EnSilica co-processor lightens the ADAS processing load
- ผู้เขียน:Ella Cai
- ปล่อยบน:2017-11-29
EnSilica’s eSi-ADAS RADAR Imaging Co-processor seeks to solve the problem of RADAR data overload and resolution by doing it in a dedicated co-processor to enable tracking of potentially hundreds of objects to be made in real time.
It is claimed to have up to 20x lower power requirements than today’s lower resolution systems and lower system costs, as well as outperforming them in terms of speed and accuracy by up to 10x.
ADAS systems are becoming increasingly complex as they advance from their simple origins of cruise control and parking assistance to fully autonomous driving. One of the challenges facing OEMs is that a combination of Video, RADAR and LIDAR is needed to cover all operational conditions, which is expensive.
This is because current RADAR sensors lack the imaging capability, 3D and 360° coverage that is available from LIDAR but LIDAR does not work well at night or in fog so the two have to be used in conjunction with Video. This creates a problem of large amounts of data that needs to be processed in real time.
“EnSilica has done a number of projects for customers in the automotive area and it became clear to us that there is a need for a solution for the autonomous/self-drive car market to effectively handle the huge amount of real time data needed to provide an accurate picture of the surroundings of the car so that it could drive itself,” says EnSilica’s David Doyle, “also, by improving the resolution of the RADAR part of the system, the need and cost of the LIDAR component could potentially be removed, providing significant system cost reduction.”
However, increasing the RADAR resolution creates even more data that uses up most of the capacity of the system’s central processor. EnSilica’s innovation is to design a dedicated co-processor that is optimised to handle this particular type of data. It is a similar solution to that used in computing of having a general processor and a dedicated graphics processor unit (GPU), and EnSilica has filed a patent to protect its innovation.
The co-processor chip offloads the RADAR target tracking from the car’s Electronic Control Unit (ECU), freeing up ECU resources to concentrate safety critical decisions such as target identification and fusing RADAR data with camera video information.
The benefits include lower system cost, lower power requirements and faster precision real-time tracking with an ultralow maximum latency of 20ms to enable critical safety decisions to be made with confidence.
As a result, EnSilica’s solution has high resolution image sampling capability, elevation and 360° coverage from multiple sensors, plus it works at night and in any weather conditions. The result is an detailed RADAR map of the kind needed for future autonomous vehicles that works in conjunction with a video system.
The co-processor chip can support up to 256 virtual antennas in a 3D array with ADC sampling rates up to 100 Mega samples per second of raw data. This provides coverage for parking and short and long range RADAR in a multi-mode sensor, removing the need for and costs of ultrasonic sensors for parking.
The co-processor has dedicated mathematical hardware engines to demodulate the RADAR signals using Range, Doppler, Azimuth and Elevation Fast Fourier Transforms which are then detected and the resulting measurements are tracked using a high performance Kalman Filter.
This means that the final output is a low data rate object list with position, speed and confidence values for over 128 objects that can be used by an ECU for sensor fusion with other RADAR object lists or images.
To provide the most precise plot extraction, eSi-ADAS operates in conjunction with modern, fast chirp-capable RF devices. EnSilica has filed patents on chirp processing techniques to overcome drawbacks usually associated with processing fast chirp modulation, notably its high computational and memory requirements.
“eSi-ADAS is easy to integrate and provides a lower system cost, especially as LIDAR is potentially no longer required, and up to 20x lower power requirements than existing solutions as well as significantly outperforming them in terms of speed and accuracy,” says EnSilica CTO David Wheeler.
The eSi-ADAS RADAR Imaging Co-processor can be bought either as IP to be integrated onto a customer’s chip or as a stand-alone chip from EnSilica.
This is a shift in the company’s business model and is designed to add regular income streams from the sales of its own proprietary products based on in-house, innovative technologies.
It is claimed to have up to 20x lower power requirements than today’s lower resolution systems and lower system costs, as well as outperforming them in terms of speed and accuracy by up to 10x.
ADAS systems are becoming increasingly complex as they advance from their simple origins of cruise control and parking assistance to fully autonomous driving. One of the challenges facing OEMs is that a combination of Video, RADAR and LIDAR is needed to cover all operational conditions, which is expensive.
This is because current RADAR sensors lack the imaging capability, 3D and 360° coverage that is available from LIDAR but LIDAR does not work well at night or in fog so the two have to be used in conjunction with Video. This creates a problem of large amounts of data that needs to be processed in real time.
“EnSilica has done a number of projects for customers in the automotive area and it became clear to us that there is a need for a solution for the autonomous/self-drive car market to effectively handle the huge amount of real time data needed to provide an accurate picture of the surroundings of the car so that it could drive itself,” says EnSilica’s David Doyle, “also, by improving the resolution of the RADAR part of the system, the need and cost of the LIDAR component could potentially be removed, providing significant system cost reduction.”
However, increasing the RADAR resolution creates even more data that uses up most of the capacity of the system’s central processor. EnSilica’s innovation is to design a dedicated co-processor that is optimised to handle this particular type of data. It is a similar solution to that used in computing of having a general processor and a dedicated graphics processor unit (GPU), and EnSilica has filed a patent to protect its innovation.
The co-processor chip offloads the RADAR target tracking from the car’s Electronic Control Unit (ECU), freeing up ECU resources to concentrate safety critical decisions such as target identification and fusing RADAR data with camera video information.
The benefits include lower system cost, lower power requirements and faster precision real-time tracking with an ultralow maximum latency of 20ms to enable critical safety decisions to be made with confidence.
As a result, EnSilica’s solution has high resolution image sampling capability, elevation and 360° coverage from multiple sensors, plus it works at night and in any weather conditions. The result is an detailed RADAR map of the kind needed for future autonomous vehicles that works in conjunction with a video system.
The co-processor chip can support up to 256 virtual antennas in a 3D array with ADC sampling rates up to 100 Mega samples per second of raw data. This provides coverage for parking and short and long range RADAR in a multi-mode sensor, removing the need for and costs of ultrasonic sensors for parking.
The co-processor has dedicated mathematical hardware engines to demodulate the RADAR signals using Range, Doppler, Azimuth and Elevation Fast Fourier Transforms which are then detected and the resulting measurements are tracked using a high performance Kalman Filter.
This means that the final output is a low data rate object list with position, speed and confidence values for over 128 objects that can be used by an ECU for sensor fusion with other RADAR object lists or images.
To provide the most precise plot extraction, eSi-ADAS operates in conjunction with modern, fast chirp-capable RF devices. EnSilica has filed patents on chirp processing techniques to overcome drawbacks usually associated with processing fast chirp modulation, notably its high computational and memory requirements.
“eSi-ADAS is easy to integrate and provides a lower system cost, especially as LIDAR is potentially no longer required, and up to 20x lower power requirements than existing solutions as well as significantly outperforming them in terms of speed and accuracy,” says EnSilica CTO David Wheeler.
The eSi-ADAS RADAR Imaging Co-processor can be bought either as IP to be integrated onto a customer’s chip or as a stand-alone chip from EnSilica.
This is a shift in the company’s business model and is designed to add regular income streams from the sales of its own proprietary products based on in-house, innovative technologies.
