, abbreviated as LOS in the remaining parts of this paper, the distance between the nodes will be incremented by 50 m over transmissions, and will be increased while the recorded PRR of the link exceeds 25%. The experiments performed around street corners, abbreviated as NLOS, will be conducted by placing the receiver at a distance of 25 m from a crossing, and the distance of the transceiver incremented by 10 m between each reading, measured from the point of the crossing. The set distance of 25 m from the crossings was chosen based on the physical deployment requirements for nodes that are to be a part of the GreenIoT testbed.
Each location evaluated in this paper will have a corresponding 3D map attached in the appendix, in order to help assess the impact of the environment on the results, as well as to support recreations of the experiments.
All experiments were conducted during daytime, with sunny weather and a temperature fluctuating around 10 ◦C.
3.3 Tools
3.3.1 Hardware
To assess the link quality between the nodes during varying conditions, two 802.15.4 enabled nodes were provided by KTH, developed by Radio Sensors AB and based on the ATMega256RFR2 chip [36]. These are the same type of nodes deployed by KTH in the GreenIoT testbed, excluding attachments for various sensors.
The nodes were connected to a host computer, which provided power and man- agement of the test session. The setup is illustrated in Figure 12.
Figure 12: From left to right - Encased node, uncased node connected to computer, laptop communicating with the node. The output from a transmission of 100 packets
can be seen in the terminal window
Both nodes were mounted to the back lid of their respective host laptops, with the antenna of the node extending above the attachment area in order to avoid dampening of the RF signal. Both of the equipped laptops used in the experi- ments are illustrated in Figure 13.
Figure 13: Nodes mounted to the backside of the laptops
The laptops were operated by two people, which coordinated amongst them- selves using wired headsets and mobile phones. The phones were also used to measure the distances between the nodes during the experiments by using built in GPS’s and the Android application Maps Measure [37].
3.3.2 Software
The Contiki OS was flashed on the nodes, as well as an app - ”PDR-Test”
[38], which was developed by researchers at KTH for the purpose of performing link-quality measurements.
The communication stack used is RIME, which provides Contiki devices with a set of lightweight communication primitives [39]. PDR-Test allows the node running it to send an arbitrary amount of raw packets of adjustable sizes over different power levels and frequencies with custom intervals between transmis- sions. The program also enables the node receiving the packets to collect the average RSSI and LQI values over the amount of received packets. This in- formation may be outputted over a serial port connection to a host computer running a serial port communications program, such as Minicom [40], which was used in this study. The output of the Minicom session was written to text files. To parse and plot the results from the experiments, a Python [41] script was developed that could load the text files and sort the results depending on distance and location. The script was also used for calculating averages and de- viations of the dataset. The graphs presented in this paper are produced using the Python library matplotlib [42].
3.4 Parameters and analysis
The transmitting node is programmed to transmit 250 packets, with each packet sent with an interval of 1/128 seconds at 3.5 dBm, the maximum output power of the antenna. The size of the packets were 128 bytes and were transmitted on channel 25 on the 2.4 GHz frequency band. The nodes are at a height of 1.5 m during transmissions, which is the minimum height requirement outlined in the deployment recommendations for the GreenIoT testbed. The experiments are repeated 10 times consecutively. The average PRR for each assessed distance will be obtained by dividing the recorded received amount of packets in the receiving sensor with the total amount of sent packets across all repetitions of the experiment for that distance.
AvgP RR = recvpkts 10 ∗ 250
The efficiency of each link will be divided into three distinct groups, shown in Table 1.
Link group PRR
Good 75% ≤ x
Intermediate 25% < x < 75%
Bad x ≤ 25%
Table 1: Thresholds for link quality classifications
To obtain the average RSSI and LQI metrics for each distance evaluated, the reported values in the received packets for each experiment iteration will be averaged. The final total average will be obtained by calculating the mean of the standard deviation obtained from the complete set of experiments for that distance.
AvgRSSI =
T otalRSSI 250
10 AvgLQI =
T otalLQI 250
10
The Pearson Correlation Coefficient was used to calculate the correlation be- tween the metrics, which is a measure for determining the linear dependency of two sets of variables using the following equation:
r = cov(X, Y ) pvar(X)pvar(Y )
The obtained data will be illustrated using various plots and tables.
4 Results and discussion
4.1 Range
Figure 14: Average PRR, RSSI and LQI, obtained at set distances in Ekonomikumparken
Firstly, the results obtained from experiments conducted in Ekonomikumparken will be discussed, which are illustrated in Figure 14. An immediate observation that can be made is that as the length of the link increases, the link quality degrades. For the first 150 m, the link maintains a high average PRR ratio exceeding 75% with very little variation in the dataset, classifying it as a good link. For this distance interval, the average RSSI drops linearly between -52 and -75 dBm, along with a decreasing LQI value fluctuating around the maximum value of 255.
When the experiments were conducted at a distance of 200 m and beyond, the average PRR decreased at a higher rate, and the variation in the dataset in- creased. This was reflected by a decrease in both the RSSI and LQI metrics, with the signal strength converging around the minimum value of -90 dBm.
Since the link maintained a high efficiency before reaching the minimum thresh- old for the RSSI, a possible cause for this performance degradation for distances exceeding 150 m could be that the antenna getting close to the limit of its trans- mission range, resulting in severe signal attenuation.
From these results, we infer that the maximum transmission range of the nodes are approximately 250 m in a favorable environment. To achieve a link effi- ciency of at least 75% in these conditions, the link length should not exceed 150 m.
To compare these results with the experiments performed in the city, we plot the average PRR, RSSI, and LQI for each assessed location and distance. The results are shown in Figure 15.
Figure 15: Measurement results from experiments in the city
Evaluating the resulting graphs, performance degradation with increased dis- tance can be observed. Compared with the links evaluated at Ekonomikum, the effect that distance has on the efficiency of the link appears stronger for the experiments performed in city environments, with lower average PRR at shorter distances and a higher variation in all observed metrics.
For distances up to and including 100 m, all the links obtained similar metrics, with a PRR between 80-90%, signal strength between -42 and -65 dBm, and LQI values at the maximum of 255. One exception was found though - the link at Kungsgatan N. While the other links were classified as good within this range, the PRR ratio of Kungsgatan N dropped in a linear fashion. This drop in PRR was accompanied by a higher variation between the obtained PRR readings in the dataset compared to the other links within this range, as well as a higher variation between RSSI readings. Despite this, the average RSSI for Kungsgatan N at these distances was higher than the other links.
One explanation for the unpredictable behavior of the Kungsgatan N link could be that the high amount of pedestrians and moving vehicles disrupted some of the transmissions. Indeed, during the experiments, it was a challenge to main- tain the line of sight between the nodes during the experiments at Kungsgatan N, as people were prone to moving in front of it mid-transmission. This proved to be a challenge in obtaining consistent measurements, and are covered more extensively in Section 5.4.
One possible cause of the lack of correlation between the low efficiency and
recorded signal strength could be that the nodes experienced interference from the nearby sensor node deployment at Kungsgatan N, which are detailed in Section 2.2.3. Since the nodes operate in the same frequency band of 2.4 GHz, this could explain why the average signal strength was increased, since the recorded RSSI would be affected by the interfering nodes signal. It is also interesting to note that while the drop in PRR was reflected in the fluctuations of the RSSI readings, the LQI metrics obtained during the same tranmissions seems unaffected.
When the link was evaluated at a length of 150-200 m, differences in the perfor- mance between all locations was observed. The average PRR, RSSI, and LQI dropped across all links, and the deviation of the respective datasets rose, ex- cept for the RSSI deviation due to the metric converging around the minimum value. The link at Vaksalagatan was the only one that managed to remain in the good link territory, while the PRR of both Kungsgatan N and S decreased to an intermediate efficiency of around 50%.
At a distance of 250 m, none of the links managed to obtain an average PRR exceeding 10%. This was an unexpected result, as it was theorized that since the metrics obtained for the experiments in Ekonomikum suggested a better connection for all previous distances compared to the other links evaluated in the city, the maximum transmission range should exceed the city-based links.
When performing experiments where the line of sight of the nodes were disrupted by buildings at street corners, the efficiency of the links was significantly lower for shorter distances compared to when the nodes maintained the line of sight.
The results are shown in Figure 16.
Figure 16: Average PRR, RSSI, and LQI, obtained at set distances in the city with obstacles in the path of the nodes
After only 30 m, each link assessed reached a reception ratio lower than 10%.
At 40 m, less than 5% of the sent packets were received across all links. The
recorded RSSI values during the transmissions indicate that obstacles in the path of the signal contribute to severe signal attenuation.
When the experiments were performed at a distance of 10 m from the corner, all links excluding Kungsgatan N achieved a PRR of less than 50%. For this distance, Kungsgatan N actually managed to achieve an efficiency fluctuating around the good link threshold of 75%.
Observing the area where the measurements for this experiment were conducted in Figure 25, it is shown that the corner separating the nodes for the first 10 m has a dent in it, which likely has allowed for a more straight path of propagation for the wireless signal compared to the other experiments. This explains the higher efficiency of this link.
4.2 Regions
To illustrate the existence and width of the transitional regions for the evaluated links, the average PRR was plotted for each experiment at every distance and location. To define, the Transitional Region begins at the point where the standard deviation of the results from an evaluated distance exceeds 20%. The resulting graph is shown in Figure 17.
Figure 17: Transitional regions of LOS links
As observed from the graph, the extent of the Transitional Region differs be- tween locations. Analyzing the results from Ekonomikum, the link enters the Transitional Region at a distance of 200 m, where the average PRR from each experiment could range between 0-40%, indicating that the Transitional Region makes up about 20% of the transmission range. In the Connected Region, the efficiency is high, with very little variation between experiments.
Analyzing the results from Vaksalagatan, there is a higher amount of variation between the obtained PRR ratios at shorter distances. The obtained PRR do not start to fluctuate severely until 200 m, where the measured efficiency of the
link was between 15-82%. The width of the Transitional Region in this case was the same as for Ekonomikum.
At Kungsgatan S, the width of the Transitional Region made up 60% of the transmission range of the transceiver, where the PRR for each experiment could be anything between 10-70%. Although the transitional region was wider com- pared to Vaksalagatan, the average PRR and its variation is very similar in the Connected Region, which was up to 100 m.
The results from the link at Kungsgatan N show that at 100 m the efficiency was between 30-82%. This means that the extent of the Transitional Region of Kungsgatan N was the widest of the assessed LOS links, making up as much as 80% of the communication range. As previously theorized in the Range section, these results could be due to interference from the sensor node that was deployed nearby. Kungsgatan N was also the location that experienced the most disturbances in the form of traffic and pedestrians during the experiments, which likely contributed to the high variation in the dataset.
Summarizing the results, the extent of the Transitional Region makes up be- tween 20-80% of the transmission range of the nodes depending on the assessed location. It is interesting to note that the range of the Transitional Region var- ied between links evaluated at the same distances from the receiver, which seems to indicate that the extent of the Transitional Region is less dependent on the length of the link, but instead correlated with the nature of the environment.
This conclusion is supported by the observation made in [27], where the results from the experiments in [7] and [30] were compared. In [7], the results showed that the Transitional Region made up as much as 50-80% of the transmission range of the nodes, while in [30] it was much lesser, around 20-30%, even though the two studies used the same equipment and settings.
Likely environmental causes for these differences in the width of the Transitional Region is the number of buildings and objects surrounding the path followed during the experiments. Observing the environmental map for Vaksalagatan in Figure 26, the link was surrounded by buildings on only one side during transmissions, while the link at Ekonomikum had a near absence of surrounding objects and buildings. Comparing this to the experiments at Kungsgatan N and Kungsgatan S, the links in those locations were surrounded by buildings at both sides. Since these links had the widest Transitional Regions, it is possible that phenomena such as scattering, reflection, and multi-path propagation led to less consistent results at shorter distances for these links.
4.3 Correlations
To see which of the metrics has the best correlation with the average PRR, the average metrics obtained for each experiment is recorded, and the PRR ratio is plotted against the RSSI and LQI. The resulting graphs are illustrated in Figure 18.
Figure 18: PRR values graphed against corresponding RSSI/LQI values. Each dot represents the outcome of an experiment with 250 packets
The results show that the obtained RSSI values are accurate when approximat- ing good quality links for the experiments. For RSSI values exceeding -70 dBm, PRR greater than 75% was consistently achieved. As the RSSI drops below this threshold, the associated efficiency of RSSI values becomes increasingly unpre- dictable, especially at the minimum value of -90 dBm, which in the experiments could mean anything between 0-70% efficiency. These results does not match those in [31], where the outcome of their experiments showed that a RSSI value exceeding -87 dBm resulted in a PRR of at least 85%. This difference in re- sults could be attributed to calibration differences in the radios, or due to other factors in the environment.
Comparing this with the obtained results from the PRR/LQI plot, it is ob- served that for experiments where the average LQI reached 255, the maximum, the PRR varied between 80-100% with a few outliers. When the average LQI exceeds 240, the majority of the experiments managed to exceed a PRR of 60%.
Between a measured LQI of 240-200, the PRR varies between 15-60%, and when the LQI dropped below 200, only two of the experiments managed to achieve a PRR exceeding 25%.
Calculating the correlation coefficient for the complete set of experiments over all distances and locations, a value of 0.179 was obtained for the PRR/RSSI, and 0.776 for PRR/LQI.
This result shows that for all experiments over distances and locations, the LQI portrays a higher linear correlation with average PRR. To analyze the corresponding correlations of the metrics for the identified regions for the links, which is found in Section 5.2, we perform the same calculation for the every region. The results are presented in the below table.
Locations and regions PRR/RSSI PRR/LQI All locations and distances 0.179 0.776 Ekonomikum 50-200 m 0.846 0.704 Ekonomikum 200-250 m 0.292 0.725 Kungsgatan 50-100 m 0.471 0.69 Kungsgatan 100-250 m 0.273 0.757 Vaksalagatan 50-200 m 0.804 0.824 Vaksalagatan 200-250 m 0.367 0.706 Kungs/Sutung 50-150 m 0.596 0.56 Kungs/Sutung 150-250 m 0.398 0.803
Table 2: Pearson Correlation Coefficients for different locations and link types
Particular insights are exemplified by studying this table. The results indicate that for links with a good and stable connection, both the RSSI and LQI in- dicate a similar degree of correlation with the recorded PRR. The LQI shows a correlation coefficient varying between 0.56-0.824 for these links, while the corresponding coefficient for the RSSI varies between 0.471-0.846.
Comparing these results with the metrics for links with intermediate quality, the LQI shows a significantly higher correlation with the recorded PRR. The correlation ranges between 0.706-0.757 for these links, while the corresponding range for the RSSI reside in a range of 0.273-0.398.
Since links with a low PRR and high variance were associated with RSSI con- verging around the threshold of -90 dBm, it is reasonable that the linear corre- lation between the PRR and RSSI metric would be low for intermediate links.
For these types of links, the LQI provided a better approximation of the PRR, almost as accurate as good quality links. The results from the experiments sup- port the conclusions made in [31] and [43] , since their experiments also implied that the average LQI was a better estimator of PRR than the RSSI when eval- uating low to intermediate quality links. Simultaneously, the results also show that for good quality links, both LQI and RSSI show a high linear correlation with the average PRR.
