0

Okay, so I am aware of the math behind perfect reconstruction in filterbank and wavelet theory when there is half-band symmetry. But I don't understand, when we have the computational power to do decent DSP to modulate the binary signal into the baseband IQ signal, why anyone puts up with any crossover from one FDM channel into another.

Consider this figure from Zhao 2014: DFT-based offset-QAM OFDM for optical communications:

frequency division multiplexing

Why use any other model than the Nyquist FDM model? Make maximum use of your channel's bandwidth and keep them other channels from fucking with the channel that your receiver will be demodulating.

Conceptually, the math is simple enough:

The discrete-time bipolar binary signal is

$$\begin{align} x[n] &\triangleq -1 + 2 a[n] \quad \in \{-1, 1 \} \\ &= -(-1)^{a[n]} \\ \end{align}$$

where $a[n] \in \{$0, 1$\}$ is our serial bit stream. And the baseband quadrature signals are:

$$\begin{align} i[n] &= \sum\limits_{m=-\infty}^{\infty} x[2m] \, p[n-2m] \\ \\ q[n] &= \sum\limits_{m=-\infty}^{\infty} x[2m+1] \, p[n-(2m+1)] \\ \end{align}$$

where

$$ p[n] = \operatorname{sinc}\left( \tfrac{n}{2} \right) w[n] $$

and

$$ \operatorname{sinc}(u) \triangleq \begin{cases} \frac{\sin(\pi u)}{\pi u} \qquad & u \ne 0 \\ 1 & u = 0 \\ \end{cases}$$

and you have a decent window function $w[n]$ such as a Kaiser:

$$ w[n] \triangleq \begin{cases} \frac{1}{J_0(\beta)} J_0\left(\beta \sqrt{1 - \left(\frac{n}{1+M/2}\right)^2} \right) \quad \quad & |n| \le \tfrac{M}{2} \\ \\ 0 & |n| > \tfrac{M}{2} \\ \end{cases} $$

$J_0(\cdot)$ is the 0th-order modified Bessel function of the first kind.

$$ J_0(u) = 1 \ + \ \sum\limits_{k=1}^{\infty} \frac{1}{(k!)^2} \left(-\frac{u^2}{4}\right)^{k} $$

$M+1$ is the number of non-zero samples or FIR taps, which gets you a finite number of terms in the summation:

$$\begin{align} i[n] &= \sum\limits_{m=\lfloor n/2-M/4 \rfloor}^{\lfloor n/2+M/4 \rfloor} x[2m] \, \operatorname{sinc}\left( \tfrac{n-2m}{2} \right) w[n-2m] \\ \\ q[n] &= \sum\limits_{m=\lfloor (n-1)/2-M/4 \rfloor}^{\lfloor (n-1)/2+M/4 \rfloor} x[2m+1] \, \operatorname{sinc}\left( \tfrac{n-(2m+1)}{2} \right) w[n-(2m+1)] \\ \end{align}$$

$i[n]$ would be sampled at every even $n$ and $q[n]$ would be sampled at every odd $n$.

This puts in a little constant delay of $\frac{M}{2}$ samples to do a nice sharply defined band for each OQPSK channel. Why would that be a problem, considering all of the other latency you will have in the signal chain? Why allow any significant leakages from adjacent channels? (Of course there should be a little bit of guard band between adjacent channels.) Why put up with any overlap?

robert bristow-johnson
  • 20,661
  • 4
  • 38
  • 76
  • 1
    Work out the bandwidth efficiency in bits/Hz for both cases and therein lies your answer (going away for a week starting tomorrow but if not done by the time I get back I will work through it analytically) – Dan Boschen Jul 17 '21 at 03:20
  • 1
    The Kaiser window that you proposed, particularly for large values of $\beta$, can indeed yield very low sidelobes in the FFT per-bin frequency response. As with any window function, however, low sidelobes come at a cost of a wider main lobe. So, you actually get more overlapping with adjacent subcarriers if you were to window the transmitted signal. The sidelobes fall down much more quickly, so there will be less interference between far-away subcarriers, but much more interference between subcarriers near each other. – Jason R Jul 17 '21 at 15:38
  • remember @JasonR that the spectrum of the window is convolved with the spectrum of the sinc function (which is a rectangular function in the frequency domain). so that main lobe is translated into the transition band between the passband and the stopband. and with larger $M$ that gets narrower and narrower. my thesis here is to choose a good $\beta$, maybe about 5 or 6 but it might be more, and a good large $M$ (maybe 16 or 32) to make a totally solid rectangular spectrum (at half width of Nyquist), no matter what the data is. – robert bristow-johnson Jul 17 '21 at 16:29
  • @DanBoschen i'll be looking forward to your analysis. because, using the channel capacity theorem, i can't think of a better way to pack in the bits of information than to pack them into a rectangular bandwidth with constant amplitude (assuming that the noise floor will also be constant amplitude). – robert bristow-johnson Jul 17 '21 at 16:31
  • 1
    Kaiser window will increase the main lobe at the benefit of low sidelobes but nothing beats the resolution of a non-windowed (rectangular window) function. It’s the fact that the signals are uncorrelated if the bin spacing is 1/T that makes it work so well. When the signal is considered over the duration T, there is no “overlap”— the overlap you see is what would occur if there was a frequency offset – Dan Boschen Jul 17 '21 at 16:43
  • @DanBoschen, i can make $M$ as big as i want, within reason. – robert bristow-johnson Jul 17 '21 at 16:52
  • 1
    In the rectangular window DFT, each bin is an independent symbol with no cross symbol interference. As soon as you apply any other window, you get leakage from one bin to the next and therefore one symbol to the next. Hence the decreased spectral efficiency. – Dan Boschen Jul 17 '21 at 16:57
  • @DanBoschen , you got this wrong. In the time-domain, it is strictly a windowed sinc() function. *None* of the other even-indexed bits will contaminate $x[2m]$ and none of the other odd-indexed bits will contaminate $x[2m+1]$. and since only the even bits go into $i[n]$ and only the odd bits go into $q[n]$, there is no ISI as long as we keep this thing phase linear and sample the received $i[n]$ precisely at even-indexed times and sample the received $q[n]$ only at the odd-indexed times. – robert bristow-johnson Jul 17 '21 at 17:08
  • I see, I’ll hold off until I can actually analyze it. By “wrong” I assume you are referring to my statement that any other window will cause bin leakage and yes i see similar to RC pulse shaping we can avoid ISI in the other domain. What isn’t wrong is my first statement so I need to see if my bandwidth efficiency hypothesis is correct (considering bandwidth efficiency at a given bits/symbol and SNR) – Dan Boschen Jul 17 '21 at 17:16
  • @DanBoschen, do you mean frequency bin leakage? if that's what you mean, the closer to a true rect() function, the better. that's one reason i am going for the Kaiser-windowed sinc(): to reduce frequency-bin leakage. the other reason for a windowed sinc is that at the sampling instances, there is no ISI. – robert bristow-johnson Jul 17 '21 at 17:19
  • Yes - Rectangular window in time so Sinc in freq (Dirichlet Kernel actually) with null at each other bin center so no leakage when we maintain that orthogonality, as in “OFDM”. I believe that any other window will widen the main lobe so don’t see how you can avoid adjacent bin leakage —- but let me study it once I can if someone else doesn’t do it first. I think the full analysis should consider/include the achieved spectral efficiency in bits/Hz given a consistent bits/symbol (so symbol rate per Hz would be fine) and if there are any power penalties affecting SNR (as a Kaiser window also has) – Dan Boschen Jul 17 '21 at 17:23
  • @DanBoschen any window times a sinc will still cross zero at integer times (except for t=0). and remember we can make $M$ quite large and that will make the rect function in the frequency domain very rectangular. – robert bristow-johnson Jul 17 '21 at 17:30
  • Definitely interesting. I don’t think that is true? Won’t the rectangular function in frequency that you approximate with your truncated Sinc get smeared by the Kernel of the Kaiser window since it would be a convolution in frequency of those two responses? – Dan Boschen Jul 17 '21 at 17:44
  • (I think I need to digest your full analysis ans what I am thinking of before commenting further) – Dan Boschen Jul 17 '21 at 17:51
  • (But if you drop the Kaiser windowing then it should end up being the same whether you decide to do Sinc functions in time or Sinc functions in frequency —- you have to overlap in one domain or the other- but by choosing the overlap in time you reduce the overall bandwidth due to the end bins which is perhaps your point overall—- so OFDM with raised cosine pulse shaping. – Dan Boschen Jul 17 '21 at 18:05
  • @DanBoschen you cannot drop the window function because we cannot add up an infinite number of terms. windowing is unavoidable. so let's use the best window, for our purposes, that we can. – robert bristow-johnson Jul 17 '21 at 18:06
  • and i am saying that this Kaiser-windowed sinc is better than raised cosine pulse shaping. – robert bristow-johnson Jul 17 '21 at 18:08
  • 1
    Yes agreed - Kaiser better and DPSS based such as this even better (barely). https://dsp.stackexchange.com/questions/70516/use-of-the-harris-moerder-nyquist-pulse-shaping-filter So if the window is much wider than the Sinc main lobe the frequency spearing of that would be minimized which is perhaps what you are doing— I am looking forward to looking at it closer. – Dan Boschen Jul 17 '21 at 18:22

0 Answers0