Many people say 2.4GHz is dead and whilst I view it as ‘best effort’ it isn’t going away and needs to be supported so I want it to be the best it can be. I use a four channel plan on our campus and my colleagues were recently informed this should never be done which inspired this post to explain the decision. This is quite lengthy but hopefully understandable. All comments and criticism welcome.
In the dim and distant past I, like many people, understood that there were 13 Wi-Fi channels available for me to use. I suspected having two APs on the same channel wasn’t a good idea, though I didn’t know how this actually worked, and so I reckoned I could deploy 13 access points all on separate channels. Great.
A tiny bit of reading later reveals there’s a good reason most Wi-Fi gear defaults to using channels 1, 6 or 11 on the 2.4GHz band. The 2.4GHz frequency space we’re allowed to use for Wi-fi is carved up into arbitrary channels of 5MHz bandwidth whereas 802.11b uses 22MHz wide channels. Because of this in order to keep the APs on separate channels that don’t overlap (which really is a good idea by the way) the IEEE says you need to have 25MHz bandwidth for a Wi-Fi channel.
The eagle eyed will spot there are actually 14 channels available. This is true, but only in some countries. In Europe we can use 1-13 whereas in the US only 1-11 are available. Only Japan gets to use channel 14, and then only for Direct Sequence Spread Spectrum radios or 802.11b…. old school.
As a result of all this is there are three non-overlapping channels available for Wi-Fi on 2.4GHz that satisfy the IEEE requirement of 25MHz bandwidth: 1, 6 & 11.
Why is this important?
If you’ve ever used a set of walkie talkies you know that you can be either talking or listening, not both. If several people are using the channel only one can talk at the same time, otherwise everyone else hears a garbled mess as a transmit collision occurs. Wi-Fi is pretty much the same. There’s a set of rules designed to avoid two stations talking at the same time and deal with re-transmitting data if that appears to have happened. A collision is assumed to have occurred if a station doesn’t receive an expected acknowledgement. Re-transmissions are costly and slow everything down. If it gets really bad a high rate of collisions can cause network throughput to drop so low that from a user perspective it isn’t working at all.
These rules mean all wi-fi devices within range of each other on the same channel have to share that channel space. The key takeaway from this is to remember the RF frequency space, the channel, is the contended resource. If you have 100 clients on a single AP using channel 1 and you add another AP also on channel 1 you may be able to spread those clients across the two APs but you probably won’t get better performance… in fact it might be worse.
Even though your additional AP is not increasing capacity at least everything should play fairly nicely. Stations transmitting data include duration information that every other station in range uses to set it’s NAV (network allocation vector) timer. Stations then consider the channel busy for this time. This is one part of the rule set for avoiding collisions known as Carrier Sense Multiple Access with Collision Avoidance or CSMA/CA.
Let’s say you do what I once did and put your second AP on channel 2. This is the worst of all worlds for Wi-Fi because these two channels mostly overlap. Stations on adjacent overlapping channels will often detect energy and hold off transmitting, but they won’t decode this as Wi-Fi traffic and understand the header information. The stations on channel 2 are no longer having their NAV timer set by transmissions on channel 1 and vice versa. This means collisions are much more likely to occur. Now, instead of adding capacity, you’ve probably made things far worse. Importantly it will work, especially when there are very few clients actively using the network, but as things get busy it will quickly fall apart.
So the above explains a tiny bit towards why putting APs on different, non-overlapping channels is important so how, when I’ve already stated there are only three non-overlapping channels, do I justify using four?
I work on a university campus where we have large buildings with dozens of APs per floor so we have to use channels more than once across the building. In theory this is straight forward. Use channel 1 for the first AP, 6 for the next, 11 for the next then by the time you use channel 1 again it’s far enough away from the first AP that devices are not in range of each other. In reality RF is messy and isn’t neatly contained in a room or floor of a building. With only three channels there’s going to be places where two APs are on the same channel and close enough that clients associated with each are in range of each other. Having four channels to play with doesn’t eliminate this but it certainly helps.
In essence there are two ways you can justify using overlapping channels. Firstly there’s a question as to just how much they overlap at all, and I’ll come to that in a moment, secondly if you space things out smartly there’s less chance of the overlapping portion of the channels causing a significant interference problem just because of attenuation of the signal through building materials and free space path loss.
DSSS, OFDM and channel plans
If you take a look at the 5GHz Wi-Fi channels you’ll see they’re 20MHz wide. Take note of that.
2.4GHz Wi-Fi before 802.11g uses DSSS (Direct Sequence Spread Spectrum) or HR-DSSS (High Rate). With 802.11g OFDM was introduced and brought with it the higher data rates enjoyed on 802.11a 5GHz W-iFi. DSSS uses 22MHz channel width. OFDM uses 20MHz channel width.
In other words if you disable the 802.11b legacy data rates of 1, 2, 5.5, 11 your 2.4GHz Wi-Fi is going to be mostly using OFDM with it’s 20MHz channel width. This means you can reasonably justify saying channels 1,5,9 & 13 don’t overlap providing you’re using OFDM.
Things are, of course, not quite so simple. Wi-Fi is very backwards compatible. That ancient laptop in the cupboard with an 802.11b PCMCIA card will still work with the latest dual band 802.11ac access point. You can disable support for the 802.11b data rates on your network but that doesn’t mean nothing is going to try talking with a 22MHz channel width from time to time and cause interference.
You can’t control other networks. I can tell my Wi-Fi to use channels 1,5,9 & 13 but I can’t control what anybody else does and neighbouring networks are probably using 1,6 & 11. This is an important point. Deviating from the industry norm is something you can do if you control the space around your network. As a university campus that’s physically separate from the rest of the city we don’t have lots of neighbouring Wi-Fi networks. What we do have is lots of people with hotspots enabled on their phone. I could consider these a problem but, to be honest, I just have to ignore them. Android phones will often select any old channel for the hotspot so they cause interference no matter what the enterprise network is doing.
Radios are not nice neat things. Even with the extremely sophisticated DSP of modern radios every radio transmission sends out power on the intended frequency, but also harmonics above and below that frequency. This means by taking away that 5MHz buffer of frequency space I increase the risk of interference.
The bottom line
I know that running a four channel plan will result in adjacent channel interference that might be avoided by sticking with 1, 6 and 11. However since moving to four channels on this network I’ve seen increased throughput (sadly I don’t have the figures for this), reduced user complaints and greater stability of the Wi-Fi (fewer power level changes with ARM) on 2.4GHz. It has demonstrably worked here. It certainly wouldn’t work everywhere. If you’re in a crowded space, it’s better if you play by the same rules as everyone else.