Engineers Push Ethernet Further on Down the Food Chain

Hannover, Germany-The "world's largest industrial fair" was a bit quieter this year, say most observers. Fear of SARS, the war in Iraq, and the general economic downturn took their toll, according to fair organizers, who noted the 200,000 attending the massive exhibition represented a 20% decline from the peak attendance several years ago.

Nevertheless, automation and control vendors were at the great German showcase with new things to show and tell. As far as communications are concerned-always a hot topic at Hannover-the big news is that it appears everyone is now on the same wavelength. Ethernet has become the high fashion of industrial comons-you really don't want to be seen without it, so while all the controller vendors were trotting out their Ethernet connections on the catwalks, companies who specialize in communications accessories showed up with delightful collections of hubs, switches, and connectors.

With Ethernet firmly established, there's no turning back, and some daring engineers are taking things to the next level. They have proposed a more radical idea: Ethernet just might work for real-time control. For many automation specialists, that's pushing things too far, but for the creative design engineers at Beckhoff, it's the starting point for a whole new line of products.

Ethernet, of all things

When Ethernet was introduced as an office network in the early 1970s, its appeal was simplicity. All you had to do to install it was to string a coaxial bus cable wherever connections might be needed. Taps at appropriate points along the cable provided network access for terminals and printers.

During the 1980s great improvements were made in Ethernet speed and cabling, but it wasn't until the mid-1990s that industrial users began deploying electronic switches to segment the bus into "collision domains" to improve performance. The star topologies added complexity to the system, but they helped to improve determinism, which was always considered the Achilles heel of Ethernet as far as industrial users were concerned.

Ethernet use in the factory has grown substantially, but always in "top down" fashion. It was first used to interconnect top-level computers with the company's enterprise network. Later the "vertical integration" crusade reached to lower levels, such as programmable controllers. Now on the minds of factory engineers is the question: just how far down to the factory floor will Ethernet go?

Beckhoff's surprising answer is: right down to individual I/O points. The new product it was showing at the Hannover Fair, called EtherCAT (Ethernet for Control Automation Technology), can scan 1,000 I/O points in 30 microseconds-an amazing speed.

And even though the system does require some special hardware-FMMU (fieldbus memory management unit) chips in each I/O terminal-the rest is standard Ethernet cards, cable, and software. All the I/O terminals can be connected in simple linear fashion, the way Ethernet was originally used, without worrying about collisions affecting the determinism.

Pure hardware

Gerd Hoppe, president of Beckhoff North America, says the high speed is the result of a "pure hardware implementation" of the system.

Instead of behaving like Ethernet TCP/IP, Beckhoff's EtherCAT sends out bursts of short "telegrams" from the PC host computer. These pass through the entire I/O system and echo back to the transmitter. During the few nanoseconds it takes for the signal to pass through an I/O terminal a special FMMU chip reads the data "on the fly." If the I/O terminal has anything to tell the host PC, it also writes this data into the telegram at this time.

Timing

On the PC control side, the Ethernet card is linked directly into the PC's memory banks with a dual port RAM. The software always finds precisely the "right time" to send a telegram out to the I/O terminals; the system is controlled right down to the microsecond which means that, if telegrams need to be sent, they never have to queue for more than a microsecond.

This is the strength of the system, says Hoppe. "It is almost as if the I/O terminals are a part of the extended memory of the PC," he says. In fact, this is nearly true, because the only thing separating the PC's memory from the FMMU chip is the DMA access hardware and a piece of wire. "This is why the system is so fast," notes Hoppe. "There are no microprocessors and consequently there is no software to be executed anywhere in the system. It is all done in hardware."

What happens if the Windows(R) operating system decides it wants to send a TCP/IP message, or any other Ethernet message for that matter? The real time system will never collide with Windows because the software will only allow Windows to transmit messages on the network if there is a sufficient gap between the telegrams. The real-time telegrams always receive the highest priority; all other Ethernet activity is buffered until a time slot is available. "In practice, this usually delays the 'other' Ethernet messages by a small fraction of a second," claims Hoppe.

The elegance of the Beckhoff design is that no matter how the terminals are arranged or physically re-arranged, the sub-telegram 'finds' its target FMMU and interchanges data with it. The data sequence is independent of the physical order of the Ethernet terminals in the network.

Unlike other recent real-time Ethernet implementations (see sidebar), which require special hardware or connections in the controller, EtherCAT manages to get along with plain vanilla $50 Ethernet cards.

High performance

The specs Beckhoff quotes are staggering: 1,000 I/O in 30 microseconds. Communication with 100 servo axes takes 100 microseconds. This type of performance approaches high end VMEbus systems with their directly connected I/O cards.

This could certainly open up potential and some new thinking about industrial PC miniaturization, Hoppe points out. "Why put all those I/O cards into PC slots if you can connect all this data to the PC through one Ethernet wire and a $30 network card?"

"The extremely high performance of the EtherCAT technology enables control concepts that could not be realized with classic fieldbus systems," explains Frank Metzner of Beckhoff. "For example, the Ethernet system can now not only deal with velocity control, but it also can handle the current control of distributed drives. The tremendous bandwidth enables status information to be transferred with each data item. With EtherCAT, a communication technology is available that matches the superior computing capacity of modern industrial PCs. The bus system is no longer the 'bottleneck' of the control concept. Distributed I/Os are recorded faster than is possible with most local I/O interfaces."

Metzner says there is no restriction on the type of Ethernet device that can be connected within the EtherCAT strand via a hub terminal. The protocol is capable of tolerating other Ethernet-based services and protocols on the same physical network, and it does so usually with a minimum loss of performance. The system uses special buffers in order to delay non-real-time Ethernet data until there is a gap in the real-time transmission.

The first EtherCAT products are scheduled to be available at the end of 2003.