The GigE Vision interface was developed a few years ago based on the technical properties of CCD and CMOS sensors at the time and represented a milestone for image processors: Being able to transfer up to 100 MByte of image data per second via Ethernet was by far sufficient for many applications at the time.

Today, the importance of CCD-based cameras is declining and multi-channel sensors with data rates in excess of 100 Mbyte/s are no longer a rarity in the CMOS sector. This means that the Ethernet interface often represents a bottleneck in data transmission.

The Canadian image processing provider Teledyne Dalsa has tackled this problem and introduced the 'Turbodrive' technology. It is based on a special type of data coding and enables the transmission of camera information at a speed that exceeds the actual limits of GigE Vision. In typical image processing applications, Turbodrive enables a throughput increase of between 120 and 235% without losing the data transmission security of GigE Vision.

... Histogram represents the pixel distribution. For each occurrence of a certain value in the image, the histogram column for this value increases by 1.

© Stemmer Imaging

As a rule, image processing cameras use an absolute coding of 8 to 16 bits to transmit image information. For example, with 8-bit coding, each pixel can have a value from 0 (black) to 255 (white). Turbodrive, on the other hand, is based on localized relative coding and examines each pixel in its context before it is coded. This results in a more compact coding of the pixel information: The identical information can be packed into fewer bits, reducing the volume of data to be transmitted.

The pixel distribution in an image can be displayed using a histogram. For each occurrence of a certain value in the image, the histogram column for this value increases by 1. In a uniform image, the intensity of all pixels is the same, which is expressed by a single peak value in the associated histogram. The entropy of this image is therefore 0. To describe it completely, only the common value of all pixels needs to be known. The coding of this image therefore only requires a few bits of information.

In reality, images are usually not that simple, but the example shows the basic idea: the lower the entropy value of an image, the more effective the achievable compression.

Relative instead of absolute coding

In order to further reduce the number of bits required to encode the pixel information without loss of information, Turbodrive also takes into account the so-called neighborhood effect. The 'neighborhood' of a pixel is all the surrounding pixels of the pixel under consideration. For most pixels, there is a small pixel-to-pixel deviation to the neighbor and thus a high redundancy. It is therefore possible to use the information of the neighboring pixels to encode the reference pixel even more efficiently.

Due to the use of image uniformity, the method uses localized relative coding rather than absolute coding. If neighboring pixels have a higher correlation, this is more efficient. The result is then used as an input value for the entropy step to further reduce the coding size of an image. This ensures a compact representation that retains all the information of the original image.

Typical industrial cameras encode the pixel information using absolute encoding. Each pixel is completely described by itself and without any additional information: The numerical value of a pixel represents its intensity. This has the advantage that the receiver can simply skip over faulty pixels in the event of a faulty transmission. The disadvantage is that this type of coding requires more bits than the principle of image entropy.

Stable with bit errors

Although absolute coding leads to more data volume, it is the most frequently used method in industrial image processing. The CameraLink interface technology, which is very popular on the market due to its fast data throughput of up to 850 MByte/s, also works with absolute coding. What is less well known is that CameraLink offers no stability if bit errors occur. If a bit is damaged during transmission, the frame grabber cannot detect this or notify the application. The result: the affected pixel takes on an incorrect value. CameraLink offers neither checksum nor data retransmission or forward error correction. The current version of CoaXPress is also limited to error detection without guaranteeing image transfer stability.

Reliable transmission channels are essential due to the dependency on the adjacent pixels. If a channel is not reliable, a transmission error in a pixel will spread to its neighbors, creating a cluster of incorrect values. With GigE Vision, USB3 Vision and CameraLink HS, every transmission error is managed on the transmission layer so that the decoding engine always receives an error-free digital signal.

Faster than GigE Vision

However, the principles described are not sufficient to exceed the maximum throughput of the camera interface. Most image processing cameras were developed to receive images at a frame rate that is below the capacity of the transmission path. The image reception is therefore not decoupled from the image transmission. This principle originates from analog and CameraLink cameras.

To fully utilize the capabilities of Turbodrive, the camera used must operate at a speed above the nominal transmission speed when absolute encoding is applied. This is known as 'burst mode'. The camera can then use the localized relative coding scheme to compress additional information on the transmission path. This speeds up the reception and transmission of images. A GigE Vision camera can achieve a speed of more than 115 MByte/s in this way, as each pixel requires less than 8 bits during encoding.

To achieve this, the camera must have integrated buffers that collect the image data. This allows deviations in the coding stage to be compensated for: Buffers correct coding deviations to achieve a good average transmission speed that is within the limits of the maximum throughput of the camera interface. In addition, the camera can use dead times between images to continue transmission and empty the internal buffers. This keeps the transmission path busy. If sufficient buffers are available, the goal is an average throughput after relative encoding, including dead times, that corresponds to the maximum transmission speed of the camera interface.

The first camera series from Teledyne Dalsa to feature Turbodrive technology is the 'Linea GigE'. This camera can exceed the typical throughput of 115 MByte/s for this product class. The 'Linea Mono 4K GigE', for example, is limited to 26 kHz due to the Gigabit Ethernet connection speed. By enabling data compression and taking into account the dead time between virtual images, the line rate for scenes with low image entropy can reach 80 kHz. This is the same line rate offered by the 'Linea CameraLink' model.

However, there are additional benefits from using the Ethernet interface: the performance of existing Ethernet-based image processing systems can be easily increased with the 'Linea' or 'Genie Nano' cameras. For users, this means that they can achieve higher frame rates and shorter cycle times. In addition, the cost-effective Ethernet structure can continue to be used, long cables can be used and systems can be redesigned quickly, thus achieving a faster time to market. As no image acquisition card is required, system costs are also reduced.

Another application of Turbodrive technology can be found in multi-camera systems. With the help of an Ethernet switch, it is possible to combine image streams from several cameras in a single network card. The aggregated throughput of these cameras after encoding must not exceed the maximum connection speed of 115 MByte/s for GigE Vision. In some vision systems, this could be more cost-effective than using multiple network cards.

The current software version of Turbodrive is available to users of the Common Vision Blox (CVB) programming library from Stemmer Imaging.

Author:
Peter Stiefenhöfer is Head of Marketing & Public Relations at Stemmer Imaging in Puchheim.