• By the way, from your original question:

    "Each [of the 14 channels] is 22 MHz wide". In general, regardless of the encoding technology.

    This is classic example of the term "bandwidth" being applied in one particular set of circumstances. In this case, if we look at Figure 6.7 on P 207, we can see that they have defined "bandwidth" in this case as being the frequency range between the first nulls on the left and right of the main carrier "lobe". We have 11 MHz on each side, giving us a total "bandwidth" of 22 MHz.

    As regards channel separation, the IEEE has said that we should have fives time the "channel value" of separation in order to get "non-overlapping channels". I.e. five times 5 MHz, giving us 25 MHz of separation. Hence, channels 1, 6 and 11.

    The output spectrum remains the same, regardless of the encoding technology, as we are looking at symbol rates and not data rates per se in terms of the final output value from the modulator.

    If I get some time in the future, I'll put some info up about symbol rates. To make matters worse, the terms "modulation, spread spectrum encoding and error detection and correction coding" are often just lumped together as "coding" ( wrong ) in many documents.


  • Thanks to you all -especially Dave!
    I think I got it at last. One thing is actual bandwith... another is "regulated" channel separation.

    So, as per pic. 6.7 on the book, actual total bandwidth for HR-DSSS is roughly 90 MHz. However, mandated channel separation is 25 MHz. This implies that 20 MHZ (= 90/2 - 25) will be overlapping, yet its effect can be negliged. I suppose that actual bandwidth should be even greater for DSSS, but I do not have the pictures.

    On pic. 6.10, only 60 MHz are drawn for OFDM, but one can visually extrapolate it up to about 100 MHz. Since in this case channel separation must be 20 MHz, then 30 MHz ( = 100/2 - 20) will overlap. This is 10 MHz more than HR-DSSS!
    I assume then that OFDM is more robust against co-channel interference than HR/DSSS so even though overlapping is 50% greater ( (30-20) / 20), results are better.

  • One of the ways that I think about it is this ( and this is very rough, non-mathematical ): Imagine the spectra in the figures mentioned above are islands as in the picture mentioned previously:

    Now, we know that unless the island is like the one in ?Lost?, it must go all the way down to the bottom of the sea ( the bottom of the noise floor ). The water ( noise ) covers up the part that we cannot see. It would be logical to assume that the island spreads out ?side-ways?, becoming wider at the bottom than it is at the top. The figures basically show the islands after all the water has been drained away. In other words, they do not show the noise floor. This is for good reason, as the relative height of the signal with reference to the noise floor would depend a little on physical temperature ( a lot in outer space ), and a lot on distance from the source, Tx power etc. So, those figures are basically islands with the water drained. If there are any overlapping bumps between islands that are close together, below the water line, that really does not matter much to us, as ( going back to the real spectra ), if something is causing problems below the noise floor ( sea level ), it really doesn?t matter, as the ?damage? is already beeing done by the noise itself. You can imagine standing up to your waist in a freezing cold swimming pool. The top part of you is fairly warm, but the bottom part is freezing. Now, a scuba diver goes under the water, and takes a bucket of freezing water with him, with a magical lid on it. He goes up against you below the water, opens the lid and ?throws? the freezing water against your legs. You won?t feel any colder because the water level is already up to your waist. Any more ?interference?, such as more cold water underneath the main water level won?t matter, as the damage has already been done by the existing cold water.
    Now, if anyone can get some spectrum analyzer plots of a couple of channels close together, then we could see some of this.

    Any kind souls out there ?

    A piece of errata: I should have said -273 degrees Celcius earlier on. That corresponds to zero degrees Kelvin.


  • This analogy leads us to a very interesting question. Let's assume that SNR is 25 dB, minimum acceptable in most cases as stated on the book. We are using, for instance, a HR-DSSS 802.11b system. Fig 6.7 pg. 207 depicts a channel of these. First sideband lobe has its maximum at +11 MHz and is 30 dB lesser than main lobe.
    Since SNR=25 dB, [b]noise base level ("water" level) is 5 dB greater than first sideband lobe, noise "hides" it[/b]. In other words, it is unusable.
    [b]This[/b] is what makes me think how much of the bandwidth is needed (depending on the coding/modulation techniques) to really be able to "reconstruct" data (i.e, retrieving it).
    In this scenario, just a small portion of the whole bandwitdh power is usable, visually "integrating" the curve (with yields power out of spectrum), about maybe just 10% or so.

    [b]Unless[/b] in SNR calculations what is meant by Signal is NOT the peak of main lobe, but some sort of average (i.e, the integral) of the whole curve. I just ignore it, I could not find a proper definition of Signal in the book. In this case, what I just said would be all false and I apologise.

  • Subscripts and superscripts don't show up well here, as I've just found out. Anything with something like "fC" actually means "f subscript C"

    The power spectral density transfer curve is an important ?tool? in digital analysis:

    Gx(f) = T [ sin pi ( f ? fC ) T / pi ( f ? fC ) T ] 2

    Where Gx(f) = single sided pulse spectral density

    T = pulse duration

    fC = Carrier center frequency

    f = Frequency offset

    Due to the variation in shapes of many digital signals, special techniques can indeed applied to integrate the area under the curve. Normalization of power over an equivalent noise bandwidth can then occur. The actual pattern that is produced at the output of a particular filter is highly dependent on the filter?s transfer function. Filter design is a complex matter with a lot of trade ? offs. The rate of change of phase with respect to offset from the carrier center frequency is an important one, for example.

    We need to get into Fourier Analysis, Autocorrelation Functions etc for a full analysis. However, many defintions of bandwidth look at the main lobe ( i.e. between the two nulls ). In FCC documents, you often see occupied bandwidth defined as ?that which leaves 0.5% of the signal power above the upper band limit, and 0.5% of the signal below the lower band limit. Occupied bandwidth = 99% of total signal energy. Some system documents will use the 3 dB bandwidth values.

    I have made many measurements of S/N using a spectrum analyzer. It gives reasonable results for a rough estimate. Filtering has to be applied ( video and noise averaging ). A noise marker is used to give noise spectral density in dBm/Hz, N0. Then, the marker can be used to measure carrier spectral density ( C0 ) plus noise spectral density ( N0 ). A ratio can be obtained of carrier spectral density plus noise spectral density to noise spectral density ( carrier spectral density plus noise spectral density being an integrated entity that cannot be ?split up? into it?s component parts ). A number of correction factors are applied ( the resolution bandwidth of the filter being an important component ), and we can end up with a value of Eb/N0 ( energy per bit to noise spectral density ratio ). Modern analyzers can often do all this automatically.


  • After trawling through what seemed to be the entire contents of the Internet, finally came across the following document. It contains just about everything I was looking for. There is a lot of bandwidth related info. The spectrum analyzer traces were made using a quality spectrum analyzer.

    1. Note the difference in spectra between 802.11b and 802.11g signals.

    2. Note the mention of different bandwidths ( 6 dB in this case, and the FCC 99% value )

    3. Note the detail ?off to the side? shown in the figure on P 48


  • Great information! That shows exactly what I wanted!

Page 2 of 2