0

How can I analyze a basic sine wave for its frequency phase and amplitude at a particular point in time in MATLAB? Are there any tools to do that? I am trying to do additive synthesis of inharmonic sounds in MATLAB, but having a great difficulty to extract the needed phase amplitude and frequency data of the partials I am trying to generate. Can someone shed some light regarding signal analysis in MATLAB?

jojeck
  • 11,107
  • 6
  • 38
  • 74
shadow
  • 1
  • 1
    The answer to this question might be helpful: http://dsp.stackexchange.com/questions/8713/doing-sine-regression-to-recover-the-paramters-from-a-given-signal – Matt L. Jul 01 '14 at 07:35

3 Answers3

1

If your the unknown signal $x(n)$ is modeled as: $$x(n)=A \sin(2 \pi f n+ \phi)+w(n)$$ and you want to estimate $A$,$f$,and $\phi$ accurately, you can use least square estimation. Unfortunately the cost function is nonlinear. You can use nonlinear least square in MATLAB to find the parameters as follows:

Make a cost function: 

f=@(A,f,phi) x[n]-A sin(2*pi*  f *n+ phi)

and use 

p0=[A0,f0,phi0];

p = lsqnonlin(f,p0);

to find the unknown parameters. Note that the optimizer will have a hard time finding $f$ as the problem is not convex. So it is best if you can give an initial estimate of frequency by using a method like fft.

If the frequency is known then the problem can be converted to a linear estimation as: $$x(n)=A \sin(2 \pi f n+ \phi)+w(n)$$ $$=A \sin(2 \pi f n) \cos (\phi) +A \cos(2 \pi f n) \sin (\phi)+w(n)$$ $$=p_1 S[n] +p_2 C[n]+w(n),$$ where $p_1=A \cos(\phi)$ and $p_2=A \sin(\phi)$ are unknown parameters, and $S[n]$ and $C[n]$ are known.

jojeck
  • 11,107
  • 6
  • 38
  • 74
Hooman
  • 321
  • 1
  • 3
1

Here's some Matlab code:

% Assume 'x' is the input time-domain sequence and 'y' is it's Hilbert transform.

Analytic_Signal = hilbert(x);
y = imag(Analytic_Signal);

Instantaneous_Magnitude = abs(Analytic_Signal);

Instantaneous_Phase = unwrap(atan2(y, x));

Instantaneous_Freq = diff(Instantaneous_Phase);
jojeck
  • 11,107
  • 6
  • 38
  • 74
Richard Lyons
  • 5,777
  • 13
  • 25
0

if the sinusoid, $x(t)$, is uncontaminated, i s'pose you can measure all three by first computing the Hilbert transform of it, $\hat{x}(t)$ and, from the two, construct the analytic signal,

$$ x_a(t) \ \triangleq \ x(t) \ + \ j\hat{x}(t) $$

the amplitude will be $|x_a(t)|$, the instantaneous phase will be $\phi(t) = \arg\{x_a(t)\}$ and, if the phase is unwrapped the derivative of the phase w.r.t. time is the instantaneous (angular) frequency is this derivative.

$$ \omega(t) \ = \ \frac{d}{dt}\phi(t) \ = \ \frac{d}{dt}\arg\{x(t) \ + \ j\hat{x}(t)\} $$

robert bristow-johnson
  • 20,661
  • 4
  • 38
  • 76
  • Would performing the Hilbert transform not introduce an implicit windowing? I'm curious how this might affect the concept of a measured `instantaneous' frequency. – Speedy Jul 03 '14 at 09:43
  • @Speedy. I believe robert bristow-johnson is correct. What do you mean by "implicit windowing"? – Richard Lyons Jul 29 '15 at 08:51
  • @RichardLyons, I was wondering whether instantaneous frequency is really instantaneous in this case. Since the computation of a Hilbert transform requires a time-base or "implicit window" - (perhaps a poor choice of words), is the computed value not meaningful only for a region of time, rather than at an instant? – Speedy Jul 29 '15 at 14:14
  • In fact I find it difficult to understand how a value of "instantaneous" frequency can be meaningful at all. – Speedy Jul 29 '15 at 14:15
  • @Speedy. Your implied question, "Does instantaneous frequency have any physical meaning?" is actually a good question. You should have a look at 'frequency modulation (FM) signals' and 'FM demodulation' tutorial material to answer that question. (If instantaneous frequency had no physical meaning we'd have no FM radio. As far as I can tell Robert B-J's scheme is a high-performance way to perform FM demodulation.)

    Recall, frequency is defined as 'a phase angle change divided by a fixed interval of time'. That is, delta phase over delta time, cycles/second, or radians/second.

     – Richard Lyons Jul 30 '15 at 12:57