One of the major reasons for choosing IIR over FIR is computational
efficiency, especially for narrowband filters. But IFIR seems to solve that
problem to a great extent. I have understood IFIR theory mainly from
Richard Lyons book and a few papers, I just wonder if IFIR can be
replacement of IIR altogether. I dont see any papers or books mentioning
what is it that is not good about IFIR. If there are none then can we
totally get rid of IIR (utopia) ? What cases IIR is better than IFIR?

I'd keep an eye on group delay vs. group delay distortion.
FIR filters - including IFIR - are most efficient, if designed for linear
phase. That means the impulse response is symmetric, and the resulting high
delay can be problematic in many applications. 
An IIR filter is faster to react to the input, but the group delay varies
with frequency. 

I could probably design an IFIR for nonlinear phase, but it would be less
efficient: The number of multiplications in the z^M section doubles
immediately, as I lose symmetry. And on top of that, the number of taps
increases as the symmetric FIR was optimal for amplitude response.

On Mon, 16 Jul 2012 01:02:56 -0500, "sudarshan_onkar"
<24599@dsprelated> wrote:

>One of the major reasons for choosing IIR over FIR is computational
>efficiency, especially for narrowband filters. But IFIR seems to solve that
>problem to a great extent. I have understood IFIR theory mainly from
>Richard Lyons book and a few papers, I just wonder if IFIR can be
>replacement of IIR altogether. I dont see any papers or books mentioning
>what is it that is not good about IFIR. If there are none then can we
>totally get rid of IIR (utopia) ? What cases IIR is better than IFIR?

Hi,
  IIR filters are very computationally efficient relative 
to all FIR filters (standard FIR or IFIR), for a given 
frequency magnitude repsonse.  Also, IIR filters require far 
less coefficient data storage than FIR filters.

The advantage of IFIR filters is when you must have a 
linear phase, high-performance (many taps), digital filter. 
IFIR filters are more computationally efficient than 
standard (single tapped-delay line) linear phase FIR filters. 

Nope, we'll rarely, if ever, see an IFIR flter replacing 
an IIR filter.

sudarshan_onkar, if you have an American of the 2nd 
Edition, or an American version or International (grey cover),
version of the 3rd Edition. of my "Understanding DSP" book, 
I can send you the appropriate errata if you can tell me 
both (1) the version type (American-blue&green cover, or 
International-grey cover), and (2) the "Printing Number" 
of your copy of the book.

You can determine the "Printing Number" book by looking 
at the page just before the "Dedication" page.  

On that page (before the Dedication) you'll see all 
sorts of publisher-related information, including the 
ISBN-10 number.  At the very bottom of the page you 
should see lines printed something like:

       Text printed in the United States on ...
       First Printing

indicating the "First Printing" of the book.  However, for 
later printings the second line above may have words 
such as: "Second Printing" or "Seventh Printing".
Send me an E-mail with the Edition and Printing numbers, 
and I'll E-mail the appropriate errata to you.

[-Rick Lyons-]

In general ifir is more computationally efficient than fir but the data storage
requirements are similar due to the fact that the first filter has many zero-valued
coefficients. If you are coding on a commercial dsp you may not care about this and the reduced
number of multiply operations will be a big win. But an iir is still way more efficient in both
storage and MIPS and is normally used if phase is not important. 

Bob

>I'd keep an eye on group delay vs. group delay distortion.
>FIR filters - including IFIR - are most efficient, if designed for linear
>phase. That means the impulse response is symmetric, and the resulting
high
>delay can be problematic in many applications. 
>An IIR filter is faster to react to the input, but the group delay varies
>with frequency. 
>
>I could probably design an IFIR for nonlinear phase, but it would be less
>efficient: The number of multiplications in the z^M section doubles
>immediately, as I lose symmetry. And on top of that, the number of taps
>increases as the symmetric FIR was optimal for amplitude response.
>

Yes , Fair point. The group delay in IIR is also high in the transition
band but in FIR or IFIR its constantly high at all frequencies. Thanks for
the reply

>On Mon, 16 Jul 2012 01:02:56 -0500, "sudarshan_onkar"
><24599@dsprelated> wrote:
>
>>One of the major reasons for choosing IIR over FIR is computational
>>efficiency, especially for narrowband filters. But IFIR seems to solve
that
>>problem to a great extent. I have understood IFIR theory mainly from
>>Richard Lyons book and a few papers, I just wonder if IFIR can be
>>replacement of IIR altogether. I dont see any papers or books mentioning
>>what is it that is not good about IFIR. If there are none then can we
>>totally get rid of IIR (utopia) ? What cases IIR is better than IFIR?
>
>Hi,
>  IIR filters are very computationally efficient relative 
>to all FIR filters (standard FIR or IFIR), for a given 
>frequency magnitude repsonse.  Also, IIR filters require far 
>less coefficient data storage than FIR filters.
>
>The advantage of IFIR filters is when you must have a 
>linear phase, high-performance (many taps), digital filter. 
>IFIR filters are more computationally efficient than 
>standard (single tapped-delay line) linear phase FIR filters. 
>
>Nope, we'll rarely, if ever, see an IFIR flter replacing 
>an IIR filter.
>
>sudarshan_onkar, if you have an American of the 2nd 
>Edition, or an American version or International (grey cover),
>version of the 3rd Edition. of my "Understanding DSP" book, 
>I can send you the appropriate errata if you can tell me 
>both (1) the version type (American-blue&green cover, or 
>International-grey cover), and (2) the "Printing Number" 
>of your copy of the book.
>
>You can determine the "Printing Number" book by looking 
>at the page just before the "Dedication" page.  
>
>On that page (before the Dedication) you'll see all 
>sorts of publisher-related information, including the 
>ISBN-10 number.  At the very bottom of the page you 
>should see lines printed something like:
>
>       Text printed in the United States on ...
>       First Printing
>
>indicating the "First Printing" of the book.  However, for 
>later printings the second line above may have words 
>such as: "Second Printing" or "Seventh Printing".
>Send me an E-mail with the Edition and Printing numbers, 
>and I'll E-mail the appropriate errata to you.
>
>[-Rick Lyons-]
>

Hi Rick,
         Thanks for the reply. Made me stop over and think before jumping
into answer. Yes i do agree that IIR are computationally better than IFIR.
But i think it will be comparable to that of IIR. By adjusting L we can 
get up to 80-85% reduction for a passband = 0.005fs. When i just did a
rough calculation it turns out that for IIR (Butterworth 7th order) we
would require 21 Multiplies and the same implemented with IFIR(~400 tap)
would require 30-40 Multiplies. So yes IIR are better but IFIR are not too
behind . So if one is ready to pay for extra multiplies IFIR may be a
better choice since we get linear phase. 

it may also turn out that an 'IIR multiplier' gets more expensive than a
'FIR multiplier', at least those in the recursive section.

On Thu, 19 Jul 2012 06:10:58 -0500, "sudarshan_onkar"
<24599@dsprelated> wrote:

>>On Mon, 16 Jul 2012 01:02:56 -0500, "sudarshan_onkar"
>><24599@dsprelated> wrote:
>>
>>>One of the major reasons for choosing IIR over FIR is computational
>>>efficiency, especially for narrowband filters. But IFIR seems to solve
>that
>>>problem to a great extent. I have understood IFIR theory mainly from
>>>Richard Lyons book and a few papers, I just wonder if IFIR can be
>>>replacement of IIR altogether. I dont see any papers or books mentioning
>>>what is it that is not good about IFIR. If there are none then can we
>>>totally get rid of IIR (utopia) ? What cases IIR is better than IFIR?
>>
>>Hi,
>>  IIR filters are very computationally efficient relative 
>>to all FIR filters (standard FIR or IFIR), for a given 
>>frequency magnitude repsonse.  Also, IIR filters require far 
>>less coefficient data storage than FIR filters.
>>
>>The advantage of IFIR filters is when you must have a 
>>linear phase, high-performance (many taps), digital filter. 
>>IFIR filters are more computationally efficient than 
>>standard (single tapped-delay line) linear phase FIR filters. 
>>
>>Nope, we'll rarely, if ever, see an IFIR flter replacing 
>>an IIR filter.
>>
>>sudarshan_onkar, if you have an American of the 2nd 
>>Edition, or an American version or International (grey cover),
>>version of the 3rd Edition. of my "Understanding DSP" book, 
>>I can send you the appropriate errata if you can tell me 
>>both (1) the version type (American-blue&green cover, or 
>>International-grey cover), and (2) the "Printing Number" 
>>of your copy of the book.
>>
>>You can determine the "Printing Number" book by looking 
>>at the page just before the "Dedication" page.  
>>
>>On that page (before the Dedication) you'll see all 
>>sorts of publisher-related information, including the 
>>ISBN-10 number.  At the very bottom of the page you 
>>should see lines printed something like:
>>
>>       Text printed in the United States on ...
>>       First Printing
>>
>>indicating the "First Printing" of the book.  However, for 
>>later printings the second line above may have words 
>>such as: "Second Printing" or "Seventh Printing".
>>Send me an E-mail with the Edition and Printing numbers, 
>>and I'll E-mail the appropriate errata to you.
>>
>>[-Rick Lyons-]
>>
>
>Hi Rick,
>         Thanks for the reply. Made me stop over and think before jumping
>into answer. Yes i do agree that IIR are computationally better than IFIR.
>But i think it will be comparable to that of IIR. By adjusting L we can 
>get up to 80-85% reduction for a passband = 0.005fs. When i just did a
>rough calculation it turns out that for IIR (Butterworth 7th order) we
>would require 21 Multiplies and the same implemented with IFIR(~400 tap)
>would require 30-40 Multiplies. So yes IIR are better but IFIR are not too
>behind . So if one is ready to pay for extra multiplies IFIR may be a
>better choice since we get linear phase. 

Hi,
  Another issue to remember is that if we use 
the "folded tapped delay line" implementation 
in IFIR filters, we can reduce the number of 
necessary multiplys (at the expense of an 
increased number of addtions), by a factor 
of two, per output sample in IFIR filters.

See Ya',
[-Rick-]

>Hi Rick,
>         Thanks for the reply. Made me stop over and think before jumping
>into answer. Yes i do agree that IIR are computationally better than
IFIR.
>But i think it will be comparable to that of IIR. By adjusting L we can 
>get up to 80-85% reduction for a passband = 0.005fs. When i just did a
>rough calculation it turns out that for IIR (Butterworth 7th order) we
>would require 21 Multiplies and the same implemented with IFIR(~400 tap)
>would require 30-40 Multiplies. So yes IIR are better but IFIR are not
too
>behind . So if one is ready to pay for extra multiplies IFIR may be a
>better choice since we get linear phase. 
>

How can you get 21 multiplies for 7th order Butterworth?