story83 wrote:
> Ok, now I am beginning to grapsp the meaning.
> Now let's take it further :
> In the following link,
> http://static.flickr.com/28/61829060_6311c246ca.jpg?v=0
>  there is a diagram of different implementations of a lowpass fir
> filter, one in floating point, and three with various wordlengths
> (12,16,64 bits etc)
> I have the magnitude response diagram and, ok what do I understand of
> it? Well only the fact that the 24 bit wordlength is very different
> from the original filter, and thus 24 not appropriate as a wordlength .
> Something else I "should" be seeing here and cant?
> Thanks again, you are very helpful

I can't read the captions or the axis labels. I see a superposition of 
several filters with nearly identical passbands and different passband 
attenuations. The worst (least attenuation) in not quite one division 
(20 dB?) above the others at some frequencies, but not much higher than 
the first peak in the stop band. Despite what you wrote -- you seem to 
have written 64 when you meant 24 -- the words curve is likely 12 bits. 
You should expect performance to improve as word size increases.

Jerry
-- 
Engineering is the art of making what you want from things you can get.
�����������������������������������������������������������������������

Ok, now I am beginning to grapsp the meaning.
Now let's take it further :
In the following link,
http://static.flickr.com/28/61829060_6311c246ca.jpg?v=3D0
 there is a diagram of different implementations of a lowpass fir
filter, one in floating point, and three with various wordlengths
(12,16,64 bits etc)
I have the magnitude response diagram and, ok what do I understand of
it? Well only the fact that the 24 bit wordlength is very different
from the original filter, and thus 24 not appropriate as a wordlength .
Something else I "should" be seeing here and cant?
Thanks again, you are very helpful

Jake wrote:
> story83 wrote:
> >
> > Hello
> > I am new in dsp and I havent understood the interpretation of a
> > magnitude response diagram. (much less how to compare two filters i.e a
> > fixed point one and a floating point one based on their magnitute
> > response diagrams)
> > I asked my prof and tried to understand it from a book but I didnt find
> > a clear example or explanation.
> > Can anybody explain in simple words how to "read" a magnitute response
> > diagram?
> > (Right now I have one in front of me,it begins at 0 db and descends
> > smoothly until frequency -60 then starts rippling ).
> >
> > Thanks a lot
>
> Basically a magnitude response diagram plots Gain (change in signal
> magnitude after filtration) versus frequency of the input signal IF the
> signal is a continuous sine wave.  The filter you described is a
> low-pass filter (lpf) because the highest gain occurs at the low
> frequencies.  The maximum gain is usually set to one (i.e., no change)
> but this is not mandatory.  In addition to the low-pass filter, there
> are high-pass filters (hpf), where the maximum gain occurs at the high
> frequencies, as well as band-pass and band-stop filters.
>
> Magnitude response diagrams are =BD of what are generally called frequency
> response diagrams.  The other =BD are phase response diagrams, and the
> same principle applies: change in phase ("delay") from filtration
> plotted against signal frequency (again, signal =3D sine wave).
>
> The gain is usually plotted not as an actual multiplier but as decibels
> (dB).  So your diagram starts not at 1.000 but at 0 dB, which is the
> same thing.

story83 wrote:
> 
> Hello
> I am new in dsp and I havent understood the interpretation of a
> magnitude response diagram. (much less how to compare two filters i.e a
> fixed point one and a floating point one based on their magnitute
> response diagrams)
> I asked my prof and tried to understand it from a book but I didnt find
> a clear example or explanation.
> Can anybody explain in simple words how to "read" a magnitute response
> diagram?
> (Right now I have one in front of me,it begins at 0 db and descends
> smoothly until frequency -60 then starts rippling ).
> 
> Thanks a lot

Basically a magnitude response diagram plots Gain (change in signal
magnitude after filtration) versus frequency of the input signal IF the
signal is a continuous sine wave.  The filter you described is a
low-pass filter (lpf) because the highest gain occurs at the low
frequencies.  The maximum gain is usually set to one (i.e., no change)
but this is not mandatory.  In addition to the low-pass filter, there
are high-pass filters (hpf), where the maximum gain occurs at the high
frequencies, as well as band-pass and band-stop filters.

Magnitude response diagrams are � of what are generally called frequency
response diagrams.  The other � are phase response diagrams, and the
same principle applies: change in phase ("delay") from filtration
plotted against signal frequency (again, signal = sine wave).

The gain is usually plotted not as an actual multiplier but as decibels
(dB).  So your diagram starts not at 1.000 but at 0 dB, which is the
same thing.

A magnitude response diagram for a filter shows how much the filter 
amplifies at different frequencies. Say you have a signal 
x(t)=x1(t)+x2(t)=cos(2*pi*f1*t)+cos(2*pi*f2*t) consisting of 2 frequency 
components f1 and f2....If you send this signal thru a filter then you can 
see how much the amplitude of x1(t) is amplified by looking at (f1, ? dB) in 
the magnitude response diagram. If ? dB is 0dB then the amplitude is left 
untouched. If the dB-factor is below 0 then the amplitude is lowered. If the 
dB-factor is above 0dB then the amplitude is amplified. Hope this helps....

Hello
I am new in dsp and I havent understood the interpretation of a
magnitude response diagram. (much less how to compare two filters i.e a
fixed point one and a floating point one based on their magnitute
response diagrams)
I asked my prof and tried to understand it from a book but I didnt find
a clear example or explanation.
Can anybody explain in simple words how to "read" a magnitute response
diagram?
(Right now I have one in front of me,it begins at 0 db and descends
smoothly until frequency -60 then starts rippling ). 

Thanks a lot