Consider the use of OFDM in 802.11a (5GHz band) as shown in this picture taken from NI's White Paper:
For a 20MHz channel with a frequency spacing of 0.3125MHz between the subcarriers, there are a total of 64 possible sub-channels. As per the standard, 52 subcarriers are used to transfer data, 4 subcarriers are used as pilot symbols, and the rest are 'null' subcarriers.
This brings us to the crux of my confusion/doubt. Are these remaining 8 subcarriers just 'null' or actually used as guard bands? As per the above picture, and CWNP's article as well, these tones are used as guard bands.
On the other hand, according to Rohde-Schwarz's White Paper and Revolution-Wifi's article, these bands are simply designated as null subcarriers.
Furthermore, if these subcarriers are indeed used as guard bands, is an additional guard band required? There's no mention about the requisite guard bandwidth in the IEEE 802.11 official standard documents.
Best Answer
The 'null subcarriers' are not driven, they have nominally zero energy in them, and are used as guard bands.
Guard bands have two uses. They protect users of adjacent systems from spillage, or spectrum spread, of our signal into their band. They protect us from the spillage of an adjacent system's signals into our band.
To protect adjacent systems, they are generated with no energy in them. Radio signals always spread somewhat due to various imperfections in the generation and transmission process. In the case of OFDM, the 64 carriers are described in the frequency domain, and then Inverse DFT'd into the time domain. The IDFT process means that all the carriers always 'exist'. Pilot subcarriers get a pre-determined description, data subcarriers get a data-dependent description, and the null subcarriers get zero. However limited numerical precision of the IDFT means there is some energy in those carriers after conversion, then signal chain analogue distortion lets a little more energy leak into them. The system has a specification for the minimum ratio of data to guard band energy that a compliant system will achieve.
Protection from the spread of adjacent systems is done by simply not attempting to use those carrier positions for data.
I think part of the confusion is that the word 'subcarrier' is used in slightly different senses in different sources. It's a definition of a frequency on the air interface, which is equivalent to an index into the signal vector prior to the IDFT. The use of the word 'tone' sounds a bit misleading, as it implies frequency and non-zero level, but in the context of a guard band, the level is always nominally zero, that is several 10s of dB lower than data or pilot tones.
The provision of guard bands is system specific. They are sufficient for adjacent 802.11a channels to co-exist. If however an operator with a system with a wider spread was to seek permission to operate adjacent to a block of 802 use, they would be required to keep out of the 802 allocation. How they did this would be up to them, but would probably implement sufficiently wide guard bands within their system to stay within their allocation.