Based on the criticality and importance of the applications, the International Society of Automation (ISA) considers six classes of wireless communication, from critical control to monitoring applications, in which the importance of the message response time and Quality of Service (QoS) requirements varies [12]. In the more critical applications, process values need to be transmitted to the destination in a reliable, timely and accurate manner. The details of the classes are shown in Table 2.

While ISA100.11a supports industrial applications from class 1 to 5, WirelessHART supports industrial applications ranging from class 2 to 5 [12]. ZigBee Pro is designed for applications which have softer real-time requirements [13]. Traditional wireless sensor networks (WSNs) are deployed in class 4–5 applications [12], where low-power consumption is given priority over providing a bounded response time delay. Such WSNs are not suitable for controlling tight control loops as nodes usually spend a large proportion of the time in a low-power sleep state.

WISA is the only wireless protocol that is suitable for factory automation applications as it can provide some strict real-time guarantees. There are related basic wireless requirements in such applications, for example, low additional latency due to wireless link (e.g., <10 ms).

We carry out a more detailed analysis of the real-time capabilities of ZigBee Pro, WirelessHART, WIA-PA, WISA and ISA100.11a later in the paper by discussing specific details relating to the MAC layer contention mechanism and priority management schemes.

As industrial processes increase in complexity, the number of points that need to be monitored and controlled increases rapidly. This makes it essential to design network architectures which are capable of scaling up. In other words, the objective is to ensure optimal network performance even when the network size or rate of data generation increases.

Current wireless technologies designed specifically for industrial applications such as WirelessHART, WIA-PA (in centralized management scheme) and ISA100.11a mostly use a centralized approach for managing resources. While centralized approaches are technically easier to develop and manage, they are unable to cope with sudden changes that might occur frequently in a harsh industrial environment. This problem is further exacerbated as the network is scaled up. For example, a motor capable of running at different speeds may cause radio interference at different frequencies as it changes its operational speed. Wireless nodes operating in the vicinity of the motor should ideally reorganize their communication protocols using distributed techniques as and when interference is detected to quickly adapt to the changing environment. Traditional centralized approaches are unable to cope with such sudden unexpected changes, as they would then require detailed network statistics to be sent back to the central system manager which would then clog up the limited network resources. Thus, the larger the scale of the deployment, the more important it is to utilize distributed approaches to ensure that the system continues to perform optimally.

Unlike traditional wireless sensor networks, power consumption has a lower priority than other performance metrics, such as reliability and real-time capability in industrial sensor networks. However, the degree of importance of power consumption varies greatly depending on the class of application. Industrial control applications can be categorized into two main classes: (i) process control, and (ii) factory automation.

Process control is typically used for monitoring fluids (e.g., oil level in a tank, pressure of a gas, etc.). Such applications which typically involve non-critical applications requiring closed-loop control usually transmit process values at regular intervals. Furthermore, due to the non-critical nature of the process control applications, latency requirements are not usually stringent (>100 ms). This allows nodes to reduce power consumption by carrying out aggressive duty cycling of their radios and sensor sampling operations. Factory automation applications, however, involve machines (e.g., robots) that perform discrete actions and are highly sensitive to message delays. Thus, such applications generate ‘bursty’ data and may require latency in the region of 2–50 ms. In such instances, reducing power consumption has a lower priority than other performance metrics such as real-time capability and reliability.

In terms of energy consumption, ZigBee Pro and WIA-PA (in the cluster/star level) do not perform as well as the other competing technologies as it carries out time synchronization using the beacon-enabled mode of IEEE 802.15.4. Using beacons introduces a large overhead in terms of higher energy consumption, as the radio needs to remain in listen mode for long periods. Conversely, TSMP solves the time synchronization problem in a more energy-efficient manner by only relying on ACKs to exchange timing offset information. WirelessHART, WIA-PA (in the mesh level) and ISA100.11a also benefit from this approach as they both utilize TSMP. Furthermore, the ISA100.11a specification allows the transmission power of individual nodes to be controlled. This can result in additional energy savings. However, the specifications do not describe any algorithms indicating the strategies to be followed to carry out adaptive power control.

Reliability is an integral part of any industrial monitoring and control system as any slight degradation in communication can potentially result in complete system malfunction. In order to ensure reliable wireless communication, various techniques can be used to mitigate communication problems such as interference and weak signals. Figure 1 gives an overview of the different classes of problems in wireless communication commonly present in industrial environments and their relevant solutions. We present some of the more important solutions developed both in academia and industry in greater detail in the following sections. The PHY aspects of reliability are addressed in Section 6.

Reliability is an integral part of any industrial monitoring and control system as any slight degradation in communication can potentially result in complete system malfunction. The industrial environment may suffer from noise, internal interferences, multipath propagation, external interferences, collision, and physical obstruction. In order to ensure reliable wireless communication, various techniques are introduced to mitigate communication problems such as interference and weak signals. We discuss the mechanisms presented in academia for addressing the reliability requirement in an industrial automation environment, such as diversity, error control schemes, multipath routing, and network coding.

In this section, we discuss MAC and routing protocols introduced in academia for addressing the real-time and latency requirements existing in WSNs applications, such as industrial process automation.

A MAC protocol can generally be designed to operate using two mechanisms: (i) contention-free (scheduled communication) and (ii) contention-based. Contention‐free approaches, e.g., dedicated timeslot-based, are more suitable for supporting real-time communication while shared timeslots (i.e., contention-based mechanisms) favor soft real-time applications.

Contention-based communication protocols, such as CSMA, are unable to provide timing guarantees when delivering messages. They are prone to packet loss by the hidden terminal problem (internal interference). Since ZigBee Pro runs on a CSMA-based MAC protocol, it is unsuitable for applications that require reliable and timely packet delivery, although WIA-PA and ISA100.11a use a CSMA-based MAC (slow hopping) for subnet discovery and retries. These latter are capable of switching to a slotted scheme where every link is scheduled to transmit at a predefined slot (TDMA) and channel offset, thereby avoiding the issue of internal interference. This mechanism is shown in Figure 2, in which the combination of slow hopping and slotted hopping is displayed. A similar form of communication scheduling is also used in TSMP, WirelessHART, and IEEE 802.15.4e.

Both WISA, WirelessHART, and WIA-PA use TDMA-based mechanisms with exclusively dedicated timeslots which do not support any variation in traffic [13]. However, ZigBee Pro, ISA100.11a, and the 802.15.4e MAC standard allow the user or centralized system manager to configure the timeslot length. This could be advantageous for coping with variable data traffic rates on the network which could be a characteristic of factory automation applications requiring real-time operations. However, as individual nodes are unable to make autonomous decisions, existing technologies are unable to provide hard real-time guarantees, especially in the presence of variable data traffic.

ISA100.11a and WirelessHART use a superframe management technique to maintain real-time communication for high traffic loads. The superframe period can determine several network performance parameters, such as packet delivery latency, energy consumption, and bandwidth utilization. ISA100.11a allows tradeoffs by enabling the system manager to determine the superframe period. A shorter-period superframe results in lower packet delivery latency and higher bandwidth utilization, but results in greater energy consumption, while a longer-period superframe has the opposite effect. ZigBee Pro and WIA-PA (in the cluster/star level) also allow the transmission of superframes with different lengths in the beacon mode.

Timely and reliable data transport is crucial for industrial automation applications. Communication networks are usually designed to meet such criteria by using certain Quality of Service (QoS) mechanisms. QoS mechanisms generally use two techniques to achieve their goals: (i) traffic classification and (ii) resource reservation.

The traffic classification mechanism can be used for channel access and packet delivery along the path between the endpoints, by labeling the packets with a priority value and placing them on the corresponding queue in the path. The resource reservation mechanism is used for allocating and reserving the resources along the path between two end-points for the specific traffic or class of traffic to achieve the desired QoS requirement.

Fox example, in the wired CAN protocol (a communication system for industrial and automotive applications), a MAC layer technique is used to resolve the contention between several nodes to access the channel. It involves bit-wise priority arbitration for collision resolution that relies on a node’s ability to transmit and receive simultaneously. Each packet has a priority value that is used to resolve the contention among different nodes trying to access the channel. The node with a higher priority label in its data packet has a higher chance of accessing the channel. Each contender node transmits its priority value and receives feedback from the channel simultaneously. A node realizes that it has lost the contention when it detects a higher priority bit on the channel compared to the bit transmitted by the node itself [59,60].

This technique cannot be used in wireless sensor networks as they typically have half-duplex transceivers. WIA-PA uses the traffic classification method for addressing different QoS requests. They define four priority levels, based on different classes of data: command packets, process data, normal packets and alarm packets. Low priority packets are declined when the device buffers become full. WirelessHART and ISA100.11a use a combination of traffic classification and resource reservation techniques for providing different QoS requests. When a device wants to establish communication with the central system manager or another device, it sends the contract request (Service request in WirelessHART), including input parameters, such as communication service type (scheduled or unscheduled communication), destination address, traffic classification (best effort queued, real time sequential, real time buffer and network control), requested period, and committed burst for non-periodic communication, to the system manager. The system manager uses its centralized optimization algorithm to determine the required allocation of the network resources (such as graphs and links) and sends a contract response to the source after all necessary network resources have been configured and reserved along the path. However, WirelessHART and ISA100.11a do not specify the specific optimization algorithms that can be used by the system manager to allocate resources.

Channel hopping is often used to mitigate external interference and multipath fading. The proper reception of wireless signals may be prevented by other radio signals generated by the devices outside the network. This kind of interference is known as external interference. Signals in the same frequency range can be generated by Bluetooth devices, microwave ovens, other external networks (such as the IEEE 802.11 network) or many unintended sources of radio interference, such as other high-power interference sources. Channel hopping techniques are a way to mitigate external interference and multipath fading.

Figure 3 provides a classification of the different channel hopping techniques as well as the standards which use each of these techniques.

There is a tradeoff between using blind channel hopping and adaptive channel hopping (ACH). In the former, if the node switches to another congested channel or switches from a good channel to a congested one, this hopping does not help to mitigate the interference and just wastes energy [30]. However, in spite of this disadvantage, blind channel hopping has less overhead as the hopping pattern is already known by the network devices. In addition, if the system manager decides to blacklist a particular channel, nodes in the network still hop to the channel, but simply remain idle in that time period. Thus the larger the number of blacklisted channels, the more time is lost by nodes idling on blacklisted channels. Blind channel hopping techniques ensure that while two communicating nodes hop in unison, neighboring node pairs never use the same frequency at the same time in order to prevent hidden terminal problems. This is shown in Figure 4.

ACH differs from blind hopping in the sense that unlike in blind hopping, nodes do not keep on changing from one operating frequency to another at regular time intervals. In other words, nodes only change their frequencies when interference is detected on the current operating channel. However, nodes need to collaborate to decide which channel to switch to and this can introduce a significant overhead since nodes need to continuously scan all channels for interference levels and also because nodes need to ensure that while communicating nodes choose the same frequency, neighboring node pairs use different channels [61]. In WIA-PA in each cluster/star network, the cluster head and each node irregularly change their channel on a link-by-link basis only when channel conditions require it to. However to the best of our knowledge, there are currently no algorithms indicating strategies to be followed for ACH approaches in the multi-hop network.

Both WirelessHART and ISA100.11a networks use blacklisting techniques to mitigate external interference or multipath fading. WISA and ZigBee Pro do not have this capability.

ZigBee Pro, however, uses “frequency agility”. This mechanism is not as tolerant to fluctuating wireless conditions as WirelessHART, WIA-PA, and ISA100.11a. In this technique, the network channel manager collects interference reports from all the nodes. If external interference is detected, the network channel manager scans for a better channel and moves the entire network to a new channel. This technique requires network formation to be carried out again and thus introduces inconvenient delays. Note that ACH is clearly a better technique as it only requires nodes facing interference to make changes to their operating frequencies—it does not affect the entire network.

ISA100.11a tries to separate the successive channels in the hopping pattern by at least 20 MHz. That means that at least a three-channel separation exists in each consecutive hop and, in the case of retries in the next hop, that they will not encounter the same IEEE 802.11 channel. The way by which this standard can coexist with the IEEE 802.11 standard has been predefined. This hopping separation is more than the coherence bandwidth value—the frequency-shift by which a link has to undergo transition from deep fade-in the case of indoor multi-path interference [26].

WISA employs frequency-hopping sequences in which the consecutive hops are widely separated in frequency and the sub-band bandwidth is more than the typical bandwidth of the coherence bandwidth. WISA uses frame-by-frame frequency hopping in which the used radio channel for retransmission is independent of the previous one, so likelihood of a successful transmission increases.

To ensure real-time capability, WISA protocols require the network to be deployed in a star topology rather than a multi-hop meshed network that is used in ZigBee Pro, WirelessHART, WIA-PA and ISA100.11a. A disadvantage of this approach is that if communication between a node and its cluster head is disrupted due to interference, alternative routes cannot be established to transport the data. The multi-hop approach of ZigBee Pro, WirelessHART and ISA100.11a prevent such problems from occurring. Additionally, to ensure robust communication in WirelessHART and ISA100.11a, the system manager defines multiple paths for each node to reach a particular destination in the network.

Large-scale, distributed, real-time control applications require data to be transmitted over long distances through a multi-hop network in a timely manner. A distributed resource reservation algorithm is needed which would allow source nodes, based on the requirements of the application and traffic characteristic, to reserve network resources for its peer communications along their paths for addressing different QoS needs. The distributed nature allows the system to adapt quickly to disturbances or changes within the network to meet timing guarantees to support real-time control operation. While such mechanisms do not exist for present day sensor nodes in a distributed manner, relevant techniques from other networking-related domains could potentially be adapted to develop solutions that are suitable for wireless sensor and actuator networks. We briefly describe some of these relevant techniques.

QoS in multi-hop networks can be supported by different mechanisms, such as circuit switching, Asynchronous Transfer Mode (ATM) networks, and internet protocols (such as Integrated Services/RSVP, Differentiated Services, MPLS and constraint-based routing). It is also supported by IEEE 802.11e [65], ISA100.11a and WirelessHART in a centralized manner.

ATM signaling protocols address certain performance issues in terms of reliability and timeliness of packet delivery that are of importance in industrial applications that require closed-loop, real-time control. The ATM protocol uses a switching technique that combines the concepts of circuit switching and packet switching. For example, similar to circuit switching, before initiating data transfer, a virtual circuit is first established between the source and destination. The protocol also includes admission control mechanisms that help determine whether the required QoS guarantees can be provided. ATM uses statistical multiplexing techniques, similar to those used in packet switching, in order to cope with variable bit rates (i.e., ‘bursty’ traffic).

Internet protocols are mainly designed for multimedia applications. In those protocols, some mechanisms exist that allow a data receiver to request a special end-to-end quality of service for its data flows or classes of data. RSVP signaling is used by several internet protocols, such as Integrated Services Architecture, differentiated service, and MPLS, through which the application can reserve the resource and set up the path between the source and destinations.

The IEEE 802.11e, based on traffic classification mechanisms, provides different degrees of satisfaction for the users of the service. They define different priorities through which traffic can be delivered in several access categories. This differentiation is achieved by considering different amounts of time for sensing the channel to be idle and by considering different lengths of the contention window during backoff. This implies that high-priority traffic can access the channel by shorter back-offs than low-priority traffic. In addition, the packets are labeled with priority value and introduced into the corresponding queue in the path. Admission control in this standard as an important component limits the amount of traffic admitted into a particular service class so that the network resources can be efficiently utilized.

WirelessHART, WIA-PA (in the mesh level) and ISA100.11a use centralized network management techniques for communication scheduling and managing routes. While such an approach may be easier in terms of implementation, they have numerous disadvantages. Centralized systems often perform poorly in terms of reaction time, as all updates need to be sent first to the centralized system manager (i.e., gateway) for further processing. The gateway node then performs recalculations and disseminates updated instructions to the relevant nodes in the network. As the round-trip time for such decision-making actions can be very high (especially when network contention is high), centralized approaches are unable to cope with highly dynamic situations (e.g., ‘bursty’ data traffic/varying link quality, and node mobility). This problem is further exacerbated as the network is scaled up. This in turn may result in problems, including increased packet loss and delayed data delivery, which increase energy consumption. The distributed nature of a distributed approach allows the system to adapt quickly to disturbances or changes within the network in real-time. However, current wireless control technologies that use distributed approaches also perform poorly in terms of reliability, efficiency and robustness.

In the previous section, it was discussed that one of the solutions to mitigate interference and multipath fading is channel hopping. In adaptive channel hopping, the channel on a link-by-link basis will be changed when interference is detected on the current operating channel. However, nodes need to collaborate to decide which channel to switch to and this can introduce a significant overhead since nodes need to continuously scan all channels for interference levels and also because nodes need to ensure that while communicating nodes choose the same frequency, neighboring node pairs use different channels. To the best of our knowledge, there are currently no algorithms indicating strategies to be followed for ACH approaches in multi-hop networks. The challenges in defining the appropriate solutions are overhead and simplicity of the algorithm.

The transmission power used by a node can have a direct impact on the radio link quality, the level of interference and energy consumption. Ideally, all nodes should always use the least transmission power that will allow them to carry out their assigned tasks effectively. While the ISA100.11a specification allows the transmission power of individual nodes to be controlled, it does not describe any specific algorithms to perform adaptive power control. Several autonomous power control strategies, developed specifically for WSNs, can be found in the literature [66,67]. However, they all have certain drawbacks which would prevent them from being used in a harsh industrial environment. For example, while the technique presented in [66] can adapt, it is not designed to handle rapid link quality fluctuations that could be caused by moving metal objects or electromagnetic interference from motors or pumps that may be common in an industrial environment. While the authors in [67] rightly point out that interference is an issue that needs to be addressed when developing adaptive power control algorithms, the presented solution does not perform optimally as it is unable to correctly distinguish between weak signals and interference. This is an area that still requires further investigation.

WirelessHART, WIA-PA (in the mesh level) and ISA100.11a use centralized network management techniques for communication scheduling and managing routes. However, those standards do not specify the specific optimization algorithms that can be used by the system manager to allocate resources. In [68,69,70] the authors have proposed the centralized scheduling algorithm in WirelessHART for convergecast by considering linear and tree networks models. ISA100.11a standard supports peer-to-peer communication, in addition to uplink and downlink traffics. This feature makes the communication scheduling and route managing algorithm more complicated than WirelessHART, in which the main concern is forwarding the traffic toward the gateway and vice versa. To the best of our knowledge, there are currently no algorithms indicating strategies to be followed for communication scheduling and route formation in ISA100.11a.

Two types of aggregation, data aggregation and packet aggregation, are supported by WIA-PA in order to reduce the number of packet transmissions. WirelessHART, WISA, and ZigBee Pro do not support this function. In certain industrial closed-loop control applications involving multiple sensors and an actuator, raw sensor readings are streamed from the sensors to the actuator. The actuator subsequently performs computations using the readings to carry out the relevant control operations. This traditional approach, however, is not suitable for multihop wireless sensor networks, since they have highly limited bandwidths. The idea would then be to allow an intermediate node to carry out the computations and only send the final control output to the actuator, thus saving network bandwidth. However, as every actuation operation may be dependent on a different set of sensors, the nodes need to autonomously decide which node should act as the intermediate aggregation node that will be responsible for computing the control output. This technique will also contribute towards improving real-time operation.

Over the last decade, multiple antenna and Multiple Input Multiple Output (MIMO) techniques have been widely discussed to increase the reliability and throughput of various wireless systems. As discussed in Section 4.1.1, the antenna diversity can also improve reliability by achieving multiple different realizations of channel. So, if one antenna gets an interfered/distorted signal, which is non-recoverable due to signal propagation through a deep faded channel, another antenna may receive a copy of the signal which is suitable for decoding. This can significantly improve link reliability. MIMO can increase the system throughput without increasing the bandwidth. Several antennas to transmit/receive a portion of a signal with spatial diversity make this possible. Orthogonal Frequency Division Multiplexing (OFDM) takes this to the next step by using orthogonal sub-carriers in a frequency band. OFDM not only increases the throughput of a system, but also enables the facility to use different levels of modulation for different sub-carriers, based on the channel state information (CSI). Such techniques have been applied in the recently developed IEEE 802.11n standard to increase WiFi throughput to a next level [64]. However, the complexity of the receivers also increases to facilitate this technology, which makes it challenging to implement in WSN applications. Building simple MIMO transceivers with a low power consumption is still in a research phase [71].