We live in a world filled with sensors. The buildings that we work in have sensors monitoring temperature, occupancy, smoke and fire, and security. Our cars contain dozens if not hundreds of sensors, monitoring engine performance, braking, and passenger safety equipment, to name a few. Manufacturing environments need sensors because you cannot control what you cannot measure. Making products, while meeting safety, quality and efficiency targets, requires a lot of sensors.
Sensors have become much smaller, less expensive and lower power in the last few decades, driven in part by both Moore’s law and the MEMS revolution. Unfortunately, the cost of installing sensors has not kept pace. The cost of running wires to carry power and data typically dwarfs the cost of the sensor itself. Take for example the closest light switch: the wiring for a $1 switch can cost $50, mostly labor, even in new construction. If you want to move that switch to the adjacent wall, the price of retrofit is much higher. In industrial process automation, the accepted rule of thumb is $10,000 to install a sensor – even a simple switch. In this cost environment, many sensors only report data to a local controller – there can be little or no “big picture” when hundreds or thousands of sensors are installed. What is needed is an inexpensive, reliable way to network sensors.
Almost since the time of Marconi, people have used wireless methods to communicate data from sensors, with mixed results. Traditionally these links have been line-powered and point-to-point, often with time-varying reliability due to environmental conditions. This is fine for some applications, but too restrictive for most.
Markets for wireless sensor networks (WSN) include building automation, industrial control, home automation, smart grid and automated metering infrastructure (AMI), industrial process automation, environmental monitoring, parking and transit infrastructure, energy monitoring, and inventory control. In most cases, these are bi-directional asymmetric data collection applications – large numbers of sense points forward data to a central host that may respond with a process set point or other configuration changes.
Customers ideally want a technology that is low cost, allows unrestricted sensor placement, receives periodic data reliably with low latency and runs for the device lifetime with no battery changes. Recent technological advances have enabled delivery of those features in many markets.
There are several technologies competing to fill this role, including satellite, cellular, WiFi, and a host of solutions based on IEEE 802.15.4 radios. These technologies allow users to form WSNs for collecting sensor data.
Satellite and cellular work well for many applications, but have the highest energy cost per packet. Data plan charges can also be prohibitive, although this is likely to change as carriers develop billing models appropriate for relatively sparse data flows. Coverage can also be an issue. Clearly it can be difficult for a satellite or cell phone signal to make its way out of a heavily obstructed structure, and the sensors generally do not have the capability of moving from side to side and asking “can you hear me now?” For an application sending at a very low data rate (e.g., one data packet per day) with good connectivity, however, satellite or cellular can make a lot of sense.
WiFi (IEEE 802.11b, g) sensors are now widely available. The energy cost for a WiFi packet is much lower than cellular, and there are no recurring fees for data. Connectivity and coverage remain important concerns, as the density of access points necessary for reliable communication with a fixed sensor is typically higher than that necessary for mobile humans with gadgets.
With reference to the OSI layer model, the 802.15.4 standard defines a physical layer (PHY) and medium access control (MAC) layer for short-range, low-power operation that is well suited for wireless sensor networks. The radio is relatively low data rate (up to 250 kbps); the packets are short (< 128 bytes) and low energy. For example, sending a few bytes of sensor data, with routing, cryptography, and other headers, takes under 1 ms. This process uses under 30 µC of charge, including receiving a secure link-layer acknowledgment (see Figure 1). Sensors can forward radio packets from peers, extending the range of the network far beyond the range of a single radio, and providing the network with immunity to any single radio link failure.
Evaluation of various WSN solutions is based on two questions, “Can I get all the data fast enough?” and “How much is it going to cost?” WSNs must be designed to work in environments with link-layer packet delivery ratios (PDR) down to about 50 percent.
When developing a wireless data-collection system, there are a few performance targets that must be met. First, the system must meet a minimum reliability goal. For industrial applications, the target is typically to receive at least 99.9 percent of the generated data, as missing data can trigger expensive alarm conditions. Second, the system must support a certain throughput, a number of sensor data packets per second. Third, these data packets are only useful if received within a maximum latency period. Many processes rely on fresh data updates — for control, stale data may have no utility. Fourth, many systems must operate in challenging environments that include wide temperature ranges and intrinsic safety restrictions. Only solutions that meet all four of these requirements are considered suitable for further evaluation.
When considering various solutions that meet the requirements, the key selection criteria become cost of ownership and flexibility. The cost of ownership encompasses several areas: product development, installation, hardware and providing power over the lifetime of the installation. Wireless technologies have reduced installation costs dramatically compared to wired solutions, but battery-powered wireless devices may require battery changes over the lifetime of the network. There is also a trade-off between building a network with a small number of high-power devices to reduce the hardware cost versus using a larger number of low-power devices. For devices powered by energy harvesting cells (e.g., solar, thermoelectric), capacitor size may determine a significant portion of the cost. Solutions with deterministic scheduling, such as Time-Division Multiple Access (TDMA), can help separate high-current events as much as possible to reduce the capacitor size requirement.
Because final deployment conditions are unpredictable, networks must be designed for flexibility. Networks must scale from small to large numbers of sensors and from low to high density. To be robust across diverse wireless environments, resource provisioning should ensure that devices reliably communicate with moderate interference and that the networks survive the loss of individual devices. Additional resources, including more wireless links, more neighbors for each device, or more signal amplification, improve reliability and latency. All these additions come at increased power costs that can be minimized with dynamic allocation. Solutions based on standards provide immunity to the supply chain vagaries of a single vendor component and the assurance that the community has agreed on the governing principles of operation, e.g., security architecture.
The wireless channel is unreliable in nature, and a number of phenomena can prevent a transmitted packet from reaching a receiver. One such phenomenon is interference. If two independent transmitters transmit on the same channel such that their signals overlap, they may corrupt each other’s signal at a receiver’s radio. This requires the transmitter to re-transmit, at the cost of additional time and energy.
Interference can come from the same network if the underlying medium access technology does not schedule contention-free communications. This is particularly problematic if the two transmitters can hear the receiver, but not hear each other – this is known as the “hidden terminal problem,” and it requires backoff and acknowledgment mechanisms to resolve collisions.
Interference can also come from another network operating in the same radio space, or from a different radio technology using the same frequency band. The latter, known as “external” interference, is especially present in unlicensed bands such as the 2.400 to 2.485 GHz Instrumentation, Scientific and Medical (ISM) band, crowded with WiFi, Bluetooth and 802.15.4.
Figure 2 was obtained by deploying forty-five 802.15.4 nodes in an office environment, and having them exchange 12 million packets, equally distributed over sixteen 802.15.4 channels. It plots the average packet delivery ratio of those packets as a function of the channel they are transmitted on; on channels overlapping WiFi channels, this delivery ratio is lower.
A second phenomenon, multipath fading, shown in Figure 3, can prevent a transmitted packet from reaching a receiver and is both more destructive and harder to quantify. Often described as “self-interference,” this occurs when the recipient receives both the signal traveling over the line-of-sight path from the transmitter as well as “echoes” of the same signal that have bounced off objects in the environment (floors, ceilings, doors, people, etc.). Since those copies travel different distances, they reach the receiver at different times, potentially interfering destructively. Fades of 20 to 30 dB are not uncommon.
Figure 3 was obtained by having a transmitter transmit 1000 packets to a receiver 5 m away, and repeating this with the receiver positioned at each point in a 35 by 20 cm grid. The z-axis represents the packet delivery ratio over that link. While the link is good at most positions, at some positions no packets are received successfully because of multipath fading.
Multipath fading depends on the position and nature of every object in the environment and is unpredictable in any practical setup. One good property is that the “topography” depicted in Figure 3 changes with the frequency. That is, if a packet is not received because of the multipath fading, re-transmitting on a different frequency has a high probability of succeeding.
Because objects in the environment are not static, e.g., cars drive by and doors are opened and closed, the effect of multipath changes over time. Figure 4 shows the packet delivery ratio on a single wireless path between two industrial sensors over the course of 26 days, and for each of the sixteen channels used by the system. There are weekly cycles where workdays and weekends are clearly visible. At any given time, some channels are good (high delivery), others bad, and still others vary highly. Channel 17, while generally good, has at least one period of zero delivery. Each path in the network shows qualitatively similar behavior, but with different channel performance and there is never any one channel that is good everywhere in the network.1 Because of interference and multipath fading, the key to building a reliable wireless system is to exploit channel and path diversity.
As stated previously, one technology well suited for solving the WSN problem is IEEE 802.15.4. Such 802.15.4 radios offer low-power, low-data rate PHYs in several unlicensed frequency bands, including the 915 MHz band, available in North America, and the 2.4 GHz ISM band, available worldwide. The 2.4 GHz band spread spectrum PHYs provide immunity to noise – a particularly important feature for a low-energy device designed to operate in a potentially crowded, unlicensed band. The standard also defines a reliable, acknowledged, packet (or frame) based MAC layer with optional encryption and authentication. This flexible solution forms the basis of several proprietary and standards-based protocols including the ZigBee protocol, which uses it to form unsynchronized single-channel networks, and the WirelessHART protocol,2 which uses it to form time-synchronized multichannel networks.
The WirelessHART protocol has an 802.15.4 2.4 GHz PHY and an 802.15.4-based link layer, which adds synchronization, channel hopping, priority and time-based authentication to the standard 802.15.4 MAC. It has a network layer that provides routing and end-to-end security, and a thin unreliable/reliable mesh transport layer. The WirelessHART standard specifies timeslot timing, how devices maintain synchronization, and how devices schedule time/channel communications opportunities by dividing time into slotted communications opportunities (timeslots) on repeating superframes. The protocol was designed to allow seamless integration of wireless devices to existing wired HART installations, widely used in industrial process monitoring and control applications. WirelessHART extends the HART application layer command set, adding commands for managing wireless resources and monitoring network health. WirelessHART networks are highly reliable meshes – even with devices that do not have line-of-sight and at tens to hundreds of meters spacing, each device has multiple neighbors to which it can send data – providing the path diversity needed for reliability. WirelessHART networks are centrally managed, with most network “intelligence” residing in a Manager. Field devices (wireless sensors) report status information that the Manager uses to groom and optimize the network, and sensor data is reported to an application proxy called a Gateway.
Earlier this year, a new 802.15.4e amendment was released, which, among other things, formalized time-slotted channel-hopping features like those found in WirelessHART at the 802.15.4 MAC layer. The standard defines the mechanisms for advertising synchronization information to allow devices to synchronize to a network, provides for time-based security, and defines slotted communications and hop sequences. It makes extensive use of data encapsulation in “information elements” – this allows for custom extensions of the MAC without having to wait for the standard to be updated. It is intended to ease development of a multilayer protocol, and was specifically designed to couple to a 6LoWPAN-compressed IPv6 network layer as defined in IETF RFCs 4944 and 6282.3
Linear’s Dust Networks SmartMesh™ product line is an example solution that contains both WirelessHART and 6LoWPAN-compliant IPv6 product offerings that leverage 802.15.4 to provide reliable, low power WSN solutions. Dust Eterna™ motes are single chip devices that couple a Cortex-M3™ microprocessor, memory and peripherals to the lowest power 802.15.4 radio currently available (see Figure 5). Designers embed a mote in their sensor package and can rely on the network to form, optimize, and carry their sensor data to their application. Managers allow for graceful scaling from tens to thousands of devices, providing data and configuration interfaces for the network. They are suitable for solving a wide array of WSN problems. Some examples of applications using motes and managers include:
Parking: Streetline4 is a smart parking provider that monitors the real-time availability of urban parking spaces. Vehicle detectors are installed underneath parking spaces, inside the pavement and flush with the roadway. This brings challenges, as the antenna for the sensor device is located underground, and then covered by a metal vehicle when the space is occupied. Wireless path diversity is essential as different vehicle positions change the path quality between device pairs. Streetline installs elevated repeater devices on nearby street lamps to obtain line-of-sight to the stall sensors. These repeaters form a multi-hop mesh to collect all the occupancy data to the local network manager, where it gets aggregated into a city-wide database available to customers and enforcement agencies. Wireless technology is critical for this application because it is intractable to wire sensors to each space, and low-power wireless decreases the frequency of battery changes.
Refinery Process Control: Chevron uses wireless networks to monitor oil extraction and refining facilities. These networks are often deployed in harsh environments (due to hazardous temperatures, chemicals or risk of explosion) where it is impossible to run conduit for wired sensors. Additionally, wireless enables monitoring of rotating structures and mobile operators. For one deployment (see Figure 6), wireless networks were installed in various locations around a large refining facility. To gather the data to a centralized control center, a Cisco IEEE 802.11a wireless mesh was used as the backhaul connection for each IEEE 802.15.4 network manager. This allowed the low-power sensor devices to report to their local manager where data was aggregated and reliably shuttled along. This deployment represents a powerful fusion between the two standards.
Energy Monitoring: Vigilent5 provides intelligent energy management systems for indoor environments such as data centers where environmental control is critical. As increased temperature at any location in the data center can cause equipment failures, air conditioning is often run continuously at full power, wasting energy. Facilities managers are reluctant to jeopardize their internal networks, so Vigilent deploys wireless devices that do not interfere with regular operation. The facilities are also sensitive to security, so the wireless protocol is required to have end-to-end encryption of all packets and additional security at the network manager. Data center sense points are typically dense, and Vigilent has had success in deploying multiple overlapping networks to achieve the required number of sensors.
Multichannel time-synchronized mesh networks based on 802.15.4 radios address many of the challenges involved in building flexible, reliable, low-power wireless sensor networks. There are a growing number of new applications that are utilizing these types of sensor networks to reduce costs and provide enhanced services around the world.
- L. Doherty, W. Lindsay, J. Simon and K. Pister, “Channel-Specific Wireless Sensor Network Path Analysis,” Proceedings ICCCN ’07, Honolulu, HI, 2007.