Includes region-of-interest encoding, intelligent digital signage, automatic object tracking, window cropping, and voice recognition… Also implementing HDMI 2.1 on the 7nm Versal ACAP for 8K applications.


Xilinx announced on the 11th that it has added machine learning capabilities to its devices targeting the professional audio/video (Pro AV) and broadcast markets. Additionally, a programmable HDMI 2.1 demo implemented on a 7nm Versal device was unveiled.

Xilinx's adaptive solutions, unveiled at Integrated Systems Europe (ISE) 2020 in Amsterdam, the Netherlands, from February 11-14, are designed to help customers address new usage models and evolving industry standards, reduce costs, and effectively future-proof their investments.


Machine learning capabilities for AI edge processing
Machine learning capabilities offered by Xilinx for pro AV and broadcast platforms include region-of-interest encoding, intelligent digital signage, automatic object tracking, window cropping, and speech recognition.

Customers can leverage these machine learning capabilities on Xilinx devices, including the Zynq UltraScale+ MPSoC platform for AI edge processing. Users can integrate real-time AV processing, AV connectivity interfaces, codecs, IP networking, CPUs, and GPUs into a scalable, adaptable, single-chip solution, reducing space, power, and cost.

Customers in the pro AV market can apply new machine learning capabilities to a variety of applications and workloads. Key examples include:

Region of Interest Encoding - Leveraging the Zynq UltraScale+ MPSoC with integrated machine learning and H.264/H.265 codecs, face and feature detection can be used to enhance video quality in those areas and apply higher compression rates to the background. This reduces overall bitrate and significantly reduces the cost of live streaming.

Intelligent Digital Signage - Machine learning models for gender, age, and gesture detection can be used to display targeted interactive ads on digital signage. This not only measures monetizable behavior but also increases advertisers' return on investment.

Automatic object tracking and window cropping - Use machine learning to output multiple HD windows from a single 4K camera. This is useful for providing switchable live event content in the control room.

Speech recognition - Automatically converts speech to text, making it ideal for taking meeting notes for conferences and collaborations, and automatically processing language translations.


First HDMI 2.1 implementation on a 7nm device
At ISE 2020, Xilinx will demonstrate programmable HDMI 2.1 on its 7nm Versal devices, an Adaptive Compute Acceleration Platform (ACAP). ACAP is a heterogeneous computing device that surpasses the performance of existing CPUs, GPUs, and FPGAs.

HDMI 2.1, which provides data transfer rates of up to 48 Gbps, supports a variety of high-resolution video and high-frame rates, including 8K60 and 4K120.

This enables more vivid visuals in fast-moving video-based applications such as LED walls, large displays, digital signage, and live sports. The Versal ACAP can transmit, receive, and process up to 8K UHD video.

The 7nm Versal ACAP delivers scalable, adaptable, and intelligent engines through a software-programmable infrastructure that tightly couples embedded software with multi-channel AV processing pipelines and AI inference.

Additionally, by integrating real-time video processing and AV connectivity interfaces, codecs, IP networking, CPUs, GPUs, and more into a single-chip solution, space, power, and cost can be reduced.