Complex bandpass vs real low-pass

GNU Radio
1

I have been in the habit of using complex-bandpass filters even when I’m
just looking for a symmetric response. I did some measurements
today and found a 40-45% performance advantage to using a low-pass
real filter when a symmetric response is desired. What’s the “crux”
of the performance difference?

Needless to say, I ran around and changed all my recent “stuff” to use
low-pass filters :slight_smile:

–
Marcus L.
Principal Investigator
Shirleys Bay Radio Astronomy Consortium

2

On Wed, May 2, 2012 at 2:55 PM, Marcus D. Leech [email protected]
wrote:

I have been in the habit of using complex-bandpass filters even when I’m
just looking for a symmetric response. I did some measurements
today and found a 40-45% performance advantage to using a low-pass real
filter when a symmetric response is desired. What’s the “crux”
of the performance difference?

Needless to say, I ran around and changed all my recent “stuff” to use
low-pass filters :slight_smile:

Complex * complex multiplies are more computationally expensive than
complex * real. For edification and titillation you can read the
volk_32fc_x2_multiply_32fc source. There’s an Orc version too which is
easier to read. The moral of the story is you should probably avoid
complex
coefficients unless you know you need them.

–n

3

On 05/02/2012 05:59 PM, Nick F. wrote:

Complex * complex multiplies are more computationally expensive than
complex * real. For edification and titillation you can read the
volk_32fc_x2_multiply_32fc source. There’s an Orc version too which is
easier to read. The moral of the story is you should probably avoid
complex coefficients unless you know you need them.

–n

Oooh, titillation. Sign me up :slight_smile:

Yup, that makes sense.

–
Marcus L.
Principal Investigator
Shirleys Bay Radio Astronomy Consortium