5

Typically windows are symmetrical and real. Are there any applications where complex windows have been used (other than applying the same real window to the real and imaginary components of a complex number).

A full complex window would have the sum and cross products between the real and imaginary terms as follows if applied as a complex conjugate product:

$$y_C[n] = (w_R+jw_I)(x_R+jx_I)^* = w_R x_R + w_I x_I + j(w_I x_R - w_R x_I)$$

What could be a possible benefit with this and is this actually used anywhere? Are there any types of signals where the typical considerations for windows (such as dynamic range versus resolution bandwidth) would be different for the real and imaginary components of the signal? Or thinking in the frequency domain, would there be any advantage or application to having an asymmetric Kernel?

Dan Boschen
  • 50,942
  • 2
  • 57
  • 135

1 Answers1

4

I can see one use case, where you'd absorb a frequency shift, i.e. a multiplication with an $e^{jfn}$ sequence, into the window. That would only be advantageous if the shift remains constant, and no phase continuity between separately windowed "frames" (or whatever you window) is needed.
Maybe some bursty FSK modem? Q-Tone modulations? OFDM with a differential PSK inside¹?

Other than that: There's GFDM systems, where the ambiguity function (Time/frequency plane of a single pulse in a multi-symbol multi-carrier frame, generalizing the OFDM frame) of a pulse only has nulls at every two subcarrier spacings, and at every two symbol durations. Through alternatingly only using real and imaginary parts of only half the subcarriers, but at a denser raster than the sinc shape of an equivalent OFDM system would allow, the orthogonality of symbols is preserved:

rhombus-shaped symbol grid From Koslowski, Sebastian: Synchronisation und Entzerrung in Filterbank-Multicarrier-Empfängern, dissertation , Karlsruher Institut für Technologie (KIT), 2018

Now, if I wanted to build something to detect symbol statistics at different subcarrier frequencies through a DFT of an undersampling of this waveform, I might want to build a window that "follows" the real and imaginary parts. But, honestly, this is as constructed as it gets.


¹ DAB(+) is an OFDM system which employs a 2048-FFT in a streaming system (not short bursty transmissions), but doesn't want to spend complexity on phase recovery, so it does differential QPSK on each subcarrier; that system complexity tradeoff never ceases to surprise me!

Marcus Müller
  • 30,525
  • 4
  • 34
  • 58
  • 2
    Ok that is excellent thinking with that frequency shift or then for that matter any sort of frequency counterpart to time domain equalization…. Thus leading to the more general case of “filtering” frequency domain waveforms with what would be a non-linear phase filter. I just don’t think of windowing as a frequency domain filter but that is exactly what it is (and a linear phase filter at that). – Dan Boschen Apr 14 '22 at 18:17
  • 1
    But this is good and insightful; I’ll select in a couple days if someone doesn’t come up with a case that is conceptually closer to the use and application of windowing as we know it. – Dan Boschen Apr 14 '22 at 18:18
  • 2
    I really hope someone does come up with more things! I was digging my memories whether I could find some Radar-related example, but none came up, although radar processing is a pretty windowingg-heavy field. – Marcus Müller Apr 14 '22 at 22:07