0
$\begingroup$

I want to know the amplitude of the first-order hold filter at the Nyquist frequency (the roll-off amplitude/gain).

I know that the Fourier transform of the reconstruction is given by:

$$\sum^\infty_{k=-\infty}F\left(\nu - \frac{k}{T}\right)\ \mathrm{sinc}^2(\nu T)$$

which corresponds to low-pass filtering of the spectrum in the frequency domain by a $\mathrm{sinc}^2$ function. This causes some aliasing due to the sinc's sidelobes. However at the Nyquist frequency $\nu = 1/2T$, approximately we have the amplitude:

$$\mathrm{sinc}^2\left(\frac{T}{2T}\right)\approx0.4053$$

Is this close to the correct value? Also, how should I display this gain as a dB reading? MATLAB's mag2db function gives -7.8, not sure if that's right.

$\endgroup$
  • 2
    $\begingroup$ @OlliNiemitalo: I guess the OP means a triangular impulse response for linear interpolation. $\endgroup$ – Matt L. Apr 10 '17 at 14:11
  • $\begingroup$ @MattL. Thanks, that's right. I forgot it's called first-order hold, and was thinking of zeroth order hold with a rectangular impulse response. $\endgroup$ – Olli Niemitalo Apr 10 '17 at 14:44
  • $\begingroup$ @OlliNiemitalo: yes, there are even n'th order holds (and some controversy about them), as discussed in this question. $\endgroup$ – Matt L. Apr 10 '17 at 15:04
  • $\begingroup$ the only controversy is simply about what the definitions are of these $n$th-order holds. at Wikipedia i have seen 3 different definitions for 1st-order hold. the last one seems a little goofy to me. $\endgroup$ – robert bristow-johnson Apr 11 '17 at 6:18
3
$\begingroup$

That's correct, $0.4053$ is the approximate magnitude of the frequency response of linear interpolation at the Nyquist frequency. With the definition:

$$\operatorname{sinc}(x) = \frac{\sin(\pi x)}{\pi x}$$

the exact value is:

$$\operatorname{sinc}^2\left(\frac{1}{2}\right)= \left(\frac{\sin\left(\frac{\pi}{2}\right)}{\frac{\pi}{2}}\right)^2 = \left(\frac{1}{\frac{\pi}{2}}\right)^2 = \left(\frac{2}{\pi}\right)^2 = \frac{4}{\pi^2}$$

or in dB:

$$20\log_{10}\left(\frac{4}{\pi^2}\right) = 20\frac{\log\left(\frac{4}{\pi^2}\right)}{\log(10)}\approx-7.844795081\text{ dB}$$

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.