USB 3.0 in surveillance and machine vision cameras
USB 3.0 is becoming one of the most promising interfaces for camera-based commercial applications as well as industrial and medical imaging. All of the leading camera manufacturers in the vision and imaging market, including Basler, Ximea, IDS Imaging, and Lumenera, have introduced USB 3.0-based cameras to their portfolio. With this growing trend of USB 3.0 adoption as a primary interface for HD cameras, it is clear that USB 3.0 is here to stay.
Camera manufacturers must find the optimal balance between performance, reliability and costs when choosing an interface. From a technological viewpoint, USB 3.0 is compelling for a number of reasons. Its transmission bandwidth is up to 350 MB/s, and USB 3.0 is completely "plug & play" ready. In addition, the real-time capabilities (low CPU load, signal latency, and jitter) and speed make USB 3.0 an attractive option to replace conventional, and slower, interfaces in the vision market, including FireWire, USB 2.0, Gigabit Ethernet and Camera Link.
Each of these interfaces offer unique capabilities but none is one-size-fit-all due to trade-offs in bandwidth, footprint, power, and cost. The following section summarizes the pros and cons of each of these interfaces, and how USB 3.0 helps to overcome these limitations.
Types of camera interfaces
FireWire: FireWire (also known as IEEE 1394) has been a very successful camera interface for many years. This success can attributed to a well-defined standard with DCAM, and ease of interoperability between software and device hardware. Though a wide variety of cameras are available in the market with a FireWire interface, its market share is shrinking rapidly because of the following disadvantages
- Bandwidth: 64MB/s as per IEEE 1394b is a serious limitation in the vision market as this drastically limits resolution and frame rate.
- Cable Length: The official cable length is also very short at just 4.5m
- Decreasing hardware and software support: Decreasing availability of hardware and increasing prices makes it difficult to source parts at a low cost.
Camera Link: Since its inception, the Camera Link interface has been popular in industrial camera applications due to its extremely high bandwidth and data security. In Full configuration mode, Camera Link supports data rates up to 850MB/s. However, all the components of a complete Camera Link solution, including cables, connectors, and frame grabbers, must adhere to this specific standard. This limits its use in applications other than image processing, keeping volumes low and making it an expensive option.
- The biggest disadvantage of Camera Link is cost due to the need for specialized and expensive frame grabbers and cables.
USB 2.0: USB 2.0 is the most ubiquitous and versatile interface standard in the computing and consumer markets. The universal adaptability of USB, along with its Plug and Play readiness, is definitely an advantage of USB 2.0 in the vision market. However, USB 2.0 falls short on bandwidth to support HD.
- Bandwidth is only 40MB/s, limiting its use in HD camera applications.
- Short cable length.
Gigabit Ethernet: Gigabit Ethernet has overcome many disadvantages of previous generations of interfaces. Bandwidth goes up to 100MB/s. It also supports the longest cable length at 100m. Gigabit Ethernet is also a universally accepted interface and doesn’t require frame grabbers. The interface is also appropriate for multi-camera applications. However, the primary disadvantage of this interface is that CPU loading must be optimized.
- CPU loading needs to be optimized.
- Bandwidth is limited to 100MB/s, limiting its use for Full HD or ultra HD image capturing
USB 3.0 overcomes the limitations of all the other interfaces by providing:
- Higher bandwidth (up to 350 MB/s) than USB 2.0, IEEE 1394b, and Gigabit Ethernet.
- Power and data are provided over a single cable at lengths up to 8m (passive).
- Lower implementation cost than Camera Link.
- Plug-and-play with easier to set up than Gigabit Ethernet.
Cameras used in surveillance and machine vision applications have an acute need for interfaces providing greater bandwidth. Greater bandwidth translates to higher sensor resolution, faster frame rates, and richer color depth to meet the demand for HD quality imaging. Table 1 shows how bandwidth requirements increase multifold across the different HD specifications.
Bandwidth = # pixels per frame (resolution) x frames per second (frame rate) x # color bits per pixel (color depth)
Slower interfaces like USB 2.0, Fire-wire, and even Gigabit Ethernet are falling short of meeting customer bandwidth requirements. As such, they pose a serious bottleneck for all camera manufacturers.
With these slower interfaces, the customer is left with the choice of either picking a lower resolution sensor or reducing the frame rate. For example, consider a surveillance camera using Gigabit Ethernet providing VGA resolution at 120 fps. If the manufacture wants to upgrade to 5 M-Pixel for next-generation designs, and then the frame rate would drop to meager 5-10 frames per second due to the available bandwidth limitations.
Alternatively, the manufacturer can use compression techniques to increase frame rate. Compression reduces the bandwidth requirements of an image using a mathematical algorithm. Modern compression algorithms are designed to progressively remove detail from an image to reduce image size to match the available bit rate.
Unlike commercial applications, where optimized compressions go unnoticed to the human eye, machine vision applications perform precise computations on captured images using advanced image analytics. Thus, it is crucial to capture raw data containing as much detail of the image as possible. In fact, compression algorithms may remove the very details that the analytics algorithms need access to.
Furthermore, compressed video is often displayed using software decoders within the host to decompress the video. This decoding can take significant time, causing extra latency while playing it. For real-time video applications like video conferencing and surveillance, such latency may be unacceptable.
Moreover, the manufacturer has to employ additional compression engines on their boards. These engines comprise compression ASICs with additional DRAM frame buffers. These additional components result in a higher Bill-of-Material (BOM) cost as well as require a larger PCB footprint. With the current trend of machine vision and industrial cameras getting smaller, camera manufacturers actually need to reduce the PCB footprint, not increase it.
With USB 3.0, designers now have additional bandwidth at their disposal. USB 3.0 is fast with a 5 Gbps data rate, a 10x increase over USB 2.0 (480 Mbps). After 8b/10b encoding, USB 3.0 has 350 MB/s available bandwidth for data. This eliminates the need for compression while providing sufficient bandwidth to support HD video without compromising resolution or frame rate.
USB 3.0 can support up to 8m of passive cable, compared to 5m for USB 2.0 and 4.5m for FireWire. It is almost comparable to Camera Link in terms of cable length (10m) but much shorter than Gig E (100m). Additionally, a USB 3.0 cable is capable of delivering 4.5W of power, enough to power a machine vision camera or surveillance camera without an additional power supply. This reduces installation complexity and cost.
Plug-and-play ready with vision standard
The USB Video Class (UVC) standard provides a readily available USB driver for video camera software applications. This driver reads data from the image sensor and sends control information to the image sensor controller. It is plug-and-play compatible with all personal computers and is commonly used in video capture applications like PC webcams.
However, the UVC class restricts images to the YUV format, thereby limiting the choice of image sensors and control signal options that can be used. To address this, the Automated Imaging Association (AIA) has developed a new ‘USB3 Vision’ standard. This standard supports image sensors that stream non-YUV images, enables more camera control feature, and allows custom driver implementation.
The USB3 Vision standard also allows re-use of as many blocks as possible from existing standards like Gigabit Ethernet Vision and CoaXPress. This significantly reduces design cycles otherwise needed for development of custom drivers to support more sensors and controls to make fit-to-purpose solutions for exact application needs. More details on designing for UVC are available in the application note from Cypress Semiconductor.
Camera interfaces that don’t require a frame grabber like Gigabit Ethernet, USB 3.0, and Firewire have differences in CPU loading. For the same image size transfer, there is lower CPU loading with FireWire and USB 3.0 because both interfaces work with DMA (Direct Memory Access) to speed memory transfers.
CPU loading with Gigabit Ethernet is considerably higher because Gigabit Ethernet requires copy processing. Consider a video stream with a data rate of 85MB/s. CPU loading for USB 3.0 would be around 1%, while Gigabit Ethernet would require around 6% CPU loading. Thus, camera implementations using Gigabit Ethernet add overhead for CPU loading optimization on the designer, requiring longer design cycle time.
USB 3.0, also popularly known as Super-speed USB, overcomes the limitations of traditional camera interfaces by providing six (over IEEE 1394b) to nine (over USB 2.0) times higher bandwidth, better error management, higher power supply, longer cable lengths, and lower latency and jitter times. USB 3.0 has also become a standard in the consumer market, with a lot of hardware supporting native USB 3.0, keeping volumes high and costs low. These advantages have made the USB 3.0 interface a de facto choice for cameras manufacturer’s in a relatively short period.