I have the following frequency transform used to design a low-pass prototype out of a bandpass specification mask: $$\omega=\frac{\omega_0}{B}\left(\frac{\omega'}{\omega_0}-\frac{\omega_0}{\omega'}\right)$$ where \$\omega_0\$ and \$B\$ are constants, \$\omega\$ represents the angular frequency of the low-pass filter and \$\omega'\$ is the angular frequency of the bandpass filter. Since the transfer functions of the filters must be hermitic the function can be reduced to:$$\omega=\frac{\omega_0}{B}\left|\frac{\omega'}{\omega_0}-\frac{\omega_0}{\omega'}\right|$$

Given that \$\omega_0/B=2.45\$, I wonder if by solving for \$\omega'\$ the equation can be rewriten like this:$$\omega'^2-\frac{\omega_0\omega}{2.45}\omega'-\omega_0^2=0$$ and from here just substitude the natural frecuencies of the low-pass prototype and get the natural frequencies of the bandpass filter since I have ignored the absolute value to solve for \$\omega'\$ and I don't know if that can be important when computing the result.

  • \$\begingroup\$ Yes - it is correct. Please note that B is the bandwidth in rad/sec. Hence: wo/B=Q). \$\endgroup\$ – LvW Apr 30 '18 at 11:31

When in doubt, plot! Matlab code :

w_array = -10:1:10;
w0_array = -10:1:10;
wl_array_1 = zeros(length(w_array), length(w0_array));
wl_array_2 = zeros(length(w_array), length(w0_array));
syms wl

for i=1:length(w_array)
    w = w_array(i);
    for j=1:length(w0_array)
        w0 = w0_array(j);
        s1 = solve(wl^2 - w0*w*wl/2.45 - w0^2 == 0, wl, 'PrincipalValue', true);
        if isempty(s1)
            s1 = NaN;
        wl_array_1(i,j) = double(s1);
        s2 = solve(w - 2.45*abs(wl/w0 - w0/wl) == 0, wl, 'PrincipalValue', true);
        if isempty(s2)
            s2 = NaN;
        wl_array_2(i,j) = double(s2);

for j=1:length(w0_array)
    plot(w_array,wl_array_1(:,j)), hold on
xlabel('\omega'), ylabel('\omega´'), xlim([-10,10]), ylim([-45,5])
title('Not considering the absolute in equation'), hold off,

for j=1:length(w0_array)
    plot(w_array,wl_array_2(:,j)), hold on
xlabel('\omega'), ylabel('\omega´'), xlim([-10,10]), ylim([-45,5])
title('Considering the absolute in equation'), hold off

enter image description here

As can be seen, the absolute ensures that solutions only exist for positive values of \$\omega\$ and \$\omega_0\$. Removing the absolute, solutions exist for negative \$\omega\$ and \$\omega_0\$. Now it's up to you to decide what it means and if this is important.


For my part, I find it easier to do it in Laplace:

$$s=\frac{s^2+\omega_0^2}{BW s}$$

where BW is the bandwidth, and \$\omega_0\$ is the center frequency. For a real root, \$\Re\$:

$$\frac{s^2+\omega_0^2}{BW s}+\Re=0 => s^2+BW \Re s + \omega_0^2=0$$

Which is easy to solve. If the root is pure imaginary:

$$\frac{s^2+w_0^2}{BW s}+j \Im=0=>s^2+j BW \Im s + \omega_0^2=0$$

and if it's complex, \$p=\Re+j \Im\$:

$$\frac{s^2+w_0^2}{BW s}+p=0=>s^2+BW p s + \omega_0^2=0$$

This one is a bit more complicated to solve, not impossible. The bare, complex roots for the equation come out as:

$$s=-\frac{BW p}{2}\pm\sqrt{\frac{BW^2p^2-4\omega_0^2}{2}}$$

and solving for complex, you get four posibilities, out of which choose, like before, only the positive frequencies:

$$Real=BW^2(\Re^2-\Im^2)-4\omega^2$$ $$Imag=2BW^2\Re\Im$$ $$Mag=\sqrt{Real^2+Imag^2}$$ $$\Phi=\frac{1}{2}\sqrt{Mag}\frac{\arctan2(Imag, Real)}{2}$$ $$s=\frac{1}{2}\left[-BW\Re\pm\sqrt{Mag}\cos(\frac{\Phi}{2})\pm j\left(\sqrt{Mag}\sin\frac{\Phi}{2}\pm BW\Im\right)\right]$$

Looks monstruous, but it works.


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.