finddelay returns identical value (0) for "best" and "worst" case.

I'm curious about what the reasoning could be for having finddelay() return 0 if no significant correlation is found. That means that the user has no way of distinguishing cases where the signals overlap with 0 delay and when the signals are not correlated at all. Why not return NaN?

Answer 1

[Brian Neiswander]

Hi Evan,

The "finddelay" function does not consider the strength of the correlation found. Given two signals, it simply returns the lag that produces the maximum absolute value of their cross-correlation. In the "More About" section of the "finddelay" documentation page below:

http://www.mathworks.com/help/comm/ref/finddelay.html

you can find a description of the algorithm used to calculate the estimated delay. Here, it states that:

"Pairs of signals need not be exact delayed copies of each other. However, the estimated delay has a useful meaning only if there is sufficient correlation between at least one pair of the delayed signals."

In other words, the output will not be significant if the two signals are uncorrelated. For the functionality you are asking for, you can use the code suggested by Wayne King in the following MATLAB Answers post:

http://www.mathworks.com/matlabcentral/answers/73802-how-do-i-know-if-the-finddelay-function-is-significant

Hope this is helpful.

-Brian

[Evan]

Are you sure? Here is the code I'm referring to, from finddelay() in R2015a:

% Set to zeros estimated delays for which the normalized cross-correlation
% values are below a given threshold (spurious peaks due to FFT roundoff
% errors).
for i=1:length(d)
    if max_c(i)<1e-8
        d(i) =  0;
        if isscalar(d) && maxlag(i)~=0
            warning(message('comm:finddelay:noSignificantCorrelationScalar'));
        elseif isvector(d) && maxlag(i)~=0
            warning(message('comm:finddelay:noSignificantCorrelationVector', i));
        end
    end    
end