Portal de Periódicos da CAPES

A mobility‐aware cluster‐based MAC protocol for radio‐ frequency energy harvesting cognitive wireless sensor networks

2021; Volume: 11; Issue: 5 Linguagem: Inglês

10.1049/wss2.12021

2043-6394

Arif Obaid, Xavier Fernando, Muhammad Jaseemuddin,

Energy Efficient Wireless Sensor Networks

IET Wireless Sensor SystemsVolume 11, Issue 5 p. 206-218 ORIGINAL RESEARCH PAPEROpen Access A mobility-aware cluster-based MAC protocol for radio- frequency energy harvesting cognitive wireless sensor networks Arif Obaid, Corresponding Author Arif Obaid arif.obaid@ryerson.ca orcid.org/0000-0002-5082-0053 Department of Electrical and Computer Engineering, Ryerson University, Toronto, Canada Correspondence Arif Obaid, Department of Electrical, Computer, and Biomedical Engineering, Ryerson University, 350 Victoria Street, Toronto, ON, M5B 2K3, Canada. Email: arif.obaid@ryerson.caSearch for more papers by this authorXavier Fernando, Xavier Fernando Department of Electrical and Computer Engineering, Ryerson University, Toronto, CanadaSearch for more papers by this authorMuhammad Jaseemuddin, Muhammad Jaseemuddin Department of Electrical and Computer Engineering, Ryerson University, Toronto, CanadaSearch for more papers by this author Arif Obaid, Corresponding Author Arif Obaid arif.obaid@ryerson.ca orcid.org/0000-0002-5082-0053 Department of Electrical and Computer Engineering, Ryerson University, Toronto, Canada Correspondence Arif Obaid, Department of Electrical, Computer, and Biomedical Engineering, Ryerson University, 350 Victoria Street, Toronto, ON, M5B 2K3, Canada. Email: arif.obaid@ryerson.caSearch for more papers by this authorXavier Fernando, Xavier Fernando Department of Electrical and Computer Engineering, Ryerson University, Toronto, CanadaSearch for more papers by this authorMuhammad Jaseemuddin, Muhammad Jaseemuddin Department of Electrical and Computer Engineering, Ryerson University, Toronto, CanadaSearch for more papers by this author First published: 10 June 2021 https://doi.org/10.1049/wss2.12021AboutSectionsPDF ToolsRequest permissionExport citationAdd to favoritesTrack citation ShareShare Give accessShare full text accessShare full-text accessPlease review our Terms and Conditions of Use and check box below to share full-text version of article.I have read and accept the Wiley Online Library Terms and Conditions of UseShareable LinkUse the link below to share a full-text version of this article with your friends and colleagues. Learn more.Copy URL Share a linkShare onFacebookTwitterLinked InRedditWechat Abstract Cognitive wireless sensor networks (CWSN) are severely energy constrained and radio frequency (RF) wireless energy harvesting (RFWEH) has been shown to improve the network lifetime. In many CWSN applications, node mobility imposes challenges owing to changing network topology. Therefore, the design of a new medium access control (MAC) protocol that can handle node mobility as well as energy harvesting is required. A cluster-based multihop MAC protocol (RMAC-M) is proposed that incorporates RF energy harvesting in a mobility-aware CWSN. Our protocol selects cluster heads using an algorithm based on an R-factor parameter consisting of residual node energy, residual node data and node speed, with appropriate weights. It then transmits data packages using a multitier super cluster head routing mechanism without the need for neighbour discovery. The multitier clustering and RFWEH mechanisms boost the energy performance of the network, increasing its lifetime. On the other hand, time slots allocated for RFWEH increase delay, thereby affecting system latency. Owing to its unique nature, the proposed algorithm has no comparable protocols in the literature. For the sake of completeness, RMAC-M is compared with well-known MAC protocols such as LEACH-M and KoNMAC that do not have energy harvesting or mobility features. Simulation results show that the proposed protocol increases the lifetime of the CWSN nodes substantially, promising a self-sustainable network in terms of energy. Furthermore, despite the allocation of time slots for energy harvesting, critical network parameters such as throughput, packet loss and average delay remain within target levels. 1 INTRODUCTION Because of spectrum scarcity and the mushrooming of Internet of Things devices, cognitive radio networks have gained in popularity. They allow users to transmit data on unused portions of the licenced spectrum allocated to another network without affecting the performance of that network [1]. A cognitive wireless sensor network (CWSN) is a wireless sensor network (WSN) with cognitive radio capability. In a CWSN, apart from performing regular WSN activities such as environment sensing, transmission, receiving, and idle listening, sensor nodes need to perform additional cognitive functionalities such as spectrum sensing, sharing information, and adapting to the radio environment [2]. Hence, the energy demand of a CWSN is much higher than that of a typical WSN, and its lifetime would be short unless some measure is taken. It has been shown that radio frequency (RF) energy harvesting is a promising way to sustain the CWSN [3]. To avoid long periods of energy harvesting that are generally required to accumulate enough energy to perform network operations, it has been proposed to harvest energy from high-power RF sources such as TV and radio antennae [4]. Furthermore, CWSN generally have lower throughput, higher latency and higher packet loss compared with regular WSN owing to limited spectrum access [5]. Traditionally, these issues have been tackled by better collision detection and avoidance schemes, prolonged sleep time, and so on, with limited success [6]. The additional allocation of several time slots for energy harvesting would cause further delay and lower throughput. Moreover, node mobility alters network topology and causes disruption to the network. As a result, the network becomes less stable and needs to self-organise to mitigate against node link stability and energy losses. Because a CWSN consists of both static and mobile nodes, it can be challenging to maintain network robustness and connectivity. Hence, a properly optimised medium access control (MAC) layer scheduling scheme is needed for RF energy harvesting in a mobility-aware CWSN; this is the objective of this study. To the authors’ knowledge, no effort has yet been made to design a suitable MAC protocol for self-sustainable mobility-aware CWSN using energy harvesting from high-power RF energy sources. The primary design parameter of the proposed RFWEH mobility-aware CWSN MAC protocol is to ensure the CWSN is sustainable in terms of energy. The R-MAC protocol that we proposed previously [7] ensures that other important network parameters such as throughput, packet loss, and delay remain within acceptable limits through multitier clustering, proper channel selection and recursive frame design for optimising harvesting gains. Here, we augment R-MAC with a clustering scheme that reduces the impact of node mobility on the stability of clusters without significantly compromising the harvesting gain. 1.1 Cluster-based MAC protocols Energy-efficient MAC protocols, with features such as longer sleep duration and minimal collisions, are essential for energy-starved CWSN. Cluster-based MAC protocols are proven to reduce energy dissipation owing to short transmission distances. Furthermore, clustering is a structured way to manage topology effectively and increase system capacity and stability [8]. LEACH, an energy efficient MAC protocol, is a flagship protocol that introduced the clustering approach in WSN [9]. Since its introduction, many improvements to LEACH have been proposed [10-12]. LEACH-M is a version of LEACH that improves on its performance using a centralised approach. Various cluster-based energy efficient cognitive MAC protocols have been proposed in the literature, such as KoNMAC [13], CogMesh [14], ClusterMAC [15], and CM-MAC [16]. However, these protocols do not incorporate energy harvesting or node mobility. We introduce a mobility-aware MAC protocol (in Section 3), which incorporates various techniques to ensure energy preservation. We employ clustering of nodes based on an R-factor parameter and higher-tier superclustering for packet routing. Our proposed protocol also harvests RF energy from high-power sources such as TV and radio station towers, enabling the network to gain energy. Furthermore, it incorporates node mobility to cater to many applications that require it. Because we cannot find such a comprehensive MAC protocol in the literature, we cannot perform a direct comparison with our work. However, we show the superiority of our proposed algorithm to existing well-known methods after incorporating energy harvesting from high-power devices within these protocols using exhaustive simulation results. 1.2 Related work Many cognitive MAC protocols have been proposed [6, 17]. Cluster-based cognitive MAC protocols are fewer, but they have multiple shortcomings. For instance, Chen et al. [14] proposed CogMesh, a popular cluster-based MAC protocol for cognitive networks. It forms clusters based on nodes that share local common channels, assuming a multichannel multiaccess network. It uses a guaranteed access period for intracluster communication. It is designed for static networks (i.e., where nodes are stationary), so its performance is degraded in dynamic networks. Along the same line, Li et al. introduced a cluster-based MAC protocol for cognitive ad hoc networks [15]. It forms clusters based on the geographical location of nodes, available primary channels, and network statistics. A database stores spectrum-occupancy statistics to support neighbour discovery and cluster formation. Nodes share their neighbour link values among themselves and form clusters based on these link values. Unfortunately, this protocol experiences energy losses owing to frequent database updates and neighbour discovery. Another cluster-based MAC protocol, KoNMAC, was proposed by Xu et al [13]. It is a schedule-based protocol developed for multihop CWSN. It works with stationary nodes and has predetermined cluster heads (CHs) and intracluster gateways. It uses a channel weighting concept to distinguish among primary network channels. Furthermore, to achieve better energy efficiency instead of sensing all available channels, nodes sense only a small number of channels. Owing to transmission scheduling and generated traffic classes, KoNMAC has poor quality of service (QoS). Hu et al. proposed CM-MAC [16], a cognitive MAC with mobility support. It is a carrier sense multiple access/collision avoidance-based protocol with a dedicated common control channel (CCC). Both the node transmitter and the receiver sense the spectrum and combine the information into a ready to send/clear to send frame, thereby sharing this information with their one-hop neighbours. CM-MAC splits data packets and transmits them on multiple channels simultaneously to improve throughput. From an energy harvesting perspective, all the work done is based on energy being harvested from low-power energy sources. For example, Lee et al. proposed a traditional opportunistic RFWEH model [18]. In this model, Secondary users (SUs) harvest RF energy from nearby primary users (PUs), while opportunistically accessing the licenced spectrum. Each PU is centred around a guard zone (for interference protection from SUs) and a harvesting zone (for RFWEH by the SUs). As a result, each SU is in transmitting mode (if it is outside the guard zones of all active PUs and fully charged), harvesting mode (if it is inside the harvest zone of an active PU and not fully charged), or idle mode (otherwise). Zhai et al. proposed a cooperative model for energy harvesting [19]. In this model, SUs allow PUs to harvest their RF energy and relay PU signal to the access point (AP) in exchange for a portion of the primary network’s bandwidth. Because the RFWEH and data decoding are performed by different circuits at the receiver, a trade-off exists between maximising RFWEH and data transmission. In each time slot, RFWEH is performed in the first τ fraction of time, whereas data transmission is done in the remaining time. Continuing this work, a hybrid RFWEH model was proposed in Obaid and Fernando [20]. In this model, SUs access the primary network opportunistically; however, a fraction of the bandwidth is given to the secondary network in exchange for SUs relaying PU data to the primary network APs cooperatively. LEACH, an energy-efficient MAC protocol, is used for better energy management. Another energy-harvesting aided spectrum sensing heterogeneous cognitive wireless sensor network is proposed by Zhang et al. [21]. In this proposal, spectrum sensors harvest energy from radio signals or ambient energy sources, whereas data sensors are battery operated. Their solution results in sustainable spectrum sensors but only minimises energy consumption of data sensors, which could die out over time. Adaptive energy harvesting is also considered a promising method to maintain a balance between extending network life and communication delay. For instance, Liu et al. [22] proposed an efficient MAC protocol with adaptive energy harvesting. The authors considered variables such as energy-harvesting time, contending time, and contending probability to obtain optimal throughput of the network. These MAC protocols are not designed to incorporate RF energy harvesting or are not designed to handle mobile nodes in CWSN. 1.3 The proposed protocol A cluster-based mobility-aware efficient cognitive MAC protocol is proposed that incorporates RFWEH. Building on our previous work on a static MAC protocol (R-MAC) [7], the proposed MAC protocol (RMAC-M) adds support for the mobility of sensor nodes. The protocol consists of four phases: first, nodes form clusters based on an R-factor criterion. Because R-factor values are not required to be shared with other nodes, energy losses are conserved. Second, a multitier architecture is proposed in which CHs themselves are allowed to form upper-tier clusters. Thus, superclusters are formed. This results in a multihop path from a sensor node to the AP through different tier CHs. Hence, a separate routing table is not generated, which saves additional energy. Third, an efficient channel selection mechanism for cognitive radio functionality is incorporated. This includes channel sensing by each node and channel selection by the CH and its CM. Fourth, an adaptive energy-harvesting mechanism is deployed such that a balance is reached between network lifetime and delay. Using a regression technique, we determine the length of time required for optimal energy harvesting while ensuring an acceptable throughput and latency. The remainder of this work is organised as follows: The system model is presented in Section 2 and a brief description of key features of the R-MAC protocol incorporated in the proposed RMAC-M protocol is described in Section 3. The RMAC-M protocol is then described in Section 4 and its performance evaluation is presented in Section 5. Finally, this work is concluded in Section 6. 2 SYSTEM MODEL The proposed CWSN is a collection of a large number of low-power, battery-operated, half-duplex sensor nodes (SUs) equipped with cognitive radio technology. The primary network consists of multiple PUs with Nc nonoverlapping orthogonal frequency channels, each with a unique channel ID. Each node is aware of its location and the location of the grid boundaries. 2.1 Network architecture An overlaid CWSN is assumed. Both the PUs and SUs are uniformly distributed inside the grid. Some SUs are assumed to be mobile whereas others are assumed to be stationary. Data packets arrive at each PU according to an independent homogeneous Poisson Point Process (HPPP) with density λp and at each SU once per frame (after each data sensing slot). Figure 1 shows the network diagram. Four types of wireless links can be seen: (1) SUs harvest energy from high-power sources such as TV and radio stations. (2) The CH in theuppermost tier (after clustering) transmits all of the network data to the AP. (3) SUs detect the presence of PU signals intermittently. (4) PUs communicate with their base station. FIGURE 1Open in figure viewerPowerPoint Network architecture. Cognitive wireless sensor network secondary users (SUs) harvest energy from TV and radio towers. The highest-tier CH node transmits all data to the access point. SUs detect primary users’ (PUs’) activity regularly. PUs communicate with their base station 2.2 Sensing model The sensor nodes perform two types of sensing: data sensing and channel sensing. Data (such as temperature and humidity) are sensed periodically by each node, typically once per frame. These data are then stored in a buffer and sent out during the allotted time slot in the time division multiple access (TDMA) frame (defined in Section 3). The sensor nodes that are SUs also sense the spectrum to detect primary channel use and use an opportunistic approach to access the spectrum according to a predefined cognitive radio protocol [1]. The primary network is assumed to be a multichannel system that consists of Nc orthogonal frequency channels. Data packets arrive at each PU according to an HPPP with density λp. A channel selection algorithm that allows each sensor node to sense only a fraction of all primary network channels periodically is used [13]. A network-wide CCC using one (or more) of the vacant channels in the primary network is assumed to manage the CWSN. Transmission on CCC is broadcast to all sensor nodes in the CWSN. In CWSN, each cluster operates on a different frequency channel. All the nodes within a cluster communicate on the same channel selected for that cluster. We use TDMA to schedule transmission of nodes within a cluster that avoids the hidden node problem. Intercluster interference is mitigated by restricting the transmit power of each node in the cluster. Furthermore, to prevent interference between PUs and SUs, a simple interference avoidance model is used [23]. 2.3 Clustering and routing model To reduce network energy losses, we adopt a clustering approach [8, 9], which happens periodically before the beginning of each TDMA frame. Nodes group themselves into clusters based on R-factor criteria. We proposed a mobility-aware clustering algorithm in Section 4.1. To achieve this, all nodes calculate their R-factor values; the nodes with the highest R-factor values then elect themselves to become CHs. Cluster members (CMs) transmit data to their respective CH during their given time slot. After the CH receives all data from its CMs, it aggregates all data into a single packet for onward transmission (full data aggregation). We allow cluster heads to form multitier clusters to further reduce power losses. Instead of transmitting directly to the AP, the lower-tier CHs form higher tier clusters and transmit data to their respective (higher-tier) CHs. This process continues until only one super-CH is left at the highest tier. The super-CH then transmits data to the AP. At higher tiers, partial data aggregation is assumed such that CM packets can be aggregated into more than one packet for onward transmission. 2.4 Energy harvesting model Sensor nodes are assumed to harvest energy from RF signals from high-power sources such as TV or radio stations [4]. The received power is stored in a rechargeable battery using RF-to-DC power conversion circuitry during the energy harvesting time slots. The number of time slots for energy harvesting changes adaptively for each superframe to optimise energy harvesting time while keeping network delay and throughput within target limits. The highest-tier node computes the number of extra time slots, △, and announces it on the CCC. Target levels for throughput and packet delay are configurable parameters derived from the users’ QoS requirements. In our simulation in Section 5, target values of 1.5 ms and 100 kbps for packet delay and throughput, respectively, were used. Once these values are set, the protocol calculates the actual values at the end of each superframe and changes the number of extra harvesting slots accordingly so as not to violate these target values [7]. 2.5 Mobility model Various mobility models exist for mobile ad hoc network simulation [24-26]. We use one of the most popular yet simplest mobility models: The random waypoint model [27], in which nodes move independently to a randomly chosen destination with a randomly selected velocity. Each mobile node selects a random location in the simulation grid as its destination. It then travels to its destination with a constant velocity, chosen uniformly from the range [0, Vmax], where Vmax is the maximum allowable velocity for every mobile node. The velocity and direction of a node are chosen independently of other nodes. Upon reaching its destination, it waits for a period, Tpause and then again chooses another random destination in the simulation field and moves towards it. The whole process is repeated until the simulation ends. As an example, the movement trace of a mobile SU on the grid is shown in Figure 2. FIGURE 2Open in figure viewerPowerPoint Random waypoint mobility model. A mobile secondary user moves to a random destination with a fixed velocity. Upon reaching its destination, the process repeats itself Figure 2 shows that the mobile SU always stays within the grid. As a result, the number of SUs on the grid at any given time is constant. Although this may be too restrictive for some CWSN applications, it is a reasonable assumption for many applications with well-defined boundaries, such as field surveillance, remote data collection, and parameter monitoring. Furthermore, we assume that the mobile SUs move slowly: that is Vmax is small. This allows for clusters to be less unstable and the network to be more robust. 3 R-MAC PROTOCOL In this section, we briefly describe the key features of the R-MAC protocol that are incorporated into the proposed RMAC-M protocol. The R-MAC protocol [7] is schedule-based and uses split phase to perform various tasks. Each channel is divided into a sequence of TDMA superframes, which are themselves divided into frames and subframes. The R-MAC protocol consists of four parts: (1) R-factor based clustering (2) tiered clustering for multihop transmission (3) spectrum sensing and channel selection and (4) RF wireless energy harvesting. A flowchart of the protocol is shown in Figure 4. FIGURE 3Open in figure viewerPowerPoint TDMA Frame Structure The TDMA frame sizes are fixed. Each superframe consists of a cluster formation (CFR) phase and f frames; each frame consists of t subframes. Each subframe has one channel sense and transmit (CST) phase and one data sense and transmit (DST) phase. Figure 3 shows the TDMA frame structure of the protocol. The superframe can also be called a round. The DST frame for CHs consists of nit CM slots followed by harvesting and transmission slots; similarly, the DST frame for CMs consists of one transmission slot and the remaining harvesting slots. The CST frame structure is the same as the DST frame structure. FIGURE 4Open in figure viewerPowerPoint Flowchart The clusters may not be equal in size, but to ensure that the frame size of the MAC protocol is constant, the cluster size at each tier must not exceed a certain limit, C M m a x t . Hence, after reaching their limit, all CHs inform all additional CMs wishing to join it to choose the next best CH. The maximum cluster size, C M max t , at each tier t, is defined as C M max t = E [ N t k t ] + △ , (1)where E[.] is the expected value, Nt and kt are the number of nodes and number of CHs in tier t, respectively, and △ is the number of additional slots. △ is an important system parameter; it should be optimal to ensure enough slots are available for RFWEH without impairing delay. We discuss this issue later in the section. The superframe structure in Figure 3 also shows the structure of the frame. Each frame consists of t subframes, one for each tier, in which tier 1 is the highest tier. Each subframe consists of a DST phase and CST phase. After the formation of the multitier architecture, each CH at each tier sends a scheduling table to its CMs in the CCC assigned for its cluster, listing the slot numbers for each CM in the frame for both data and channel sensing transmission. The DST phase has SDt fixed length slots. One slot of a CH DST is allocated to each CM of its cluster. The final slots are used for CH transmission and the remaining middle slots are used for RFWEH. Immediately after transmission of a packet by the CM to its CH, the CM waits for an acknowledgement packet from the CH on the CCC. In case the packet is determined to be lost, the CM attempts retransmission in a harvesting slot using CSMA/CD. However, in case there are no slots left, the CM retransmits the packet in the next frame or it drops that packet altogether. Retransmission reduces the number of harvesting slots. Similar to the DST phase, the CST phase has SCt fixed length slots. Each CM gets a slot CST phase of the subframe. All nodes sense and transmit spectrum information to their CHs in their scheduled time slot. At the end of this phase, the CH uses data fusion to combine the information received from all nodes to update its channel weight table and then transmits it to its CMs. The CMs then use this table to select the channel for data transmission in the DST phase while dynamically updating the table. The R-MAC implements an energy-efficient channel sensing algorithm that requires every node to sense a subset of Nc frequency channels during the CST phase. First, channel weights are calculated for the Nc frequency channels by all nodes in the cluster. Then, the CMs transmit them to the CH and the CH forms a channel weight table. An updated channel weight table is transmitted by the CH to all CMs afterward. The table is used to distinguish the channels and to help the nodes choose the better channel for transmission. The channel weights change according to the states of the channel, which is described in Obaid et al. [7] In R-MAC, communication within a cluster is scheduled. Hence, there is no channel interference within a cluster. However, communication may be affected if nodes in adjacent clusters choose to transmit on the same channel, causing intercluster interference, or a PU appears to use the same channel in the primary network. The intercluster interference problem is mitigated by the following: (1) nodes in each cluster use a transmit power level such that transmission range is generally restricted to their own cluster, and (2) nodes in adjacent clusters tend to select different channels for transmission by adjusting channel weights [7]. It is possible that after clustering, the number of CHs are still large. Instead of letting all of them directly transmit to the AP (like the LEACH protocol), the R-MAC allows hierarchy of clusters, which significantly saves energy [28]. A node participates in multitier cluster formation in the CFR phase by running a separate instance of Algorithm 1 for each tier. Increasing tiers to save more energy will affect throughput and latency by adding more hops and more time slots. Therefore, the number of tiers in R-MAC is carefully chosen such that this trade-off is considered. In practice, the number of tiers will be small [7]. 3.1 Radio-frequency wireless energy harvesting Slots that not dedicated to CMs in both DST and CST phases are used for RFWEH to compensate for energy losses in the network and are called harvesting slots. Energy harvesting can be done in both DST and CST phases as well as by both CHs and CMs. Although the harvesting slots tend to increase delay and reduce throughput, which can be considered the cost of introducing them, the cost is compensated for by energy harvesting during that time. All sensor nodes in CWSN perform data sensing once per frame. At any tier, a node can be a CH, a CM, or neither (i.e., idle). It is clear that if it is idle, it incurs no energy loss. If there are a total of Nn nodes in the network, and in tier t, there are Nt CMs and kt CHs, the total energy loss per frame becomes E FrameLoss = ∑ t N t E CMLoss + k t E CHLoss . (2)where ECMLoss and ECHLoss are the energy loss by the CM and CH per subframe, respectively. At each tier, the amount of energy harvested per subframe for CM, CH and idle nodes becomes E FrameHarvest = ∑ t N t E CMharvest + k t E CHharvest + ( N n − N t − k t ) E Idleharvest . (3)where ECMharvest, ECHharvest and EIdleharvest are the energy gains due to CM, CH and idle node harvesting per subframe, respectively. The energy loss and gain model in Obaid et al. [7] provides derivations for Equations (2) and (3). Then, if there are f frames per round, the total energy loss (or gain) per round is E TotLoss = f ( E FrameLoss − E FrameHarvest ) . (4) As can be seen from Equation (4), energy consumption is compensated for by energy harvested in each round. A significant part of our design goal is to minimise total energy loss in Equation (4) by maximising the amount of harvested energy, such that throughput and delay remain close to acceptable target values. This can be achieved by increasing the number of harvesting slots in both DST and CST phases of a subframe [7]. The number of additional harvesting slots, △ in a subframe is recursively changed after each round to maximise harvested energy while maintaining system throughput Rs and packet delay Dp close to predefined acceptable target values Rtarget and Dtarget, respectively. Adding harvesting slots increases packet delay; hence, Dtarget is used to and achieve energy-delay trade-off by adjusting the number of additional harvesting slots: