If you'd like to think analog, an OFDM signal can be written as a sum of weighted complex sinusoidals:
$$
x(t)=\sum_{k=0}^{N-1}X_k \exp{\left( \mathrm j 2\pi\frac{kf_\mathrm{s}}{N}t \tag{1}\right)},
$$
where $N$ is the number of subcarriers, $f_\mathrm{s}$ is the sampling frequency and $X_k$ denotes the subcarrier values. For a digital implementation, (1) is sampled at $t=n/f_\mathrm{s}$:
$$
x(n/f_\mathrm{s})=\sum_{k=0}^{N-1}X_k \exp{\left( \mathrm j 2\pi\frac{kn}{N} \tag{2}\right)},
$$
which corresponds to the inverse discrete Fourier transform (IDFT), implemented by an IFFT. The IFFT has $N$ input values and $N$ output values. That's ok, because the IDFT is N-periodic and thus any additional sample would be redundant. Consequently, every subcarrier in (1) is represented by $N$ samples, regardless of its number of periods during the OFDM symbol length.
If you try to imagine the digital OFDM time domain signal as a superposition of $N$ sampled sinusoidals, the figure you posted has two problems, in my opinion:
- The first subcarrier in the image obviously has frequency $f_\mathrm{s}/N$ and the fourth subcarrier has frequency $f_\mathrm{s}$. That's wrong. From (1) you can see that the first subcarrier should have frequency 0 and the fourth should have frequency $(N-1)f_\mathrm{s}/N$. You can reorder the subcarriers (e.g. from negative to positive) but the DC subcarrier must be included and there must not exist a subcarrier with frequency $f_\mathrm{s}$. (One could argue that subcarriers with frequency 0 and $f_\mathrm{s}$ are actually the same but I think it's a poor choice for an illustration)
- The image only shows the real (or imaginary) part of the sinusoidals. That could give you the impression that some aliasing is going on when sampling. I will explain in a second what I mean.
Please consider the following figure that I have created instead
It contains the subcarriers from $k=0$ to $k=3$ (lines) and their samples (dots). It also shows the real part (blue) and imaginary part (red). Now consider the red line in the first and the third subfigure: their samples are exactly the same. However, the blue (imaginary) part is different. That is, $N=4$ samples are sufficient here to represent all subcarriers.
Here is the Matlab code I used to create the above figure:
N = 4; % num subcarriers
os = 128; % oversampling
t = linspace(0, 1-1/(N*os), N*os); % time
n = 0:N-1; % discrete time
figure;
cnt = 1; % helper
for k = 0:N-1;
subcarrier = exp(1i * 2*pi*k*t);
subplot(N,1,cnt);
plot( t, real(subcarrier), 'b-');
hold on;
plot( t(n*os+1), real(subcarrier(n*os+1)), 'b.', 'MarkerSize', 20);
plot( t, imag(subcarrier), 'r--');
hold on;
plot( t(n*os+1), imag(subcarrier(n*os+1)), 'r.', 'MarkerSize', 20);
if(k ~= N-1)
set(gca, 'XTick', []);
else
xlabel('t/T');
end
ylim([-1 1]);
title( sprintf('k=%d', k));
cnt = cnt + 1;
end