1. Limited time only! Sign up for a free 30min personal tutor trial with Chegg Tutors
    Dismiss Notice
Dismiss Notice
Join Physics Forums Today!
The friendliest, high quality science and math community on the planet! Everyone who loves science is here!

Physical meaning of autocorrelation

  1. Jan 9, 2013 #1
    Hi All,

    I was in a process of processing my vibration-test data. I now generated a plot of the autocorrelation function of the object acceleration. Please see the attachements (the second attachment is the close-up for small tau's).

    The x-axis in the plot is the time delay tau. You can see from the second attachment that, at tau=0,the autocorrelation reaches its maximum.

    My question is, as for the signal like this, in the long time delay (tau=60s, 70s), the autocorrelation still does not decay too much. What could this behavior imply? In terms of the noise frequency, what can we say?

    Any comments will be welcome. Thanks in advance.


    Attached Files:

    Last edited: Jan 9, 2013
  2. jcsd
  3. Jan 9, 2013 #2

    Jano L.

    User Avatar
    Gold Member

    The autocorrelation function gives the amount of correlation of values of the original function at times delayed by ##\tau##. If the autocorrelation function vanishes quickly, it means that the value of the function at some time cannot be predicted with success from neighboring values.

    If it does not vanish quickly, it means that the behavior of the function is less random; from the value at time 0 one can estimate possible values at future times with better success. For example, if the function is periodic, after one wave one can make reasonable estimate what the next values will be, and the autocorrelation function comes out also periodic. But if the function is position of Brownian particle, the next values are hard to predict successfully, and also the autocorrelation function decays very fast (exponentially).
  4. Jan 9, 2013 #3


    User Avatar
    Science Advisor
    Gold Member
    2017 Award

    If the autocorrelation does not die away then there must be some significant periodic elements in the signalbecause the delayed signal correlates well with an earlier version of itself.
  5. Jan 9, 2013 #4

    Andy Resnick

    User Avatar
    Science Advisor
    Education Advisor

    A few comments:

    I think of the autocorrelation as a measure of how deterministic the system is- a perfectly deterministic system will always have a normalized autocrrelation of '1', while a perfectly random system will have have an autocorrelation that looks like a delta function. For real systems with dissipation present, the autocorrelation function will generally follow a decaying exponential, and the width of the exponential is a measure of the (de-)correlation time (or length, if we extend the analysis to cover spatially random systems such as reflection from a rough surface). Multiple dissipation processes result in multi-exponential autocorrelation functions- the dynamic light scattering field is a good place to get familiar with interpreting autocorrelation curves.

    Without knowing any details about what generated the data (what is vibrating? How was it excited?) it's tough to parse your data. However, when we see a low-frequency component like your first graph has, it's due to drift in our detector.
Know someone interested in this topic? Share this thread via Reddit, Google+, Twitter, or Facebook