By: by Daniel M. Dobkin and Lew Adams, GainSpan
With the advent of low-power, low-cost CMOS-based radios and efficient protocols, data acquisition tasks that have traditionally employed analog or serial wired connections can now be implemented wirelessly.
Installing sensors for data acquisition has traditionally been a labor-intensive process, involving analog or network wiring from sensor locations to aggregation points. The high cost of installation, maintenance and configuration changes limits the use of such sensors to high-value applications.
Recent progress in wireless technology has made it possible to replace these older wired networks with wireless sensor nodes. In wireless sensor networks, installation labor is limited to the placement of the sensors themselves; node configuration is largely automated and can be managed from a remote network server. Maintenance may consist only of battery replacement once every few years, in the case of battery-powered sensors. Sensors using energy harvesting techniques may require no scheduled maintenance at all.
To implement such networks requires physically small, energy-efficient, computationally capable wireless devices. Today, such devices can be implemented at low cost using advanced System-on-Chip (SoC) designs, in which advanced microprocessors, SRAM and non-volatile memory, local data interfaces and fully capable wireless modems are combined on a CMOS chip. A complete sensor node can be constructed from the SoC, with the addition of the sensor itself, a battery or other power source, an antenna, and in some cases an optional RF power amplifier.
As with any system based on digital communications, a protocol is required for such wireless nodes to communicate with back-end devices so that the data can be collected and accessed. Today’s wireless sensor designs use both proprietary and standardized communications protocols. Popular choices for this purpose include IEEE 802.11, often referred to as Wi-Fi, and IEEE 802.15.4, often referred to as Zigbee, though that term should properly be reserved for the Zigbee ad-hoc networking layer that operates over the 802.15.4 physical and link layers.
Sensor Network Design and Architecture
Wireless links are more flexible but more complicated than wires. The ability of two devices to communicate depends on a number of factors:
For example, the GS1011 SoC CMOS transmitter (Figure 1) can deliver about 9 mW (9 dBm) of output power in the 2.4 GHz ISM band. In an outdoor line-of-sight environment with a moderate-gain receive antenna, this signal power is sufficient to maintain reliable links at distances of hundreds of meters. In a typical indoor environment, range is extremely dependent on the placement of the node and receiving antenna, and the exact layout of the building walls and interior objects: the same setup that yields a range of 300 meters outdoors may be limited to 30 meters indoors. The expected range determines how far sensor links can be from an aggregation device, such as an access point (Wi-Fi) or full-function device (802.15.4). Wi-Fi devices operate at 2.4 GHz. The 802.15.4 protocol allows for operation in the 868 and 915 MHz bands. Lower fre-quencies suffer less from absorption and diffraction by obstacles, but have correspondingly less bandwidth available for use. It is also more difficult to fabricate efficient small antennas when the wavelength of transmission increases.
Interfering transmissions on the same or nearby bands may also limit the ability of a sensor node to communicate. Most interferers are time-dependent, so protocol adaptation may be used to communicate successfully in spite of interference. For example, the 802.11 protocol uses a carrier-sense multiple access with collision avoidance (CSMA-CA) approach. The modem tracks other received Wi-Fi transmissions and postpones initiating a transmission until it is expected that other stations have completed their messages, using the network allocation vector (NAV) contained in each packet. When the medium is believed to be clear, the node attempts to send a packet; if it does not receive an acknowledgement (ACK) from the destination, typically an ac-cess point, it waits a random backoff time and tries again. Wi-Fi can provide robust communications, except at very high traffic levels, and requires no central coordination or reservation system.
The required average data rate determines how many sensor devices can be assigned to a single aggregation point. For many sensor network applications, average data rate is very low: for example, in many temperature-sensing applications a packet might only be sent every few minutes. This permits a large number of nodes to share a single aggregation device.
On the other hand, the instantaneous data rate determines the actual transmit time needed to send a given amount of data. Typical sensor node packets are short: tens to a few hundred bytes. At data rates of hundreds to a few million bytes/second, the actual transmit time for a packet is a few tens of milliseconds. The duty cycle of the node is thus often much less than 1%. In order to operate efficiently, the node must be capable of using very little power when it is not active, and rapidly enter an active state where it can send and receive messages. For example, in low-power applications, the GS1011 spends most of its time in an ultra-low-power standby state, with total power dissipation of around 10-15 microwatts. When a timer expiration or alarm input signals the need for an action, the processors are powered up and memory state restored in about 10 milliseconds, after which a measurement can be made or a transmission initiated. The SoC also supports a variety of clock-gate and sleep states, which allow power consumption to be minimized while still providing sub-microsecond latency.
Network and Node Configuration
In order to actually send data to a remote server, network configuration is required. In traditional analog sensor applications, configuration was performed manually, by wiring a given sensor to a known measurement port and manually recording the association. Modern wireless networks support self-registration and automated configuration.
The process of joining a Wi-Fi-based network is summarized in Figure 2. The node must first discover what suitable networks can be heard. In 802.11, access points send special beacon packets at regular intervals, containing the information required by a Station that wishes to associate with their network. A Station can also send a probe request, which will cause any access point that receives it to send a reply with the needed information. The node must then request an association, with optional authentication. Note that authentication may be required as a result of network policies, even when the data sent by the nodes is not in itself proprietary. The authentication process also often involves exchange of information required to provide physical-layer security through encryption.
Association alone allows only local communications within layer 2 of the OSI communications protocol reference model. An additional network configuration process is needed to support wide-area or global communications (Figure 3). Wi-Fi-based nodes are capable of standard IP networking. A node undergoes a conventional DHCP Discover process and message exchange, leading to the assignment of a locally or globally unique IP address. Once network configuration is complete, the node can communicate with a data server located in any reachable network segment, allowing a flexible network architecture not constrained by the physical sensor configuration. Network joining operations are relatively energy-intensive, but in most cases will occur only once or at long intervals, and thus do not materially shorten operating life.
Once a node has joined the network, it can enter a low-power state from which it can awaken to send data and perform other operations. An example sequence is depicted in Figure 4. After power is applied and network joining operations are completed, the node enters a loop in which most of its time is spent in standby. The RTC block provides an always-on timer function that can be used for scheduled wakeup; alarm inputs can also initiate asynchronous bootup.
A flexible sensor network should support changes in node configuration, but a network with a large number of low-cost nodes does not allow for practical manual configuration. Remote management capabilities are therefore indispensable. In order to support remote configuration without excessive energy use, nodes can periodically waken and notify a configuration server that they are ready for management messages. For example, if the Simple Network Management Protocol (SNMP) is used, nodes can send unsolicited trap messages to a configuration server; the server can respond with Set and Get commands if a configuration operation is pending. Since reconfiguration is likely to be an infrequent event in most working installations, the time period between configuration operations can be hours or days, minimizing the average power devoted to node management while providing acceptable latency.
It may also be necessary to send messages to maintain a network connection. In 802.11, access points may disassociate a station if they have not heard from it for some time; this behavior is not defined by the 802.11 standard, and varies from one commercial access point to another. DHCP leases may also terminate if not renewed. If data transmissions are not sufficiently frequent, the node may need to send linkup or heartbeat messages whose purpose is simply to maintain link and network connectivity.
Powering Sensor Nodes
Sensor nodes operating as described above use very little average power. For example, in the Standby state, a GS1011 SoC consumes about 15 microwatts. A typical AA Lithium battery can supply about 26,000 joules. If the remainder of the components constituting a node based on the SoC are turned off in standby, that’s enough to power the node in standby for about (Equation 1):
Current batteries won’t last that long even with no load! If Standby power is low enough, energy consumption is dominated by wakeup events even when they are infrequent.
A typical sensor wakeup is depicted schematically in Figure 5. The sensor spends a few milliseconds stabilizing the clock oscillator and booting up the firmware, at relatively lower power consumption. This is the stage where sensor data may be acquired or forwarded from off-chip storage. Once the packet is assembled, the radio is turned on. The receiver first listens to see if any other device is transmitting—clear-channel assessment (CCA). If the channel appears vacant, a packet is sent. The radio may remain on for a short time to receive an ACK packet from the access point. After successful ACK, state information can be saved and an orderly return to Standby conducted. The whole process takes 10-20 milliseconds and consumes around 1-5 milli-joules. Even using the worst-case values, a battery can support about (Equation 2):
Which is 10 years at 1 wakeup per minute. While achievable lifetimes are shorter than this due to battery leakage, power consumed by the actual sensor hardware, and other operations such as linkup and configuration messages, years of operation can be achieved from a battery.
In some applications, battery cost may be prohibitive, or maintenance even at intervals of years may not be acceptable. Harvesting energy from the environment may be sufficient to power a sensor node, but applicability of this approach is highly sensitive to the specific environment in which the sensor is to be placed. In a lighted indoor environment, a 1 cm2 photovoltaic cell can produce about 10-30 microwatts. This is insufficient to support transmit power on an instantaneous basis; the energy must be stored in an ultracapacitor or other device. A cell of around 10 square centimeters area can support a transmission every few minutes when the ambient is lit. Outdoor sunlight can provide 10 times more energy.
Mechanical energy can be substantial when vibration is continuous; harvested energy levels on the order of 1 milliwatt are achievable. Peltier effect elements (basically a thermocouple) can be employed to obtain energy from temperature differences—for example, if cold and hot process piping are located close to one another. Output voltages are very low and DC-DC boost is needed to produce voltages sufficient for standard CMOS operation. Specialized devices with ultra-low-power consumption have been developed for operation with harvested energy, but at the cost of simplified link-layer operation that sacrifices standard networking protocols. Wireless sensor networks can be readily constructed using commercially available components and standardized link and network protocols. Wireless link ranges of tens to hundreds of meters can be achieved depending on the environment and sensor configuration. Operation over standard networks allows the data architectures to be independent of the physical sensor arrangement. Sensor nodes can operate for years from a conventional AA battery; energy harvesting techniques can be substituted for battery power in some applications.
San Jose, CA.
© 2009 RTC Group, Inc., 905 Calle Amanecer, Suite 250, San Clemente, CA 92673