Don't Believe the 5G Streaming Video Hype
Type "5G better streaming" into your favourite search engine.
The result? Seemingly endless arguments that 5G will lead to better streaming. Now inspect a few. When you look under the hood, these arguments for 5G benefitting streaming are—to a streaming professional —very weak. They typically centre on "speed." Last time I checked, the speed of electric signalling and propagation delay of such signals was still pretty close to the speed of light whatever network you use. So a packet traversing a 5G network travels at the same "speed" as 4G, 3G and all those before.
What they might mean is that the peak "throughput" is ostensibly higher on 5G than 5G, and that is arguably a truism, but the devil is in the detail.
There are some caveats: To get those high bitrates you need a "high range" frequency for the 5G radio carrier. These 20GHz to 100GHz bands are not yet broadly available for license other than for astrophysics or military use. 60GHz is the most desirable for 5G because high oxygen absorption at this range means the least interference between masts operating on the same frequency. Around 10Ghz and below—the frequencies currently being licensed—throughput is considerably less, and is comparable between 4G and 5G.
So 5G as it stands might offer higher throughput in the lab, but real-world results are far off.
5G also typically has a range of ~800m at the licensed frequencies. This means more transmitters. Each transmitter can connect to a group of remote devices. With more transmitters there is thus a possibility to connect more devices to the backhaul networks than with 4G and its relatively sparse 10km range between backhaul access points.
This means that over a given area the relative "contention" (the number of users sharing a connection) on a given backhaul fibre is potentially lower. So more users can stream per mast at a given time. But to be honest, 4G is not struggling today. Consumers may demand lower contention ratios if their signal is always buffering, but most buffering is happening because of the distance from a mast, rather than because of the contention for the backhaul. Also note that backhaul is nearly always dark fibre, and so capacity is "easy" to increase at the contention point.
If you talk to executives across the mobile telecoms space, they will confirm that the contention is not on the backhaul, but on the radio spectrum. Only so many users can share that 2/5 /10Ghz carrier. Think about the comparable domestic wi-fi channels, typically 11 of them. You can't really fit more in without spreading out from your licensed frequency too far and straying into unlicensed territory. That's bad for a telco.
Even if these issues were not a problem, how many people download movies these days? And a perfectly watchable stream is viewable on a phone at 5Mbps. We already have that on 4G. So let's take the promise of speed being a benefit to the streaming consumer with a pinch of salt.
5G will definitely have benefits to CDN operators in that a new rollout offers the opportunity to install flexible computing architecture distributed across the network much deeper into the edge, following modern distributed compute architecture that is evolving to be knows as "fog" architecture as it escapes the public clouds into the telco infrastructure.
And that is cool.
But it could be done in 4G too, if anyone could raise investment to do so in a market drunk on 5G kool-aid. Following the same logic, it is difficult for network operators to invest in ubiquitous 4G rural access when they are being pushed by their shareholders to have a 5G story in the densely populated metropolitan networks.
So it boils down to latency. Latency is a hot topic in the streaming sector. Barely a conference goes by that's not dominated by the topic. Why has this happened? Because the CDNs decided to move the narrative to latency and away from price because to not do so was creating a painfully crippling battle on their margins. So they adopted a mindset of "Let's fix the price at $0.001/GB and fight over millisecond latency instead," and that's what the CDNs have done for the past decade.
This despite the fact that 1ms difference in a video stream that is typically buffered for 2s, 4s, or even 30s makes no difference.
But let me back up a little. We need to step back from the streaming space, and even back from the CDNs - the petri dishes in which streaming is evolved - and look at the carrier and telco strategy that underpin these network applications.
Telco strategy can be split into two groups:
- Industrial Automation
- Information Society
Let us first look at what I include in these two groups.
We have seen the prefixes "cyber," "e-," and "i-" in front of nearly every new generative wave of technologies over the past few decades. Today's trending prefix is "smart."
Here in the UK we are building smart motorways that can support IoT devices that will utilise 5G networks to deliver autonomous driving and better monitoring of the roads.
We see endless discussion about smart cities, every one of which is predicated by a 5G network.
Likewise we see smart factories, smart delivery by drone, etc.
"Smart" has (for those who know the term "5G" to be a bit more than a news story about Huawei) become synonymous with 5G in common vernacular.
These applications indisputably benefit from many of the capabilities that 5G brings: Lower contention, more devices connected in concentrated spaces. On top of this the fact that the architecture welcomes edge compute "in" the factory or "at" the roadside means that ultra low latency is no longer a thing of dreams. In a factory that outputs 1600 bottles of cola in a minute, being able to make a computational decision in 10ms rather than 100ms means a tenfold reduction in waste. Why wouldn't you?
Even more impressive is that if you can similarly make a braking decision for a car, or a "stop'' decision for a drone flight 90ms faster, then you are reducing real risk against life.
That is awesome, and if we go down that path, I get it. But these are NOT consumer-facing streaming applications.
Let us now look at the other group that i split my telco/carrier strategy focus into.
I use this group to collect information that is exchanged by humans. It is the very reason that the telcos and carriers have seen unprecedented growth since the late 1800s.
To further collect the groups of applications that drive this network growth, I am going to use a very simple model that looks at the applications that run on the networks with regard to delivery of information. These applications are the very capabilities that the content delivery networks provide.
Every CDN product or service will fit into one of the following models:
- One-to-one: For example a telegram, a phone call, a fax, or a BBS connection.
- One-to-many: TV, radio and satellite (and most cable) broadcast networks, OTT services, webcasts, webpages, RSS feeds, and so on.
- Many-to-many: Consumers may stream multiple services at the same time, some over the same access network, but most of these services will arrive with audiences from many different networks behind our access network; it may be that audiences watch TV over the air, stream OTT on their DSL, and refresh Twitter over their wi-fi all at the same time.
- Many-to-one: This is futurist, but can be seen emerging in aggregator MCRs, NOCs, and playout facilities, and on a micro scale on a video vision mixer: many images arrive at one location at the same time.
So how applicable is 5G to each of these models?
I would argue that throughput is a commercial issue, and there are a number of ways to address this. 5G is not a panacea for the reasons I outlined at the start of the article.
Edge compute is useful to the CDN and the content publisher, but in those models I would argue that the operation of the business is a type of industrial automation, and so the edge compute benefits are largely helping the telco and carrier or CDN rather than the consumer (beyond facilitating the delivery of the content at a high quality).
Density of devices really doesn't add up. 5G solves a problem we don't really have—wherever there will be good 5G connectivity the consumer will usually be able to watch a stream on a fixed wired/fibre-connected device. Obviously there will be folks who are mobile who want to stream, but I am far from a believer that there are people on the commuter train wishing they could stream more. The issue is access, not throughput, and again I have spoken about access and how much investment 5G has to make to even dream of parity with 4G let alone surpass it. I do not consider access or density of device connection and resulting contention to be a USP for 5G as far as streaming is concerned.
In the case of one-to-one, the only frame of reference is the remote user. Within 250ms a normal voice conversation can easily be managed, and it is tolerable up to 650ms. Getting down below the 120ms that we usually see over IP end-to-end is really not going to make much difference, and the highly regarded 10ms that 5G offers is really really not worth the effort—particularly when point-to-point streams may be buffered for a few hundred ms anyway before they are decoded. So one to one information society communication is NOT latency sensitive to the extent that it will benefit notably from 5G.
In the case of one-to-many, likewise we do not know nor care about the latency. Typically these are broadcast or multicast applications, or application layer multicasts such as CDNs or P2P networks offer. For decades, consumers didn't really know that there was a 3- to 10-second delay between the filming of a live sports stream and seeing the image on thier TV. Why? Because they only had one source and one stream globally, so they all got the same, and there was no other frame of reference.
The next model, many-to-many, is roughly where we are in the consumer market today: I can watch a sport stream on my phone and on a cable TV service at the same time. Since there are now many sources (reaching many users) it is easy for me to be a reference and to see that one stream is far behind another. My awareness now makes me want a parity that the traditional TV market never had to worry about: As a consumer I know that my OTT stream is behind and so I want lower latency. Will that hinder growth? Perhaps. Certainly latency is a big topic today because of this very issue. But is it a showstopper for a consumer? Does it really matter? Have we nearly moved things into an acceptable range on all current IP models (4G / DSL etc)? Frankly, yes. Just read up on all the recent developments in the streaming sector, and we are ultimately far past the point where the latency is good enough for companies like ESPN and DAZN and FUBO to generate huge subscriber bases. Will 5G make a difference to their subscribers? I would venture the difference wouldn't be enough for these operators to demand 5G is rolled out faster.
So then we have to look at the last model: many-to-one. Where many streaming signals are aggregated and need to be frame synchronous - such as in playout or in vision mixing - there is a real need for massive throughput and low latency. So yes for production and monitoring purposes there is an argument for 5G. But show me a playout facility that will drop it's fibre connectivity to the internet and replace it with 5G! Perhaps a group of cellmux linked outside broadcast camera teams may benefit from the higher throughput, although they have been getting away with it well on 4G for a good 15 years! And even if this demand was made it would barely be a consumer scale market - these types of operation are few and far between. Even TikTok and Facebook Live streamers are really one to one streaming to a CDN endpoint. The consumer side is finding the current status quo more than good enough. I don't see the demand for upstream streaming or multiple stream aggregation being a killer app for a 5G network investor looking for scale.
- One-to-one: Not latency sensitive. No real value from 5G.
- One-to-many: Not latency sensitive. No real value from 5G.
- Many-to-many: somewhat latency sensitive but ‘current' is ‘good enough' - no real value from 5G.
- Many-to-one: Yes latency and throughput sensitive and will directly benefit from 5G, although a tiny "production-centric'' niche use case, and so limited to industry that it really should be in the industrial automation group and not in the general consumer centric information society group.
There you have it: Smart society will demand 5G to build smart factories that will product smart IoT devices to power the smart road, smart home, and smart deliveries.
But smart society's human information consumers will barely see any benefit from 5G. If 60 to 70% of all network traffic is now video, then two-thirds of all network traffic will see little or no benefit whatsoever.
So don't believe the hype when you read 5G means better streaming for the average consumer who is being asked to pay for it. It doesn't. Articles inferring that streaming will be better with 5G are purely hyping the consumer into paying for a network that is actually mainly going to benefit a few corporations who drive industrial automation.
In my personal streaming-centric opinion it would be far better to invest in ubiquity and fog capabilities in the 4G network, and improve the availability of services like LTE-B and ATSC 3.0 to ensure that the 4G networks reach everywhere and deliver efficiently.
InterDigital, one of the chief architects of generational telecoms developments, says true immersive video experiences are a bust at 5G
An ambitious demo for IBC's Showcase brought together multiple broadcasters and vendors to demonstrate 5G's potential for remote production, but showed that it's a long way from ready for primetime
Speeds of 4.3Gbps could download an hour-long 4K episode in just 14 seconds; the old record was 2.92Gbps