I'll try to give you a good example:
Assume the following baseband stream:
1,1,-1,-1,1,1,1,-1,1,-1,-1,-1,-1,1,-1,-1,-1,1,…
We split this main stream to sub-signals. The serial to parallel conversion is given in the following table, every sub-channel will have lesser bandwidth (1/4 of main stream Bandwidth).
I will arrange the stream within sub-carriers as:
C1 C2 C3 C4
1 1 -1 -1
1 1 1 -1
1 -1 -1 -1
-1 1 -1 -1
-1 1 1 -1
-1 -1 1 1
Assume the carrier frequency C1 = 1 Hz, the harmonics are C2=2 Hz, C3 = 3 Hz and C4 = 4 Hz.
Now we multiply every row by the vector [cos(2*pi*1*t) cos(2*pi*2*t) cos(2*pi*3*t) cos(2*pi*4*t)]T to obtain an OFDM signal. Right ?
The row 1 of the table represents the amplitudes of a certain range of sinusoids because they'll be multiplied by sinusoids to obtain OFDM! Thus, the IFFT would retrieve a time domain signal (the OFDM signal). For example, at the first N instants, we capture the amplitude of a low frequency (C1) and a higher freq C2, C3 and so on to compose an OFDM signal.
Every row in the table can be considered as a spectrum !. Each row has only 4 frequencies because we multiply every element by a different sinusoid/carrier . Actually those rows are not spectra, they are extracted from a time domain baseband signal but the IFFT is a mathematical concept that doesn’t care what goes in and out, does it ?
Each of these rows can be converted to a time domain signal (OFDM) through IFFT, I know that this row is originally N adjacent baseband samples, but they are equal to the spectrum of the desired OFDM.Thus, the input to IFFT is a time domain signal disguising as spectrum.
Hope you got it
:|