When CDNs cache A/B variants using playlist embedding in watermarked content, how does this reduce latency?

If you are a service provider, you can perform A/B watermarking in the head-end or distribution network, which is two steps. Streaming sessions are split into two subsets, called “A” and “B,” and each subset is given a different watermark. The A and B segments of each of these streams are then combined to create a single stream with a unique A and B segment combination. No two OTT subscribers will receive the same sequence of videos.

For each video source to be distributed simultaneously, two encoders are needed, as well as enough storage and origination resources. A/B variant video watermarking is a resource-intensive process because of this dual stream approach. A CDN caching infrastructure is needed to store the two versions of a video on the origin server before they can be accessed via the CDN. Because of the way A/B watermarking works, each chunk of data can only contain one WMID (watermark identifier). For some applications, such as live streaming of DRM protected content, this can cause a delay.

With A/B video watermarking, the manifest can be used to restore some caching capabilities of the video delivery network and achieve good scalability. Only segments containing the recipient’s watermark identifier are delivered in serialised playlists. If the playlist is being consumed over a network, the client can only ask for variant segments that encode the WMID for each one of the associated WMIDs. This method allows the video delivery network to store both the A and B versions of the stream.

The requested variants are served by a streaming server that hosts all of the pre-computed variant segments. Because the segments are already in the cache, they are not sent to the origin server, thus reducing latency. An HLS or DASH-based segment granularity playlist entry is needed for this purpose. A playlist is protected from malicious access and manipulation by obfuscating the segment URIs. Segment delivery is determined by the manifest manipulation component.

The scene is represented just by its 2D projection, which are photos acquired by cameras. It is possible to watermark image sequences that record a 3D scene and extract the watermark from any rendered image generated for any arbitrary view angle, as opposed to the first two methods, which only  protect the watermark information for the two key components of 3D scene representation (geometry and texture). If you’re using dynamic watermarking, you may embed information on the video asset while it’s being played back at the user’s end, such as the user’s email, date  and time of watching, their IP address, or even their business logo. Because of their dynamic nature, they provide additional protection for confidential content that  is not intended to be shared or altered. DAI (dynamic ad insertion) is also activated via dynamic watermarking in order to optimise addressable ad income.

DRM  video protection techniques such as watermarks are not sufficient on their own, but when used in conjunction with other measures, they can help to safeguard the  intellectual property of the content owner and aid to trace the source of any alleged infringement. They also serve as a helpful reminder to users about their own  and others’ rights to the content they’re using.