Post by jlee on Nov 16, 2008 2:29:59 GMT -5
Dear Charan,
In the Matched Filtering tutorial I understand why an integrate and dump process is best for a rectangular pulse, but I'm confused as to why for other pulse shapes like the triangle or raised cosine mentioned that it is optimum to just do a single sampling event at the center point for best detection.
I understand that center point is where the instantaneous peak amplitude out of the matched filter is with no noise. And I agree that the SNR has the most likely maximum at this point in time, but couldn't it also be at it's minimum if we were unlucky since noise has a large crest factor in a practical system? I don't quite understand why it still wouldn't be better to integrate all the possible energy over the symbol time and compare to a threshold. Even though the noise may not have a zero mean over that particular symbol time we know that for a binary shaped NRZ pulse that a 1 has a positve bias and a negative bias for 0; therefore yielding the typical 0 threshold. Isn't this how we overcome long distance links where the SNR is poor? Don't you just increase the bit time so that over the observation the likelihood of having a zero mean noise is better while integrating the positive or negative data bias longer?
Thanks for tutoring,
Jeff Lee
In the Matched Filtering tutorial I understand why an integrate and dump process is best for a rectangular pulse, but I'm confused as to why for other pulse shapes like the triangle or raised cosine mentioned that it is optimum to just do a single sampling event at the center point for best detection.
I understand that center point is where the instantaneous peak amplitude out of the matched filter is with no noise. And I agree that the SNR has the most likely maximum at this point in time, but couldn't it also be at it's minimum if we were unlucky since noise has a large crest factor in a practical system? I don't quite understand why it still wouldn't be better to integrate all the possible energy over the symbol time and compare to a threshold. Even though the noise may not have a zero mean over that particular symbol time we know that for a binary shaped NRZ pulse that a 1 has a positve bias and a negative bias for 0; therefore yielding the typical 0 threshold. Isn't this how we overcome long distance links where the SNR is poor? Don't you just increase the bit time so that over the observation the likelihood of having a zero mean noise is better while integrating the positive or negative data bias longer?
Thanks for tutoring,
Jeff Lee