I would have to disagree with some of the above.
Narrow RF channel yields lower speeds (lower capacity) plus tolerance of weaker signals. These are from the laws of physics.
Wider channels yield the opposite of the above.
Actually it's Shannon that discovered that - there's a limit named for him, and yes, the wider the spread of the channel, the more bits can be carried on that channel - never said it didn't
and outside of 802.11 - other access technologies have EC methods that ensure this is the case for any channel width - but the general rule is that wider channels do take more power to push the same number of bits...
In 802.11 space - the wider channels are limited by the same EC methods as the narrow channels.
I've seen real world use cases where a narrow channel in 802.11n at MCS 14 vastly outperform a wide 2 stream channel at MCS 12 or 13 - why? Less errors at a given range - the RF wasn't bad, but the noise killed the wider channel and higher frame and bit errors reduced the throughput at the network layer.
802.11b/g/n all use 20MHz channel width; 40MHz is new and not practical for 2.4GHz, growing use in 5.8Ghz. The net throughput of b is less than g is less than n, though all use a 20MHz channel. It's from "adaptive modulation" which means the speed in that 20MHz varies according to the two-way signal strengths. The techniques for adaptive modulation, and "diversity" vary among b/g/n. That is, how they use the fixed 20MHz differs.
Analogy: Weak/noisy spoken communcations, crowded room/long distance, you speak more slowly to be understood. That's adaptive.
802.11n (and 802.11ac) are much better at the near/far problem, and adaptive rate setting can be done with some chipsets on a client by client basis (and given that it's enabled in the driver).
We're singing the same tune here...
with 802.11n - I prefer throwing more streams at the problem first - better pre-coding at the AP, more EC, more actual "gain" due to the effect of MIMO precoding.
then throw bandwidth at it, but there's a point where things basically fall apart in the noise domain, and Tim's chart's show this in the 20/40MHz location values.
So the addendum to what I said earlier...
Coverage - narrow channels - better range, less bandwidth
Bandwidth - wide channels - less range, better bandwidth
and more streams are always better in both cases...
I'm trying to keep this simple, as it can and will get over the head of most folks pretty quickly, even though I think you and I could have an awesome conversation on this discussing the relative merits, and more often than not agree with each other
Wait until we see third gen 802.11ac chipsets - they're going to rock! Just like 802.11n as the silicon and drivers matured and more optional features were turned on.
sfx
