“Back in the days of analogue TV,” we'll one day tell the children, “we used to curse 'ghosts' in the image – but that was before we found out how to use the ghosts to carry information.”
Welcome to the emerging world of MIMO – multiple in, multiple out – wireless systems, where the same phenomenon that causes ghosting in analogue TVs is being used to expand the wireless channel and deliver higher throughput. What was once a fairly obscure study in mathematics is becoming part of the consumer and business landscape in both cellular and LAN applications.
The emerging 802.11n standard is today's case in point: part of the speed improvement over previous WiFi standards comes from 802.11n's use of MIMO as a “spatial multiplexer”, and since this is key to getting the most out of 802.11n, it's worth getting a basic understanding of the use it makes of MIMO.
But let's start with the standard itself, first.
The Timeline, Today
The IEEE is still some way from finally ratifying 802.11n. Although products are now shipping under Draft 2.0 of the proposed standard, the IEEE's own timeline gives the end of 2009 as the current target for completion of the standardisation process. Products on sale at the moment represent a bet (albeit a good bet, vendors tell us) on what the final standard will look like. Vendors believe that any changes between now and ratification will be able to be accommodated in software alone.
At its best, 802.11n represents a big jump in speed. The signalling rate on the air interface – which should not be confused with the effective throughput – is a maximum 248 Mbps (the theoretical 'best case' is even higher than this).
But as with all networking technologies, there are caveats to be considered.
The first is that the top signalling rate needs a 40 MHz radio channel, achieved in current products by 'bonding' two 20 MHz channels together. If 802.11n products are used in 20 MHz mode, the signalling rate is limited to 124 Mbps.
And the real throughput – the data rate available for “real” data transfer – is far lower than the signalling rate. This isn't unusal: you only need to look at the difference between signalling rate and throughput for all previous WiFi standards; or, going back further in history, the difference between old-fashioned shared Ethernet's 10 Mbps signalling rate and the effective maximum throughput of around 4 Mbps in a real-world network, since as usage rose, so did collissions.
So the “real” numbers for 802.11n are 60 Mbps peak with a 20 MHz radio channel, and 90 Mbps with a 40 MHz radio channel.
Since each base station has to share its capacity among the active clients, and since the peak data rate is subject to how much noise is on the channel, these numbers represent the best-case throughput.
And, as I mentioned earlier, 802.11n is dependent on the MIMO concept.
Types of MIMO
MIMO refers to the use of multiple radio paths between transmitter and receiver, and can be used in various ways.
Each application of MIMO depends on both transmitter and receiver being able to use multiple antennas so they can “see” multiple radio paths. When signals are reflected around different objects (for example, off buildings in outdoor applications, or walls when indoors), they take slightly different times to make their trips (the phenomenon that causes 'ghosting' in a TV signal).
Normally, this is treated as noise by the receiver – and by a simple application of Shannon's Law, “noise” reduces the amount of channel space available to carry data. The theory behind MIMO is that the existence of multiple radio paths can be used to improve performance, instead of degrading it.
The simplest application is to improve the reliability of a channel, by allowing the transmitter and receiver to choose the radio path offering the best performance. In this scenario, the performance benefit derives only from the ability to use the lowest-noise radio channel – all others are ignored.
A second possibility is to use multiple radio paths to transmit the same information.
Here, the use of different paths means that a noise event will affect different parts of the same signal – and that means you can reconstruct the original signal by correlating the different received signals (this is illustrated below).
This is useful, since it effectively reduces the noise of the radio channel (and therefore increases the capacity of the channel). However, it's processor-intensive, and that increases the latency of the radio system.
Then there's a third kind of MIMO – the kind that 802.11n uses. In the “spatial mulitplexing” used by 802.11n, the transmitter assigns different signals to different antennas whose signals travel different paths to the receiver. This way, the same block of spectrum can be used to carry as many streams as there are high-quality paths between the transmitter and reciever, thereby multiplying the effective bandwidth between the two (see Figure 2, below).
The problem with this is that MIMO is dependent on various reflections between transmitter and receiver. The system can only multiply the capacity by the number of spatial radio paths it's able to identify.
What most of us would think of as an “ideal” WiFi setup – where there's a nice clean path between the base station and the user – is not the ideal setup for 802.11n. If there's no reflections, there's no alternative paths, and performance will suffer.
Moreover, the alternative paths themselves may be subject to interruption and interference – and any time that happens, users will suffer performance drops.
There are swings and roundabouts here. Some of the factors interfering with 802.11n paths might be transient, and may only affect some of the alternative paths rather than the whole channel. The upside to this is that while there's a performance drop, there mightn't be a complete interruption of the transmission. If the system is smart enough, it will see the interruption and retransmit the data over a reduced channel – before something happens that the user notices (such as a VLAN session being dropped due to an interruption).
The downside, of course, is the loss of performance.
As long as the obstruction or interference is transient, this isn't critical. Intermittent ups-and-downs aren't the problem – but if a permanent obstruction gets in the way of one of the 802.11n paths, then the users will suffer permanently.
Previous WiFi versions suffered from obstructions, of course – but because people could assume that their WiFi needed a fairly clear path between (say) a laptop and a base station, they could understand that putting a metal structure between the two is a bad idea.
The mundane, day-to-day risk in 802.11n environments is that people won't understand how changes to office layout will affect the indirect paths that give MIMO the reflections it needs.
So people planning 802.11n coverage need to keep in mind that the rules for WiFi coverage no longer apply.
As is already well-known and thoroughly discussed, backwards compatibility means users will have to choose between performance and cost. A forklift upgrade will get the greatest speed benefit quickest – but most companies will probably choose to sacrifice the short-term performance uplift so as to maintain compatibility with existing 802.11g kit.
But even in an all-802.11n network, there's still the channel space to consider. With channel bonding switched on to provide maximum performance, 802.11n greatly reduces the number of radio channels each base station can use. Even with 20 MHz bands, the 2.4 GHz spectrum only leaves room for three non-overlapping channels, and with a 40 MHz band, there's only one channel available (per antenna).
To reduce the chance of collissions somewhat, the 802.11n specification reduces the overhead at the physical layer, with “frame aggregation” improving the overhead-to-data ratio. This means fewer conversations are needed to carry the same amount of data.
Intelligence in the systems – particularly in the antennas – will also help, but you're going to need intelligence in the deployment and coverage design.
But that's probably a story for another day.