Monday, December 17, 2012

Difficult optimisation for 3G vs 4G vs WiFi

For the last few days I've been playing with an iPhone5 on EE's new LTE network in London. Rapidly launched after the regulator Ofcom controversially allowed EE (EverythingEverywhere) to refarm its 1800MHz 2G spectrum, it has allowed EE to steal a march on its UK rivals in launching a 4G service. 

(As backdrop - the UK auction of 2.6GHz spectrum has been plagued with delays for the past 5 years. I worked on a couple projects for UK government bodies looking at options for the band around 2007-8, and since then we've been stuck in an endless cycle of lawsuits, consultations and slow government action. EE short-circuited the endless loops by preempting the auction with refarmed 1800MHz. This has led to a rapid settling of differences by everyone else, and now a combined is 2600+800 sale is coming up soon).

My initial thoughts on UK LTE are pretty positive - using Speedtest, I've been getting 20-25Mbit/s down (up to 33Mbit/s at one point), and 13-20Mbit/s up, along with 40-50ms ping times. I've most noticed hugely faster email downloads, instant initial web connections and blazing-fast maps (obviously using the new Google Maps app, rather than the dreadful Apple version included on the iPhone).

However, I've also noticed pretty poor indoor penetration - weirdly, often getting a UMTS2100 connection, dropping back 3G, rather than the LTE1800. I assume that means that EE hasn't put LTE cells alongside all of its 3G cells in central London, as otherwise I would have expected the lower-frequency band to give better coverage, not worse.

That would also explain why part of EE's press announcement about build-out last week referred to densification of its network in existing covered markets.

In my view, that's absolutely essential. At the moment, the user experience is of blazing fast outdoor speeds, and much more variable indoor ones - often bumped down to 3G - which is just where people are much more likely to use video, m-commerce and so on. Outdoors, people mostly use maps, email, social network stuff and messaging. Maybe some content downloads or streaming on public transport. Clearly, having fast speeds where you don't really need them, and slow speeds when you do, is sub-optimal.

That has another implication - people indoors will still prefer WiFi if this situation is maintained. If I had a decently large monthly quote, and a good chance of having indoor LTE, I'd probably start switching back my usage from WiFi to cellular, unless I was doing something very data-intensive (in volume terms) indeed. At present, on an unloaded network, LTE gives better average performance than WiFi, which in turn gives better average than 3G. Obviously it depends a lot on venue - some WiFi networks are congested and almost useless, but overall I'd say 4G>WiFi>3G, which also reflects backhaul.

That's a problem for operators, who have to contend with user behaviour of choosing when/where to  use WiFi (no, it won't be "seamless" access to carrier WiFi, in 90% of cases). Without a suitably dense network or sub-1GHz bands, they risk failing to capitalise on LTE's speeds to change back user perception, of relatively desirability of cellular vs. private WiFi.

This also leads to a secondary problem: if MNOs are going to want to use WiFi for offload, or even as an extra value-added service, they're probably going to have to install fibre connections for fixed broadband to support it, as otherwise it's likely to offer worse performance than LTE. That's feasible (but expensive) for own-build networks, but likely to be near-impossible for roaming partners or venues that just operate their own broadband.

This suggests to me that connection management is going to become ever more complicated. Not only is there a WiFi/cellular problem to optimise, but it's also going to be dependent on local availability of 3G vs. 4G (which is driven by frequency & cell grid density) and possibly on small vs. macro cell availablity. And in additional factors such as application preferences and whether the user is actually mobile (ie moving about) and there is going to be an almost unsolvable level of complexity.

Imagine trying to optimise connection between:

- 3G small-cell coverage on 2100MHz
- LTE macro coverage on 1800MHz
- 3G macro coverage at 900MHz
- MNO WiFi from the small cell outside the building, with fast microwave backhaul
- 3rd-party WiFi + MNO roaming, but with a 10Mbit/s ADSL backhaul
- Another 3rd-party WiFi provided by the venue, with no MNO roaming or auto-authentication, but with fibre backhaul

Oh, and the user is trying to play gaming (for which latency is the #1 concern) and also downloading email attachments in the background (cost/speed as #1).

Then, just for the fun of it, imagine that you also have the option to run cellular+WiFi concurrently and either bond the two connections together, or split them entirely via the handset OS.

Sounds like a combinatorially-hard problem, not solved easily in the network, the OS or a single app. Should be a lot of opportunities for software and infrastructure vendors - but also some serious pitfalls such as the ill-fated ANDSF technology, which tries to make everything operator-driven. It does point to the huge value of sub-1GHz spectrum, though - which should make the UK Treasury happy about the 800MHz auction.

1 comment:

  1. I don't know anyone who would pay extra for 4G over 3G. How many people need fast data outside the home? When travelling outside towns it's a moot point anyway, as you can rarely get a connection. Inside towns, what are people going to be doing on their devices that needs 4G?

    My only issues with data on the move are (a) Outside town I can't get anything much. (b) When I can get a connection latency is pretty bad. Of these two, the former is the biggest problem, and 4G will do nothing to address that.

    I think it is nothing more than a slow evolution of quality, it certainly is no revolution, and I would rather we had a more ubiquitous network.

    The carriers all need to merge their masts to work on coverage and manage as virtual networks within that. If they had done that ten years ago we wouldn't have the crappy system we have now.

    I am not looking forward to another 10 years (yes I think it will be 10 years) of 4G rollout and adoption that will basically leave us in the same situation we are on now, only with faster speeds that we don't need and the same lack of coverage. The UK (European?) industry needs to get a better vision than this terribly short-sighted one it has now.