Untangle industrial network selection

There's no universal answer for figuring out the benefits and trade-offs that enter into network selection, but here are some observations that can help automation designers.

By Clark Kinnaird

AN IMPORTANT TREND in factory automation is the increase in networked connectivity between sensors, controls, actuators and other system components. Automation designers face many challenges and tradeoffs in the development of a successful network solution.

This article, the substance of which was presented as a White Paper, “Specifying Industrial Field Bus for Automation,” at the 2005 SPS Electric Automation America Conference in May, discusses the choices necessary to select a network technology appropriate for specific applications. We’ll compare the technical constraints of several common network options, primarily for discrete applications, and discuss guidelines for selecting media and protocol.

The intended audience are system designers who develop networked automation, but who might not be familiar with the details of data transmission design. The takeaway should be a general overview of some of the common network options available, and some knowledge of the kinds of questions to pose when choosing an architecture, a protocol, and an implementation.

There’s Ideal, and There’s Real
There are at least two scenarios that can arise when selecting a bus network for an application. In one, the engineer starts with a blank sheet of paper and can specify any network that best meets the needs of the application at hand. In the other scenario, the engineer is constrained to specify a network compatible with an existing installation.

In an ideal world, all networks would be compatible, and it would be easy to translate from one to another. In actuality, interconnecting different networks can be very difficult, so we’ll leave that can of worms alone. If we assume the choice of network for the second scenario essentially is pre-determined, it makes sense to focus this article on the first scenario.

As we look for ways to compare different network choices, we find a long list of possibilities. Specific parameters of comparison and design concerns include:

  • Data rate and data latency
  • Physical interconnect medium
  • Noise immunity
  • Bit error rate and bus faults
  • Allowable interconnection length
  • Allowable number of network nodes
  • Ease of adding additional nodes
  • Power consumption, cost, reliability and isolation requirements

We’ll look at several of these, discuss why each of the characteristics is of concern for automation networks, and what tradeoffs come into play for making appropriate selection. We’ll see how there is no universal answer but, in optimizing one parameter, there are tradeoffs and compromises involving other parameters.

The Likely Suspects

Available network choices include:

  • 4-20 mA--This analog current loop network is slow but simple. It is limited to one transmitter per loop, but can have several receivers. The analog format limits the higher-level functionality, but this still is a widespread implementation choice for communicating simple sensor measurements to a central controller.
  • HART--The Highway Addressable Remote Transducer network augments the 4-20 mA loop with a modulated signal. This allows transmission of digital information, although the data rate is relatively slow.
  • RS-232--This standard interface has been around for a long time and still is used in many simple interfaces for initial setup of systems, diagnostics and other non-time-critical functions. A single-ended network, RS-232 does not have the same noise immunity as other standards that take advantage of differential signaling.
  • RS-485--An outgrowth of RS-232 and RS-422, the RS-485 electrical specification is the basis for several industrial network standards, including Profibus, Interbus, Modbus and others. The strengths of RS-485 are its immunity to noise and ground offsets, bidirectional and multiple-driver capability, and party-line simplicity.
  • Interbus--A ring-based network, Interbus uses RS-485 signaling with point-to-point connections, and full-duplex operation to make an adaptable bi-directional structure. Other variations of Interbus use fiber or infrared media for signaling.
  • Modbus--This bus has several variations, the most common is based on RS-485 signaling. Other implementations use Ethernet or RS-232. In addition to industrial automation, it also is used in building control applications.
  • Profibus-DP--Based on RS-485 signaling technology, Profibus-DP (Process Fieldbus, Decentralized Peripheral) is commonly used for factory automation, especially in Europe. The Profibus standard specifies the protocol, electrical layer, termination, signaling rates and grounding/isolation schemes. There are other variants of Profibus for fiber media, intrinsically safe applications, and motor control applications.
  • DeviceNet--Based on the Controller Area Network (CAN) signaling specification, the DeviceNet standard specifies the electrical layer (voltages, current loading, termination and isolation/grounding) and protocol requirements for a device-level network.
  • ControlNet--With a high signal-to-noise ratio and coaxial media, ControlNet is a robust, relatively high-speed industrial network. Its strengths are deterministic timing, robust electrical characteristics, and simplicity of expansion.
  • Industrial Ethernet--Several variations of Industrial Ethernet now are available. These include Profinet, EtherCAT, EtherNet/IP (Industrial Protocol), Ethernet Powerlink and others. Each is based on the IEEE 802.3 Carrier Sense Multiple Access with Collision Detection (CSMA/CD) Standard for Local Area Networks. Each variation has differences and is not easily interconnected with the others, mainly because of the manner by which the requirement for known data latency (deterministic timing) is handled. Versions with 10 Mbps and 100 Mbps are commonly used for industrial automation applications.

Node Limits
Automation applications often require connecting many sensors and actuators, as well as controllers and human-machine interface panels. The maximum allowable number of nodes might be limited by system architecture, electrical or optical characteristics of the physical layer, or by the addressing scheme inherent in the network protocol (Click here to view a .pdf of Table I).

Network architecture can be bus, ring, star or other arrangements. Bus networks such as DeviceNet or Modbus accept nodes anywhere along their length, but usually have restrictions on the spacing between the nodes. Ring networks such as Interbus form a closed chain with point-to-point links between adjoining nodes. Star networks such as Ethernet allow a hierarchical structure with many stars connected in various patterns. Loop connections such as 4-20 mA may have several receivers on one loop, but are limited to one transmitter.

Observe Posted [Data] Speeds
The basic function of any communications network is to move data from place to place. Data rate is a measure of how much data can be moved in a certain amount of time, but data rates of different networks might be measured in different ways. For analog communication such as 4-20 mA, the bandwidth of the circuit elements limits the rate. For digital communication, the rate is determined by how many bits (binary digits) can be communicated per second, and what fraction of the communicated bits actually have meaning for the application.

Another parameter related to the speed of the network is data latency, or the time interval between data being sent from one node and received at another. Data rate affects data latency, as does transceiver propagation delay, media propagation delay, and protocol overhead.

Transceiver propagation delay typically is on the order of 1 microsecond or less; the propagation delay through the media (fiber or copper) is limited to some fraction of the speed of light, and is therefore on the order of 3-5 nsec per meter of cable. Therefore, only in very long cables or very fast networks will the media delay be significant.

The protocol delay accounts for the time added by the protocol overhead (parity, addressing, error check and handshaking bits) required, plus the data payload on any message. This varies by network standard, but is significant for the higher-level protocols with complex formats.

An illustration of the difference between data rate and data latency might be helpful. A fast data rate/long data latency is like listening to a pre-recorded tape of a Spanish-language sports show. A slow data rate/short data latency is akin to listening to James Earl Jones live on the phone.
Table 2 surveys the raw data rates and message protocol latency for several networks. (Click here to view a .pdf of Table 2)

Demand Immunity
The industrial environment presents challenges from high-current components such as motors, pumps, switching power supplies, welding equipment and robotics. Immunity to these noise sources is necessary to ensure reliable network operation. At the physical layer, this requires proper attention to grounding, shielding and transceiver features. High signal levels on the network will increase the signal-to-noise ratio. The highest possible receiver threshold levels (often called “squelch”) will discriminate proper signals from noise.

Hysteresis in the receiver thresholds decreases the possibility of erroneous switching due to noise during signal transitions.

Each improvement in noise immunity comes with some cost. High signal levels require more power and might generate noise to other components. High receiver thresholds mean the system is less tolerant of signal losses in the media, which can decrease allowable network length. Receiver hysteresis can introduce propagation delay and pulse width distortion if not properly balanced. Note in Figure 2 how the receiver output with hysteresis is slightly delayed in response compared to the output without hysteresis.

Network Boundaries
Another measure of a network is how far the data can be communicated. Industrial networks often require long interconnections, more so than consumer, computer or automotive applications. Factors that limit allowable network length are losses in the media and electrical noise pickup, both of which affect signal-to-noise ratio, and propagation delay through the media, which affects latency.

Losses in the media (attenuation) occur whether the medium is copper, fiber or wireless. Optical fiber losses are very low (0.3 dB/1000 m at λ=1310 nm), allowing for very long connections using optical networks. Typical twisted-pair copper wire has higher losses, in the neighborhood of 1.5-5 dB per 100 m at 1 MHz. 

Network standards take the media losses into account by requiring much higher transmit-signal amplitudes compared with receive levels. For example, RS-485 signaling requires at least 1.5 V driver output and receiver thresholds of 200 mV. This gives a driver output to receiver factor of 7.5, or equivalently 17.5 dB of margin, which allows cable lengths to about 1,200 m at low signaling rates. Table 3 presents example calculations for typical industrial networks. (Click here to view a .pdf of Table 3). Note that the losses in the media may not be the only consideration in determining maximum allowable network length.

For Example
An example solution using an RS-485 physical-layer-based network will show how the network implementation is critical to support application performance. The contrasting requirements of a precision motion-control application and a multi-sensor process control application illustrate how system requirements affect network design choices (Click here to view a .pdf of Figure 1).

In the motion control example, tool speed and encoder precision combine to dictate a high data rate. The number of nodes is limited, as is the distance of the network. Data latency also would be critical for the position and velocity information used to close the servo loops. In this application, some of the slower networks such as DeviceNet or HART would not be appropriate because of the data rate requirements. Networks such as 100BaseTX Ethernet, Profibus, or ControlNet are possible choices for this application.

In the process control example, overall signaling rate is less important because the process time constants are much longer than the motion control application. This network might span a much longer distance and have several dozen nodes interconnected. Ground offsets between the nodes might result from localized high-current surges. The high node count makes Profibus a less-attractive choice, while slower networks such as DeviceNet and Interbus are now contenders for this application.

For either application, there almost certainly will be constraints regarding total cost of the solution, interoperability with other systems and legacy hardware, and motivation to standardize with other applications already supported in the field.

  About the Author

Clark Kinnaird is systems engineer for Texas Instruments, Richardson, Texas. You can contact Clark at c-kinnaird@ti.com.


Click the Download Now button below to view a .pdf version of the tables referred to in this article.