You will note that I said ?should? and not ?will?. Nearly all channelized systems have some form of leakage from adjacent and sometimes non adjacent channels. The relative amount of interference is what concerns us here. More details on the actual detailed processes carried out by the Wi-Fi Alliance etc can be obtained from them.
Test procedures involve packet error rates whose ?acceptable values? are highly dependent on the applications in use. For example comparing voice and video.
It will be interesting to see the spectral responses plotted on a spectrum analyzer for all the devices used in your tests, along with ranges and S/N values etc.
From the spectral plots ( and using appropriate filtering so as to minimize the noise floor thermal noise effects etc ) taken close to the clients as well as the APs, we should be able to see what is happening. Without the spectral masks, it will be difficult to make any conclusions.
If you don?t mind me asking, what brand of spectrum analyzer did you use to record the spectral output in your last tests ?
Bye the way, one of the reasons that I'm asking about the Brand/Model number of the analyzer is that in over 25 years of using and teaching how to use physical and "software" spectrum analyzers, I have been amazed at how many do not have a warning of uncalibration re the relationship between sweep time and resolution bandwidth. As you know, sweep time is inversely proportional to the square of the resolution bandwidth. Some devices do not have an "auto" feature to lock the two together. I have known several cases where clients have sent me plots showing some very unusual displays. I then discovered that the sweep time was not appropriate for the particular resolution bandwidth utilized. Once that was corrected, the displays were totally different. Most modern analyzers of high quality have taken care of that. The other issue is that of resolution bandwidth itself whereby some analyzers lack the granularity to allow the "detailed features" of a spectral mask to be clearly displayed. Due mainly to cost, some use a minimum value of RBW in Hz that will mask adjacent components when the filter sweeps across the visible display. This is one of the reasons why Intelsat, the Wi-Fi alliance etc spec out particular values of RBW, VBW, span, averaging etc, in order to allow "apples" to be compared with "apples".
Nice blog by the way.
Most documentation with respect to adjacent channel issues tends to focus mostly on the transmit spectral masks ( and with good reason ). Another area that is often neglected are the actual receive response characteristics of the devices. In radio design, the overall noise figure and hence final S/N are highly dependent on the input filter characteristics. This is a difficult area. The filter should be narrow enough that the wanted signal is still allowed to pass through, but narrow enough that unwanted energy ( from adjacent channels etc ) is rejected. You must also allow a certain amount of "leeway" to account for the fact that the transmit carrier frequency from the other end will have some amount of variation. The more expensive the electronics, the better ( usually ) the stability. The filter edges have to be carefully shaped as rapid rates of change of phase with respect to the slope of the filter characteristic are very important ( all 802.11 modulation schemes involve phase changes, either alone or in combination with amplitude shifting as in QAM ).
Better ( read expensive ) systems will have sophisticated filters that acheive the balance mentioned above.
What starts off as a relatively easy looking issue can become quite complex when we start taking all the factors into account.
The testing organizations such as the Wi-Fi alliance also make partial assumptions of a fairly balanced link budget. In other words, if we have a situation whereby we have a tiny signal at the edge of the coverage area, and another "booming" signal adjacent to it, that situation will be different from a normally balanced setup.
The Wi-Fi alliance are unable to test every single possible RF scenario, but they have done their best in showing that under a fairly balanced system with good quality receivers that the adjacent channel interference present in 5GHz links should not be a major degradation factor due to spectral width alone.
Even with the famous 1, 6, and 11 channels of 2.4 GHz, sidebands can be seen to "spread" from these channels into the others. However, under normal circumstances, with a well site surveyed system, this should not be a show stopper ( assuming of course that APs and clients are not "one on top of the other", no neighbours blasting away illegally etc ).
Could we design a system where there was no chance of adjacent channel interference ? Yes, we could, but then we would have to cut down dramatically on the number of channels available, with all the implications that would have on roaming etc.
In terms of adjacent channel interference under the scenarios mentioned above, the 5GHz channelization plan seen on the UNII bands etc is a good one.
Compromise has to be acheived in all areas of Wi-Fi design and in other areas such as VOIP. The larger the frequency range desired ( audio frequencies ) in order to pass an inteligible coversation, the greater must be the bandwidth ( in terms of bits/second ) required. A compromise was reached in the original G.711 specification which gave us 64 kbit/s of bandwidth ( I hate using the term bandwidth for bits/s. It is wrong. Bandwidth refers to the difference in frequency from lower to upper edges of a range. However the term "Bandwidth" when refering to bit rate has become commonplace, so when in Rome ...). Nowadays, with compression using thing such as Code-Excited Linear Prediction ( utilizing codebook entries ) etc we still have reduced bandwidth rates, but still have a compromise with signal quality and bandwidth.
Whenever I see cases of problems with adjacent channel interference, I usually check to see whether APs are "one on top of the other" etc in terms of spatial separation etc. If there is interference from other systems ( such as a neighbor ), then all bets are off, as that is one of the disadvantages of the unlicensed spectrum. You can perform a beautifully carried out site survey and set up your APs well with good results, then your next door neighbour blasts out on an adjacent channel.
However, there are a few possible issues with the devices themselves. Is there a bad batch of amplifiers with poor spectral regrowth characteristics that has gotten through the net ( has happened with some manufacturers ) etc. The only to determine that ( as best you can without an RF lab setup ) is to check the output spectral mask against the appropriate standard. More tricky than it appears. The old way was with a marker pen on the screen. Nowadays, plots can be produced from many spec ans and physical marking with pen and ruler can be used. If you're really lucky and get your hands on a device that shows the filter mask on the printout, then that is the best option. Rare though. Noise spectral density tests should be performed ( dBm/Hz ) of the noise floor, first with all devices off ( hopefully no interference....if there is, move the system for the test period ) then gradually switching devices on and off in order.
Spectral regrowth occurs rapidly beyond a certain point with many devices.
Another area that can affect the ability of the receiver to reject interference ( in general ) is whether we are using single conversion or superheterodyne double downconversion methods prior to demodulation. The latter ( more expensive ) gives us much greater image frequency rejection, where the image frequency is separated from the wanted frequency by twice the IF.
One of the best tools in the battle "against" ACI is a good site survey and a good site survey/predictive survey program. Physical changes can and do occur in a real world environment. A "rerun" is often a good thing every now and then. Calibrated spectrum analyzer plots acting like a baseline at the startup of the project can be worth their weight in gold when problems occur later on. Sadly, there are many cases when this is not done.
As a summary:
Going back to the original issue of 40 MHz operation, even with a well conducted site survey, operating two 40 MHz systems at 2.4 GHz is just not practical. The 5 GHz band gives us much more ?space? in which to plan. Depending upon which country you are in, up to 20 channels can be available. Having two 40 MHz systems operating in the 5 GHz band is highly feasible.
One of the main goals of any site survey/predictive analysis at 5 GHz is to try to minimize the effects of co and adajacent channel interference. (The term adjacent channel does not necessarily have to refer to the physical channel directly beside the one we are interested in, but may be one or more away, there being no term for ?the second to furthest away, the third from furthest away etc?. ).
In an environment where there are no neighbours, it should be perfectly possible to arrange APs such that the effects of adjacent channel interference are minimized. It must be remembered ( very important point ) that the sidebands which are further away from the center frequency, in most cases are going to be ?buried? under the noise floor anyway. If you have a large amount of adjacent channel interference, the first thing to do is to check the spacing etc between APs. If you space them well and still have problems, there is the possibility ( although highly rare nowadays ) that there is a faulty device or devices. Spectral regrowth can occur under abnormally high power levels. We?re ignoring automatic power control systems at this point.
If you take two 5 GHz APs and a spec an, and put them very close to each other, you will probably be able to see the effects of adjacent channel interference. However, one of the purposes of the site survey is to place the devices so that doesn?t happen. If you cannot place them any farther apart, then altering the transmit power should help. Another goal is to minimize the number of APs ( and hence the cost of the network ).
Taking a look at P 209 of the CWNA book, we can see that if we choose the two end channels of the of say the UNII 3 band for 20 MHZ operation, we should have no overlap at all. With a proper site survey, it is perfectly feasible to have 5 GHz channels directly adjacent to one another.
Again, if there are problems with ACI ( not from a neighbour ), usually spacing/power issues are present.
Refer to the following:
There is more spacing with two 802.11n 40MHz BW channels than for three 802.11g 22MHz BW channels.
The frequency separation between channel 1 and 6 is 3 MHz.
The frequency separation between channel 1 + 5 and 11 + 15 is more than 20MHz.
Recall 22MHz channels are +-11MHz, but 40MHz channels are not +-22MHz, but instead plus 30MHz minus 10MHz (or maybe it's +33MHz -11MHz) either way you get the point.
Is this correct?
You have hit upon one of the biggest problems in representing a Wi-Fi signal by means of a diagram. The ?real shape? of a transmitted Wi-Fi signal can be seen with a spectrum analyzer. If devices are not Wi-Fi certified or certified via a local or governmental agency etc., then the output spectral mask of one device by one manufacturer could vary from that of another. The IEEE 802.11 ? 2007 specification gives us spectral masks which should be adhered to.
If we look at the spec or in the CWNA book, we can see that the shapes are more complex than the pictures given in the Wiki illustrations. ?The Wiki? is a great thing. Sometimes, however, illustrations are simplified. The one given is a bit misleading. In other words it is too simplistic. When we look at the diagrams in the IEEE docs and the CWNA book, we can see multiple sidebands. Do we see these sidebands all the time with a spectrum analyzer ? The answer is not necessarily. It depends on a multitude of factors such as the power being emitted by the device, how far away it is from the spec an etc.
Think of the island in ?Lost?.
We can see certain features. However, we know that there is more of the Island below the surface. It?s just that we can?t see it. It probably spreads out horizontally. The part below the surface is obscured by the water level. If the island were to start to rise up above the water level, we would see more of the parts that ?spread out? below the water surface.
So it is with a W-Fi signal. Imagine the Wi-Fi signal as being the island, and the noise floor as being the water around the island. We know that there is more to the island, but we just can?t see it under the ?noise/water?. If nasty Ben or someone lifted the island upwards, we could see more. Conversely, we could let the water level drop. We can drop the noise level, but it involves literally physically cooling the receive amplifier using liquid helium. Not very practical, but still done by NASA on the tracking stations for their deep space missions.
As we move away from the island ( think of when Sawyer was offshore??if you?ve not seen Lost, please do so immediately and without hesitation !!!...it?s good stuff and sadly missed ), the island looks smaller and smaller. As we move away from an AP etc, the Wi-Fi signal appears smaller and smaller and the sidebands seem to ?disappear? in the noise.
In a nutshell, I would only use the ?official? illustrations in the IEEE docs or the CWNP books as a guide. Dual 40 MHZ 2.4 GHz does not work well at all. If you have neigbours ( home or enterprise ), 40 MHz at 2.4 GHz 20 MHz or 40 MHz is not a good idea.