中文责编:英 子; 英文责编:雨 辰
Zhang Dian, Ming Zhong, Liu Gang, Lu Kezhong, Mao Rui, Feng Yuhong, and Chen GuoliangCollege of Computer Science and Software Engineering, Shenzhen University, Shenzhen 518060, P.R.China
在真实室内环境中,用MICA2节点设计分析影响无线接收信号强度(radio signal strength,RSS)的实验,发现其影响因素不仅包括发送接收方(transmitter-receiver,T-R)之间的距离,且MICA2节点的工作频率和供电电池电压、发送接收方节点差异、天线角度和高度,以及环境中的时空因素和动态环境等都会影响无线接收信号强度.经分别测试这些因素,建议传统无线信号传播模型和信号校准算法应综合考虑各项影响因素.
A series of experiments using MICA2 nodes in real environments was designed to investigate parameters which are able to affect radio signal strength(RSS). The frequency, variation of transceivers, antenna orientation, battery voltage of each MICA2 node, temporal-spatial properties of environment, and dynamic environment can affect the performance of RSS along with the transmitter-receiver(T-R)separation distance. Their impacts on RSS are measured comprehensively and independently. The experimental results have demonstrated that the traditional radio propagation model should consider all these parameters and the impact factor of each parameter. The desirable and practical RSS-related solutions in sensor networks are presented in this paper.
Recent advances in embedded systems, wireless communications, and micro-electromechanical systems(MEMS)have enabled the development of wireless sensor networks(WSNs). Sensor networks can bring a wide range of promising applications such as volcano monitoring, structural monitoring, hospital monitoring, city monitoring and so on. Radio is an essential component of a sensor node. The basic characteristic of radio is radio signal strength(RSS). Usually, signal strength is expressed in dBm units, which is a logarithmic expression of power referenced to 1mW so that higher dBm value corresponds to higher power. In current sensor nodes, such as MICA2, their radios can report the received signal strength indicator(RSSI)for each received packet in dBm units. RSS is such an attractive property for researchers because it can be obtained without requiring additional hardware. In other words, RSS is a “free” resource.
Generally, the RSS will decrease gradually as the receiver moves away from the transmitter. The relationship between RSS and transmitter-receiver(T-R)separation distance is described as a propagation model. Many propagation models have been developed to represent different RSS characteristics under various kinds of environments. Based on propagation models, RSS has been employed in analysis and simulation of wireless networks, localization, routing protocols to measure the link quality and filter out the bad communication links.
Most papers assume RSS is only determined by T-R separation distance, the communication range between the transmitter and receiver based on a RSS threshold(receiver sensitivity)is calculated. This derived connectivity model is called the disk model. However, it is well known that RSS is quite complicated in real environments due to signal reflection, diffraction, and scattering. In this paper, we carefully design a series of experiments using MICA2 nodes in real environments to investigate other parameters besides T-R separation distance-frequency, variation of transceivers, antenna orientation, battery voltage, temporal-spatial properties of environment, and dynamic environment. The novelty of our work can be summarized as follows.
Firstly, to the best of our knowledge, this is the first work to investigate these parameters in a comprehensive and quantitative way. Secondly, in contrast to previous papers with electrical engineering background that focus on analyzing the reflection, diffraction, and scattering in signal transmission path, the parameters considered in this paper, which are related to sensor nodes or environments, are on a higher level and more familiar to researchers with computer science background. It bridges the gap between the two groups of researchers. Finally, quantifying and comparing their impacts on RSS independently, we could identify the critical parameters that affect RSS significantly. These parameters should be added in consideration of node deployment, network modeling, and protocol design in sensor networks.
Usually, there are three kinds of RSS-based localization technologies: leveraging reference nodes, building a radio map and building propagation models.
For those technologies using reference nodes, the users have to deploy a large number of nodes in advance. Therefore, the deployment cost is high.
For the radio map based approaches, laborious training work should be performed to construct the map. When the environment changes, the map has to be rebuilt. Reference  investigated the antenna radiation pattern of MICAz nodes in an interference-free laboratory and measured the mapping between the reported RSSI value and the actual signal strength. Compared with Chipcon CC1000 radio used in MICA2 nodes, it is also a low-power baseband radio operating in the same frequency band so that their radio characteristics are expected to be similar. However, the interference-free environment in the laboratory is far from a real environment.
For those technologies building propagation models, some papers take insight into the relation between packet error rate(PER)and T-R separation distance. The communication area of a node is very different from a perfect disk. Enhancing the disk model, some works[13-14] also introduced a parameter to model the irregular border of communication area. However, the parameters suffer from their unclear physical meaning. Furthermore, the communication area of a node changes frequently in a dynamic environment.
Rather than building propagation models, Cho et al, references [15-16] proposed calibration procedure called linear calibration to resolve the linear variation among different transmitters and receivers. At first, a reference transmitter is selected to calibrate all receivers at a known distance. Then, one of the calibrated receivers is chosen to calibrate the rest of transmitters sequentially by adjusting their transmitting power settings to match the reference transmitter.
In real environments, besides the T-R separation distance, there are many parameters affecting the radio signal strength. To investigate these parameters independently, each time we change only one parameter under the same T-R separation distance while keeping the other ones unchanged. In contrast to the assumption of the disk model in which RSS is only determined by T-R separation distance, we quantitatively measure the variation under each parameter so that their impacts on RSS can be compared. Before we jump into the details of our results, we review some classical propagation models that lead to the disk model. We also discuss the experimental platform and instrumentation.
As we know, the strength of a signal falls off as the signal propagates further. The transmission path between the transmitter and the receiver can vary from simple line-of-sight to one that is severely obstructed by buildings, mountains, and foliages. Various propagation models have been developed to predict the local average received signal strength for an arbitrary T-R separation distance under different environments.
In most propagation models, the received power is given as a function of the T-R separation distance raised to some power. The free space model is the most famous one in which the transmitter and the receiver have a clear, unobstructed line-of-sight path between them. The free space power received by a receiver that is separated from a transmitter by a distance d, is given by the Friis free space equation
where Pr is the received power; Pt is the transmitted power; Gt is the transmitter antenna gain; Gr is the receiver antenna gain; d is the T-R separation distance; λ is the wavelength.
However, a single direct path between the transmitter and the receiver is seldom the only propagation path. The ground reflection model considers both the direct path and a ground reflected propagation path between the transmitter and the receiver. The received power at a T-R separation distance d can be expressed as
where ht is the height of transmitter antenna in meters and hr is the height of receiver antenna in meters.
The two models are typical ideal propagation models, in which RSS is only determined by the T-R separation distance. The difference between them is the rate of signal attenuation, which leads to different communication ranges in the disk model.
Another well-known propagation model is the log-distance path loss model, the path loss PL measured in dB at a T-R separation distance d is
where PL0 is the pass loss at the reference distance d0; γ is the path loss exponent; XG is a normal(or Gaussian)random variable with zero mean.
We conduct the experiments in the Pervasive Computing Lab of our department covered by carpet. Its floor plan is shown in Fig 1. We use the popular MICA2 nodes with Chipcon CC1000 radio chips and monopole antenna. Unless stating explicitly, we deploy all sensor nodes on the ground since it is hard to deploy supportive planes such as pillars under sensor nodes to raise their height, especially in ad hoc deployment of sensor nodes. Except measuring the impact of radio frequency, we choose 870 MHz as the frequency for all the nodes. During the whole experiment, we try our best to keep the environment static so that RSS are only affected by the parameter we want to investigate until we introduce environmental dynamics intentionally. This statement is supported by the stable RSS values observed at all sensor nodes in our experiments. In all static cases, the standard deviation of RSSI readings at a node is less than 2 dBm in our experiment period.
The frequency of Chipcon CC1000 radio chip ranges from 300 MHz to 1 GHz. However, due to the setting of MICA2 motes, they can only work either in the 433 MHz band, or in the 868/916 MHz band. We find out our MICA2 motes can operate in a broader band, which ranges roughly from 850 MHz to 970 MHz. The default transmission power is 0. The interval between beacons is 50 ms.
The CC1000 chip has a built-in RSSI output function, which gives an analogue output at its RSSI/IF pin. Following the equation provided in the CC1000 manual, we could convert this analogue value to dBm units. To avoid packet collision that can affect the received signal strength, each receiver only measures the RSSI values on successively received packets. Each node is assigned with a fixed node ID. When a node receives a packet successfully, it will report its own node ID, the node ID of the packet sender, and the RSSI reading of the packet to the base station.
After discussing our experimental methodology, we investigate a set of parameters besides the T-R separation distance, which are not considered in the ideal propagation models. In contrast to analyzing reflection, diffraction, and scattering in the transmission path at signal level, we focus on high-level parameters that are either related to the sensor nodes, or related to the surrounding environment. Rather than testing many different T-R separation distances within the communication range, we are interested in comparing the effects of these parameters at the same distance.
As is mentioned above, our MICA2 sensor nodes operate in the 850/970 MHz band. With the same transmitting power, different frequencies lead to different received signal strengths at the receiver. Ideally, the higher the frequency is, the faster the signal attenuation is. In other words, lower frequency achieves higher received signal strength and longer communication range. However, due to the complex signal behavior in real environments, the relationship between RSS and the frequency is irregular. We test the frequencies from 850 to 970 MHz with each step being 20 MHz. The T-R separation distance ranges from 1 to 4 m. Based on 100 RSS samples, the average recorded RSS values with intervals are shown in Fig 2(a). Since the intervals of RSS values are quite small, it proves that the environment is quite static. It is obvious that RSS roughly goes down with a few exceptions when the frequency becomes higher. For example, when the distance is 3 m, the RSS value at 910 MHz is lower than that at 930 MHz. Furthermore, RSS seems to become more irregular as the T-R separation distance increases. This can be explained mainly by the increasing complexity in the signal transmission path. We try several pairs of nodes and get similar results.
From another perspective, the signal strength is decreasing when the T-R separation distance increases. Once again, it is not very consistent at several frequencies, such as 870, 890, and 910 MHz. Moreover, the signal at 970 MHz only covers a very short range that is less than 1.5 m, while the signal at 950 MHz cannot reach as far as 4 m. When the stability of RSS values is concerned, we also analyze the standard deviation of RSS values at different frequencies. We omit the figure due to the page limit. Most of the standard deviations are less than 1 dBm. At the distance of 1 and 2 m, higher frequency generally has larger variance. Nevertheless, it becomes irregular at the distance of 3 and 4 m. In the following experiments, we choose 870 MHz frequency since it has the lowest average variance.
It is well known that there is variation among the same type of transceivers due to the difference in radio circuits. Ideally, a receiver should report the same RSS values for the packets from different transmitters put in the same position. However, the variation of transmitters leads to the difference among actual transmitting powers. Similarly, due to the variation of receivers, different receivers at the same place may report different RSS values even they receive the packets from the same transmitter.
In our experiments, we put one node as the receiver in a fixed position and put 40 other nodes one by one as the transmitters in another fixed position. Two T-R separation distances, 2 and 4 m, are tested. Base on 100 RSS samples, the recorded average RSS values with intervals for different transmitters at distance 2 m are depicted in Fig 2(b). It is obvious that there is a distinct variation among the transmitters. We alo notice that the variation is not quite consistent over distances since the standard deviation at 4 m distance is quite larger than that at 2 m distance.
Following the same process to measure the variation of the transmitters, we exchange the roles of transmitter and receiver to measure the variation of receivers. We tested 100 RSS values for different receivers at the T-R separation distance of 2 and 4 m. One result is shown in Fig 2(c). The situation is similar to the variation of transmitters. An interesting point is that as the behavior of node 35 is far from other nodes, we find out that the node does not work correctly.
Several calibration algorithms were proposed to fight against the variation of transceivers. As we mentioned in section 0, they usually assume that there is the linear variation among different transceivers. However, we find that the linear model does not fit well. As shown in Fig 3, when the RSS data at 2 m is used to calibrate the RSS data at 4 m, the standard deviation after the calibration just decreases a little bit both for the variation of transmitters and receivers. It clearly points out that, besides the linear variations among transceivers, more parameters should be added in consideration of calibration algorithms.
When an omnidirectional antenna is applied in a transceiver, the power of the radio signal should be equally distributed in every direction so that the received signal strength remains the same regardless of the antenna orientation. Surprisingly, we found that this is not true for MICA2 sensor nodes.
In order to investigate the influence of antenna orientation independently, we must keep other parameters unchanged. Since we only compare the RSS values from the same transceiver, the variation of transceivers are excluded. Furthermore,it is well known that the received signal strength can be affected by the environment so that the received signal strengths vary at different places with the same distance to the transmitter. As we only rotate a sensor node round its antenna to change the antenna orientation, all the positions of the antennas are fixed.
At first, we study the transmitter orientation while fixing the receiver orientation. The four placements for the T-R pair is illustrated in Fig 4(a), in which the transmitter is put on left side and rotated by 0°, 90°, 180°, and 270°, respectively. We refer the transmitter orientation as the rotation angle of the transmitter. We sequentially test each node as the transmitter at the T-R separation distances of 2 and 4 m. Each following test is based on 100 RSS samples.
Similarly, we put one node as the receiver in a fixed position, and test 8 nodes as the transmitters. To avoid making the picture confusing, we only show the RSS data of 4 transmitters at the T-R separation distance of 4 m in Fig 4(b)due to the page limit. The RSS data of other transmitters is similar.
Following the same procedure to study the transmitter orientations, we exchange the roles of transmitter and receiver to measure the effect of receiver orientations. The RSS data of the same four nodes at the T-R separation distance of 4 m are depicted in Fig 4(c).
We conclude that antenna orientation does have a significant impact on RSS, because the difference of RSS values for different antenna orientations at 2 and 4 m can reach 5 and 7 dBm, respectively. Furthermore, all the curves in these figures nearly have the same shape. Especially, when the received signal strength is strong at 2 m distance, the four nodes nearly behave the same. The RSS values at 0° are similar to that at 90°, and the RSS values at 180° are similar to that at 270°. However, there is a big difference on the RSS values between the two groups.
Examining the four different placements in Fig 4(a), we found out that the effect of antenna orientation could be mainly explained by the node board that resides under the antenna and affects the signal transmission path. At the orientation of 0°and 90°, the node board of the rotating node lies outside of the line between two antennas. While at the orientation of 180°and 270°, the board of the rotating node lies inside of the line between the two antennas. The effect of antenna orientation comes as a blow to calibration algorithm because antenna orientation can hardly be calibrated in ad hoc node deployments.
It is very common that sensor nodes have different battery voltages. Even all the nodes are equipped with new batteries at the start, the variation of workload on the nodes lead to the variation of battery voltages. In this subsection, we measure the impact of battery voltage on RSS.
In the experiments, we deploy two sensor nodes, both of which periodically transmit packets to each other with the distance of 2 m. One node is provided with steady power supply, while the other node is powered by batteries. The two nodes keep transmitting and receiving packets for more than 12 h. Its battery voltage ranges from 3.10 to 2.85 V during the period.
The RSS values recorded at the node with steady voltage and the node with batteries are shown in Fig 5(a). In the former picture, the node with batteries acts as the transmitter. While in the latter one, it acts as the receiver.
It shows that the received signal strength approximately decreases as the battery voltage declines either in transmitter or in receiver side. However, the variance is quite small(<0.5 dBm)according to different battery voltages. We get similar results when different pairs of nodes are tested.
After investigating the parameters related to the sensor nodes, we start to consider the parameters related to the environment. It is well known that RSS varies with the passage of time even in static environments. A T-R pair has been tested for more than 7 h during the night when there are no people in our lab. The two nodes are supplied with steady voltage. The T-R separation distance is 2 m. The recorded RSS values are given in Fig 5(b). As we expected, the RSS values nearly keep the same.
Transmission of radio signal highly relies on the environment due to the well-known multi-path phenomenon.
Especially, the indoor environment is usually more complicated than outdoor environment because of furniture, equipment, and surrounding walls. After exploring temporal variation, we study spatial variation that involves changing node positions. The change of node positions leads to the change of the signal transmission path, which definitely will influence RSS. Nevertheless, the signal transmission path is always kept unchanged in previous experiments. We divide the spatial variation into three cases:
Firstly, we investigate the effect of different directions on RSS. Here, direction means the direction of the line passing from the transmitter to the receiver. Three directions illustrated in Fig 5(c), called direction 1, 2, and 3, have been tested. The first direction is along one side of the wall, the second is along the other side of the wall, and the last has 45° azimuth with both two walls. We place a T-R pair along three directions. The two nodes should be rotated around their antennas according to different directions so that their relative antenna orientations are kept unchanged. For example, when the two nodes are moved from direction 1 to direction 2, they needs to be counterclockwise rotate 90°. Ten different pairs of nodes are tested.
To avoid making the picture confusing, we only show the RSS values of four pairs in Fig 5(d), in which the T-R separation distance is 2 m. Obviously, there is a distinct variation of RSS values among the three directions. Compared with directions 1 and 2, direction 3 has a positive influence on the signal strength. The situation is nearly the same when we increase the T-R separation distance to 4 m. It seems that direction 3 still strengthens RSS a lot. However, the difference between direction 1 and direction 2 is not very consistent with the difference between them at 2 m distance.
From the two figures above, we could find out that, though the T-R separation distance stays the same as well as the environment of the room, the signal strength can vary up to 10 dBm among different directions.
Secondly, we simultaneously move a T-R pair to four different places illustrated in Fig 5(e)while keeping the direction as direction 2, which is along one side of the wall. The same ten pairs of nodes are tested. The RSS values of the same four pairs at the four different places are shown in Fig 5(f). The four curves have similar shapes. Clearly, different places have different impacts on the received signal strength due to the different signal transmission paths affected by the surrounding objects around the places. The signal strength at place 1 is the weakest among the four places, while the place 3 strengthens the signal strength most.
From our experience, in the indoor environment, the reflection effect of the walls, ground and the objects play a big part in the influence. However, it is hard to separate different objects and investigate their effects on the received signal strength.
Finally, we study the reflection effect of ground by adjusting the height of nodes. We simultaneously increase the height of the T-R pair so that they can be maintained in the same horizontal plane. Nine different heights, from right on the ground to 1.2 m with each step being 15 cm, are measured. The T-R separation distance is 2 m. As the height of T-R pair changes, the most changed factor is the ground reflected path of the signal. The same ten pairs of nodes are tested. The RSS values of the same four pairs are shown in Fig 6(a). There is a significant variation among the RSS values at different heights. Furthermore, the four curves are quite similar. The RSS values can reach up to 10 dBm higher as we increase the height. Such huge RSS difference is due to the multi-path phenomenon especially indoors. The receiver may receive many reflection paths from the source. As the reflection effect of ground really affects the received signal strength, it implies that RSS could be very complex and different when 3-dimension deployment is considered.
Previous experiments were all done under static environments, or we can say that the environment is nearly static. For example, there is no person moving around who can interfere with the signals. However, the environment is hard to be controlled in real deployments. Sensor networks may reside in a dynamic environment with some moving objects. In this section, we want to estimate the influence of the dynamic environment on RSS.
We introduce one person, who moves around in carefully selected zones, as environmental dynamics to the T-R pair. The four dynamic zones, called zones A, B, C, and D, are demonstrated in Fig 6(b). Zones A and B are outside the line segment between the T-R pair. The dynamics in zone C breaks the line-of-sight between the T-R pair. While zone D is along the line segment between the T-R pair without obstructing the line-of-sight. Four different T-R pairs are tested. The average RSS values of two pairs with intervals according to the dynamics in different zones are shown in Fig 6(c)in which S refers to the RSS values in the nearly static environment. As is expected, compared with the static case, RSS values begin to fluctuate when dynamics are introduced. RSS is significantly affected when the line-of-sight is broken. The variance of RSS under the dynamics in zone D is slightly larger than that in zones A and B.
This paper, conductes a series of experiments to investigate the effects of eight parameters on RSS comprehensively and independently, which are ignored in the ideal propagation models. We show that, besides the linear variation among different transceivers, more parameters should be added in consideration of calibration algorithms.
Surprisingly, we find that the sensor boards attached under the antennas stand in the signal transmission path and make the omnidirectional antenna look like directional antenna. However, the antenna orientation is not considered in current calibration algorithms. The height of antenna could affect RSS significantly, which could cause problems when we extend the current RSS-related solutions from 2-dimensional space to 3-dimentional space. Dynamics in environments can also cause problems for RSS-related solutions, especially when the line-of-sight between the T-R pair is blocked. This can restrict the application of RSS-related solutions in some harsh environments.
The highly irregular characteristics of RSS could challenge the assumptions in many current papers. The communication area of a sensor node is always an irregular shape, which could be a concave one. Furthermore, the area could change frequently due to the environmental dynamics.
It leads to a large portion of asymmetric links and dynamic hop distances, which are always not considered in many papers. Localization is also highly affected since the ranging information could be very inaccurate. As far as link quality is concerned, our experimental results support the existence of the gray area, in which PER varies irregularly.
In future work,we can explore these directions based on our experimental data.Moreover,as the temperature of lab is always controlled by air-conditioning,temperature is a parameter we could investigate in future.We notice that temperature can also affect RSS when we deploy sensor nodes in outdoor environments.As planned,the impact of each parameter on the RSS will be investigated in our future work.