+ Post New Thread
Results 1 to 3 of 3
  1. #1
    Junior Member level 1
    Points: 148, Level: 1

    Join Date
    Jun 2016
    Posts
    16
    Helped
    0 / 0
    Points
    148
    Level
    1

    fir filter difference in 8-bit and 12-bit modes

    hello
    i designed a low-pass filter in fdatool like this:
    Click image for larger version. 

Name:	FDA_good.png 
Views:	4 
Size:	59.7 KB 
ID:	133021
    when i use it in 12-bit configuration output is like below:
    Click image for larger version. 

Name:	LPFIR12bit.png 
Views:	2 
Size:	8.9 KB 
ID:	133022
    this output looks like my input and seems to be correct.
    but when i use that filter in 8-bit configuration the output is like this
    Click image for larger version. 

Name:	LPFIR8bit.png 
Views:	3 
Size:	9.1 KB 
ID:	133023
    my output is some steps!!!
    is this normal?
    why it happens?
    tanx in advance

    •   AltAdvertisement

        
       

  2. #2
    Advanced Member level 5
    Points: 37,911, Level: 47
    Achievements:
    7 years registered

    Join Date
    Jun 2010
    Posts
    6,874
    Helped
    2019 / 2019
    Points
    37,911
    Level
    47

    Re: fir filter difference in 8-bit and 12-bit modes

    This will be from clipping and saturation. It's pretty normal.



    •   AltAdvertisement

        
       

  3. #3
    Advanced Member level 3
    Points: 5,833, Level: 18

    Join Date
    Feb 2014
    Posts
    832
    Helped
    273 / 273
    Points
    5,833
    Level
    18

    Re: fir filter difference in 8-bit and 12-bit modes

    Rounding and precision issues can be very significant in digital signal processing.

    For example I inherited an IIR filter in a control application and eventually realized how poorly configured it was for the task. Small signals (and as an error amplifier that's all it should see) multiplied by small coefficients rounded to zero, essentially negating that coefficient and meaning that the transfer function/step response varied significantly with signal amplitude.

    The solution was to add additional fractional bits. But with fixed point math and/or large dynamic range in coefficient values these problems can be very significant. It's very important to evaluate real life performance which simulate the constraints of the real system.



--[[ ]]--