• This email address is being protected from spambots. You need JavaScript enabled to view it.

ADAS Signal Processing

The first stage in the autonomous vehicle system is the perception stage.
The environment both near and further away is combined with localization data.

Next is the planning and decision stage.

The higher the level of autonomy the more complex the computing will get

ADAS Signal Processing

AD/ADAS Signal Processing

The autonomous vehicle must be able to work under all types of road, weather and light conditions. This includes nighttime, low light conditions, fog, rain, snow, black ice etc.

To be able to properly plan the right course of action the vehicle must perceive its location and itsenvironment. The environmental sensing is derived from inputs from its many sensors - cameras, radar, lidar, ultrasonic, GNSS etc. The data from the sensors is then either pre-processed locally or sent as raw data to the tracking system. The fused data for the tracked object(s) will then be handled by the next stage of the system.

VSORA's algorithm agnostic architecture makes it easy to implement the algorithms using high-level (Matlab-like) language.


High processing power

High processing power eliminates need for specific coprocessors and hardware accelerators ensuring greater flexibility.

Customizable hardware

Customizable hardware with a configurable number of ALUs per core and unlimited number of cores.

User defined quantization

Floating point precision can be user specified to optimize the system.

Optimal algorithm mapping

Optimally maps algorithms, including non-parallel algorithms.

Automatic interconnect handling

Automatic handling of the interconnect between the different cores.

Low power

Lower energy consumption / power constraints.

High efficiency

  • Signals handled in hardware.
  • Signal memory bandwidth scales with MPU processing power.
  • Multi-instructions per cycle / rich set of instructions.

Handles higher processing requirements

Higher processing power to handle MiMo, beamforming and carrier aggregation requirements.


Futureproof should standards evolve.

Easy system programming

Compilation platform separates codes running in the different cores.

Platform independent design flow

High-level, platform independent design flow (C++/Matlab-like).

Algorithms mapped to different cores

Easy mapping of algorithms to different cores.

Optimized silicon area

Silicon area optimized to required processing power.

Combining Track Lists

Track-to-Track Fusion

Sensors observing the vehicle surroundings pre-process the data using tracking algorithms, extracting the information required. Only the extracted information is passed on to the next stage for the environmental perception. This means that the information from the various sensors are combined as lists of tracked objects. This is commonly called "track-to-track" fusion.

Combining Track Lists

Central Fusion

If the sensors pre-process the data using predetermined tracking algorithms some information will be lost, as only the list of tracked object will be passed on. In some cases having the ability to combine the raw data directly from the sensors may allow for a more comprehensive and complete understanding of the environment, which then will make the next stage involving (mission, behavioural and motion) planning and control easier. Central fusion requires higher computational power though.

Combining Track Lists

L4/L5 Fusion

Level 4 and Level 5 autonomy expects the vehicle to have the same mobilty as a human driver - unlimited! The vehicle must adjust to all driving scenarios, regardless of road, weather and traffic condiftions.

This requires the computational unit to receive a large amount of sense information. This data needs to be fused to be able to make/take decisions. Decisions based on both past experience as well as built in imagination/programming direction. All of it in just a few milliseconds!

It is no longer enough just to do AI sensor fusion or DSP signal fusion. What is required is the ability to use both AI and DSP when needed and required following advanced mathematical and AI algorithms. But it must be done with a minimum of latency, so the more parallelism that can be integrated the faster the system can work. 

This is where the ADxxxx family excels!