Use GMSL to Reliably Meet Industrial and Automotive High-Bandwidth Video Requirements

Industrial and automotive applications increasingly depend on high-resolution imaging systems that must deliver real-time, high-bandwidth video data reliably and efficiently. While GigE Vision is well understood and widely used, the demands of newer applications are prompting a search for alternatives. Gigabit Multimedia Serial Link (GMSL) technology is one such alternative, offering support for multiple cameras, stringent real-time processing, reduced complexity, determinism, low power consumption, and a compact form factor.

This article provides a brief overview of the key differences between GigE Vision and GMSL. It then introduces GMSL solutions from Analog Devices and shows how they can be used to significantly reduce system complexity, enhance reliability, and enable efficient real-time video transmission.

How camera interface technology impacts performance

Different interface technologies offer solutions for extending the distance between camera sensors and the host processor to meet the basic requirements of many imaging applications. Based on Gigabit Ethernet (GbE) technology, the GigE Vision camera interface standard has earned widespread adoption. GigE Vision cameras typically rely on a signal chain comprising three main components: an image sensor, a processor, and an Ethernet physical layer (PHY) interface (Figure 1).

Figure 1: Ethernet cameras use a processor-based signal chain that buffers and processes image sensor data before transmission. (Image source: Analog Devices, Inc.)

On the sensor side, GigE Vision cameras can use their internal processor to support custom sensor interface protocols. On the transmission side, by using standard Ethernet, GigE Vision cameras provide compatibility with a wide range of host devices. For example, personal computers and embedded systems typically include a GbE port as a standard interface. If the GigE Vision camera supports a universal driver, which is generally available with these systems, it works as just another plug-and-play peripheral.

Ethernet-based solutions can be advantageous for single-camera applications but require additional hardware for use in multiple-camera applications. Typically, these applications require an additional dedicated Ethernet switch or network interface card (NIC) to handle the multiple data streams. The inclusion of these devices in the video data path can potentially compromise throughput and latency between the cameras and the host.

Alternatively, Analog Devices’ GMSL technology employs a point-to-point serial link approach, offering an efficient solution for applications that require multiple cameras with minimal latency. Originally designed for automotive applications, GMSL cameras are being increasingly adopted outside of the automotive space as an alternative to Ethernet-based cameras.

In a GMSL-based application, multiple compact GMSL cameras can connect to a single GMSL host without compromising throughput or latency, provided the host system on chip (SoC) supports the full bandwidth of all the cameras (Figure 2).

Figure 2: GMSL multi-camera applications use simple cameras (left) with individual GMSL links that converge on a single host (right). (Image source: Analog Devices, Inc.)

Cameras using GMSL typically employ a simplified signal chain comprising an image sensor and a GMSL serializer. GMSL serializers support two standard sensor interfaces:

• First-generation GMSL (GMSL1) devices support the parallel low-voltage differential signaling (LVDS) interface.
• Second-generation GMSL (GMSL2) and third-generation GMSL (GMSL3) devices support the popular Mobile Industry Processor Interface (MIPI) standard, enabling the use of a wide range of leading image sensors in GMSL cameras.

In most applications, raw data from the image sensor is serialized and sent over a GMSL link in its original format. By eliminating the need for a processor and other support components, GMSL cameras are simpler to design and manufacture. They also offer a more effective solution for applications that require a compact camera form factor and low power consumption.

The host for a GMSL link is typically a custom embedded system that combines one or more hardware deserializers. A few lines of code on running on the host are generally sufficient to access these hardware deserializers and acquire data. In cases where a driver exists for the image sensor, developers only need to set appropriate registers to read the video stream from the camera. Analog Devices’ GMSL device evaluation kits include the software required to access these devices and explore their capabilities. For additional GMSL development support, Analog Devices provides an open-source software repository for GMSL technology.

Tackling multi-camera application configurations

GMSL’s performance advantages arise from the way this technology handles the transmission of a video stream (Figure 3).

Figure 3: Upon image sensor exposure and readout (top), a GMSL camera serializes and transmits packets of raw video data before entering an idle state until the next frame (middle); a GigE Vision camera buffers, processes, and transmits data in Ethernet frames before it enters an idle state (bottom). (Image source: Analog Devices, Inc.)

For each video frame, a global shutter image sensor reads out data immediately after the exposure period and then enters an idle state until the next frame (Figure 3, top).

When the camera readout period begins, GMSL and GigE Vision cameras handle data transmission differently. In GMSL cameras, the GMSL serializer immediately serializes and transmits the image sensor data, then returns to an idle state until the next readout period (Figure 3, middle).

In GigE Vision cameras, the processor buffers and often processes the data before building and transmitting Ethernet frames (Figure 3, bottom).

Understanding the factors underlying video system performance

In practice, the performance of a camera system depends on multiple factors, including some of these key characteristics:

Link rate: In both GMSL and Ethernet-based cameras, the maximum data transmission speed, or link rate, varies by camera type; however, each type of interface technology relies on a set of fixed link rates. Ethernet-based GigE Vision cameras adhere to Ethernet standards for link rates, which are specified in a series of discrete steps, ranging from 1 gigabit per second (Gbit/s) for GigE Vision cameras to 100 Gbits/s for state-of-the-art 100 GigE Vision cameras.

The link rates for GMSL vary with the technology’s generation. GMSL1 supports 1.74 and 3.125 Gbit/s serial-to-deserializer link rates, while GMSL2 and GMSL3 support 6 and 12 Gbits/s, respectively.

Effective data rate: In any data communications application, the effective data rate describes the data rate capacity, excluding the protocol overhead. This concept also applies to video data communications, where the effective amount of video data being transferred is equal to the pixel bit depth × pixel count in the payload of a packet or frame.

GMSL cameras transmit video data in packets. The use of fixed packet sizes in GMSL2 and GMSL3 devices results in a well-defined effective data rate. For example, when GMSL2 devices use a 6 Gbit/s link, the recommended video bandwidth is no more than 5.2 Gbits/s. Since the link also includes protocol overhead and blanking intervals from the sensor’s MIPI interface, the effective data rate of 5.2 Gbits/s represents aggregated data from all input MIPI data lanes, rather than purely video data.

As with other Ethernet-based devices, GigE Vision cameras transmit video data in frames, using a frame length optimized for the specific application. Longer frames improve efficiency, while shorter frames reduce delay. The use of higher-speed Ethernet helps mitigate the risks associated with using long frames to achieve a better effective video data rate.

Both GMSL and Ethernet-based technologies exhibit bursty transmission patterns. The burst time for GMSL cameras depends solely on the video sensor’s readout time, so the burst ratio (burst time/frame period) in real applications can potentially reach 100% to support its full effective video data rate. In a GigE Vision camera system, the burst ratio is often low to avoid collisions between video data and other data typically present in an Ethernet-based network environment (Figure 4).

Figure 4: A GMSL camera’s video data burst can occupy a full video frame period (top), while an Ethernet-based camera’s data burst shares the network with bursts of data from other sources (bottom). (Image source: Analog Devices, Inc.)

Resolution and frame rate: Both GMSL and Ethernet-based cameras exhibit tradeoffs in resolution and frame rate, which are two of the most critical specifications for video cameras and key drivers of higher link rates.

As noted above, GMSL devices do not include frame buffering or processing capabilities. Consequently, resolution and frame rate in these cameras depend solely on what the image sensor or its internal image sensor processor (ISP) can support within the link bandwidth. Typically, performance in these systems is a straightforward exchange between resolution, frame rate, and pixel bit depth.

GigE Vision cameras present a more complex performance model stemming from their internal buffering and processing capabilities. These cameras may feature a slower usable link rate than GMSL cameras, but they might also support higher resolutions, higher frame rates, or both, with additional buffering and compression.

Latency: In both automotive and industrial applications, reliable system operation and user safety depend on the ability to acquire and process video stream data in real time with minimal and deterministic latency.

In Ethernet-based cameras, internal buffering and processing capabilities that support higher resolution and frame rates can degrade latency performance and deterministic response. With these cameras, however, system-level latency may not always be longer, as the cameras’ internal processing capabilities can result in a more efficient system image pipeline.

Latency in GMSL cameras is simpler to analyze. GMSL camera systems have a short signal chain from the image sensor output to the receiving SoC input (see Figure 2 again). Because this signal chain simply conveys raw video data from a serializer on the sensor side to a deserializer on the receiving side, the latency of the video data remains minimal and deterministic.

How additional GMSL technology capabilities enhance applications

Transmission distance: GMSL serializers and deserializers are usually designed to transmit data up to 15 meters (m) using coaxial cables in passenger vehicles. In practice, transmission distances can exceed 15 m, provided the camera hardware complies with the GMSL Channel Specification.¹ Advanced GMSL devices, such as Analog Devices’ MAX9295DGTM/VY+T GMSL serializer and MAX96716AGTM-VY GMSL deserializer, employ adaptive equalization capabilities. This enables coaxial cable lengths beyond 15 m.

Power over Coax (PoC): GMSL technology supports the transmission of power and data over the same cable. This PoC capability is typically used by default in camera applications using coaxial cable and requires only a few passive components to complete a PoC circuit. In this configuration, power and data run on a single wire in the link.

Peripheral control and system connectivity: GMSL technology is designed to support dedicated camera or display links, rather than a wide variety of peripheral devices; however, GMSL devices often provide connectivity support for standard interfaces. For example, Analog Devices’ MAX9295DGTM/VY+T and MAX96716AGTM-VY support tunneling or pass-through operation of multiple standard interfaces, including general-purpose input/output (GPIO), inter-integrated circuit (I2C), and serial peripheral interface (SPI) interfaces. For large applications that employ GMSL cameras, developers typically use lower-speed interfaces such as a Controller Area Network (CAN) bus to exchange control signals or other data.

Camera triggering and synchronization: With GMSL devices, GPIO and I2C tunneling occur within a few microseconds in both forward and reverse channels. This capability enables triggers to originate from either the image sensor on the serializer side or the SoC on the deserializer side, supporting a range of low-latency triggering and synchronization requirements.

Conclusion

While GigE Vision has a well-deserved place in industrial and automotive imaging, GMSL technology offers a robust solution for applications that require minimal latency, low complexity, compact form factors, and determinism. Built with Analog Devices GMSL serializers and deserializers, GMSL-based camera systems enable streamlined designs that simplify multi-camera applications while maintaining the performance required in demanding real-time environments.

Reference

1. GMSL2 Channel Specification User Guide