Road-Tested GMSL Cameras Drive into New Markets

作者:Pete Bartolik

投稿人:DigiKey 北美编辑

Technologies developed for automotive applications frequently transfer to other markets due to automobile manufacturers' rigorous requirements for reliability, performance, and the need for fast data rates in an electronically hostile environment. That's why Gigabit Multimedia Serial Link (GMSL™) cameras are finding ready markets for vision applications in areas such as automation and robotics, smart agriculture, digital healthcare, avionics, robotaxis, and retail and warehouse inventory management.

Initially introduced for addressing applications for high-speed video and data transmission in vehicles, Analog Devices GMSL is a widely adopted and proven technology for bringing new levels of performance to high-speed video links and enabling multi-streaming over a single cable.

Vision applications require very large data streams to ensure high-quality video. A full HD image is comprised of 1080 rows by 1920 columns. That amounts to 2 million pixels, each of which consists of a red, green, and blue element, resulting in 6 million elements. Each element represents 8 bits of data, so every frame results in nearly 50 Mbps of data. At 60 frames per second, the required data rate for one camera is over three-and-a-half Gbps.

First-generation GMSL, first available in 2008, utilized the low-voltage differential signaling (LVDS) standard to deliver parallel data downlink rates up to 3.125 Gbps. That was particularly suited for conveying data from multiple camera systems and other advanced driver assistance applications (ADAS), as well as the growing use of in-car, high-definition flat panel displays.

A second generation, GMSL2, was introduced in 2018, increasing data rates up to 6 Gbps and supporting more standard highspeed video interfaces, including HDMI and the MIPI interface standard, a popular image sensor interface for consumer and automotive cameras. These advances accommodated full high definition (FHD) displays and cameras with resolution up to 8 MP.

GMSL3, the next generation, can deliver data up to 12 Gbps over a single cable, supports multiple 4K resolution streams, the daisy-chaining of multiple displays, and aggregation of multiple cameras such as those located on the front, back, and sides of a vehicle to provide a 360° viewing capability. Today, increasing numbers of automobile manufacturers supplement rear and side-view mirrors with cameras, utilize forward and rear-facing cameras for collision avoidance, and internal cabin cameras for monitoring driver and passenger safety. GMSL3 can aggregate data from multiple video feeds as well as LiDAR and radar.

With cameras scaled down to the level of CMOS sensors, they can produce what once was considered incredible quality at low cost and with low power demands. Image sensors have millions of receptor elements, each of which converts measurements into digital values to be streamed via serial data lanes of a parallel interface, along with synchronization information.

Both GMSL2 and GMSL3 utilize MIPI interface standards that provide designers and vendors access to a wide range of image sensors for GMSL cameras.

GMSL versus GigE

Engineers starting out on vision applications will no doubt quickly face a decision on whether to use GMSL or gigabit Ethernet (GigE) vision technology. GigE is widely used in industrial applications due largely to its reliance on Ethernet network infrastructure and standards.

GigE Vision cameras with 2.5 GigE, 5 GigE, and 10 GigE are commonplace in applications today, and 100 GigE state-of-the-art cameras can utilize up to a 100 Gbps data rate. GMSL is designed to transmit data over coaxial cable or shielded twisted pair cable at up to 15 meters, compared to 100 m for GigE, although both may be exceeded under certain conditions.

Each technology is capable of transmitting data and power through the same cable: GMSL uses Power over Coax (PoC) so video, audio, control, data, and power can be transported on a single channel. Most GigE Vision applications rely on Power over Ethernet (PoE) for 4-pair Ethernet, or less commonly, Power over Data Line (PoDL) for Single-Pair Ethernet (SPE).

System requirements and application needs will determine which vision technology is most appropriate. GigE Vision, for example, may offer some advantages for single-camera applications, particularly where they connect directly to a PC or an embedded platform with an Ethernet port.

When using multiple cameras, GigE Vision applications will require use of a dedicated Ethernet switch, a network interface card (NIC) with multiple Ethernet ports, or an Ethernet switch IC. That switching requirement can potentially reduce the maximum total data rate and introduce unpredictable latency between the cameras and the terminal device, whereas GMSL provides a simpler, more direct architecture.

GigE Vision devices may support higher resolution and a higher frame rate—or both simultaneously—with additional buffering and compression. Frame buffering and processing are not provided by GMSL devices, so resolution and frame rate depend on what the image sensor can support within the link bandwidth. Engineers will need to determine a simple trade-off between resolution, frame rate, and pixel bit depth.

GMSL simplifies high-speed video architecture

GigE Vision cameras typically utilize a signal chain that includes an image sensor, a processor, and an Ethernet physical layer (PHY) (Figure 1). Raw image data from the sensor is converted by the processor into Ethernet frames, often relying on compression or frame buffering to fit the data rate of the supported Ethernet bandwidth.

Diagram of key signal chain components on the sensor side of Analog Devices GigE Vision camerasFigure 1: Representation of key signal chain components on the sensor side of GigE Vision cameras. (Image source: Analog Devices, Inc.)

The GMSL camera signal chain utilizes a serializer/deserializer (SerDes) architecture that avoids the use of a processor (Figure 2). Instead, image sensor parallel data is converted by the serializer into a high-speed serial data stream. On the back end, a deserializer converts the serial data back into parallel form for processing by an electronic control unit (ECU) system-on-chip (SoC).

Diagram of Analog Devices GMSL cameras utilize a simpler signal chain architectureFigure 2: GMSL cameras utilize a simpler signal chain architecture on the sensor side than GigE Vision. (Image source: Analog Devices, Inc.)

The GMSL camera architecture makes it simpler to design small form factor cameras with low power consumption. Serializers can directly connect to cameras through standard MIPI CSI-2 interface and transmit packetized data through the GMSL link.

A typical host device is a customized embedded platform with one or more deserializers that transmit image data through MIPI transmitters in the same format as the image sensor MIPI output. New GMSL camera drivers are required for customized designs, but if there is an existing driver for the image sensor, it can be utilized with just a few profile registers, or register writes to enable a video stream from cameras to a control unit.

GMSL components

ADI offers a comprehensive portfolio of serializers and deserializers to support a variety of interfaces. These feature robust PHY designs, low bit error rates (BER), and backward compatibility. Any video protocols can be bridged together—for example, HDMI to the Open LVDS Display Interface (oLDI).

Engineers will need to select the best components based on application needs, such as device interfaces, data rates, bandwidth, power consumption, environmental conditions, and cable length. Other factors include EMI, error handling, and signal integrity. Some examples of ADI's GMSL components include:

  • MAX96717, a CSI-2 to GMSL2 serializer (Figure 3), operates at a fixed rate of 3 Gbps or 6 Gbps in the forward direction and 187.5 Mbps in the reverse direction.

Schematic illustrating the data stream utilizing Analog Devices MAX96717 serializersFigure 3: A schematic illustrating the data stream utilizing MAX96717 serializers. (Image source: Analog Devices, Inc.)

  • MAX96716A, which converts dual GMSL2 serial inputs to MIPI CSI-2. The GMSL2 inputs operate independently and video data from both can be aggregated for output on a single CSI-2 port or replicated on a second port for redundancy.
  • The MAX96724, a quad tunneling deserializer, converts four GMSL 2/1 inputs to 2 MIPI D-PHY or C-PHY outputs. Data link rates are 6/3 Gbps for GMSL2 and 3.12 Gbps for GMSL1, and reverse link rates of 187.5 Mbps for GMSL2 and 1 Mbps for GMSL1.
  • The MAX96714 deserializer converts a single GMSL 2/1 input to MIPI CSI-2 output, with a fixed rate of 3 Gbps or 6 Gbps in the forward direction and 187.5 Mbps in the reverse direction.
  • The MAX96751 is a GMSL2 serializer with HDMI 2.0 input that converts HDMI to single or dual GMSL2 serial protocol. It also enables full-duplex, single-wire transmission of video and bidirectional data.
  • The MAX9295D converts single- or dual-port 4-lane MIPI CSI-2 data streams to GMSL2 or GMSL1.

ADI also offers several development tools, such as the MAX96724-BAK-EVK# evaluation kit for the MAX96724 devices.

Conclusion

With their reduced complexity, GMSL cameras are more compact and generally able to provide a more cost-effective solution compared to GigE Vision. GMSL provides reliable transport of high-resolution digital video with microsecond latency for a growing range of camera and display-based applications, from machine learning and autonomous operations to infotainment and safety. Millions of GMSL links are enhancing the driver experience on the road today, attesting to their reliability and performance.

免责声明:各个作者和/或论坛参与者在本网站发表的观点、看法和意见不代表 DigiKey 的观点、看法和意见,也不代表 DigiKey 官方政策。

关于此作者

Pete Bartolik

Pete Bartolik 是一名自由撰稿人,二十多年来一直从事有关 IT 和 OT 问题及产品的研究和写作。他曾任 IT 管理刊物《计算机世界》的新闻编辑、一家终端用户计算机月刊的主编和一家日报的记者。

关于此出版商

DigiKey 北美编辑