Keeping Pace with In-Vehicle Data Speed
Higher speed and bandwidth for in-vehicle networks, switches, and connections that carry data demand new development and testing solutions.
Next-generation advanced driver-assistance systems (ADAS) require camera and radar systems with increasingly high resolution. That means more speed and higher bandwidth for networks, switches and the connections that carry the data. Date rates greater than 1 Gbps over existing cabling infrastructures, with lower latency networking, will play a critical role in addressing time-sensitive and complex automotive technologies to come.
High-bandwidth technologies also are important for keeping weight down to maximize fuel (or battery) efficiency. Many of these requirements can be met by automotive Ethernet with a bandwidth of up to 10 Gbps. But with some cameras requiring up to 3,500 Mbps, we must also consider another technology to move that data around.
To better understand bandwidth requirements, remember that the approximate bit rate of a video stream can be calculated as: Frame Size = Resolution x Color Depth, and Bit Rate = Frame Size x Frame rate. So, for an ADAS camera capturing a 1080p image, with a color depth of 24-bits and transmitting at 30fps, the bit rate to be supported equals: Frame Size = 1920 x 1080 x 24 = 49,766,400; and Bit Rate = 49,766,400 x 30 = 1,493 Mbps.
The following table shows typical volumes of data from the different sensors involved in autonomous driving:
Zonal architecture and SerDesNetwork architects need to understand how the vehicle will be required to scale as the technology improves. The current expectation is that vehicles remain in use for 10 to 15 years. If cost-effective interconnect solutions can provide support for additional bandwidth, they may need to consider designing them today to allow for added ADAS/AV-heavy features customers will want during the life of the vehicle, particularly as safety becomes an important consideration.
Reducing complexity is an ongoing focus in in-vehicle networking. An abstract version of a vehicle using many different data rates in the backplane is illustrated in Figure 1. It is an oversimplification, but it helps us imagine how some of these technologies and standards work together.
A zonal architecture will pull together multiple inputs and ultimately lower the complexity, cost and weight of the wiring harness, transitioning from a “many to one” architecture to a daisy-chained, one-to-one architecture. This is an example of a zone-based architecture, while others are considering a domain-based architecture. Both will aggregate camera and sensor data, where Ethernet acts as an interconnect between each zone or domain. Because a central computing complex is linked to the sensors and devices through networked zonal gateways, a zonal approach can provide better scalability as well as improved reliability and functionality.
In today’s infotainment systems, it is common for in-vehicle cameras and displays to be connected to the image-processing electronic control unit (ECU) via a SerDes (serializer/deserializer) connection. Today, they are delivered by individual vendors using closed, proprietary standards. Extending the reach of feature-rich SerDes links can require operating at lower Baud rates and higher order modulations (e.g. PAM-4). In addition, it will require higher bandwidth Ethernet links as primary interconnects between zones perhaps with 802.3ch supports up to 10 Gbps throughput.
Emerging SerDes standards like mobile industry processor interface (MIPI) A-PHY (MIPI A-PHY is a physical layer specification targeted for ADAS/ADS surround sensor applications and infotainment display applications in automotive), and Automotive SerDes Alliance (ASA) will be implemented by multiple silicon vendors. This will create a competitive market that acts to drive down the cost while delivering application-specific features.
There is also a desire to have standardized test methods throughout the ecosystem that establish interoperability requirements. For implementers and test vendors, this would unify requirements for silicon, tier ones and OEMs. Unified test requirements allow the suppliers and OEMs to accelerate their development cycle, lower costs and improve interoperability with other commercial devices.
Some of the features of next-gen SerDes support the Service Oriented Architecture of the future with protocol tunneling and adaptation, which will enable emerging SerDes standards to forward legacy automotive protocols along daisy-chained links to the appropriate ECU or bridge device. Stream duplication provides a means for safety-critical systems to duplicate themselves if the primary link fails. Daisy-chaining will allow multiple SerDes ports to be connected back-to-back, aggregating data on the link, before it arrives at the ECU. Finally, functional safety is being addressed by providing end-to-end protection mechanisms that comply with ISO 26262.
These features are welcome in the next generation of ADAS/AV developed vehicles, but there are also challenges to overcome. These include different media dependent interface (MDI) cables and connectors, securing the network, interoperability with other vendors and technical concerns of Tx testing ensuring linearity and PSD of PAM-N networks. It will also be critical to validate the robustness of receivers against electromagnetic interference (EMI) to ensure operation in the harsh automotive environment. This is a complex measurement that involves injecting pre-defined, calibrated levels of noise at the RX pin of the SerDes while monitoring its ability to clock symbols within acceptable error limits.
Interoperability is a real concern. Transceivers are sensitive devices that must operate in the notoriously harsh automotive environment that includes heat, vibration, electro-static discharge (ESD) and EMI. This can be broken into three different areas of testing. First, transmission ensures that what is sent meets expectations. Second, receiver capability establishes how reliably a device (gateway, module, switch, PHY) receives the correct signals. And third, the link segment shows the performance of the passive interconnect between transceivers known as the link segment. Physical layer validation includes all three of these elements.
Ultimate goals for all this testing are interoperability between vendors of different devices. There could be over 100 different vendors that contribute to one car and there are standards organizations that create specifications. These applications are a way to evaluate against known standards to ensure the data integrity is maintained.
In the case of the transmitter, we are looking to ensure that the signal characteristic is good. So, we use an oscilloscope that acts as a receiver. The device under test (DUT) is put into a series of known states and the acting receiver makes sure the signal is ‘valid’. Figure 3 is an example of a backup camera view with lines in it. The lines equate to gaps in the transmission, dropped packets. One or two and we can still see the image, but we certainly don’t want it to blink black when there is a child behind the vehicle.
The camera, cable, switch and GPU or ECU, and the vehicle braking system are each made by different vendors. They need to work together, thereby underlining the importance of interoperability. In addition, the data rate is going up >100X->1000X faster than CAN and growing a lot more complexity when there is a higher speed signal. The modulation type has become increasingly complex.
Legacy standards like CAN use NRZ or PAM-2, as compared to PAM-3 or PAM-4 for Automotive Ethernet and automotive SerDes. So, these Tx tests also need to check data integrity which include jitter tests; power spectral density (a noise measurement over a frequency range), and a linearity test to look for variation in amplitude levels, which can cause bit errors at the receiver.
Ultimately, we need to ensure that the data does not cause radiation emissions, reflections, or attenuations, and doesn’t interfere with other circuits. If it doesn’t pass one of these tests it will lead to symbol or packet errors that lead to dropped frames at the receiver, or lines in the display like we saw before.
The cable, connector, fixture or harness connecting these devices together is the link or the channel. A vector network analyzer (VNA) can characterize the impact the channel has on the signal, making sure signal integrity is maintained between the transmitter and receiver. Given the cable lengths used in the harsh automotive environment, it’s crucial to look at impedance versus frequency to predict how the channel will perform within the vehicle.
A link segment consists of cables plus inline connectors, along with mating connectors at either end. Ultimately the wire harness is responsible for moving control and payload data, as well as for providing DC power to remote sensors. Channel characterization for SerDes link consists of both time domain and frequency domain analysis. This requires looking at the cabling system, the MDI, and the fixturing and test setup requirements.
The actual MDI connector is not standard, but there are some rigid specifications to help ensure that interactions between the MDI and cable are minimized. Figure 4 (top) provides an example of an H-MTD connector that is being used for multi-gig automotive Ethernet and could also be used for emerging SerDes standards. When we look at the channel tests, we are looking for errors such as impedance mismatch; signal distortions or defects, and cross talk between the cables.
Receivers are responsible for making sense of the data sent over the link, then passing it along for further processing in an ECU or display device. Bit errors at the receiver result in lost or corrupted data coming from safety-critical sensors like camera, radar and lidar.
Proper receiver functionality becomes increasingly difficult for complex modulations like PAM-4, especially when sent over long channels exposed to many simultaneous sources of noise. To characterize the receiver’s capabilities, one must measure error levels in the presence of multiple noise sources, including narrow band interference; bulk current injection; transients on-line and alien cable bundle crosstalk.
The measurement setup can include noise sources, amplifiers and coupling circuitry that allow precise levels of noise to be injected onto an active SerDes link. The DUT’s signal quality registers are then queried to verify whether the receiver could interpret symbols correctly in the presence of noise. The emphasis in receiver testing is to stress the receiver to ensure it can still maintain BER rates.
The future of mobility involves more cameras, more connections and more sensors with greater accuracy, less weight and increased safety. Undoubtedly, there will be a need for an in-vehicle network to seamlessly handle these challenges. Those in-vehicle networks will need to be tested, they will need to be interoperable, and they will need to be secure.
Carrie Browen is the autonomous vehicle business line product manager for Keysight Technologies. Kevin Kershner is responsible for defining the requirements for Keysight’s future IVN solutions and is an active participant in MIPI A-PHY, ASA and VESA standards meetings.