AWS Takes SDI in the Cloud to Next Level
In September, AWS unveiled AWS Cloud Digital Interface (CDI), a new network technology enabling customers to migrate to uncompressed live video workflows to the cloud.
AWS' ambition has taken a further step with a proof-of-concept live production workflow built with Grabyo's cloud-based video production kit and the production expertise of LTN Create.
Beyond the immediate demonstration, AWS is planning to create a standard for interoperable video transport for live workflows in the cloud. SDI is the analogy, but SMPTE ST 2110 already exists. Just like ST 2110, AWS wants to maintain all the useful attributes of SDI (namely interoperability and reliability between machines) but do so in a way that makes sense for moving data around in the cloud. There's more on this below, and it's arguably the bigger story.
"We want CDI to have the same interoperability quality as SDI in the cloud," said David Griggs, senior product manager, AWS Media Services. "If we don't have that all we end up doing is building processing islands that don't talk to each other and we miss. So, we're really serious about CDI becoming a product that brings level of interop."
Uncompressed Live Demo Explained
But first. The CDI demo (not shown live) was of a hybrid architecture where live HD MPEG video feeds are brought in over the LTN transport network from a venue (in this case from an American football stadium) and decoded (in this case in a fixed location facility run by LTN, but it could be done in the cloud). The LTN Create team switches and integrate this into an uncompressed clean feed which is handed back to the network and delivered into the AWS cloud via MediaConnect (AWS' live transport stream over IP service).
In this case, MediaConnect feeds a live production solution provided by Grabyo in the cloud, and represented the first time the partners had shown a full end-to-end live sports workflow.
Grabyo CEO Gareth Capon explains, "Once the feeds are received by the public cloud into Grabyo we replicate that feed three times to create three different graphics outlets. These could be for regionalised production or different distribution platforms. These are sent back uncompressed using CDI for transport. So, we're essentially taking one high fidelity stream, replicating it three times, moving that into three different production instances at low latency and repackaging that with different graphics outputs and sending it back over the AWS network into LTN for presentation."
Since the LTN Network is multicast-enabled, it could potentially take each of those three unique feeds into a hundred locations or thousands of locations whether local broadcaster, MVPD, VMPVD, digital owned-and-operated platform. or a social network.
"A very common next step in the process for sports would be to handoff to a copyright protection system before handing off downstream," says Rick Young, SVP, head of global products at LTN Global. "The video is replicated in uncompressed fashion so there's no generational loss from continued encoding and decoding."
The demo diagram represents three typical handoffs
Benefitting the Future of Live from Sports to Education
The vendors talked up what this technology could do for future live production. From a presentation standpoint, for example, the number of variations is limitless and could be used to create tens or hundreds of different outputs from a single event.
"From a control point of view, you can do this with a much smaller number of people who could be based anywhere," said Capon. "Using CDI in the cloud offers significantly more flexibility at a much lower price point than traditional host broadcast production out of a truck."
Indeed, just three people worked on this production, each in different locations.
Other things that could be achieved far more efficiently at the scale CDI promises includes changing the audio for multi-regional distribution. There are format implications for social media, too, which will make it much easier to reframe for Instagram Live or Facebook Live.
"You can start to use lightweight workflows to change the output presentation which does have really positive upsides for commercialisation," Capon said. "With these types of workflows, you can put together more live content in more places for more people which democratises the content opportunity whether sport, music, news, corporate or education."
AWS' Griggs added, "Previously, attempts to do this would have been limited because at some point you hit a resource constraint like CPU memory or I/O bandwidth. CDI allows you to spread that processing across multiple instances. In this case each replicated feed is being processed on its own instance so ultimately the scalability of the deployment is far less constrained.
"That means you can grow and shrink your cloud-based broadcast infrastructure based on the sophistication of the event instead of having to pre-provision and buy infrastructure you think you'll need at some point in future. It turns the whole operational paradigm on its head."
In this demo the sources were 1080i 59.94fps but there should be no real infrastructure restriction with regards to data. On the transmission side, LTN is a MPEG TS native network at low latency handling 50Mbps to 100Mbps feeds, and so is agnostic to the underlying format.
"A bigger challenge is the ability to perform frame accurate synchronization in the cloud across a range of different input devices to platforms and locations," said Capon. "It has held us back but that is coming into production now."
"Another important challenge with some of our cloud productions in the last 12 months has been having heavily compress the outputs of vision mixer in the cloud when needing to scale. By using CDI and moving video around uncompressed it protects the fidelity of the production as you scale it up and frees up compute resource (more feeds, more inputs).
"We're not outputting uncompressed video but in future there will be lots of opportunities for uncompressed video on the distribution side."
Griggs followed up, "These pre-transmission workflows are largely dependent on data rates supportive of uncompressed video and for some time AWS has not been able to offer its customers a path to translate these workflows to cloud environments. With MediaConnect and CDI we are paving the way and this demo is just the tip of the iceberg.
"The move from traditional FPGA hardware to software-defined workflows makes that transition to higher resolution, higher frame rates so much less challenging. Once we've made that pivot as an industry, the incremental change to embrace high fidelity standards is a lot less painful."
The focus of this demo was not on end-to-end latency but on the scalable capability of live workflows running in the cloud.
SDI in the Cloud and Data Interop
One challenge that perhaps vendors should think about is how to monitor for compliance and integrity as the number of outputs rises. As you start to automate all of these functions you need to know what is happening and how to make changes—things that live event producers have not had to think about when only outputting one version of a mixed feed.
Capon said, "We did quite a lot of work with the edports industry over the summer, and what we saw there was you had 60 to 100 players in an event at any one time. It's rare that you see video of them all in production—and they may all be in different places given everything's behind closed doors- so to scale up to that workflow using traditional methods is hard. This type of cloud-based network makes it much easier."
The production possibilities in the cloud only make sense with multi-vendor involvement which is why AWS made an SDK for CDI available which Grabyo used for this demo.
AWS clearly sees a substantial market in being the backbone for live events and an important part of that is having multi-vendor's tools work interoperably.
Griggs calls this "SDI in the cloud." He explains that CDI's technology stack is proprietary and allows it to move data around at rates that are equivalent to SDI (it can move a frame of video in 8-12 ms, which is well within SDI specs). "We had to use proprietary technology that knows how to survive and thrive in our multi-tenant network; ST 2110 just doesn't fly in those kinds of environments."
Separate from the technology stack of CDI is the audio video metadata (AVM) schema, which is not proprietary. This is designed to be an open standard and is in the SDK. AWS want to promote that through industry bodies and get buy-in from multiple vendors so that it ends up with an interoperability standard.
"A key part to CDI is an interoperability of the data plane that enables tools from vendors to communicate," he explains. "To do that, we're talking to as many partners as we can across the spectrum of live production, playout and master control to ensure we get momentum to deploy their products and services in the cloud. That is essential to the success of cloud production. We need a healthy and vibrant vendor community that is embracing the cloud.
"Secondly, whilst there's a lot of proprietary tech underneath CDI to provide 'SDI in the cloud' with the same quality and latency characteristics as on-prem technologies we don't want to have proprietary byte-packing. This is the way that bytes are transferred from host to host. We think that necessitates the input of the broadcast community."
To that end, AWS is taking its ideas to the VSF as part of its Ground to Cloud, Cloud to Ground initiative (GCCG). "We are not dictating anything," Griggs declared. "We're saying 'Here's a byte-packing order' and 'Here's an AVM layer' that sits on top of the CDI technology that we think makes sense but [industry bodies] shape it, improve it, make it better and we will reimplement into the CDI tech stack.
"We can have an interoperability standard that takes away the friction and the concern around how to extend your existing broadcast infrastructure into an Amazon EC2 environment using CDI. The two will work together just like SDI does on-prem."
UK's largest commercial broadcaster offers six linear channels on the ITV Hub app, delivering them directly to third-party distribution partners like Amazon Prime
Zixi's Software-Defined Video Platform and live protocol are being used to deliver the 8K stream with AWS and Verizon for a major global broadcaster
The remote-control device for real-time video transport is the size of a Kleenex box, completely silent, and requires no onsite configuration.