Sign in

username:

password:



Not a member?

Search Online Books



Search tips

Free Online Books

Ads

Chapters

Spectral Audio Signal Processing
    Spectral Interpolation
      
          Maximum Likelihood Sinusoid Estimation

Chapter Contents:

Search Spectral Audio Signal Processing

  

Book Index | Global Index

Would you like to be notified by email when Julius Orion Smith III publishes a new entry into his blog?

  

Maximum Likelihood Sinusoid Estimation

The maximum likelihood estimator (MLE) is widely used in practical signal modeling [107]. A full treatment of maximum likelihood estimators (and statistical estimators in general) lies beyond the scope of this book. However, we will show that the MLE is equivalent to the least squares estimator for a wide class of problems, including well resolved sinusoids in white noise.

Consider again the signal model of (4.1) consisting of a complex sinusoid in additive white (complex) noise:

$\displaystyle x(n) \isdef {\cal A}e^{j\omega_0 n} + v(n) \protect$ (5.7)

Again, $ {\cal A}= A e^{j\phi}$ is the complex amplitude of the sinusoid, and $ v(n)$ is white noise. In addition to assuming $ v$ is white, we add the assumption that it is Gaussian distributed5.7 with zero mean; that is, we assume that its probability density function (see Appendix D) is given by5.8

$\displaystyle p_v(\nu) = \frac{1}{\pi \sigma_v^2} e^{-\frac{\vert\nu\vert^2}{\sigma_v^2}}. $

We express the zero-mean Gaussian assumption by writing

$\displaystyle v(n) \sim {\cal N}(0,\sigma_v^2). $

The parameter $ \sigma_v^2$ is called the variance of the random process $ v(n)$, and $ \sigma_v$ is called the standard deviation.

It turns out that when Gaussian random variables $ v(n)$ are uncorrelated (i.e., when $ v(n)$ is white noise), they are also independent. This means that the probability of observing particular values of $ v(n)$ and $ v(m)$ is given by the product of their respective probabilities [107]. We will now use this fact to compute an explicit probability for observing any data sequence $ x(n)$ in (4.7).

Since the sinusoidal part of our signal model, $ {\cal A}e^{j\omega_0 n}$, is deterministic; i.e., it does not including any random components; it may be treated as the time-varying mean of a Gaussian random process $ x(n)$. That is, our signal model (4.7) can be rewritten as

$\displaystyle x(n) \sim {\cal N}({\cal A}e^{j\omega_0 n},\sigma_v^2) $

and the probability density function for the whole set of observations $ x(n)$, $ n=0,1,2,\ldots,N-1$ is given by

$\displaystyle p(x) = p[x(0)] p[x(1)]\cdots p[x(N-1)] = \left(\frac{1}{\pi \sig... ..._v^2}\sum_{n=0}^{N-1} \left\vert x(n) - {\cal A}e^{j\omega_0 n}\right\vert^2} $

Thus, given the noise variance $ \sigma_v^2$ and the three sinusoidal parameters $ A,\phi,\omega_0 $ (remember that $ {\cal A}= A e^{j\phi}$), we can compute the relative probability of any observed data samples $ x(n)$.

Order a Hardcopy of Spectral Audio Signal Processing

Previous: Sinusoidal Frequency Estimation
Next: Likelihood Function
written by Julius Orion Smith III
Julius Smith's background is in electrical engineering (BS Rice 1975, PhD Stanford 1983). He is presently Professor of Music and Associate Professor (by courtesy) of Electrical Engineering at Stanford's Center for Computer Research in Music and Acoustics (CCRMA), teaching courses and pursuing research related to signal processing applied to music and audio systems. See http://ccrma.stanford.edu/~jos/ for details.


Comments

No comments yet for this page


Add a Comment
You need to login before you can post a comment (best way to prevent spam). ( Not a member? )