Originally published Jan 12th 2023 on my LinkedIn Newsletter - see here for comments
Sometimes, upgrading the network isn't the answer to every problem.
For as long as I can remember, the telecom industry has talked about quality-of-service, both on fixed and mobile networks. There has always been discussion around "fast lanes", "bit-rate guarantees" and more recently "network slicing". Videoconferencing and VoIP were touted as needing priority QoS, for instance.
Cloud gaming would need millisecond-level latency, connected cars would send terabytes of data across the network and so on.
We see it again today, with predictions for metaverse applications adding yet more zeroes - we'll have 8K screens in front of our eyes, running at 120 frames per second, with Gbps speeds and sub-millisecond latencies need to avoid nausea or other nasty effects. So we'll need 6G to be designed to cope.
The issue is that many in the network industry often don't realise that not every technical problem needs a network-based solution, with smarter core network policies and controls, or huge extra capacity over the radio-network (and the attendant extra spectrum and sites to go with it).
Often, there are other non-network solutions that achieve (roughly) the same effects and outcomes. There's a mix of approaches, each with different levels of sophistication and practicality. Some are elegant technical designs. Others are best described as "Heath Robinson" or "MacGyver" approaches, depending on which side of the Atlantic you live.
I think they can be classified into four groups:
- Software: Most obviously, a lot of data can be compressed. Buffers can be used to smooth out fluctuations. Clever techniques can correct for dropped or delayed packets. There's a lot more going on here though - some examples are described below.
- Hardware / physical: Some problems have a "real world" workaround. Sending someone a USB memory stick is a (high latency) alternative to sending large volumes of data across a network. Phones with dual SIM-slots (or, now, eSIM profiles) allow coverage gaps or excess costs to be arbitraged.
- Architectural: What's better? One expensive QoS-managed connection, or two cheaper unmanaged ones bonded together or used for diverse routing? The success of SDWAN provides a clue. Another example is the use of onboard compute (and Moore's Law) in vehicles, rather than processing telemetry data in the cloud or network-edge. In-built sound and image recognition in smart speakers or phones is a similar approach to distributed-compute architecture. That may have an extra benefit of privacy, too.
- Behavioural: The other set of workaround exploit human psychology. Setting expectations - or warning of possible glitches - is often preferable to fixing or apologising for problems after they occur. Skype was one of the first communications apps to warn of dodgy connections - and also had the ability to reconnect when the network performance improved. Compare that with a normal PSTN/VoLTE call drop - it might have network QoS, but if you lose signal in an elevator, you won't get a warning, apology or a simplified reconnection.
These aren't cure-alls. Obviously if you're running a factory, you'd prefer not to have the automation system cough politely and quietly tell you to expect some downtime because of a network issue. And we certainly *will* need more bandwidth for some future immersive experiences, especially for uplink video in mixed reality.
But recently I've come across a few examples of clever workarounds or hacks, that people in the network/telecom industry probably wouldn't have anticipated. They potentially reduce the opportunity for "monetised QoS", or reduce future network capacity or coverage requirements, by shifting the burden from traffic to something else.
The first example relates to the bandwidth needs for AR/VR/metaverse connectivity - although I first saw this mentioned in the context of videoconferencing a few years ago. It's called "foveated rendering". (The fovea is the most dense part of the eye's retina). In essence, it uses the in-built eye tracking in headsets or good quality cameras. The system know what part of a screen or virtual environment you are focusing on, and reduces the resolution or frame-rate of the other sections in your peripheral vision. Why waste compute or network capacity on large swathes of an image that you're not actually noticing?
I haven't seen many "metaverse bandwidth requirement" predictions take account of this. They all just count the pixels & frame rate and multiply up to the largest number - usually in the multi-Gbps range. Hey presto, a 6G use-case! But perhaps don't build your business case around it yet...
Network latency and jitter is another area where there are growing numbers of plausible workarounds. In theory, lots of applications such as gaming require low latency connections. But actually, they mostly require consistent and predictable but low-ish latency. A player needs to have a well-defined experience, and especially for multi-player games there needs to be fairness.
The gaming industry - and also other sectors including future metaverse apps - have created a suite of clever approaches to dealing with network issues, as well as more fundamental problems where some players are remote and there are hard speed-of-light constraints. They can monitor latency, and actually adjust and balance the lags experienced by participants, even if it means slowing some participants.
There are also numerous techniques for predicting or anticipating movements and actions, so network-delivered data might not be needed continually. AI software can basically "fill in the gaps", and even compensate for some sorts of errors if needed. Similar concepts are used for "packet loss concealment" in VoIP or video transmissions. Apps can even subtly speed up or slow down streams to allow people to "catch up" with each other, or have the same latency even when distributed across the world.
We can expect much more of this type of software-based mitigation of network flaws in future. We may even get to the point where sending full video/image data is unnecessary - maybe we just store a high-quality 3D image of someone's face and room (with lighting) and just send a few bytes describing what's happening. "Dean turned his head left by 23degrees, adopted a sarcastic expression and said 'who needs QoS and gigabit anyway?' A cloud outside the window cast a dramatic shadow half a second later". It's essentially a more sophisticated version of Siri + Instagram filters + ChatGPT. (Yes, I know I'm massively oversimplyifying, but you get the direction of travel here).
The last example is a bit more left-field. I did some work last year on wireless passenger connectivity on trains. There's a huge amount of complexity and technical effort being done on dedicated trackside wireless networks, improving MNO 5G coverage along railways, on-train repeaters for better signal and passenger Wi-Fi using multi-SIM (or even satellite) gateways. None of these are easy or cheap - the reality is that there will be a mix of dedicated and public network connectivity, with cities and rural areas getting different performance, and each generation of train having different systems. Worse, the coated windows of many new trains, needed for anti-glare and insulation, effectively act as Faraday cages, blocking outdoor/indoor wireless signals.
It's really hard to take existing rolling-stock out of service for complex retrofits, install anything along operational tracks / inside tunnels, and anything electronic like repeaters or new access points need a huge set of certifications and installation procedures.
So I was really surprised when I went to the TrainComms conference last year and heard three big train operators say they were looking at a new way to improve wireless performance for their passengers. Basically, someone very clever realised that it's possible to laser-etch the windows with a fine grid of lines - which makes them more transparent to 4G/5G, without changing the thermal or visual properties very much. And that can be done much more quickly and easily for in-service trains, one window at a time.
I have to say, I wasn't expecting a network QoS vs. Glazing Technology battle, and I suspect few others did either.
The story here is that while network upgrades and QoS are important, there are often highly inventive workarounds - and very motivated software, hardware and materials-science specialists hoping to solve the same problems via a different path.
Do you think a metaverse app developer would rather work on a cool "foveated rendering" approach, or deal with 800 sets of network APIs and telco lawyers to obtain QoS contracts instead? And how many team-building exercises just involve hiring a high-quality boat to go across a lake, rather than working out how to build rafts from barrels and planks?
We'll certainly need faster, more reliable, lower-latency networks. But we need to be aware that they're not the only source of solutions, and that payments and revenue uplift for network performance and QoS are not pre-ordained.