'xcorr()' giving some negative values

I have two arrays A and B both having only positive values with some being zero. I am using c=xcorr(A,B) to get the cross correlation. I would expect the result to be all positive but getting some negative values. What is the reason?

[Jan]

Could you post the concerned data and the code? Perhaps it is only it is only a typo and wild guessing the reason is a waste of time. Thanks.

[dpb]

More than likely it's just rounding error...for example if use the example data for xcorr

>> load sensorData s1 s2
>> plot(s1)
>> hold all
>> plot(s2)
>> cor=xcorr(abs(s1),abs(s2));
>> max(cor)
ans =
  0.0025
>> min(cor)
ans =
-3.3204e-19
>> sum(cor<0)
ans =
 255
>> length(cor)
ans =
      5241
>>

the data going in were all >=0 by abs() operation and still some 5% of the resulting autocorrelation values are negative. Now the largest in absolute value of those is only on order of machine double precision, but it's still negative.

[Biplab Roy]

Thanks all. As dpb correctly placed, is a classic example of some results being negative even though all inputs are positive. And i also feel it is the rounding error. Could not this be corrected?

[dpb]

Sure

cor(cor<0)=0;

if you feel you must. But, in general, when you use floating point, rounding errors are inherent with the beastie.