Back to future with an article from Airheads Online, dated July 22nd 2009, by Peter Thornycroft (aka. Antar). Enjoy the humor :)
Elizabeth:
Dear David, the subject for today is whether a four-channel plan at 2.4 GHz is a good idea. I hope you have done your homework?
David:
No homework necessary, a one-pipe problem in Sherlock Holmes’ terms. This is such an easy one to answer: if three channels good, four channels must be better. No need for much further discussion on that although, dear colleague, you will undoubtedly find some obscure points to argue. So first, let me explain why we should care. There’s a limited amount of spectrum available, compared to the number of people wanting to use it, and this is the way to ensure fair access for different users and applications, while preventing different devices in the same spectrum from interfering with each other…
Elizabeth:
Spare me the blather, dear David, our readers don’t need two pages of your pontificating on regulation and spectrum sharing. If you want to do that, get your own solo column.
David:
Pontificating is a harsh term, dear colleague. Surely our esteemed readers hang on my every word? But I’ll take your advice and cut to the chase. While the 5 GHz spectrum available to Wi-Fi extends to over 20 non-overlapping channels (for 20-MHz channels, it would be less for 802.11n at 40MHz), there is only about 80 MHz allocated in the 2.4 GHz band. Take out some ‘guard band’ space and we can only fit three non-overlapping channels in this band.
Why would three channels be restricting? In a word, co-channel interference. When we design enterprise Wi-Fi networks, we need to put access points quite close together to make sure the coverage offers high capacity and that there aren’t any dead spots. An enterprise WLAN is a bit of a jungle – it’s not uncommon for a client to hear 15 or more access points at a given location. And since those access points only have three channels to use, there have to be a good number of them on the same channel. So this Wi-Fi jungle is incredibly noisy; you can hear all kinds of creatures calling to each other, both near and far. The chimpanzees call to other chimps, monkeys to other monkeys and the parakeets communicate among themselves. With all the calls in the air, it’s difficult for a humble Morton Stanley to distinguish the calls directed at him from those to the other animals.
Elizabeth:
If this is an attempt to surpass the cocktail party analogy of co-channel interference, it’s ineffective. And surely you know that simian similes aren’t cricket?
David:
Quite so, quite so. Now, in Wi-Fi terms, co-channel interference is the normal condition and over-riding capacity limitation in an enterprise WLAN. The 802.11 CSMA/CA contention algorithm handles this very well in practice, as any station wanting to transmit has to wait until the air is clear, regardless of whether other stations are on its own access point or on others. But it can reduce capacity by slowing down
Elizabeth:
You may be right at that – I’ve run site surveys where upwards of 50 SSIDs are audible through most of a building... that would include multiple SSIDs per radio and two radios per access point, but it’s still a lot. A significant conclusion, then, is that one should reduce the number of access points transmitting on each channel, and the way to do that is to move to a four-channel plan in 2.4 GHz. There – I give you credit for a clever idea, and I saved our esteemed readers another two pages of your painfully slow but logical explanation of open and hidden nodes, collision domains and their ilk.
In simple terms, if a building has 20 access points, we could put seven on each channel with a three-channel plan, but reduce that to five access points per channel if we go to four channels. And since all the devices on a channel are effectively sharing the same bandwidth (if they’re close enough to hear each other), we squeeze out an extra 20-30% of throughput across the network.
David:
There you are, dear colleague, in a nutshell, cut and dried. Four channels are better than three.
Elizabeth:
Now, where are your caveats?
David:
Nothing to caveat: four channels good, three channels bad. Would you like to hear a farmyard metaphor for Wi-Fi?
Elizabeth:
Our esteemed readers have no time for your rambling, David. Neither do I. Now, let’s look at where a four channel plan could get us into trouble. First, there’s a reason Wi-Fi channels use a guard band. The spectral mask for a transmission doesn’t have barn-door sides, but rolls off – steeply but not vertically. With a guard band, the signal from transmissions in an adjacent channel will be low enough to avoid causing interference, but as you move to a plan that uses channels 1, 4, 8, 11 rather than 1, 6, 11, the channel spacing is 15MHz less, so there’s a considerable amount of overlap.
The immediate result of the increased interference from adjacent channels is that in many cases a transmission in, say, channel 8 will interfere with one in channel 11 to the extent that the wanted transmission is corrupted. This effect is statistical; it gets worse as the spectral overlap increases and it depends on the relative transmit power of the different transmissions and the network and client topology.
David:
But the frame will go through, eventually? Wi-Fi frames are all ACKed, and if an ACK is missed, there is a retransmission of the original frame.
Elizabeth:
And that’s the point. Increased retransmissions mean the data throughput goes down. So while the increased number of channels increases data capacity, the higher retransmission rate acts in the opposite direction.
And before you point this out, there’s a different mechanism involved with simultaneous transmissions in overlapping channels, rather than the same channel. Let’s consider identical network topology and client distribution, but in the first case the access points are on a three channel plan. On a given channel, all devices will be able to decode Wi-Fi frame preambles from other devices, so the 802.11 CSMA/CA contention algorithm is used… when a device senses another transmission, it uses the random backoff timers to delay its next attempt. In this way it avoids transmitting over the top of the other frame. Although there’s some dead air time when nothing is transmitted, because the backoff is sometimes longer than necessary, it’s quite efficient in using available bandwidth.
Now, take the same access points and distribute them across four channels, with a considerable overlap. Simultaneous transmissions on adjacent channels will cause interference. Depending on the relative power levels at the intended receiving devices, one or both frames will be corrupted and not correctly received. Note the difference between this and the three channel case: here, a lost frame is detected when the transmitter receives no ACK, and subsequently retransmits the frame, usually at a reduced modulation rate (taking longer on the air). Thus every collision results in at least one extra (re)transmitted frame. If we had a three-channel plan, collisions are avoided by the CSMA/CA protocol, and the backoff algorithm takes care of interleaving frames from different devices on the air.
It’s difficult to quantify the difference – but my friend Ken says these effects tend to cancel out, so there’s no net benefit.
David:
That’s deep, and I’m always suspicious of complicated explanations. But I can’t fault the reasoning, dear colleague. And Ken’s seen a lot of three and four-channel plans in action.
Elizabeth:
But there’s more. A four channel plan is a ‘bad neighbor’. If other enterprise WLANs are near enough to hear your signals, a four-channel plan gives interference on their network in a way they will find difficult to detect, if they are set up with three channels, as are most networks. Overlapping a four-channel plan with a three-channel, you end up with a mix of channels 1, 4, 6, 8 and 11. The maximum overlap is between 4 and 6, or 6 and 8, where the effect of a transmission on one channel on the next is quite severe – worse than within a four-channel plan, where the worst overlap is 1 to 4 or 8 to 11. And out-of-channel interference is one of the more difficult impairments for Wi-Fi access points to detect.
David:
But, dear colleague, in Wi-Fi everyone is entitled to set up their equipment according to the rules, and that includes using any of the designated channels – 11of them in the U.S.?
Elizabeth:
I’m not saying it’s illegal to use a four-channel plan near others, or anything like that, but it could cause your neighbors some trouble – and of course their networks will affect yours in the same way. Perhaps that’s a better way to look at it – if there are three-channel plans in the vicinity, it’s not a good idea to use four.
David:
That makes more sense. Now, what else does your friend Ken say? He seems to know what he is talking about.
Elizabeth:
I’m glad you asked, dear David, because there’s one more effect I find quite interesting. Recall that the four channel plan includes asymmetric neighbor overlap: for instance, channel 4 overlaps on one side with channel 1 and on the other with channel 8, while channel 1 has only one neighbor, channel 4. This means that – with an even distribution of access points and client devices over the four channels - there will be more adjacent channel interference on 4 and 8 than on 1 and 11. So the optimum distribution for data capacity, given that more channel overlap means more retransmissions, is to redistribute traffic so slightly more channels use channels 1 and 11.
Now, most enterprise WLANs using automatic RF planning algorithms of varying sophistication. Ken has observed that his algorithms take account of this asymmetric overlap effect by setting more access points to channel 1 and 11 than 4 and 8, whereas in a three-channel plan they tend to be evenly distributed. This is a result of an ‘interference’ measurement by the access points that is included in the algorithm, and while it results in an uneven channel distribution, it is an optimum solution in terms of WLAN data capacity.
David:
Dear colleague, we promised to spare our esteemed readers both impenetrable jargon and fractured meter. ‘Asymmetric neighbor overlap’ takes the biscuit. But it’s a fascinating observation, supporting evidence for the interference theory.
While I have the floor, let me sum up the arguments. I say that using four channels provides more data capacity in the network, because fewer devices share each channel, so each can transmit more. This still seems a simple and logical argument.
But you pointed out that when adjacent channels overlap, there will be more frame corruption due to other Wi-Fi transmissions; although it’s statistical, of course. And because the corruption-retransmission effect is more wasteful of airtime than CSMA/CA collision detection-backoff, this effect subtracts from the capacity gains we would have expected from using more channels. Ken’s observations support this theory, but it’s difficult to quantify.
And there’s the secondary effect where overlapping networks using three and four-channel plans can cause each other a considerable amount of grief. So I think you have convinced be that there’s little, if any benefit to network data capacity from moving to a four-channel plan.
Elizabeth:
A brilliant summary, David, that might have been more entertaining than re-runs of the Goon Show.
David:
Needle nardle noo, it was the dreaded batter pudding hurler of Bexhill on Sea. Or was that Captain Heath?
Elizabeth:
We must move on, swiftly, but here’s a last thought. While we have focused on the four channel plan for capacity improvement, concluding that it probably doesn’t help, there may be other benefits. For instance, I’ve been wondering if Wi-Fi phones in a lightly loaded network might be able to make better handover decisions when they see fewer access points per channel? I suspect not, and as load increases, the extra retransmissions in a four-channel network will probably be bad for voice traffic, but it might be worth some further testing. The fundamental conundrum is that in a lightly-loaded four-channel WLAN, each client gets more bandwidth. But if all clients use that bandwidth, the increased interference effect cancels out any overall capacity gains.
David:
Have a gorilla.