I've been given a list of 5 composite signals, where each is composed of 10 sinusoids of different frequencies. For instance, the first composite signal $S_1$ is given by

$$ S_1 = \sum_{i=1}^{10} A_i \sin (2 \pi f_i t- \phi_i) $$ where $A_i, f_i$ and $\phi_i$ are known values. The other 4 composite signals are similar to $S_1$. In total, I have 50 different values for $A_i, f_i$ and $\phi_i$ (10 for each $S_i$).

My goal is to find the orthogonal basis functions for $\{S_1, S_2, \cdots, S_5\}$. I've decomposed them into $\sin$ and $\cos$: $$ \sum_{i=1}^{10} A_i \sin (2 \pi f_i t- \phi_i) = \sum_{i=1}^{10} A_i [\sin(2\pi f_i t)\cos(\phi_i) - \cos(2\pi f_i t)\sin(\phi_i)] $$ and from here I can see that $A_i, \cos(\phi_i), \sin(\phi_i)$ are constants, and a spanning set would be

$$ \bigcup_{k=1}^{50} \{\sin(2\pi f_k t), \cos(2\pi f_k t) \} $$ so I end up with a 100-element set. I presume these are my orthogonal basis functions, since each $f_k$ is unique. Can I check if my train of thought is right?


Your $S_1,\ldots, S_5$ already are your orthogonal bases, exactly the way your first equation writes them down.

To check that: write down the definition of your inner product, and insert, for example, $S_a$ and $S_b$. You'll see that $\left<S_a,S_b\right>=0 \iff a\ne b$.

This follows quite directly from the linearity that a (real) inner product needs to have:

\begin{align} \Big<\sum_i A_{a,i}\sin(2\pi f_{a,i}t -\phi_{a,i}), \sum_k A_{b,k}\sin(2\pi f_{b,k}t -\phi_{b,k}) \Big>&\\ = \sum_i \Big< A_{a,i}\sin(2\pi f_{a,i}t -\phi_{a,i}), \sum_k A_{b,k}\sin(2\pi f_{b,k}t -\phi_{b,k}) \Big>&\\ =\sum_i\sum_k \Big< A_{a,i}\sin(2\pi f_{a,i}t -\phi_{a,i}), A_{b,k}\sin(2\pi f_{b,k}t -\phi_{b,k}) \Big> \end{align}

and sines of different frequencies are orthogonal.

  • $\begingroup$ Could you include a more detailed and less hand-wavy proof of your claim that the 5 signals are indeed orthogonal? Let's simplify the question even more with each $S_i$ being just one sinusoid instead of the sum of 5 different sinusoids. Why are $\sin(2\pi 1000t)$ and $\sin(2\pi (1000\sqrt 2)t)$ orthogonal signals? And over what time interval are they orthogonal? $\endgroup$ – Dilip Sarwate Mar 5 at 20:30
  • $\begingroup$ @DilipSarwate in undergrad courses, harmonic oscillations are often axiomatically introduced as base functions. However, showing the orthogonality of your example is relatively straightforward through properties of any inner product: Iff $x,y$ orthogonal, then $<x+y,x+y>=<x,x>+<y,y>$. The classical power signal space $\tilde L_2(\mathbb R)$-inner product $<x,y>=\lim_{T\to\infty}\frac1{2T}\int_{-T}^Tx(t)y^*(t)\,\mathrm dt$ fulfills exactly that. We can also take the cross-spectrum detour, showing their Fourier-domain product is a constant 0, but I'm sure you'll appreciate the direct 0-integral. $\endgroup$ – Marcus Müller Mar 5 at 21:01

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.