20.01.2015 Views

njit-etd2003-081 - New Jersey Institute of Technology

njit-etd2003-081 - New Jersey Institute of Technology

njit-etd2003-081 - New Jersey Institute of Technology

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

90<br />

when there is significant correlation between the two signals. In practice<br />

used to indicate the regions where Фx1(f) has a valid interpretation. Phase estimates<br />

can be interpreted as the phase difference between harmonics <strong>of</strong> N 1 and x at frequency f.<br />

The arctan function can be used to obtain the argument <strong>of</strong> the cross-spectrum, resulting<br />

in a phase estimate over the range [—Tr / 2,71- /2] radians. However, the signs <strong>of</strong> the real<br />

and imaginary parts <strong>of</strong> Px1(f ) can be used to determine in which quadrant the arctangent<br />

falls, so extending the range to [<br />

-π, π] radians.<br />

Phase estimates can <strong>of</strong>ten be interpreted according to different theoretical<br />

models. A useful model is the phase curve for two signals, which are correlated with a<br />

fixed time delay, where the theoretical phase curve is a straight line, passing through the<br />

origin (0 radians at 0 frequency) with slope equal to the delay, and a positive slope for a<br />

phase lead, and negative slope for a phase lag (see [67]). In situations where there is<br />

significant correlation over a wide range <strong>of</strong> frequencies and a delay between two signals,<br />

it is reasonable to extend the phase estimate outside the range [-π,π] radians, which<br />

avoids discontinuities in phase estimates. Such a phase estimate is <strong>of</strong>ten referred to as<br />

an unconstrained phase estimate. The representation <strong>of</strong> phase estimates is discussed in<br />

[66]. Different theoretical phase curves for other forms <strong>of</strong> correlation structure are<br />

discussed in [67]. Details for the construction <strong>of</strong> confidence limits about estimated<br />

phase values can be found in [64]. In situations where the correlation structure between<br />

two signals is dominated by a delay it is possible to estimate this delay from the phase<br />

curve. Such an approach, based on weighted least squares regression, is described in the<br />

Appendix in [62]. This method has the advantage <strong>of</strong> providing an estimate <strong>of</strong> the

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!