IIR filter - sample instants to flush out existing data

Hi,

I have a question on IIR filters. I have an IIR filter running on a
TMS DSP.

Suppose I have an FIR filter with 'a' taps. If the filter is running
and I want to flush the existing data at each of the taps out, it
would take me 'a' sample instants. Is that correct?
Now, if I have an IIR filter with 'a' feedforward and 'b' feedback
taps, roughly how many sample instants would it take before I could
safely say that the filter has been flushed off all previous samples?
i.e. how long before I can say that the filter o/p is due to the new
input data?

In theory, due to the feedback paths, it would take me infinitely
long. But is there a reasonable approximation for the number of sample
instants I can arrive at?

Any inputs on how I go about calculating the approximate number of
sample instants (through matlab code etc.) would be greatly
appreciated.

Thank you,
Madhu

On Apr 5, 5:11 pm, "Madhu" <balding_spald...@inbox.com> wrote:
> Hi,
>
> I have a question on IIR filters. I have an IIR filter running on a
> TMS DSP.
>
> Suppose I have an FIR filter with 'a' taps. If the filter is running
> and I want to flush the existing data at each of the taps out, it
> would take me 'a' sample instants. Is that correct?
> Now, if I have an IIR filter with 'a' feedforward and 'b' feedback
> taps, roughly how many sample instants would it take before I could
> safely say that the filter has been flushed off all previous samples?
> i.e. how long before I can say that the filter o/p is due to the new
> input data?
>
> In theory, due to the feedback paths, it would take me infinitely
> long. But is there a reasonable approximation for the number of sample
> instants I can arrive at?
>
> Any inputs on how I go about calculating the approximate number of
> sample instants (through matlab code etc.) would be greatly
> appreciated.
>
> Thank you,
> Madhu

It depends on the impulse response of the filter. If you think of each
sample you put in as an impulse, the decay time of the impulse
response to an insignificant level is how long that sample influences
the output.

John

Madhu wrote:
> Hi,
> 
> I have a question on IIR filters. I have an IIR filter running on a
> TMS DSP.
> 
> Suppose I have an FIR filter with 'a' taps. If the filter is running
> and I want to flush the existing data at each of the taps out, it
> would take me 'a' sample instants. Is that correct?

Why do you need to "flush" the filter instead of zeroing all the data 
memory at once?

> Now, if I have an IIR filter with 'a' feedforward and 'b' feedback
> taps, roughly how many sample instants would it take before I could
> safely say that the filter has been flushed off all previous samples?
> i.e. how long before I can say that the filter o/p is due to the new
> input data?
> 
> In theory, due to the feedback paths, it would take me infinitely
> long. But is there a reasonable approximation for the number of sample
> instants I can arrive at?

It depends on the length of the impulse response, but you are at liberty 
to set all the storage elements to zero (or any other suitable initial 
condition) as fast as you can address the storage elements.

> Any inputs on how I go about calculating the approximate number of
> sample instants (through matlab code etc.) would be greatly
> appreciated.

If this is a practical question, don't bother to flush it; bash it.

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

On 2007-04-05, Madhu <balding_spalding@inbox.com> wrote:
>
> Any inputs on how I go about calculating the approximate number of
> sample instants (through matlab code etc.) would be greatly
> appreciated.

Others already told you to look at the sample response.  In Matlab
you can do that with impz.

Someone mentioned 'smashing' the values as a practical solution, but
there's still the question of how long after you start feeding real
data into an IIR filter before the output is not strongly affected
by the startup conditions.

-- 
Ben Jackson AD7GD
<ben@ben.com>
http://www.ben.com/

Madhu wrote:

> I have a question on IIR filters. I have an IIR filter running on a
> TMS DSP.
> 
> Suppose I have an FIR filter with 'a' taps. If the filter is running
> and I want to flush the existing data at each of the taps out, it
> would take me 'a' sample instants. Is that correct?

Yes.

> Now, if I have an IIR filter with 'a' feedforward and 'b' feedback
> taps, roughly how many sample instants would it take before I could
> safely say that the filter has been flushed off all previous samples?
> i.e. how long before I can say that the filter o/p is due to the new
> input data?
> 
> In theory, due to the feedback paths, it would take me infinitely
> long. But is there a reasonable approximation for the number of sample
> instants I can arrive at?

As Jerry suggests, zeroing-out the filter's memory might be the best 
option in many applications.  However, to answer your question 
specifically, what I would do is:

1) Determine a threshold at which you deem the filter to be flushed out. 
  For example, if your full-scale input is 1.0, you might set 1.0e-6 as 
the threshold.
2) Zero all memory elements, put in a full-scale impulse, and run the 
filter until the magnitude of the output is less than the threshold for 
some number of samples, N.

This procedure allows you to set set a practical limit on the IIR 
filter's otherwise "infinite" impulse response.  The value of threshold 
and N are empirical values that you must determine relative to your 
application.

=g2

Madhu wrote:

> I have a question on IIR filters. I have an IIR filter running on a
> TMS DSP.
> 
> Suppose I have an FIR filter with 'a' taps. If the filter is running
> and I want to flush the existing data at each of the taps out, it
> would take me 'a' sample instants. Is that correct?

Yes.

> Now, if I have an IIR filter with 'a' feedforward and 'b' feedback
> taps, roughly how many sample instants would it take before I could
> safely say that the filter has been flushed off all previous samples?
> i.e. how long before I can say that the filter o/p is due to the new
> input data?
> 
> In theory, due to the feedback paths, it would take me infinitely
> long. But is there a reasonable approximation for the number of sample
> instants I can arrive at?

As Jerry suggests, zeroing-out the filter's memory might be the best
option in many applications.  However, to answer your question
specifically, what I would do is:

1) Determine a threshold at which you deem the filter to be flushed out.
  For example, if your full-scale input is 1.0, you might set 1.0e-6 as
the threshold.
2) Zero all memory elements, put in a full-scale impulse, and run the
filter until the magnitude of the output is less than the threshold for
some number of samples, N.

This procedure allows you to set set a practical limit on the IIR
filter's otherwise "infinite" impulse response.  The value of threshold
and N are empirical values that you must determine relative to your
application.

=g2

Grant Griffin wrote:
...

Sorry for the double reply...haven't used this darn thing for awhile. ;-)

=g2

Grant Griffin wrote:
> Grant Griffin wrote:
> ...
> 
> Sorry for the double reply...haven't used this darn thing for awhile. ;-)

Grant,

I'm delighted that you're using it now. How's the 12-step program 
coming? :-)

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

Jerry Avins wrote:
> Grant Griffin wrote:
>> Grant Griffin wrote:
>> ...
>>
>> Sorry for the double reply...haven't used this darn thing for awhile. ;-)
> 
> Grant,
> 
> I'm delighted that you're using it now. How's the 12-step program 
> coming? :-)

Thanks, Jerry.  :-)

Actually, I think I'm doing better than most addicts ever could hope: I 
seem to be able to imbibe every once and awhile, yet not get addicted again.

That said, I'm glad to see that you, r b-j, Eric, Rick, Randy, et. al. 
(sorry if I've left anybody out--those are just the old friends I've 
seen here lately) have never succeeded at all 12 steps. ;-)

=g2
_____________________________________________________________________

Grant R. Griffin
Publisher of dspGuru                          http://www.dspguru.com
Iowegian International Corporation            http://www.iowegian.com
See http://www.iowegian.com/img/contact.gif for e-mail address

Thanks for your inputs, everyone. I will let you guys know how things
go.

Thanks,
Madhu