Butterworth Filter Calculation Time in LV

[bilik]

I have found very strange behaviour of LV butterworth filter. If I use signal some AC signal first, followed by DC (zero) samples, then calculation time increases from aprox. 0 ms to 10ms. Using WinXP or Win2K and LV6.1 or LV7.0 with the same results.

Please see example at:
http://dvi.elcom.cz/~bilikp/TestFilter.vi

or make your own example. Does anybody have idea what can be wrong? Thank you in advance for any help.
Petr Bilik

[LesHammer]

I'm using LV7.0 on WinXP.
Your example shows a flat "0" in the CalcTime display.

Les Hammer

[bilik]

I tried it again on 5 indenpendend PC computers, always with same results in LV6.1, LV7.0, LV7.1. I even changed my example to find out behaviour of different filter types. High calculation time phenomena is just a question how long this example runs. Anyway on normal PC (P4/2GHz) you can see it in 30seconds.
Please see new example at:
http://dvi.elcom.cz/~bilikp/TestFilter2.vi

Petr Bilik

[bilik]

In the meantime I tested this example under Linux version of LV7.0, using Linux Mandrake 9.1. Again I got high calculation time after few seconds of run.

[LesHammer]

Ah, I see what you mean. Your original description was confusing. I thought it was the 0->1 transition that was the problem, not 10 seconds later.

Sorry, I don't have an answer for that one yet.

[LesHammer]

And the high calculation time happens sooner if you initialize the arrays to a larger size, and later or not at all if you make the arrays smaller.

Looks like a memory manager issue to me.

Les

[DSPGuy]

I looked at the state of the filter when the change in execution speed occurs. The state is very small (~ E-322), and does not change. It looks like the filter output is decaying to zero, and reaches a point where the filter numerics underflow. If you add a very small amount of noise to the zero-input-signal the behavior disappears.