[urgent] Matched filter (Pulse shaping)

[salam2000]

hi
I have a project for design receiver.
the fs/symbolrate is irrational, examp =7.5 sampel per symbol. I want to add the matched filter, please help me to add it.
I use interpolation and timing recovery in a loop.
1-where I replace MF? If I should take it before loop or in the loop with interpolation.?
2-if I put in the loop we should have a lot of complexity because of calculate interpolated coefficient and convolution; if I take it before loop I should solve the irrational problem.
please help me
Thanks

 

[cheggy]

Re: [urgent] Matched filter

I haven't understood everything clearly, but if your only problem is to create filter that works with non-integer number of samples per symbol, you may operate at frequency of 15*symbol rate (you sample your signal, then add zero in pattern, then sample again, then add zero again, etc.), then perform LP filtration (like "rised cosine"). This would be end of interpolation process. After that, send your signal to matched filter, but keep in mind that it should work at 15 samples per symbol.

 

[salam2000]

we said nsamp=fs/symr=7.5 as an example. this factor can be a number p/q=7.5223486643455778644356746746... for example.
but I think two ideas where two have owns problems:
1- we use a Pulse shaping with a lot of coefficient, then we use quantization and set a offset for choose coefficients, example : coefficient(offset:10:end). this method need a big lookup table and we want the vector size not be large.
2- we use an interpolation inside the loop and then become nsamp to 8 with interpolation. but this has a lot of complexity.

Is there better solution.