[問題]digital filter的問題

[gggould (evanescent)]

兩個問題 不過講的可能是一樣的東西...

1. 如果濾波器輸出是10 bits, 濾波器內部設計需要超過10 bits 嗎

2. An FIR filter with 10 bits in and 10 bit out, representing integer numbers
between 0 to 2^10. To convert the coefficients from floating point to fixed
point, can I truncate the fractional bits of the coefficient without losing
the resolution? (No but why??)
How about IIR? (No but why??)

thanks

--
※ 發信站: 批踢踢實業坊(ptt.cc) 
◆ From: 71.204.133.130
→ uxijgil:1. Dynamic range... accumulative sum may overflow
→ uxijgil:2. Because there is exponent part in floating point ?
推 bxxl:it would be better if you think both data and coefficient
→ bxxl:as fractional numbers, not integer
→ bxxl:in/out/coef/internal nodes usually need different format
→ gggould:thx.But why diff format? Why not all 2's complement?
推 bxxl:Oh, I mean different number of int/frac bits