With advances in autonomous navigation, positioning, and in general robotics technologies, small to miniature-sized unmanned aerial vehicles (UAVs, or colloquially called drones) are witnessing their ever-increasing use in engineering practice. Small UAVs are low-cost, agile, and flexible in imaging payloads as remote sensing platforms when compared with traditional space- or airborne platforms (e.g., satellites or aircrafts) [1
]. Today’s UAVs have incorporated the latest GPS technology. Many small UAVs, especially the multi-motor ones, can fly following predetermined GPS waypoints. Some advanced drones have been equipped with lost-cost radar or vision sensors, acquiring a minimum level of flying beyond (visual) line-of-sight (BVLOS or BLOS) due to their sense-and-avoid capabilities [2
]. This potentially would further render small UAVs an attractive remote sensing platform for numerous different applications.
On the other hand, wireless sensing network (WSN) technology has matured in recent years, with applications found in many scientific and engineering projects. Many WSN applications focus on ad-hoc tasks, wherein the local sensors are deployed with a goal of completing the task in a short time. Therefore, the energy consumption is not excessive. However, for deploying WSNs over a geospatially large or spatially complex space with an expected long-term monitoring mission, both WSN implementation and energy efficiency become the primary challenges. One possible application of sensing is in farming land, wherein precision agriculture (PA) practice demands data sensing at different granular (spatial and temporal) scales [5
]. Another application scenario is performing structural health monitoring (SHM) for civil structures and lifeline infrastructure systems that are often massive and spatially complex (e.g., urban buildings, long-span bridges, and power transmission lines/towers, etc.). For structures that are critical to society, sensors, and especially WSNs, can be installed in these structures in their lifetimes to achieve ‘smart structures’. Continuous health monitoring through these WSNs provides stakeholders a basis for ensuring public safety and for decision-making when dealing with unexpected natural or technological disruptions [6
Taking the two areas of PA and SHM as the application settings (Figure 1
) it is asserted that, in both situations, the necessity of combining UAV-based remote sensing and WSNs is straightforward. In PA, the traditional practice relies on sensing data such as space- or airborne imagery for decision-making and management in farming [7
]. However, the high cost and the long revisit period of satellite or aerial imagery may prevent applying precision agriculture solutions at any location and time around the world. Images taken by low altitude UAVs give an alternative solution in the emerging precision agriculture practice [10
]. In addition, since microelectromechanical systems (MEMS) technology and, particularly, the emerging Internet of Things (IoT) sensing technology have been rapidly improved in recent decades, many researchers have proposed and implemented different ground-based wireless sensor solutions for facilitating precision agriculture [13
]. To achieve data fusions and more intelligent and tactical operations in these sensing modalities, integration of UAVs and ground-based WSNs becomes a rational choice. In the arena of SHM, the traditional approach is to deploy wired or wireless sensors to obtain the real-time responses of structures to environmental or hazard-induced vibrations [18
]. In reality, however, for structures with slight to moderate damage, such as local cracking and corrosion, visual or remote sensing-based inspection is the most efficient approach to date. In recent years, as UAV technology has penetrated into many industrial sectors, small UAVs have enabled remote sensing that is low-cost, highly mobile, and is being treated as an emerging tool that expands the SHM technology inventory [22
]. This further corroborates the necessity of combining WSNs and UAV-based remote sensing technologies.
Reflecting on the trends in PA, SHM, and other similar geospatial-scale monitoring applications for critical missions, we have proposed and developed a prototype to realize a wireless aerial imaging and ground-sensing network; in short, it is termed the aerial–ground wireless sensing network (AG-WSN) [25
]. First, this AG-WSN features the use of one or multiple UAVs as the primary imaging nodes, which will serve as the gateway to the ground sensors; second, the wireless sensing units are deployed (by UAV delivering or manual installation) in the ground (or ground structures) over a geospatially large or a spatially complex space. The combination of low altitude imaging and ground sensing provides the power of fusing remotely captured images with a high resolution and point-based ground-truth data in the field. The high mobility of the UAV allows it to be deployed opportunistically according to scheduled tasks or in response to unexpected urgencies (e.g., disasters). Combining the collaborative aerial and ground sensing and the opportunistic operation modes (e.g., a ground node may be only active when the UAV hovers above it and collects data from it), we state that the proposed AG-WSN can potentially provide the most high-fidelity and most flexible sensing solutions to many monitoring problems arising from the need to assess geospatially large and complex built/agriculture environments.
In this paper we first address the opportunistic nature of the AG-WSN by reviewing the related UAV-WSN integration efforts and proposing a conceptual operation design, which further motivates the proposition of sensor activation for network energy efficiency. Centering around sensor activation, we propose to develop a sensor wake-up solution. Related work is reviewed that shows the benefits and drawbacks of different wake-up designs and the rationale for choosing an active radio frequency (RF) mechanism. Subsequently, a general out-band RF wake-up mechanism is developed and demonstrated. For a comparative purpose, the infrared wake-up prototype is implemented too. We further conduct a comprehensive study on energy conservation, followed by the conclusions and remarks for this paper.
The scholarly contribution in the design of this work is two-fold: first, it provides a realistic design of active RF-based wake-up mechanisms for sensor activation in a novel aerial–ground sensing network (AG-WSN), which bears the goals of rapid prototyping and having a realistic application in urban- or farming-scale monitoring tasks; second, the experimental evaluation reveals the superb effectiveness of the proposed wake-up method and its appropriateness for the AG-WSN system. The technical contribution in the experimentation of this work also includes (1) the use of a digital imaging approach to measure the wake-up time delay and (2) the resulting time-dependent rates of the battery-based power consumption using three different wake-up methods.
2. Unmanned Aerial Vehicle Wireless Sensing Network (UAV-WSN) Integration, Opportunistic Sensing, and Research Needs
To our best knowledge, there were only a few efforts that attempted to integrate UAVs with wireless sensor networks. In [26
], UAVs were considered as mobile sinks for ground sensor data dissemination. This approach intended to optimize the route from a given sensor node on the ground to a few mobile sinks that move in the area. Authors in [27
] presented a different approach that kept the sensor network continually connected. It used multiple UAVs to establish a reliable relay network to guarantee the delivery of data produced by the wireless network nodes on the ground to the users. Given these few simulation-based and conceptual efforts, even fewer efforts are found that physically realized a UAV-based sensing network system. In a recent effort, the authors developed a WSN using a fixed-wing UAV as the aerial gateway for marine data collection [28
]. In our recent effort, we further investigated the interference between the WiFi-based video transmission link and the ZigBee-based ground-data transmission links [25
The use of flying single or multiple UAVs, either as a mobile sensor node or a data sink, triggers the need to optimize network efficiency between sensors and sinks. Energy cost is an inevitable constraint, considering that both the UAVs and ground sensors are, to date, usually battery-powered. An opportunistic network is the emerging technology that solves such an optimization problem. In [29
], protocols were proposed to better exploit the durations of high-quality channel conditions. Based on that, authors [30
] proposed routing protocols that increase the throughput of large unicast transfers in a multi-hop wireless network. There are also a number of research efforts on optimizing resource and performance in wireless sensor networks. In [31
], the authors considered a different scenario where the paths from message sources to their destinations do not always exist. Then the authors analyzed protocols that alleviated the problem of chronically disconnected paths by having a node store the packet, carrying it until meeting another relay node and forwarding the packet to the other relay node. In a more recent effort, researchers also developed middleware that implemented the opportunistic network into mobile social networks [32
]. It is noted that for general opportunistic networking systems (without using a UAV as a gateway node), different protocols are proposed, including the flooding protocol and the history-based protocol (e.g., [33
]). Another possible approach at a higher level is to apply the software defined network (SDN) concept into an opportunistic system. In a recent work [35
], the author proposed an energy efficient SDN framework for wireline–wireless cross networks, which can potentially be helpful for large-scale deployment. Last, it is stated that these optimization or software-based schemes mostly focus on designing improved communication protocols by assuming that either the UAVs or the sensors are not constrained by the battery-based power.
To illustrate such energy constraint, Figure 2
illustrates a conceptual AG-WSN, where (besides being the imaging and computing hub) the UAV is designed as a robotic vehicle that flies to ground sensors at tactical locations. This operational mode, and furthermore, the possible loss of sensors, sensor malfunctions, and out-of-range communication, renders the underlying network as opportunistic, which affects energy consumption in the UAV and sensors.
In Figure 2
a four subnets are shown, which indicate four physically isolated sensor networks in the fields, except that the UAV can fly to each subnet to execute opportunistic sensing. Figure 2
b indicates the idealized situation where sensor failure (or other malfunctions) and energy consumption does not need to be considered. Hence, assuming each node Nmn
can communicate with its neighboring nodes Nm±1,n±1
through a wireless connection, when the UAV flies into this sub-network some of the nodes are in the communication range of the UAV (within the red dashed-line circle), while some are not. The UAV will pick one of the nodes in the range as a relay node, herein which is N21
, and collect data from any other node in this subnet. Figure 2
c illustrates when sensor failure happens; the UAV can move from one disconnected zone to another one. Figure 2
d demonstrates a different scenario where all sensor nodes are far away from each other and no wireless connection can be established between them. In this case, the UAV will reach within each node’s communication range to communicate with individual nodes. Then, the existing optimal communication protocols can be used.
When the energy consumption of either the UAV or sensor networks is considered, optimization in the physical layer (rather than in the communication protocols) needs to be addressed. Two apparent scenarios exist: (1) through spatial path-energy optimization the UAV finds the optimal flying path through the geospatially deployed ground sensors, for which it belongs to a typical traveling salesmen problem (TSP) and is being tackled in our recent research [36
]; and (2) through a sensor activation approach, as being concentrated on in this paper, such that sensors are only active when the UAV is in its neighborhood.
3. Sensor Activation and Related Work
Although solar power or other intermittent energy supply techniques exist, battery power continues to be considered as the most reliable source for powering sensors and robots. By implementing the commonly adopted duty-cycle method, wireless sensor nodes could be pre-programmed to wake up and communicate with the gateway, then go back to sleep after communicating. This approach for extending battery life has been treated as a default function in many commercial wireless sensors. A number of researchers also attempted to optimize power management to further extend the battery life of WSNs [37
]. However, one key problem that prevents us from realizing long-term aerial–ground sensing is the opportunistic nature of deploying the UAV (gateway) and the sensor nodes. In an AG-WSN, the gateway (as a payload of the UAV) is deployed spatially to proceed toward the ground sensors on a non-scheduled basis or randomly upon abrupt events. This further implies that the ground sensors do not have ‘knowledge’ or are not programmable to realize duty-cycle sensing. If the ground sensors are turned on to include at least the microcontroller and communication units (whereas the sensing units may be on or off according to the duty cycles), the battery of the sensor nodes may be drained quickly.
One straightforward approach to such energy inefficiency issues is to wake up ground sensor nodes when the UAV is deployed as needed to approach the sensors without any preprogramming. In this paper, we first propose to use a radio frequency (RF)-based out-of-band wake-up mechanism. Then, comparative studies are conducted to investigate their energy saving performance against two other wake-up mechanisms. Using a traditional star-like sensor network, the analytical and experimental studies show solid evidence that the RF-based wake-up mechanism outperforms the other two solutions on energy consumption.
Earlier efforts reveal that data transmission in a WSN is generally very expensive in terms of energy consumption, whereas data collection (or the sensing itself) consumes significantly less [41
]. For this reason, various methods are developed to extend the life of battery-powered WSNs by reducing the power consumption of the wireless modules. A significant number of efforts were found that focused on developing lower level network protocols by adopting duty-cycle based solutions [42
]. These studies aimed to optimize the network protocols, specifically through reducing the energy consumption during the idle or the listening time of the wireless modules. For example, the authors in [45
] proposed an adaptive medium access control (MAC) protocol, which introduced a flexible duty-cycle method and claimed to reduce 96% of energy use compared with traditional protocols. However, the core concern for these duty-cycle solutions is that the wireless modules do not know when the data transmission is coming or required, and the node must listen periodically to limit data latency; thus, the duty-cycle ratio cannot go arbitrarily low [46
]. Also, duty-cycle methods may have problems with delay and synchronism and, hence, the protocol is relatively complicated. As such, using waking-up mechanisms to answer this concern has been extensively studied.
A number of sensor activation methods have been proposed to date. Essentially, such an activation approach features a waking-up mechanism for activating sensing modules in an as-needed (or on-demand) basis. There are two categories of methods when considering wake-up mechanisms for use in wireless networks: in-band and out-band. If an in-band method is used, a special value is transmitted through the data channel to send out the wake-up signal. By contrast, a separate channel is needed to realize such a waking-up mechanism in an out-band method. Using in-band methods can reduce the complexity and cost of the implementation. A recent study on the in-band wake-up method [47
] claimed that, by using both game theory and reinforcement learning techniques, it achieved very effective sleep/wake-up scheduling. However, it kept the wireless communication channel busy and may have required more energy consumption. From an energy-efficiency perspective, the out-band approach is more suitable for the proposed concept that emphasizes opportunistic aerial-ground sensing.
There are many studies that employ the out-band wake-up mechanism [48
]. In this paper, they are categorized into two groups according to their communication medium: (1) non-RF based and (2) RF-based. In a non-RF based mechanism, researchers proposed wake-up methods using infrared (IR), optical, and acoustic signals. The authors in [49
] developed an IR LED-based wake-up mechanism, in which the receiver was a photo-detector that received IR signal and then generated an interrupt. The authors stated that their IR design only consumed 12 μW while listening. It is noted that the obvious drawback of this prototype is the sensitivity of the circuits to external light and vulnerability to ambient noise. In [50
], the authors presented a home-energy management system using infrared, signal-based control over a Zigbee network. In this system, an infrared receiver was attached to the Zigbee gateway. The Zigbee gateway was responsible for communicating with other home appliances, whereas the infrared remote control was the out-band wake-up channel used to wake up the Zigbee network. Unfortunately, this paper did not mention the power consumption of the IR receiver. To our understanding, this type of IR receiver in the paper is commercially available and similar to the one used in our experiment, as shown in this paper, which has a better resistance to noise at the cost of a much higher power consumption, and it may require up to 45 mW according to our experiment.
Optical communication is another non-RF option for the secondary wake-up channel. Two efforts in [51
] used free-space optical (FSO) communication as the transceiver. The receiver, at idle listening, consumes 317 μW and 695 pW. However, the transceiver and receiver both need to be placed in line-of-sight (LOS), and the data rate is slow. It is impractical for use in a UAV since it mostly does not stay in a position that accurately faces the transmitter. Thus, this option is not suitable for our application. The AG-WSN scenario may also limit the use of acoustics as wake-up methods [53
] due to the noise produced by the UAV blades. Ultrasonic, as stated in [55
], may avoid the noise made by the UAV. It uses a piezoelectric transducer that converts the mechanical energy into electrical energy for generating wake-up interrupts. However, most ultrasonic communication and ranging efforts, to date, are applied in indoor (short-range) or LOS scenarios [57
In comparison to the non-RF based wake-up mechanisms reviewed above, RF-based communication has the advantages of not requiring LOS, it has better noise and interference tolerance, a higher data rate, and it is more cost-effective. Research on the RF-based wake-up mechanism can be divided into two designs: passive wake-up and active wake-up, both of which have been well studied in the laboratory environment. In a passive design, the RF receiver harvests energy from the transmitter to power itself, thus, requiring no power supply [59
]. In [62
], the authors simulated a passive RF wake-up receiver, in which they indicated that comparing with the existing duty-cycle method, their RF wake-up could significantly enhance energy efficiency by up to 70%. There are also simulations on both passive and active RF wake-up circuits, such as [63
]; the authors of these efforts later implemented the passive RF circuit into a sensor network with a multi-hop capability [65
]. However, among these passive RF-based methods, information on the communication range between the transceivers was not found. In addition, it was reported that the harvested energy by the receiver may decrease with increasing distance between the receiver and the transmitter. The authors in [66
] showed that the hardware setup could only reach a maximum distance of 4 m for a successful wake-up. Considering the AG-WSN scenario proposed in this work, the passive RF-based wake-up design is not suitable.
Regarding the active RF wake-up design, as mentioned in [46
], there are 13 active RF-based wake-up methods using discrete components, whereas there are 29 methods using complementary metal–oxide–semiconductor (CMOS technology. The most significant parameters relevant to these designs and prototypes for the interest of the proposed AG-WSN configuration are power consumption, range, address decoding capability, wake-up latency, and balancing. For example, the author in [68
] configured the wake-up receiver using discrete components and claimed to achieve 120 m of communication range. However, the receiver consumed 1620 μW at the state of idle-listening, which is too high for the battery-powered nodes. There is a low-power design in [69
], which only consumed 52 μW. Unfortunately, the authors did not provide a range test. A favorable design was presented in [70
] recently. It achieved a communication range of 50 m at idle-listening with a power consumption of 1.2 μW. Unfortunately, at the time of our experiment, there was no market-ready product or porotype based on this design.
In this paper, the use of off-the-shelf components is stressed in our prototyping and experimental validations, with the goal of rapidly putting the proposed AG-WSN into practice. As many researchers have similarly done, the AS393X wake-up receiver has been used in many efforts and designs [71
]. Among these researchers, the author in [74
] used AS3933, which is the same chip in our experiment, to prototype the receiver circuit to have an 87 m communication range, at the cost of more than 5000 μW of power consumed when decoding the wakeup signal. Also, in a recent paper, the authors compared the RF wake-up mechanism and the low-power listening techniques [76
]. They concluded similarly what we achieved in our energy evaluation results in this paper. However, the authors of this paper did not measure the delay caused by the RF wake-up transmission, and their power consumption measurement was not based on batteries, but a constant power supply, hence, it lacked a realistic configuration.
7. Conclusions and Remarks
In this paper we reviewed the concept of using an aerial-ground wireless sensing network (AG-WSN) in a remote and geospatially large, complex space. We particularly recognized the need of implementing such sensing solutions in precision agriculture and structural health monitoring practices. We then recognized the technical challenges in achieving energy efficiency in the ground sensors. Different wake-up mechanisms are then reviewed and compared. Among those mechanisms, we chose an active radio frequency (RF)-based wake-up method and implemented it physically. The focus was on evaluating their performances to achieve energy efficiency in the battery-powered ground sensors. The following findings were achieved through the experimental evaluation in this work:
The experimental results in this paper indicated that the RF-based out-band wake-up mechanism can save a great amount of energy compared with the other two solutions (the infrared wake-up and the default duty-cycle methods). A direct comparison between the RF-based solution and the infrared-based solution indicated that the RF-based wake-up mechanism had a noticeably better performance in the wake-up range, and had a tremendous improvement in power consumption. Specifically, the results showed that the RF-based wake-up mechanism could potentially save more than 98.4% of the energy that the traditional duty-cycle method would otherwise consume, and 96.8% if an infrared-receiver method was used.
The energy consumption for different RF wake-up distances was evaluated in this paper. The results indicated that as the distance between the transmitter and receiver increased, the receiver consumed more power (around 8.4 μJ). However, it was argued that this value could be ignored compared to the energy consumed in the listening mode, which was at least 103 times higher.
The evaluation of wake-up time delay by using a variety of different wake-up signal codes indicated that the time delay was below 80 ms; hence, the delay will not affect most opportunistic sensing applications (wherein the sensors sense the data at one time and transmit at a later time, then the sensors go back to sleep mode until another abrupt event). However, it was pointed out that a stricter time delay evaluation needs to be conducted if synchronization is critical between the sensors.
Given the three findings, it was concluded that the RF wake-up mechanism was the first candidate for implementing the proposed wireless aerial-ground sensing network for monitoring applications in large-scale geospatial or challenging spaces. The technical contribution also included the use of a digital imaging approach to measure the wake-up time delay, and the resulting time-dependent rates of the battery-based power consumption using three different wake-up methods. This experimental and empirical knowledge may be extrapolated in similar sensing network research where sensor activation needs to be integrated.
Last, we point out a few limitations in this effort that warrant future exploration. The first limitation is the RF wake-up range with the device used in this effort, which was relatively short (0‒7 m) compared with the Xbee communication range (about several hundred meters). One practical solution for this is to increase the power of the wake-up transmitter and to consider advanced antenna design (such as using the multiple input, multiple output, or MIMO technology) to increase the sensitivity of the wake-up signal. This may introduce more power consumption on the UAV side. With this potential long-range RF-based wake-up mechanism, the UAV no longer needs to reach the proximity of the sensor nodes. The UAV path planning then becomes a traveling salesman problem with neighborhoods, which was covered in our recent research [36
The second limitation is the star topology assumed for the ground-level WSN. It is arguable that a mesh topology with communication relays can be used. Potentially, we believe that this decision depends on application context. In our context the UAV had two roles: the gateway to wake up sensors and collect data, and a remote sensing platform to collect geospatial images of the ground. Given this mobility and its dual role, a star topology is considered more appropriate. Nevertheless, if a mesh or other dynamic topology is considered in conjunction with the UAV-based sensor activation, this will demand a new research endeavor. A different scenario that uses the star topology for collecting data in a large geospatial area was discussed in our recent work [36
]. In this case, the whole WSN can be segmented into a few subsets with a star topology, and the UAV will stay hovering in each subset for both wake-up and data collection procedures. Once the task is performed, the UAV can fly to the next subset. This practical treatment is believed to be a feasible solution to use the UAV with the aforementioned dual roles in an aerial–ground network in larger outdoor environments.