Video: Line by Line Processing of Video on IT Hardware

If the tyranny of frame buffers is let to continue, line-latency I/O is rendered impossible without increasing frame-rate to 60fps or, preferably, beyond. In SDI, hardware was able to process video line-by-line. Now, with uncompressed SDI, is the same possible with IT hardware?

Kieran Kunhya from Open Broadcast Systems explains how he has been able to develop line-latency video I/O with SMPTE 2110, how he’s coupled that with low-latency AVC and HEVC encoding and the challenges his company has had to overcome.

The commercial drivers are fairly well known for reducing the latency. Firstly, for standard 1080i50, typically treated as 25fps, if you have a single frame buffer, you are treated to a 40ms delay. If you need multiple buffers for a workflow, this soon stacks up so whatever the latency of your codec – uncompressed or JPEG XS, for example – the latency will be far above it. In today’s covid world, companies are looking for cutting the latency so people can work remotely. This has only intensified the interest that was already there for the purposes of remote production (REMIs) in having low-latency feeds. In the Covid world, low latency allows full engagement in conversations which is vital for news anchors to conduct interviews as well as they would in person.

IP, itself, has come into its own during recent times where there has been no-one around to move an SDI cable, being able to log in and scale up, or down, SMPTE ST 2110 infrastructure remotely is a major benefit. IT equipment has been shown to be fairly resilient to supply chain disruption during the pandemic, says Kieran, due to the industry being larger and being used to scaling up.

Kieran’s approach to receiving ST 2110 deals in chunks of 5 to 10 lines. This gives you time to process the last few lines whilst you are waiting for the next to arrive. This processing can be de-encapsulation, processing the pixel values to translate to another format or to modify the values to key on graphics.

As the world is focussed on delivering in and out of unusual and residential places, low-bitrate is the name of the game. So Kieran looks at low-latency HEVC/AVC encoding as part of an example workflow which takes in ST 2110 video at the broadcaster and encodes to MPEG to deliver to the home. In the home, the video is likely to be decoded natively on a computer, but Kieran shows an SDI card which can be used to deliver in traditional baseband if necessary.

Kieran talks about the dos and don’ts of encoding and decoding with AVC and HEVC with low latency targetting an end-to-end budget of 100ms. The name of the game is to avoid waiting for whole frames, so refreshing the screen with I-frame information in small slices, is one way of keeping the decoder supplied with fresh information without having to take the full-frame hit of 40ms (for 1080i50). Audio is best sent uncompressed to ensure its latency is lower than that of the video.

Decoding requires carefully handling the slice boundaries, ensuring deblocking i used so there are no artefacts seen. Compressed video is often not PTP locked which does mean that delivery into most ST 2110 infrastructures requires frame synchronising and resampling audio.

Kieran foresees increasing use of 2110 to MPEG Transport Stream back to 2110 workflows during the pandemic and finishes by discussing the tradeoffs in delivering during Covid.

Watch now!
Speaker

Kieran Kunhya Kieran Kunhya
CEO & Founder, Open Broadcast Systems

Video: CDNs: Building a Better Video Experience

With European CDN spend estimated to reach $7bn by 2023, an increase in $1.2 in only three years, it’s clear there is no relenting in the march towards IP. In fact, that’s a guiding principle of the BBC’s transmission strategy as we hear from this panel which brings together three broadcasters, beIN, Globo and the BBC to discuss how they’re using CDNs at the moment and their priorities for the future.

Carlos Octavio introduces Globo’s massive scale of programming for Brazil and Latin America. Producing 26,000 hours of content annually, they aim to differentiate themselves as much with the technology of their offerings as with the content. This thirst for differentiation drives their CDN strategy. Brazil is a massive country, so covering the footprint is hard. Octavio explains that they have created their own CDN to support Globo Play which is based on 4 tiers from their two super PoPs in Rio and Sao Paolo down to edge caches. Octavio shows that they are able to achieve the same response times as the major CDN companies in the region. For overflow capacity, Globo uses a multi-CDN approach.

Bhavesh Patel talks about the sports and news output of beIN, both of these being bursty in nature. Whilst traffic for sporting events can forecast, with news this is often not possible. This, plus the wide variability of customers’ home bandwidth are drivers in choosing which CDNs to partner with. Over the next twelve months, Bhavesh explains, beIN’s focus will move to bring down latency on their system as a whole, not on a service by service level. They are also expecting to continue to modify their ABR ladders to follow viewers as they continue their shift from second screens to 60 inch TVs.

The BBC’s approach to distribution is explained by Paul Tweedy. Whilst the BBC is still well known as a linear, public broadcaster, it has been using online distribution for 25 years and continues to innovate in that space. Two important aspects to their strategy are being on as many devices as practical and ensuring the quality of the online experience meets or is comparable to the linear services. The BBC has been using multiple CDNs for many years now. What changes is the balance and what they use CDNs for. They cover a lot of sports, explains Paul, which leads to short-term scaling difficulties, but long term scaling difficulties are equally on his mind due to what the BBC calls the ‘glide path to IP’. This is the acknowledgement that, at some point, it won’t be financially viable to run transmitters and IP will be the wise way to use the licence fee on which the BBC depends. Doing this, clearly, will demand IP delivery of many times what is currently being used. Yesterday’s article on multicast ABR is one way in which this may be mitigated and fits into a multi-CDN strategy.

Watch now! Free registration

Looking at today’s streaming services, Paul and his colleagues aim to get analytics from every player on every device wherever possible. Big data techniques are used to understand these logs along with server-side, client-to-edge and edge-to-origin logs. This information along with sports schedules can lead to capacity planning, though many news events are much less easy to plan. It’s these unplanned, high-peak events which drive the BBC’s build up of internal monitoring tools to help them understand what is working well under load and what’s starting to feel the strain so they can take action to ensure quality is maintained even through these times of intense interest. The BBC manage their capacity with their own CDN, called BIDI, which provides for the baseline needs and allows an easier-to-forecast budget. Mulitple, third-party CDNs are, then, the key to providing the variable and peak capacities needed.

As we head into the Q&A Limelight’s Steve Miller-Jones outlines the company’s strengths including their focus on adding abilities on top of a ‘typical’ CDN. For instance, running applications on the CDN which is particularly useful as part of edge compute and their ability to run WebRTC at scale which not many CDNs are built to do. The Q&A sees the broadcasters outlining what they particularly look out for in a CDN and how they leverage AI. Globo anticipate using AI to help them predict traffic demand, beIN see it providing automated highlights whilst the BBC see it enabling easier access to their deep archives.

Watch now!
Free registration
Speakers

Carlos Octavio Carlos Octavio
Head of Architecture and Analytics,
Globo
Bhavesh Patel Bhavesh Patel
Global Digital Director,
beIN MEDIA GROUP
Paul Tweedy Paul Tweedy
Lead Architect, Online Technology Group,
BBC Design + Engineering
Steve Miller-Jones Steve Miller-Jones
Vice President of Product Strategy,
Limelight Networks

Video: Making a case for DVB-MABR

Multicast ABR (mABR) is a way of delivering traditional HTTP-based streams like HLS and DASH over multicast. On a managed telco network, the services are multicast to thousands of homes and only within the home itself does the stream gets converted back unicast HTTP. Devices in the home then access streaming services in exactly the same way as they would Netflix or iPlayer over the internet, but the content is served locally. Streaming is a point-to-point service so each device takes its own stream. If you have 3 devices in the home watching a service, you’ll be sending 3 streams out to them. With mABR, the core network only ever sees one stream to the home and the linear scaling is done internally. Not only does this help remove peaks in traffic, but it significantly reduces the load on the upstream networks, the origin servers and smooths out the bandwidth use.

This video from DVB lays out the business cases which are enabled by mABR. mABR has approved the specification which is now going for standardisation within ETSI. It’s already gained some traction with deployments in the field, so this talk looks at what the projects that drive the continued growth in mABR may look like.

Williams Tovar starts first by making the case for OTT over satellite. With OTT services continuing to take viewing time away from traditional broadcast services, satellite providers are working to ensure they retain relevance and offer value. Delivering these OTT services is, thus, clearly beneficial, but why would you want to? On top of the mABR benefits briefly outlined above, this business case recognises that not everyone is served by a good internet connection. Distributing OTT by satellite can provide high bitrate, OTT experiences to areas with bad broadband and could also be an efficient way to deliver to large public places such as hotels and ships.

Julian Lemotheux from Orange presents a business case for next-generation IPTV. The idea here is to bring down the cost of STBs by replacing CA security with DRM and replacing the chipset with a cheaper one which is less specialised. As DASH and HLS streaming are cpu-based tasks and well understood, general, mass-produced chipsets can be used which are cheaper and removing CA removes some hardware from the box. Also to be considered is that the OTT ecosystem is continually seeing innovation so delivering services in the same format allows providers to keep their offerings up to date without custom development in the IPTV software stack.

Xavier Leclercq from Broadpeak looks, next, at Scaling ABR Delivery. This business case is a consideration of what the ultimate situation will be regarding MPEG2 TSes and ABR. Why don’t we provide all services as Netflix-style ABR streams? One reason is that the scale is enormous with one connection per device, CDNs and national networks would still not be able to cope. Another is that the QoS for MPEG2 transport streams is very good and, whilst it is possible to have bad reception, there is little else that causes interruption to the stream.

mABR can address both of these challenges. By delivering one stream to each home and having the local gateway do the scaling, mass delivery of streamed content becomes both predictable and practical. Whilst there is still a lot of bandwidth involved, the predictable load on the CDNs is much more controlled and with lower peaks, the CDN cost is reduced as this is normally based on the maximum throughput. mABR can also be delivered with a higher QoS than public internet traffic which allows it to benefit from better reliability which could move it in the realm of the traditional transport-stream based serviced. Xavier explains that if you put the gateway within a TV, you are able to deliver a set-top-box-less service whilst if you want to address all devices in you home, you can provide a separate gateway.

Before the video finishes with a Q&A session, Williams delivers the business case for Backhauling over Satellite for CDNs and IP backhaul for 5G Networks. The use case for both has similarities. The CDN backhauling example looks at using satellite to efficiently deliver directly to CDN PoPs in hard to reach areas which may have limited internet links. The Satellite could deliver a high bandwidth set of streams to many PoPs. A similar issue presents itself as there is so much bandwidth available, there is a concern about getting enough into the transmitter. Whether by satellite or IP Multicast, mABR could be used for CDN backhauling to 5G networks delivering into a Mobile Edge Computing (MEC) cache. A further benefit in doing this is avoiding issues with CDN and core network scalability where, again, keeping the individual requests and streams away from the CDN and the network is a big benefit.

Watch now!
Download the slides from this video
Speakers

Williams Tovar Williams Tovar
Soultion Pre-sales manager,
ENENSYS Technologies
Julien Lemotheux Julien Lemotheux
Standardisation Expert,
Orange Labs
Xavier Leclercq Xavier Leclercq
VP Business Development,
Broadpeak
Christophe Berdinat Moderator: Christophe Berdinat
Chairman CM-I MABR, DVB
Innovation and Standardisation Manager, ENENSYS

Video: Encoding Vs Compute Efficiency in Video Coding

Ioannis Katsavounidis from Facebook joins us to talk us through his work finding the best balance between computation and encoding. He explains how encoding has moved from real-time, hardware-based encoding in the late 80s and 1990s through to file encoding, chunk-based encoding and now shot-based encoding. Each of these stages has brought opportunities to speed up encoding, but there has always been a fundamental reason why encoding can’t simply be sped up by the advance of IT.

Moore’s law posits that every year, the number of transistors in chips doubles. Whilst this has continued to be true until recent years, transistors have always been a proxy for processing power. For many years now, the way to keep the computational ability of CPUs high has been not to increase clock-speed as it was twenty years ago, but to add cores to the chip. As each core acts as its own CPU, this gives the ability to execute code in parallel with a thread of code running separately on each core. Whilst 12-20 cores are typical for servers, there are CPUs which deliver up to 128 cores.

Ioannis explains why DCT-based codecs are resistant to multi-thread encoding by showing how some of the encoding decisions are based on the previously decoded video frame so the encoder needs to decode the video before it has the information it needs to make the next encode decisions. An example of this motion estimation where you need to understand what a macroblock looks like in order to detail if and how it can be moved to form part of the macroblock currently being encoded.

It turns out that some of the information you need to calculate can be found from the original video. Whilst this doesn’t provide full parallelisation, it does help in freeing some of the computation to be done in parallel thus reducing the length of time spent on the linear encoding stage. As the design of the codec itself is limited in its ability to be parallelised, the best way to speed up encoding has been to split up the original video and encode these, now separate, sections independently.

Speeding up video encoding has therefore focused on splitting up the video into different sections and encoding those in parallel rather than trying to parallelise the encoding itself due. Encoding each frame separately is one way to do this, but sacrifices encoding efficiency. Splitting each frame up into sections (tiles or slices) is another way, though this also sacrifices either quality or bitrate. The most successful encoding parallelisation has been chunked encoding. As streaming applications use chunks, typically around 2 seconds nowadays, there’s no reason not to just cut your video up into small sections and encode those separately; the whole of this video focuses on non-live video.

If there’s a shot change in the middle of your chunk, this is likely to look very bad since the motion estimation will fail to produce good results and there may not be enough bitrate budget to compensate. Therefore it’s best to drop in an IDR frame at the shot change or to actually change your video chunks to match shot changes. Simply encoding these chunks in parallel would speed up the encoding, however, it misses an opportunity to optimise quality vs bitrate.

Ioannis explains an experiment to determine the best operating point for chunks. He does that by reminding us that all encoders have certain ‘speed’ settings which control how much computation, and therefore time, is required for each encode. The ‘very fast’ setting in x264 will encode at the highest speed possible, but the quality will be worse or a certain bitrate compared to the ‘very slow’ setting. Ioannis’s experiment encoded each chunk at every speed setting for a variety of resolutions and bitrates. Each encode was then analysed for quality using PSNR, MS-SSIM and VMAF.

From Ioannis’ work, we can see how the bitrate setting affects both the encode time and the quality and we can observe that the slower speeds tend to have minimal quality advantages for the significant extra time involved in the encoding. Each curve has a steep part and a shallow section with the transition between known as the ‘convex hull’. Choosing a setting on the convex hull portion of the line is the optimal balance between quality and encoding time and is where, says Ioannis, most people should aim to operate.

The talk finishes with a summary of the conclusions which can be drawn from this work looking at the use of convex-hull which we’ve just discussed, the best type of parallel processing, whether oversubscription of CPU cores is helpful or not and an interesting observation that it’s often the metrics which put a significant burden on encoding rather than the video encoding itself, particularly for lower resolutions.

Watch now!
Speakers

Ioannis Katsavounidis Ioannis Katsavounidis
Research Scientist,
Facebook