Is it better to use a crowded 2.4GHz Wi-Fi channel 1, 6, 11 or "unused" 3, 4, 8, or 9?

Cisco tested this. The result is that if you use an overlapping channel (anything other than 1,6,11), you get terrible performance and you make everybody else's performance worse. The problem is that any time an AP on the overlapping channel broadcasts, you get stepped on. And because the channels overlap rather than coincide, other network's transmissions are seen as noise, not signal, and don't trigger the bandwidth sharing built into the design.

Non-overlapping channels (1,6,11) work better than overlapping channels. With overlapping channels, you step on each other and can't do anything about it. With non-overlapping channels, you see each other and share the bandwidth.

For more recent devices, your best option is to get to the 5 Ghz spectrum, especially if all of your equipment can support 802.11ac or newer. Soon, we will also have 6 Ghz spectrum to play with.

But for the question as it relates to the 2.4 Ghz band:

Stick to 1, 6, or 11!

The thing to understand is the channels are only 5 Mhz wide. Channel 1, for example, centers at 2412 Mhz, and Channel 2 centers at 2417 Mhz... only 5 Mhz later. But wifi uses at least 20 Mhz of spectrum. So a wifi radio using 20 Mhz centered on Channel 1 will have signal going up to 2422 Mhz, well into Channel 3. A wifi radio centered on Channel 6 (2437 Mhz) will range down to 2426 Mhz, below Channel 4, and as high as 2448 Mhz, past Channel 8.

And that assumes only 20 Mhz channel sizes. 40 Mhz is also common in the 2.4 Ghz range. If you're using 40 Mhz channels (or greater), things are limited even more.

Best results come when your wifi signals do not overlap, and using only 1, 6, 11 with 20 Mhz channels gives the maximum potential. This is especially true in high-density areas, such as large apartment buildings, so for best results get your neighbors to do the same. Note that 20 Mhz channel widths will reduce the maximum theoretical speed, but it makes it more likely to have consistent reliable throughput, especially if your neighbors are on board.

Of course, if you live by yourself in the middle of nowhere, feel free to run a single radio using 80 Mhz signaling on whatever channel you want.

Even if other channels seem less crowded, remember that because channels overlap you still have to deal with interference from those busier channels as well. Your "clearer" channels will still have interference originating from the busy channels, so there is little to gain. What happens when you put your system in between two of the "standard" channels is that now you get interference from both of them. So, if you were to use, say, channel 3, you might now get interference from radios on both channel 1 and radios on channel 6 (and everything in between). More than that, you will yourself now cause interference with people using both of those channels. Whenever that happens, those other users will have to re-transmit their message, making the wireless signal in your area even busier.

There are a few studies indicating that, under the right circumstances, it may be possible to get more throughput using a four-channel scheme (such as 1,4,7,11, 1,4,8,11, or 1,5,8,11). However, for this work everyone in your area would have to agree on it. Until you can get everyone cooperating on that scheme, you will get best results by using the least busy of 1,6, or 11. Even then, this was only shown to help for certain kinds of loads and densities.

Finally, be careful when deciding which of 1,6, or 11 is least busy. Tools like InSSIDer will not help you here. They will only show you which neighbors have the strongest signal available on which channels, based on beacons from the access points/routers. They will not tell you how much those neighbors are using the signal. If you have someone next door with a strong access point on channel six, but they hardly ever use it, and other neighbors down the way with weak access points on channels one and eleven, but they use them to work from home and are on them all the time, you may be better off using channel six, even though it might look "bigger" in a tool like InSSIDer.

So how can you know which channel is least busy? This article on the serverfault blog may help:

http://blog.serverfault.com/2012/01/05/a-studied-approach-at-wifi-part-2/

It's the 2nd part of two part series, but the first part is less important to this discussion. The main thing is they recommend a tool called Vistumbler that will allow you see not just signal strength, but also actual traffic. It's takes a bit of doing, but you can use this to really know, not just guess, what channel is typically least busy in your area.

The proof of the pudding is in the eating!

1-6-11 is often worse in moderately congested areas

The 1-6-11 recommendation contained in Cisco's whitepaper about IEEE 802.11 deployment in the corporate environment certainly does not apply to all circumstances, especially in non-corporate settings! For example, in moderately congested neighbourhoods, one stands a very good chance to benefit from not sticking to this proposed scheme. So, don't be a monkey and consider this:

• First, note that the signal of a device on a partially overlapping channel is merely noise to the device on the overlapped channel. This is entirely intentional by design. The technique employed by 802.11b is called spread spectrum, or rather direct-sequence spread spectrum (DSSS) to be precise. 802.11g circumvents in-channel noise through orthogonal frequency-division multiplexing (OFDM) of a multitude of narrow (hence slow but more reliable) carriers.
• However, the situation usually gets worse when one voluntary abides to the 1-6-11 non-overlapping channel scheme. Doing so will expose your devices to the IEEE 802.11 RTS/CTS/ACK (Request to Send / Clear to Send / Acknowledge) of alien devices, effectively silencing your devices and hence forcedly lowering your bandwidth. This problem is known as the exposed node problem. In a corporate setting this problem can be solved by synchronising the nodes. In the wild, this is not readily achievable.
• In the end, Shannon's theorem is what dictates the maximum achievable information transfer rate of a channel in function of the noise level on that channel.
• Your antenna might provide more gain on certain channels and/or in certain directions, both greatly affecting your signal-to-noise ratio.

Hence, I call for actually measuring one's own signal-to-noise level. On a busy time of the day, try a number of seemingly quiet channels in between the most busy channels and far away from the strongest alien signals.

On a GNU/Linux system you can list all access points seen by your WLAN device as follows:

sudo iwlist wlan0 scan

Your own network will also be listed with a Quality value, roughly proportional to the signal-to-noise ratio. Try to maximise this value by changing channels and/or improving your base station antenna gain in your direction (e.g. by using a sector antenna at the edge of your home). Note that antennas often provide a little bit less gain at the band edges (channels 1 & 13/14). Maximum Quality is what you are looking for. The Quality value takes into account noise from overlapping channels.

Channel:3
Frequency:2.422 GHz (Channel 3)
Quality=70/70  Signal level=-40 dBm

If 2.4GHz is too crowded, you may consider falling back to RTS/CTS/ACK channel sharing in the 1-6-11 scheme. Still better; do yourself a favour and upgrade your devices to 5GHz. Much more bandwidth is available on 5GHz and overlapping does not exist.

The important lesson here is: bandwidth is a finite resource. It is especially scarce on the lower frequency bands (2.4GHz). Like with any scarce resource in life, there are only a limited number of possible approaches, listed here using metaphors:

• The 1-6-11 non-overlapping channel scheme would be the equivalent of a state-sanctioned communist planned economy (i.e. all too often like internal corporate culture).
• Signal-to-noise optimisation is blatant libertarianism and probably more efficient.
• And migrating to 5GHz should than be something like... colonising Mars.