Wireless Mesh Networks in Industry: How Radios Really Perform

It is now possible to embed radio technology in a wide variety of inexpensive battery-powered devices. In the industrial world, many new products are being developed to place inexpensive sensors wherever there are interesting data. This emerging technology, unconstrained by expensive wiring, has the potential to save industry a lot of lost productivity.

Two new and closely related standards are fueling interest in wireless sensor networking: IEEE 802.15.4 and ZigBee. IEEE 802.15.4 is a physical and MAC layer standard for low-rate RF networks, whereas ZigBee is a standard proposed by the ZigBee Alliance, which provides a recommendation for the network and application interfaces and operation of low-rate wireless networks. Market forecasts for IEEE 802.15.4 transceivers project hundreds of millions of units over the next five years. Before it is used in the kind of applications where millions of dollars are at stake, the technology must be proven and well tested. As a first step toward this goal, we've been carefully validating the real-world performance of these early offerings under a Department of Energy (DoE) grant.



Evaluating the Radios

Current standards-body compliance procedures are limited to interoperability testing, rather than testing whether the radios meet their specifications under a reasonable range of operating conditions, such as multipath, temperature, and vibration. Multipath is a phenomenon that occurs when multiple copies of the same information, following different paths at different times, arrive at a single point; it can be both destructive as well as constructive. Temperature can cause issues with receiver sensitivity, such as transmitter-to-receiver clock skew, as well as transmitter output degradation. Vibration can cause random variations in crystal frequency, affecting both radio performance and system timing. With dozens of IEEE 802.15.4 radios coming on the market over the next several months, we need industry-wide benchmarks for radio performance under well-defined conditions. Without benchmarks of this sort, users cannot make informed comparisons.

In early 2004, there were only two IEEE 802.15.4 radio transceivers available, one of which was still in the beta stage. To validate the radios' specifications, we first tested their performance in a wired bench configuration. We found the beta product to have 4.5 dB less sensitivity under ideal conditions and 6.5 dB less sensitivity under controlled multipath conditions, despite similar specifications. According to vendor claims, only about 2 dB of this difference could be attributed to the use of beta-level technology. However, even accounting for beta-level technology, this still leaves the beta units at a 4.5 dB disadvantage under multipath conditions—which are frequently encountered in indoor applications.

Theoretical Performance in the Factory

We then explored performance in factory environments. We characterized three basic factors: channel fading, channel response, and interference.

Channel Fading. Channel fading is the effect of multipath on the received signal. Factory radio environments are far from ideal, with many metal surfaces to block and reflect radio signals. In practice we've observed that small changes of position (a few inches) or small changes in frequency (a few MHz) will radically change the signal level and radio connectivity.

The graph in Figure 1 shows how received signal strength at a single position on a factory floor can vary across the radio band from 2400–2483 MHz, which is one of the unlicensed bands in which IEEE 802.15.4 operates.

 Figure 1. This graph of data from a factory floor shows how received signal strength varies across the 2400–2483 MHz radio band. The blue dots show the signal variation with a narrow-band radio, while the pink line models that for an 802.15.4 direct-sequence radio. Note how the profile changes with even tiny variation in frequency. Signal level is also sensitive to small changes in position and to random environmental changes over time.
Figure 1. This graph of data from a factory floor shows how received signal strength varies across the 2400–2483 MHz radio band. The blue dots show the signal variation with a narrow-band radio, while the pink line models that for an 802.15.4 direct-sequence radio. Note how the profile changes with even tiny variation in frequency. Signal level is also sensitive to small changes in position and to random environmental changes over time.

The blue dots show the received signal level variation with a narrow-band radio. At some frequencies, we observed fading of 40 dB in line-of-sight tests. The pink line, showing 20 dB fades at various frequencies, models the received signal strength for an 802.15.4 direct sequence signal. Tiny variations in position (<2 in.) or movement of objects in the environment can cause radical changes in the profile.

Channel Coherence. This is the tendency of the radio channel to change over time, position, and frequency. We were particularly concerned that the channel might not be stable enough to support reliable communications. For example, a motor operating at 3600 rpm might cause rapidly oscillating changes in the radio path at a sub-millisecond period. Since IEEE 802.15.4 packets are about 1–3 ms in length, we'd want the radio channel to be coherent over that period. In typical factory environments and for a fixed wireless installation we measured coherence times on the order of 0.1 s, indicating that channel coherence time may not be a significant issue. At the time, we did not measure the coherence time on a vibrating platform because a robust enough hardware platform was not available. Testing on a vibrating platform might have shown a significantly lower coherence time, which in turn can limit the maximum reliable packet length.

Interference. This is the radio's sensitivity to noise in the spectrum, especially cross-interference from co-located 802.11 (WiFi) transmitters or Bluetooth devices. The 11 available 802.11 channels are relatively wide and overlapping. In contrast, the 16 available 802.15.4 channels are relatively narrow and distinct from one another. Due to the very different nature of the signals, 802.15.4 radios are relatively insensitive to interference from 802.11 devices, and vice versa. We have verified this through simulation as well as with controlled experiments on physical devices.



It is important to note that current 802.15.4 radios transmit very low signal levels of ~1 mW (0 dBm). This is one or two orders of magnitude below the signal level typical for 802.11 or some classes of Bluetooth devices. When the radios operate at these extremely low power levels, interference from Bluetooth devices becomes a serious issue. In addition, typical indoor path loss calculations predict a range limitation of ~25 m at 0 dBm. For these reasons, we decided to amplify the radios to about 15 dBm in products that we have under development.

IT managers will need simple installation guidelines to ensure that 802.15.4 will not interfere with mission-critical wireless LANs in the same facility. In applications we've envisioned, 802.15.4 will be a good neighbor in the radio spectrum due to the extremely low duty cycles of wireless sensors. The low duty cycle is intrinsic to the operation of devices that need to run for years on a small battery; these devices simply lack enough of an energy source to significantly impact the radio environment.

Actual Performance in the Factory

For system-level validation, we tested 802.15.4 networks programmed to exchange many thousands of packets on all 16 channels over extended periods of time in an industrial facility with restricted access. No attempt was made to place the nodes to best facilitate transmission; they were placed in positions that would be appropriate for industrial temperature and vibration sensors.

In our tests there was good connectivity among all nodes in the network, even though several had large pieces of machinery blocking their transmissions. Analysis of lost packets showed that some channels performed very well while others showed significant packet loss. The problematic channels were specific to particular pairs of nodes and could be overcome with a frequency-hopping protocol. This performance was consistent with the channel coherence bandwidth measurements we made in these locations and shows that multipath can be both a positive as well as a negative effect and is essentially impossible to predict or control.

Conclusion

Based on our testing to date, we conclude that a combination of the following implementation strategies will provide optimal performance of IEEE 802.15.4 radios in factories.

Well-Characterized Radios. With more than a dozen new 802.15.4 transceiver chips coming on the market, vendors will be under pressure to release their offerings as soon as possible. Without standardized testing, data sheets will not necessarily reflect operating realities.

Frequency and Path Diversity. Radio connectivity varies over time and is very sensitive to position. Network reliability can be remarkably improved with a protocol that adaptively and seamlessly changes frequencies and connections from minute to minute in response to interference and other environmental conditions. With additional output power, network density can be reduced while maintaining reliability.

Higher-Powered Radios. Today, most IEEE 802.15.4 radios are limited to about 1 mW, or 0 dBm. More margin is needed for a system that can be flexibly installed in spaces that may span many acres. We are using nodes that are amplified to 15 dBm (plus antenna gain), which provides a reasonable tradeoff of range and battery power.

Daniel Sexton is Program Manager, GE Global Research, Niskayuna, NY; 518-387-4121, sextonda@crd.ge.com, www.crd.ge.com

Jay Werb is CTO, Sensicast Systems, Needham, MA; PHONE, EMAIL, www.sensicast.com.