We continue our coverage of this week’s CES 2024 event with NXP’s announcement today of a single-chip radar for software defined vehicles (SDVs).
The goal of automotive vision systems is not to equal human perception, but to exceed it. That is much of the impetus behind NXPs new SAF86xx system on chip (SoC) which monolithically integrates a high-performance 77 GHz radar transceiver with on-chip processing capability via NXP’s S32 architecture.
The new part further allows for seamless integration of multiple sensors to create a 360 degree “network of sensors.” It communicates with MACsec (IEEE 802.1AE) networking for secure communications over Automotive Ethernet.
The NXP SAF86xx
The part builds on the earlier generation, SAF85xx, released a year ago, with higher speed and greater integration. Like the SAF85xx, the 86 is built on NXP’s 28 nm RF CMOS process node, delivering a 40% increase in compute performance over prior generation products. All About Circuits covered the NXP SAF85xx a year ago, after CES 2023.
The SAF86xx SoC is designed to create a network of radar imaging sensors and deliver a 360-degree picture surrounding a vehicle. The networked sensors have individual built in processing capability to allow for a distributed vision processing system.
Such a system provides more information faster with less burden on a CPU and has high speed communications to allow for both edge computing and centralized computing. Further, the system is designed to allow over-the-air field upgrades to allow for in-application upgrades.
Software Defined Vehicles
In our interview with Matthias Feulner, Senior Director ADAS at NXP, Feulner explained that the automotive industry is moving to add advanced comfort features and field upgradability to the safety component of advanced driver assistance systems (ADAS).
Car manufacturers have been largely focused on safety features, like automated emergency braking, blind spot detection but they want more now and in the future. According to Feulner, a number of European manufacturers are announcing level three ADAS features in their premium vehicles.
Level 3 is known as “conditional automation” and is the first of the six levels of ADAS where the system starts to take over responsibility for environmental detection from the human driver. Level 3 is still two steps away from full automation and requires human override, and it depends on extremely sophisticated sensors and software.
The ability to upgrade in the field to accommodate bug fixes, new regulations and field gained knowledge is critical for this stage, given the infancy of the ADAS industry. This leads to the software defined vehicle concept. The hardware must be flexible and reprogrammable enough to accommodate rapidly changing technology.
Network of Sensors for Improved Recognition
The SAF86xx sensors are designed around an AI processing capable central ECU (engine control unit), but they maintain the integrated processing unit from the prior SAF85xx generation. This dual capability allows for compatibility with edge processing vehicle systems, central processing systems and hybrids of the architecture, according to Feulner.
“It’s not like all the intelligence is moving out of the sensor into a centralized controller. Rather, it’s like a repartitioning. It’s a distributed architecture and smart partitioning will be key.”
Distributed radar array increases virtual aperture
With this distributed smart architecture, the sensors are networked with high-speed automotive Ethernet and can operate in a collaborative mode sharing antennas. When collaborating, the sensors receive both their transmission and transmission from neighboring sensors. This gives an overlapping 3D view of an object from multiple sensors.
Distributed aperture radar (DAR) effectively increases the radar antenna aperture which helps to overcome some of the vehicle-based limitations that would otherwise lead to smaller apertures and reduce sensitivity. The virtually increased aperture is critical when discerning between different sized and different speed obstacles, such as the bicycle and car in the above example.
Integrated 3D Waveguide
In one application, the SAF86xx has a built-in PCB antenna but the part takes the integration a step farther with an integrated 3D wave guide. In what NXP calls launcher in package (LiP), the chip directly radiates into a 3D waveguide antenna. The waveguide antenna is directly coupled without need for a signal to be routed through the PCB. This can double the effective range of the sensor for certain objects.
LiP waveguide architecture reduces losses, increases detection ranges.
AI Sensor Fusion with Distributed and Centralized Processing
With the new system, there is a greater level of centralized processing for the computationally heavy software functions. The sensors have built-in computational power, so pre-processing of the data can be done on the edge.
With 100 Mbit Ethernet now and Gigabit Ethernet in the near future, AI algorithms can be used in the more powerful centralized ECU computer system for AI-based sensor fusion and recognition.
360 degree radar imaging with distributed and centralized processing
Sensor fusion takes place in the ECU. The AI algorithms will benefit from having received clearer data from the SAF86xx sensors, especially with the sensors sharing antennas for larger virtual apertures. The total system delivers a 360-degree sensor coverage with collaborative sensing, AI processing and ECU hosted software features.
Designed for Today and Tomorrow’s Automation
The SAF86xx is designed to integrate into the current generation of ADAS and be upgradable for future more advanced versions. By providing integrated radar, processing on chip, high-speed chip to chip communications and field upgradability, the system allows for greater flexibility in a volatile market.
You can visit NXP’s booth at CES this week at Central Plaza, booth CP-18 in the Las Vegas Convention Center.
All images used courtesy of NXP
