Since your time vector is not guaranteed to follow a regular pattern, I think you need to use an iterative approach and perform the averaging manually.
The following code works by accumulating a sum when in an appropriate "region" of the data and then calculates the mean when exiting the region. The averages and NaNs are then inserted after the loop using logical indexing.
The approach does not assume any number of samples on either side of midnight. Getting the averageCount based on the number of midnights lets us be sure we've calculated all the averages when the loop is finished.
accumulated = 0;
rollingCount = 0;
averageCount = sum(EChr == 0);
averageCounter = 1;
averages = zeros(1, averageCount);
accumulating = false;
for i=1:numel(EChr)
if EChr(i) <= 3.0 || EChr(i)>= 22.0
accumulated = accumulated + A(i);
rollingCount = rollingCount + 1;
accumulating = true;
elseif accumulating == true
averages(averageCounter) = accumulated / rollingCount;
averageCounter = averageCounter + 1;
accumulated = 0;
rollingCount = 0;
accumulating = false;
else
%do nothing
end
end
% Edge Case - Was still accumulating at end of loop,
% and there is still an average left to compute
if accumulating && averageCounter < averageCount + 1
averages(averageCounter) = accumulated / rollingCount;
averageCounter = averageCounter + 1;
end
A(EChr==0) = averages;
A(EChr~=0) = NaN;
