Additional filtering of the complex baseband signal

[cochenob]

I suppose this is more of a DSP question rather than the USRP hardware itself...but still quite relevant, so I'll give it a go.

In this widely linked to example (https://decibel.ni.com/content/docs/DOC-25716), the authors simultaneously measure the signal of interest, and a 'local reference'.  They do this by placing each signal at an IF relative to the center frequency of the USRP, and then using bandpass filters to isolate each signal in order to the proceed with calculating relative phase.

Using the VI's under the Signal Processing->Filters pallate...I started playing around with filters and observing what the filtered spectrum looks like.  I selected the input type as "Complex Double" (presumably, under the code interface node, the VI is filtering the I and Q components seperately, and then recombining after filtering).  I observe that the filtering is symmetric around the center frequency.  i.e. a BPF from 200kHz to 300kHz also BPFs from -200kHz to -300kHz.  I suppose this makes sense....but raises a fundamental question:

Can I design a filter that operates on complex data that doesn't exhibit this symmetry?  Or am I fighting Math/Physics?

If not...this would seem to suggest in the above example, that the signal of interest and local reference calibration signal are located at different IF's from the center frequency in order to resolve this issue.  I tried to verify this myself by looking at the MUSIC code...to see exactly where their filters are centered...but I am unable to click into and expand the Matlab code blocks (do these become clickable only if you have Matlab installed)?

Hoping someone can give some advices as to how this simultaneous capture and subsequent filtering is actually carried out.  Thanks!

---

BC

[PeanutButterOven]

cochenob,

Symmetry within a filter exsists for a lot of reasons, in lumped element design physical symmetry can reduce the amount of necessary components by half, in EM simulators and design environments symmetry often leads to a reduction in processing time and memory usage (I've heard by factors as large as x16).

You are asking specifially about response symmetry of filter design, where the first thing that comes to my mind is implications with respect to the transfer function and the effects on phase. A symmetrical filter will maintain a more linear group delay and lessen distorions created from the signal passing through the filter. Creating asymmetrical digital filters is entirely possible (asymetrical Hilbert Transform for example) but will result in non-linearites within the phase response. If you want to know more there is a plethora of information on the web concerning asymetrical digital filter design. 

As for the m files I believe that you would need Mathworks Inc Matlab ^® software to run the .m files but any text editor should allow you to view them. 

[cochenob]

Thanks.  I've been searching around a bit and trying to catch up on Complex Band-Pass Filters.  Indeed, the Hilbert Transform shows up a lot.  I didn't consider the non-linear phase response, which may be problematic for phase sensitive measurements.

Regarding trying to view the Matlab code within LV....I don't believe the example I linked to above includes the .m files for the filter blocks that show up in the LV block diagrams.  Rather, the code blocks behave like un-editable pictures.  No response to left/right clicking, resizing, etc..  Is there any way to view this code without Matlab?  Seems like there should be.

---

Brandon

[cochenob]

Thought I'd see if LabView has any built-in solutions for this yet.

I want to bandpass filter my baseband IQ data.  i.e. - I have two data streams...one in a band >0 Hz and one in a band <0 Hz.  Is there a straightforward way to separate them, so should I be looking to do some more frequency translation to the get negative frequencies positive, and then use the standard filters?

Anyone have experience with this?

---

Brandon