In the beginning—about 12 years ago—there was analog. The landscape of camera interface technology was simpler and easier to understand, with most vision systems consisting of an analog camera connected to a frame grabber. Then digital interfaces like FireWire, Camera Link, USB, and GigE were introduced, and the debate began. Now, with a plethora of devices popping up everywhere from the local Best Buy to megastores in Japan, USB 3.0 seems set to add fuel to the fire.
To understand how USB 3.0 will impact vision in the future, it’s important to understand the road that led to its development. The Universal Serial Bus is the most common serial peripheral interface in the history of computing. Present in virtually 100 percent of all computers, it is the standard for most computing peripherals such as mice, keyboards and hard drives, and sells billions of units every year. The USB standard was originally developed by Intel, Microsoft, Compaq, and others for peripheral components. These same member companies formed the USB Implementers Forum (USB-IF) to provide a support organization and forum for the advancement and adoption of Universal Serial Bus technology.
The USB 1.0 specification was released in 1996 and ran at 1.5 Mbit/s (low-speed) and 12 Mbit/s (full speed). While useful for lower data rate peripherals, it was not until the USB 2.0 specification (high-speed USB) was introduced in 2001 with a maximum raw data throughput of 480 Mbit/s (60 MByte/s) that the standard became useful for applications such as video and data storage. At a time when many industrial camera manufacturers defined themselves by the interface they supported, some companies saw a chance to differentiate themselves using this new technology. This led to the birth of the first USB 2.0 digital video cameras. Of course, like any interface, USB 2.0 has its strengths and limitations when applied to machine and computer vision. It provides sufficient bandwidth for many applications, has built-in support on virtually every computer, and is typically very low cost (no frame grabber required). Conversely, USB 2.0 bandwidth may not be enough for many of the new high data rate image sensors becoming available, provides limited power, and is not as efficient in its signaling and data transfers as some other digital interfaces.
USB 2.0 employs a host-directed (aka master-slave) architecture where every transaction either goes to or comes from the master, which is typically the host computer. Communication is half-duplex, allowing data to flow in only a single direction at a time. The host initiates all data transfers, handles all arbitration functions, and dictates data flow to, from and between the attached peripherals. This adds additional system overhead and can result in slower data flow control. In addition, USB 2.0 uses polling as the primary signaling method, meaning a USB 2.0-enabled camera must be constantly polled by the host software to check for activity. The USB 2.0 specification offers two different data-transfer mechanisms: bulk and isochronous. Bulk transfers are guaranteed delivery, but not bandwidth, and provide error correction in the form of a CRC field on the data payload. They also have error detection/re-transmission mechanisms that ensure data is transmitted and received without error. Bulk transfers have a theoretical maximum rate of approximately 40 MByte/s, which is more than sufficient to handle the resolutions and frame rates of the more common 1/4-inch, 1/3-inch or 1/2-inch image sensors on the market. Isochronous transfers, on the other hand, are guaranteed bandwidth, making this mechanism well-suited to the transmission of real-time data. Isochronous provides low-latency data distribution, enables the latency of that data to be deterministic, and provides error detection via a CRC. However, USB 2.0 isochronous transfers are limited to roughly 24 MByte/s.
Fast forward to 2008. The USB 3.0 Promoter Group, comprised of HP, Intel, Microsoft, NEC, ST-NXP Wireless, and TI, finishes development of the USB 3.0 specification and transitions its management to the USB-IF. The design goal was simple: build on the strengths of USB 2.0 while addressing many of its limitations. The USB 3.0 specification increases raw data throughput to 5 Gbit/s (640 MByte/s). Though 8b10b encoding sets a practical limit of about 500 MByte/s, this still represents a substantial performance improvement over USB 2.0. USB 3.0 adds five wires for a total of nine wires in the connectors and cabling and utilizes a unicast dual-simplex data interface that allows data to flow in two directions at the same time; an improvement over USB 2.0’s unidirectional communication model. The USB 3.0 specification preserves the legacy bulk, isochronous, control and interrupt transfer types, but significantly increases isochronous throughput with three bursts of 128 MByte/s per service interval, for a total of 384 MByte/s. The USB 3.0 architecture has many similarities to PCI Express (PCIe), and although there are obvious functional differences between them, they both aim to increase bandwidth and lower power consumption. USB 3.0 is still a hosted device protocol and maintains much of the existing USB 2.0 device model. One important change, however, is the signaling method. The USB 3.0 specification uses asynchronous signaling, which allows a device to notify the host when it is ready for data transfer. This significantly reduces system overhead and CPU usage compared to the polling mechanism in USB 2.0. A variety of other protocol improvements, such as streaming support for bulk transfers and a more efficient token/data/handshake sequence, are designed to improve system efficiency and reduce power consumption.
In addition to an improved architecture and higher bandwidth, USB 3.0 also provides more efficient power management and increased power delivery over USB 2.0. The amount of current draw for USB 3.0 devices operating in SuperSpeed mode is now 900 mA, resulting in an increase in total power delivery from 2.5 W to 4.5 W (at 5 V). USB 3.0 also offers an improved mechanism for entering and exiting low-power states, depending on whether a device is active or not, and eliminates continuous powerconsuming polling. Although the USB 3.0 cable contains five new wires, it is still backward-compatible with USB 2.0, allowing consumers to continue to utilize their existing peripherals with a USB 3.0-enabled computer, or USB 3.0 devices with a legacy computer (see Figure 1). USB 3.0 Standard-A receptacles are backward-compatible with USB 2.0 but add new pins for USB 3.0 signals. The new Standard-B and Micro-AB receptacles are also backward-compatible (see Figure 2).
An abundance of USB 3.0 devices are already available, ranging from motherboards and notebook PCs, to interface cards, cables, hubs and hard drives. Work on the silicon required to provide USB 3.0 connectivity for these devices has been in development for some time. Companies like NEC and Fresco Logic provide the required Extensible Host Controller Interface (xHCI) chips, while others, like Texas Instruments, provide the much-needed low - level physical layer (PHY) chips. USB 3.0 technology has also been showcased at a multitude of conferences and industry tradeshows , including SuperSpeed USB Developers Conferences, the Intel Developer Forum (IDF), and the Consumer Electronics Showcase (CES). At CES the USB-IF announced the first 17 consumer products, from some of the biggest electronics manufacturers in the world, that have passed USB 3.0 compliance and certification testing. Most of these products are already shipping, such as ASUS’s P6X58D motherboard, SIIG’s 2-port PCIe add-in card, and Western Digital’s My Book external hard drive.
Figure 3: Point Grey's USB 3.0 demo camera streams uncompressed 1920x1080 video at 60 FPS.
Though it is clear that consumer market acceptance of USB 3.0 will be ahead of the vision industry, USB 3.0 promises to open up new applications in machine and computer vision, as well as in non-industrial markets where USB 2.0 already has widespread acceptance. Rumors are already circulating that several key camera manufacturers, and not just those who currently provide USB 2.0, are planning to introduce USB 3.0 products. Point Grey Research, Inc., revealed the world’s first USB 3.0 digital video camera at the 2009 Intel Developer Forum (IDF) in San Francisco (see Figure 3). The high-definition demo camera was shown streaming 120 MByte/s of raw, uncompressed 1080p60 video, generated by a high-performance Sony CMOS image sensor, to a Fresco Logic host controller. Many other USB 3.0 products made their debut at CES in 2010, including the FireNEXuLINK from Newnex Technology (see Figure 4), the world’s first USB 3.0 active repeater capable of extending USB 3.0 signals up to 12 meters in length, and Total Phase Inc.’s Beagle USB 5000 SuperSpeed Protocol Analyzer, an affordable real-time bus monitor that captures and displays USB traffic and bus-states. Many of these devices will be used within the vision industry.
When compared to the existing lineup of digital interfaces, USB 3.0 sits at the intersection of high data rate, power and data transmission over a single cable, ease-of-use, and cost-effectiveness. The increased 500 MByte/s throughput and improved 4.5 W of power delivery is well-suited for many of the high-speed, multi-megapixel area scan image sensors on the market today, such as Kodak’s TRUESENSE 5.5 micron Interline Transfer CCD Platform, a series of quad-tap CCDs with resolutions from 1 mp at 120 fps to 8 mp. USB 3.0 is also nicely matched to several upcoming sensors, such as the CMOSIS CMV4000 and CMV2000 global shutter CMOS sensors, featuring 2K x 2K resolution at 170 fps and 2K x 1K at 340 FPS, respectively. A variety of new HD 1080p60-capable sensors, such as Sony’s future 2.8 mp 50 FPS quad-tap EXview HAD CCD, may also find their way into USB 3.0-enabled cameras.
Data rates for many new image sensors start in the 120 MByte/s range, which is beyond the limits of IEEE 1394b (aka FireWire-b) or GigE but well within the capabilities of USB 3.0 (see Figure 5). While Camera Link is still the bandwidth king at approximately 680 MByte/s with a full eight-tap configuration, many customers may choose to sacrifice some pixels or frames per second in exchange for the easier to use and more cost-effective USB 3.0 alternative. Like FireWire, the USB 3.0 specification provides power and data over a single cable; has guaranteed, truly isochronous bandwidth; and is well-matched to applications requiring small size and low cost. Though USB 3.0 is almost 10 times as fast as FireWire and GigE, FireWire provides more power (up to 45 W) and GigE will always win when it comes to cable length. FireWire also uses a more efficient peerto- peer network architecture, in which the peripherals are intelligent and can negotiate bus conflicts to determine which device can best control a data transfer.
There are also other practical factors to consider when evaluating USB 3.0 for vision applications. An important one is the control protocol implemented on the camera. FireWire cameras use the 1394-based Instrumentation and Industrial Digital Camera (IIDC) standard, which enables any IIDC-compliant camera to be used with any vision software package that also supports IIDC. GigE Vision is the common standard for many Gigabit Ethernet cameras and provides the same benefits as IIDC in terms of software compatibility. USB 2.0, on the other hand, has no such common protocol. The USB Video Class (UVC) is not appropriate for industrial digital cameras, leading some manufacturers to create their own proprietary camera control interface and others to use IIDC. The Automated Imaging Association (AIA), which historically has had no involvement in USB 2.0, announced at its January business conference a new USB 3.0 standard committee to evaluate appropriate protocols like IIDC and GenICam.
Another consideration is cable length. The maximum length is not explicitly specified in the USB 3.0 standard. However, the standard does describe the relationship between wire gauge and maximum length in order to achieve USB 3.0 voltage drop requirements. For example, a cable can be up to 5.3 meters long when using an American Stranded Wire Gauge (AWG) of 20. In the majority of cases, the host computer system is located within this distance. A variety of high-performance and cost-effective solutions will rapidly become available to address situations where it is not. For FireWire-based applications, for example, where the maximum recommended distance for FireWire cables is 4.5 meters, products that increase the distance between the camera and host system are now commonplace. These include optical repeaters for spans greater than 100 meters, longer cables up to 20 meters, and more recently, 1394- over-GigE technology. Similar USB 3.0 solutions will be introduced. USB 3.0 hubs and repeaters already are in production, and work on signal-corrected long-distance cables, equilizer technology like EqcoLogic’s EQCO5000, and long-haul optical solutions is in progress. Other USB 3.0 cable and connector products geared toward industrial and machine vision are under development, including screw-locking connectors, high-flex chain cables, and so on.
Obviously if one interface can cover all requirements, ranging from size of the camera, bandwidth, cable length, price, and so on, it would be the clear technical winner. But in reality customer requirements are too diverse to be met by just one interface. In fact it is often what the interface does not do that makes the decision, rather than what it does do. Customers want more pixels at faster frame rates, superior reliability, a migration path for future retrofits, and perhaps most importantly, lower overall system cost. CCD and CMOS sensor manufacturers are providing more high-quality multi-megapixel image sensors that can run at increasingly fast speeds. While there is no single digital interface that works best for all vision applications, on technical merits alone USB 3.0 will be a strong contender, and will certainly become an important camera interface in the years to come.
Michael Gibbons is Product Marketing Manager at Point Grey Research, Inc. (Richmond, BC), where he is responsible for marketing and product management. He has more than seven years of experience in the field of advanced digital imaging, and is a graduate of the University of British Columbia (Commerce) and the British Columbia Institute of Technology (Embedded Systems).