Progression through Regression — Taking backwards compatibility to a whole new level
We are proud to announce yet another step in our continuous work to improve our video streaming service. Thanks to the hard work of the Video Core team, and their collaboration with client teams as well as digital infrastructure teams, we have made a major improvement to how we transcode video. In the coming weeks, we will start to transcode and distribute video using the ISO/IEC 13818–2 standard alongside the associated Layer 3 audio!
We believe that this will immensely improve the efficiency of our production workflow, both in terms of transcoding time and used storage, while at the same time increasing our backwards compatibility for earlier device generations. Although we hope that the reader is just as excited as we are for this particular development, let me spend a few paragraphs to really hit home as to why this is so important.
At SVT we transcode several hundreds of videos on a daily basis for our OTT VoD service, and for each and every one of those videos we face the same challenge:
How do we balance visual quality, distribution costs and complexity of compression?
You see in a very practical sense we can only choose two out of three for any given video. That is to say, if we want high quality video and low bitrate we naturally suffer a higher complexity of compression, which in turn leads to longer transcoding times and a more taxing decode on the receiving device. Alternatively, if we keep the complexity and bitrate as low as possible the visual video quality will suffer.
In reality however, there is a fourth relation to this engineering triangle for us at SVT, and that is device compatibility. Whereas a commercial actor can simply choose not to support a certain device (or family of devices) if the number of potential customers it would bring does not compensate for the amount of work it would need, a public service broadcaster cannot disregard that particular group. Instead, we must always strive to support every kind of device regardless of the number of individual viewers that uses it, now there are obviously limitations to this but generally that is the case. This conundrum presents a very hard challenge (although not strictly a limitation) when it comes to advancements in video and audio transcoding. Thus, we at SVT have chosen to go a slightly different route for our development moving forward, namely improvements through backwards compatibility!
Enter ISO/IEC 13818–2 (Yes we are aware that the standards actual name is: ICS:35:35:35.040:35.040.40 — ISO/IEC 13818–2:2013)! This backwards compatible standard of video transcoding provides several benefits to our workflow. First and foremost, it will decrease the overall complexity of our video compression, stuffing less information into the same number of bytes. Simultaneously we will distribute the ISO/IEC 13818–2 content using a lower bitrate compared to our current AVC and HEVC streams, lowering distribution costs and saving on the individual data plans of our viewers. This also enables us to transcode the same content using less pixels, resulting in a higher “NSR”-ratio. Now arguably, such a reduction comes at a price, the acute reader will be quick to notice that such a change would result in loss of details in low, medium and high frequency ranges, however it is important to consider that the average PSNR will be also be lower.
There is more however, using ISO/IEC 13818–2 there is another avenue of changes made possible that might not be obviously beneficial at first or later. Today most of our streaming is done through the help of two standardized streaming protocols, namely MPEG-DASH and HLS. While both of these are indeed great pieces of technology, there is a poor argument to be made in favor of omitting them entirely. You see ISO/IEC 13818–2, as with most other codec standards, allows for the traditional albeit sometimes deprecated notion of progressive download. As the positive connotation of the term suggests the concept of progressive downloading is that the viewer downloads the whole file upon attempted playback, rather than small chunks of the video. This would mean that we can increase buffer times substantially, although admittedly risking a higher loss of data. As a viewer, you would simply start a video, wait for 15% of the viewing time to buffer and then be able to enjoy all of that content at your leisure.
Finally ISO/IEC 13818–2 brings some changes to the way we transcode audio. Instead of having different channel layouts and bitrates for regular, multi channel and low bandwidth transcodes, we can simply choose to use the same audio configuration for all content. Again, ISO/IEC 13818–2 allows us to lower the bitrate for audio while removing only 2/3s of the information compared to current audio transcodes.
So in summary, using ISO/IEC 13818–2 will lead to the following benefits:
- Lower complexity of compression than AVC/H.264 and HEVC/H.265
- Lower bitrate than our current ABR-ladders
- The ability to display the same content with fewer pixels
- No need to package content in MPEG-DASH or HLS, instead content can be progressively downloaded
- Achieving increased buffer times at the cost of slightly higher data loss
- Streamlined audio with the same kind of stereo track on every device
To showcase what the future will bring, in true progressive download-fashion, we are happy to provide an example of the three video qualities and that we are going to distribute, as well as the streamlined audio track:
Finally here is a single frame from our new top quality stream:
Until next time!
