December 2009 Archives

I purchased a D-Link DWA-556 PCI-Express card. This is a pretty standard draft-N wireless device. Once it showed up, I threw it in a computer and tried to get it to function as an access point.

It turns out that 802.11n is a bit more complicated than "standard wireless" (802.11a/b/g). Two of the biggest changes from previous wifi standards would be the addition of multiple input/output streams and the ability to operate with channels 40MHz wide, as opposed to the current 20MHz.

Now, 802.11n runs over the exact same frequencies as 802.11a/b/g. This has some pretty obvious implications on existing access points when you both double the channel width and starting using multiplexing streams on the same channel.

Probably the biggest impact to existing networks is the addition of the 40MHz wide channels. Existing 802.11b/g has 11, 13, or 14 channels depending on your region. Each channel is 22Hz wide, but are only 5MHz apart from each other. This means in every country but Japan, there are a grand total of three channels that don't overlap (1, 6, and 11 -- Japan also has 14 that does not interfere with 11). Now, strictly speaking, 802.11n doesn't have to use a 40MHz wide channel. It can sit perfectly content with a 20MHz wide channel. The obvious advantage to running in the 40MHz mode is double the bandwidth as compared to 20MHz. Twice the channel width, twice the bandwidth. The obvious disadvantage is that instead of bleeding over into 4 other channels (channel 1 bleeds over to channels 2-5) you're munching on 8 channels minimum.

To be precise, the 40MHz mode of operation uses two distinct 20MHz channels of operation in order to provide backwards compatibility with clients that are not 802.11n aware/capable. The second channel is used for clients who can use the 40MHz features only, and is placed +/- 20MHz away from the main channel. If you instructed your card to operate on channel 1 in 40MHz mode, it would also allocate channel 5 for use as well. If you selected channel 11, it would use channel 7 as well.

The obvious problem here is that we can't have any more than one 40MHz 802.11n access point in b/g spectrum without overlapping channels. If you're living in an apartment complex that is already covered with wifi, one single individual is going to be able to take up to 82% of the total spectrum by running one single access point on "one" channel.

Further, 802.11n gets a fair bit more range than 802.11b/g. While existing wireless gets about 150 feet/45 meters indoors and 300 feet/90 meters outdoors from the AP, 802.11n gets about 300/70 meters feet indoors and 250 meters/820 feet outdoors. Not only can an 802.11n signal in the b/g spectrum take 80%+ of the available spectrum, it's going to "pollute" twice the distance at a minimum.

The other big impact on existing networks is the addition of MIMO streams over a single channel. While this has less of an impact than 40MHz channels, it will still be felt, and even more so in 40MHz mode.

Multiple input/multiple output, or MIMO, multiplexes multiple streams over the same channel. This is the primary change from 802.11a/b/g, and it manifests itself as increased noise on existing connections. Much in the way that having 6 access points on the same channel in the same corner of a room degrades your ability to use even one, MIMO throws more data over the same frequency resulting in more noise -- lots more noise -- for anyone that isn't you (or is you, if you're not the one running the 802.11n access point).

Now, 802.11n supports up to four streams being multiplexed over one channel. The most advanced 802.11n hardware I can currently find maxes out at 2, which means roughly double the noise for anyone who isn't you.

Stream count is adjusted largely by the MCS index. The MCS index is a number, 0 through 31, which tells the hardware primarily ultimately what bitrate it is going to be capable of. MCS 0 has a rate of 6.5Mbit on a 20MHz channel, on up to 15Mbit with a 40MHz channel. Two streams starts at MCS 8, which provides 13Mbit at 20Mhz up through 30Mbit at 40MHz. it should be noted, however, that MCS 7 provides 65Mbit at 20Mhz and 150Mbit at 40MHz, which is still more than our aging 54Mbit, even using the standard 20MHz channels. MCS 12 is where two streams will beat out single streams, providing 78Mbit/180Mbit. 

Most of the (current) higher end hardware supports MCS 0 through MCS 15. MCS 15 provides 130Mbit at 20MHz and 300Mbit at 40MHz. The maximum that 802.11n provides is MCS 31, which provides 4 streams at 260Mbit/600Mbit.

As usual, wikipedia has a great table with all of the stream count/channel width/bitrate possibilities and the relevant MCS index.

That is a lot of noise for other clients and a lot of spectrum taken. That's also a lot of potential speed, given the right circumstances.

The 5GHz frequency can solve a lot of the above issues. As mentioned, three channels is not a lot of channels, and 802.11n over the 2.4GHz band can take two of them at once. The 5GHz frequency, or 802.11a, provides 20 non-overlapping channels in the United States. This is enough for 10 802.11n access points in the same area with 0 signal overlap (11 total if you add one in the 2.4GHz range). This is much, much more acceptable. In addition, the increased transmission power that 802.11n provides alleviates the primary complaint about 802.11a: extremely limited range. While 802.11b/g provides up to 150 feet/45 meters indoors, 802.11a maxes out around 50 feet/15 meters indoors (and 100 feet/30 meters indoors, ouch!). This is primarily due to the different frequency -- 5GHz as opposed to 2.4GHz. However, the 5GHz spectrum is rarely used in general because of this.

This leaves a lot of people in a bad place with regards to 802.11n. On one hand, using any new features of 802.11n is going to muck up the 2.4GHz band very rapidly. On the other, the 5GHz band is pretty much perfect for 802.11n operation, but range limited. 

The "recommended" course of action is to place standard 802.11b/g traffic in the 2.4GHz band and start operating 802.11n traffic in the 5GHz band. This provides the maximum amount of backwards compatibility with existing clients (by not changing a thing) while still using 802.11n features with newer clients in the clean part of the spectrum.

In the end, I'm not sure what I'll do. Unless the ath9k driver is buggy right now and the newegg page for the D-Link DWA-556 is lying to me, it does not support the 5GHz frequency, and hence is quite useless to me as far as 802.11n is concerned.