Convolution of square root raised cosine is not an ISI free filter!?

HI all,

I am trying to calculate the ratio between the symbol power to the intersymbol interference (ISI) noise power of the raised cosine filter. These practically should be infinite.

So what I do is I create an srrc filter first. After that I convolve that one with itself to create the matched filter. And after that I sum up all the values at sampling points.

So if I created an SRRC filter with 5 symbol lengh and 8 oversampling factor than I have an tap length of the filter of 2*5*8+1=81 elements in vector. After the convolution I will have 161 elements in the matched filter vector. So what I do is I sum up sum(matched_filter(1:8:end))-matched_filter(81) which should be 0, but it isn't.

I tried to do the same thing with the ideal case when the tx filter is sinc. If you run the script as below:

x=-5:0.1:5; tx_filter=sinc(x); matched_filter=conv(tx_filter,tx_filter); symb_to_ISI_ratio_tx=tx_filter(51)/(sum(tx_filter(1:10:end))-tx_filter(51)) symb_to_ISI_noise_ratio=matched_filter(101)/(sum(matched_filter(1:10:end))-matched_filter(101))

you will see that the tx_filter has the infinite ratio which is not the case with the matched filter after the convolution. So it means that the convolution of two sinc functions in matlab is not anymore sinc!?!?!?!

So am I doing something wrong or what is the question here!? Or it is all about the precision of matlab that it can not create the sinc function again after convolving two sinc functions!?

Regards, Jet