Description
Development of Electronic Control Units (ECUs) for Automotive, Defense, Robotics etc. nowadays has to rely on data from generative AI but also on real world recorded scenarios. Such recorded scenarios are valuable for for ECU developers if they add significant coverage to the situations that ECUs must be able to handle in real life and demonstrate in certifications.
Not only the quality of the content must be high, the bandwidth demand has also increased with ever higher camera resolutions and frames-per-second (FPS). PLC2’s ADL-1000 has been developed as a high-bandwidth, low-power recording solution with Multiple 10G Ethernet and PCIe inputs and recording to NVMes. Augmented by AI inference algorithms to classify and detect, the triggering upon analysed features in data to be stored becomes the selective process for high-quality content and associated metadata can provide pre-labeling.
Methodology
The ADL-1000 is a low-power, high-bandwidth platform where data can be recorded, stored, and analysed. This is possible with a data rate up to 60 Gbit/s per Ethernet inputs, or even PCIe. The data is recorded on up to 8 NVMe devices. The following external data sources can be connected to the ADL-1000:
- GigE Vision cameras
- Ethernet/UDP data
- Raw data
In the following demonstrator use case, GigE Vision cameras are directly connected to the ADL-1000. The camera data is recorded, analysed (detection of objects) and stored on the NVMe devices. While the camera data is recorded, it is also provided to an AI acceleration device, which is performing the AI inference of the camera data. In parallel, the camera data is also provided to an external small cube PC, where a Prometheus database and a Grafana server are running. In addition to the camera data, sensor data (metrics) are also provided to the connected PC. With the help of the PC, an external monitor and the Grafana web interface, it is possible to monitor health data such as temperatures, the recording speed of the NVMe devices, and, the results of the AI detection and classification (and more).
In the following diagram, the setup of the described use case is shown:
ADL-1000 Demonstrator Use Case
As mentioned before, there are other use cases, where the data is provided from external sources mixed with or other than GigE Vision cameras. This could for example be a RFSoC board, where it is necessary to record a huge amount of raw ADC data. But there are also generic low latency interfaces / protocols possible, so that any type of data can be recorded efficiently.
Conclusion
PLC2’s product ADL-1000 represents a low-power, high-bandwidth recording solution that can selectively record high quality data when combined with AI inference.
Various embedded programmable logic and software IPs, developed along the video capturing, processing and recording pipelines provide evidence of the deep knowledge and experience PLC2 has gained and can apply in development service projects.
Please visit the ADL-1000 website for more information.
