Camera Link and GigE improve image speeds

As cameras used for inspection must produce ever-higher-quality images and as image-processing speeds increase dramatically, data transmission rates have remained a bottleneck for vision systems. Communication standards designed for more leisurely data environments have not proved adequate for handling the demands of inspection applications. To address this situation, two vision standards have emerged to provide faster, more reliable data transmission—Camera Link and GigE Vision

The Camera Link standard was developed by camera manufacturers themselves, specifically for connecting cameras to frame grabbers. Kinney explained, "We didn't try to adapt existing standards like USB and FireWire that had been created for consumer applications."

The standard is based on an implementation of technology from National Semiconductor that dramatically reduces the number of conductors in the cable. Camera Link uses the same standard connector for both frame grabber and camera, regardless of the (conforming) manufacturer.

Before Camera Link, industrial cameras had already advanced from analog to analog-progressive, to 8-to-10-bit (monochrome) imaging. Connecting cameras to frame grabbers involved a host of proprietary protocols and connectors, along with random pinouts, rather than a single standard. The cameras might feature a 31-pin miniature connector, while the frame grabber might use a 100-pin SCSI. Cables were unique to each brand of camera and each brand of frame grabber. A cable designed to connect a Coreco frame grabber to a Pulnix camera, for example, wouldn't work with a frame grabber from another manufacturer. Low volumes and the plethora of cable architectures meant that cables could cost $500 or more apiece.

The industry's transition to 24-to-30-bit color cameras aggravated the situation. Adapting existing techniques would have required 90-pin cables that could easily cost $1000 each. So, a group of 12 industry experts met in 1998 to find an alternative.

Fig. 1  A differential clock serializes 28 bits (24 data bits and 4 bits for camera timing), transmits them across four parallel data lines and a clock line, and converts them back to the original 28 bits at the other end. Courtesy of JAI Pulnix.

Fig. 2  The “double-base configuration” allows a single frame grabber to support two base configuration connections. Courtesy of JAI Pulnix.

Kinney contends that Camera Link remains the best way to build the highest-quality cameras with the fewest parts. "Camera Link provides an interconnection that is independent of image resolution, video format, and frame rate," he said. "Camera Link offers real-time asynchronous reset and camera signaling directly over the interface. That capability, along with high bandwidth, gives Camera Link distinct advantages compared to the other standards. FireWire and USB make no provision for direct real-time data transfer, signaling, or camera control."

In addition, designing a universal cable around a 3M MDR 26 connector dramatically reduces costs. A 2-m Camera Link standardized cable usually costs less than $100.

The original Camera Link standard provided a straightforward solution that has not changed much. Initially, communication was done through a manufacturer-supplied .dll file that provided such basic commands as "open port" and "write to port." Version 1.1 merely cleaned up some verbiage, added a communications link to the controlling PC, and allowed for error handling. The improved communication link provided a supervisory layer as a means to control several different Camera Link communications ports in a system with multiple frame grabbers installed.

In 2005, a miniature version of the original connector was proposed to the Camera Link Committee. With some minor changes to the specification, the committee approved the new connector, and it has begun to appear on many manufacturers' products, including a new line of cameras by Sony.

The latest update to the standard calls for incorporating power into the Camera Link cable, which would eliminate the need for a special power cable. Although the idea has been well received, some manufacturers have expressed concern over backward-compatibility. The current proposal before the committee—scheduled for final action by May of this year—accepts the concept, provided that prior-generation nonpowered equipment would incur no damage because of the change.

The mini-connector currently includes four ground pins at the corners of a rectangle, and the new version would convert two of those to power pins. The remaining ground pins would serve as the connector's inner shield. To maintain compatibility, a powered frame grabber would detect nonpowered equipment or cable faults out on the line before applying power. Camera manufacturers would include a resistance of between 50 Ù and 200 kÙ on the powered lines to assist the protection logic.

According to Kinney, Camera Link minimizes component count, reduces power consumption, and remains the most straightforward camera-to-frame-grabber communication standard to implement. It "is the lowest-cost, highest-performance solution for connecting video components," he said. Many OEM and military applications already require Camera Link compliance.

As with Camera Link, developers of the GigE Vision standard recognized the bandwidth, scalability, networking, and processing-flexibility limitations of standards such as FireWire (IEEE 1394b) and USB for real-time applications. The developers based their standard on Gigabit Ethernet (GigE), which overcomes those limitations.

"It seemed an ideal fit for real-time machine-vision applications," said George Chamberlain, president of Pleora Technologies and former co-chair of the GigE Vision committee. "It had plenty of bandwidth and supported a wide range of networking and processing architectures—including distributed processing and pipeline processing—using standard PCs and switches. It wasn't a daisy chain configuration, either. It offered dedicated, full-duplex connectivity over either copper or fiber, so there was no bandwidth sharing between links, which is important for real-time operation. And 10GigE, the next generation, will support transmission speeds of 10 Gbps, more than enough for evolving throughput requirements."

The goal of the GigE Vision committee, which began its work in 2003, was to standardize communications over GigE for vision applications from the camera head all the way to the target application. "We wanted to promote software and hardware interoperability independent of vendor. This would reduce support requirements and shorten time-to-market for OEMs who incorporate compliant products into their systems," said Chamberlain. "The idea was to deliver a framework that would support low-cost as well as high-value cameras without imposing excessive restrictions on their designs." Chamberlain contends that GigE, along with the emerging GigE Vision standard, stands to benefit both manufacturers and users of machine-vision equipment by reducing the cost and complexity of the final application.

The first version of the GigE Vision standard, now almost finished, is expected to be released by the Automated Imaging Association (AIA) in mid-2006. Although the standard does not discuss applications, and thereby offers manufacturers a broad latitude, it nevertheless addresses traditional machine-vision applications that must deliver high-speed image data. The standard includes four basic elements:

• The GigE Vision Control Protocol (GVCP)—which runs on top of UDP (Universal Datagram Protocol) IPv4—defines how to control and configure compliant devices (such as cameras), specifies stream channels, and provides mechanisms for cameras to send image and control data to host computers.
• The GigE Vision Stream Protocol (GVSP) defines data types and describes how images are transmitted over GigE.
• The GigE Device Discovery Mechanism defines how cameras and other compliant devices obtain IP addresses.
• An XML description file based on the emerging GenICam standard developed by the European Machine Vision Association (EMVA) provides the equivalent of a computer-readable datasheet to allow access to camera controls and image streams.

With regard to GenICam, Chamberlain commented, "The GigE portion of the standard fulfills its original goals. GenICam, however, places constraints on camera design that are inconsistent with the implementation style of many camera manufacturers."

Fig. 3  All Ethernet packets consist of headers, payloads, and trailers. Courtesy of Pleora Technologies.

Chamberlain said that the GigE Vision standard offers advantages over Camera Link. It can, for example, transmit low-latency sustained video data up to 100 m without repeaters—and can transmit it further with low-cost GigE switches or optical-fiber cable. GigE Vision uses industry-standard PC and LAN equipment—such as GigE network interface cards or network-interface chips (NICs) in place of specialized frame-grabber boards—which reduces the resulting network's complexity and implementation costs. Making use of commercial software reduces costs further.

The 1-Gbps data rate of GigE translates to an image transfer rate of up to 115 Mbytes/s, which supporters contend meets 90% of today's image-transfer requirements. Full-duplex (bidirectional) data transfers enable users to control cameras or video devices like any other IP-connected network devices. GigE Vision also supports a wide range of networking options, including single camera to single PC, multiple cameras to a single PC, a single camera to multiple PCs, and multiple cameras to multiple PCs.

Table 1 compares GigE, Camera Link, FireWire, and USB 2.0. In raw speed, Camera Link comes out the winner, easily accommodating high-performance vision applications. It can stream data at rates up to 6.12 Gbps over dedicated point-to-point copper links of 10 m or less. The 10-m maximum, however, requires tethering PCs to cameras or incorporating repeaters or more expensive fiber-optic cable. Another drawback: Camera Link isn't flexible enough for interconnecting several cameras or for centralizing control and maintenance. Because it runs over specialized cables and terminates on PCI frame grabbers, the standard enjoys few economies of scale.

FireWire, on the other hand, which evolved for consumer applications, offers "plug-and-play" capability, and uses a readily available low-cost PC interface. With its bus topology, up to 63 devices share 800 Mbps (up to 512 Mbps for a single camera) in a "daisy chain." Adjacent devices can be separated by up to 4.5 m to a maximum of 72 m over twisted-pair copper cable.

FireWire does not, however, include error-checking algorithms, and its relatively limited bandwidth must be shared among all cameras on the network. In addition, FireWire's Windows driver monopolizes the PC during data transfer. (Some manufacturers have addressed this by developing proprietary drivers.) The cost of its copper cable represents another drawback. Less expensive options are available, but at a considerable sacrifice in bandwidth.

Like FireWire, USB also developed as a standard for connecting peripherals to PCs. As such, it can deliver up to 480 Mbps, much slower than either GigE or Camera Link, and the bandwidth must be shared among up to 127 connected devices. Its chief advantage rests in its ubiquitousness, but it lacks sufficient power to succeed in most inspection-imaging applications.

In addition to data packets, a GigE Vision system includes two other key elements—the camera interface and the physical infrastructure. All GigE video or imaging cameras must include an Ethernet interface. It provides full-duplex (bidirectional) data, allowing image data to flow from the camera to the PC as control data flows from the PC to the camera. System integrators can implement the interface in two ways. An IP/Ethernet communications software stack can run on a microprocessor with an embedded operating system. Chamberlain contends that at high data rates, typical implementations of such systems can consume as much as 25 W of power at GigE's full data rate. Thus, this approach serves best in low-performance applications.

Alternatively, a manufacturer can develop hardware that contains hard-coded packet-processing functions that eliminate the embedded operating system. This approach packetizes data with clock-cycle accuracy, delivering latencies as low as 500 ìs. It also consumes much less power—less than 2.25 W in some implementations—and more easily integrates into cameras.

Clearly, of the currently available options, GigE and Camera Link exhibit distinct advantages over the more consumer-oriented USB and Firewire 1394b. Still, they both present benefits and drawbacks. In the end, as always, application needs will dictate which transport framework to use.