Decimation filter bandwidth

[cochenob]

Does anyone have information as to what the decimation filter BW is given a particular IQ rate? 

For my application, it is more important to be able to vary/control the BW around the carrier.  It would help to know how to "work backwards" from IQ rate since this is the variable that the driver uses.

I'll assume that the IQ rate and filter BW have a fixed relationship regardless of the decimation factor, such that the relationship at high IQ rate (low decimation) is the same at low IQ rate (high decimation)?  Is there a "rule of thumb" calculation that would get me this info?

For what it's worth....defining the IQ rate doesn't seem to be a very useful parameter. Rather, you're more likely to know the bandwidth that you want to capture, and would then just let the decimation occur based on that.  Otherwise....if you're not careful (or don't know what you're doing) you might unintentionally filter/decimate and end up looking at a smaller chunk of BW than you think....all in the name of reducing the rate data is streaming from the USRP.

The NI A/D's with DDCs seem to do this already.  You specify the data BW, and the DDC configures the decimation based on the closest integer factor of the A/D sample rate.  There's a handy little table in the data sheets that shows for a given BW, what the resultant IQ data rate is.  I'm basically looking for something similar for the USRP.

Thanks!

---

Brandon

[Anthony_F · Accepted solution]

This might be a subtle understanding I've gained from using RF equipment bu thet IQ rate is what ultimately limits your maximum bandwidth. How much of that bandwidth is useable depends from device to device. (True for both acquisition and generation)

In the ideal world, your IQ rate would "equal your bandwidth" (acquisition).  This is the general rule of thumb.  If you take an FFT on some number of samples at a rate of 10 MSa/s, you will end up with a frequency plot that spans from -5 MHz to +5 MHz. Whether or not your signal uses that entire 10 MHz span isn't implied but you would not be able to fully acquire a signal with a bandwidth larger than 10 MHz for sure.

The USRP has specific IQ rates it supports, so I suggets always reading back the coerced IQ rate.  If you need to change your filter specifications or need more flexibility than just the decimate and filter process alone, I woud suggest implementing a filter in software -- your choices open up considerably as well by being able to choose a butterworth over a chebyshev and specifying the order of the filter, etc.

I would also suggest reading this article as well from the GNU Radio page, it has a little more depth and background.

[cochenob]

Thanks Anthony-

This was helpful.  I suppose I could always try to measure the magnitude response of the filters myself by varying the input carrier a bit if I wanted to see how close to the filter 3dB point is to a certain IQ rate/decimation factor.

Thanks!

---

Brandon

[TYzhang]

Hello Anthony,

I'm using my USRP2920 to  record RF signal and I wonder how USRP  implementing NCO and filters in DDC.Is NCO using cordic algorithm or Table lookup?Are filters CIC&halfband filters or just FIR?My system need accurate phase so I need to know how it works in hardware.I wonder if you can give me

some information about this if  it's possible.I don't  need any other details at all.

Thanks!

TYZhang

[Nathan-P]

The USRP is an open hardware platform, and the hardware schematics can be found here:

http://code.ettus.com/redmine/ettus/documents/19

https://www.ettus.com/content/files/07495_Ettus_N200-210_DS_Flyer_HR_1.pdf

Additional Resources:

http://digital.ni.com/manuals.nsf/websearch/1BA4C1B73D184AA886257AB4006F7F0D

[TYzhang]

Thanks,Nathan! I find what I need at here http://code.ettus.com/redmine/ettus/projects/uhd/repository In folder fpga there are some files named "cordic" and "cic",that's what I need. Thanks! TYZhang