GigE Vision boosts inspection networks

Two key issues that affect all Ethernet networks are CPU loading and latency.

By Ann R. Thryft, Contributing Technical Editor -- Test & Measurement World, 8/1/2010 12:00:00 AM

GigE Vision has brought numerous advantages to machine-vision systems, including a 100-m cable length and higher bandwidth. Other advantages are a variety of lower-cost, readily available, standard hardware components, and the ability to easily manage a network of components, including cameras, switches, and routers. Disadvantages have included Ethernet's inherent latency and the host CPU loading issues that arise with the elimination of frame grabbers, yet some component manufacturers and system integrators are finding they can work around those drawbacks. Today, demands for higher data rates are pushing some to work on methods for increasing bandwidth in GigE Vision above 1 Gbps (see "Next step: 10 GigE? ").

In this example of an inspection system for flat-panel displays, imaging data is transferred in real time to a GigE switch, then multicast to rack-mounted PCs, which analyze the images for different types of defects. Courtesy of Pleora.

The demand for networking-based connectivity is rising along with the rise in vision system complexity, said George Chamberlain, president of Pleora, a co-founder of the GigE Vision standard. "Networking enables a wider range of applications, such as a single point of analysis for multiple cameras and distributing video to multiple end points simultaneously," he said. The GigE Vision 1.2 specification's support for nonstreaming devices brings much more than just traditional camera-computer connectivity, said John Phillips, Pleora's senior product manager. "Now, we have access to multiple device types along an Ethernet backbone, so we can build a more fully featured and robust system that's more tightly integrated."

Until recently, camera interfaces were limited to older analog technology or Camera Link. Although the digital Camera Link is technically an open standard, it is not used outside machine vision. By contrast, the GigE Vision, FireWire (IEEE 1394a/b), and USB digital interfaces are based in the much larger, high-volume consumer and PC industries, said Arndt Bake, GM of Basler Vision Technologies. These interfaces let vision system designers take advantage of the lower product costs made possible by those high volumes to build larger, faster, more comprehensive vision systems.

Although every PC now has a GigE interface, that's no longer true for FireWire, said Bake. In its Scout camera line, the company offers both interfaces, while prices and all other features are identical. In the past year or so, said Bake, most customers buying these cameras have chosen GigE Vision rather than FireWire. He added that FireWire's growth probably peaked in 2008 while GigE Vision is still growing.

In 2010, Basler estimates that, for all vendors, volumes of machine-vision cameras with a GigE Vision interface will be 50% more than in 2009. The company expects GigE Vision to bypass FireWire and analog in a couple of years as the most popular interface in machine-vision cameras and to reach its peak in about six to seven years.

Compared to other camera interfaces, GigE Vision has almost all the advantages, including bandwidth, cable length, the ability to network cameras, and the fact that it's a universal standard, said Ravi Guntupalli, business manager for Princeton Instruments. Camera Link is more mixed: It supports the highest bandwidth for real-time data transmission and has low latency, but cable lengths are limited and the frame grabber is expensive and not easily integrated.

When GigE Vision debuted, everyone thought it would replace Camera Link right away because of its high data rates, said Bake. But it does not reach Camera Link's 6.8-Gbps bandwidth. Only 10% of Basler's GigE Vision cameras are sold as Camera Link replacements, whereas 90% replace either FireWire or analog cameras.

CPU loading and latency
Two key issues that affect all Ethernet networks are CPU loading and latency. Initial concern about the effects of CPU loading on GigE Vision applications was due to the overhead caused by Ethernet's use of packets, especially in high-speed data networks. But the machine-vision industry has proven that neither issue is a problem, said Paul Kozik, product manager for Allied Vision Technologies. Filter and performance drivers have minimized CPU loading, while successful installations and a track record of reliability in GigE Vision systems have eliminated concerns about determinism and latency.

Compared to Camera Link, the CPU load is higher with GigE Vision, but it's not a major concern for Princeton Instruments' industrial machine-vision customers, said Guntupalli. That's because the built-in FPGAs in the company's cameras not only perform onboard image data handling, but also handle some of the post-acquisition image-processing load, such as white balancing, pixel correction, and flat-field removal, before data reaches the CPU.
Some, however, say that Ethernet latency may be a problem at real-time inspection rates. Guntupalli said that many of Princeton's customers acquire images at up to 30 fps, and the company has had no complaints. "If it was hundreds of frames per second, the latency might be a concern to them," he said.

Electronics inspection needs very fast image transfer, especially in automated optical inspection systems, said Bake. "A GigE Vision camera would fit here from the data transfer angle, but other requirements such as latency make Camera Link the better interface choice," he said. "This is especially true when you start to scale things up. At the slower end, you can go with a GigE Vision camera, but the higher the data rate, the more the requirement for real-time behavior increases."

Latency can occur at the camera, network component, and PC driver levels, said Pleora's Phillips. Although vendors have limited control over latency introduced at the network level, the GigE Vision standard lets them differentiate their products in terms of latency introduced at the other two levels. "For example, we can introduce as little as 525 µs of additional latency with a 2-Mpixel camera over a 1-Gbps Ethernet link, with Pleora eBUS drivers on the PC and a Pleora iPORT IP engine in the camera," he said. Pleora's drivers bypass the operating system's network stack and are optimized for GigE Vision, so they not only help reduce latency, but also reduce the burden on the CPU, giving it more head room for image-analysis applications.