Forums › BB Series Discussions › noise equivalent bandwidth of zero span IQ recordings
- This topic has 5 replies, 4 voices, and was last updated 5 years, 5 months ago by
Dan0.
- AuthorPosts
Dan0ParticipantDear Signal Hound,
Possibly this is a Spike-related question, rather than BB60 specific, but I’ll ask it here to start.
We are performing RF noise measurements and I/Q recording in zero span mode on the BB60C, and we are interested in determining the noise equivalent bandwidth of our I/Q recordings in zero span. Our frequencies of interest range from 30 MHz up to 1 GHz and we are generally using decimation 32 to achieve a nominal 1 MHz bandwidth for recording and analysis. We are also using external preselection bandpass filters, and are trying hard to make sure the I/Q recording bandwidth falls entirely within the passband of the external filter.
Looking at the spectrum display within zero span mode, it appears that there is some kind of roll-off at each edge of the analysis bandwidth… I assume this is not an ideal “brick wall” kind of thing, but probably has some filter shape that we’d need to account for if we wanted to determine the noise equivalent bandwidth of a zero span recording.
Can you shed any light on this? Thanks in advance.
AndrewModeratorHi Dan0,
The spectral window in zero-span mode is plotted using a width equal to the sample rate. The rolloff you see is the software filter we apply to the data. The filter cutoff (6dB point) is selectable in the zero-span controls. The filter is a standard windowed sinc filter using the blackman window. The size of the filter does vary based on decimation and there can be cascaded filters at higher decimations.
Let me know if you have follow up questions.
Regards
dbretonParticipantHi Andrew,
Thanks for your post above, and I apologize for the change in login name — I forgot my password. I’m going to walk through what I understand so far, and hopefully you can help me pick up the pieces after that.
In the case above with decimation of 32, this implies a sample rate of 40 MS/sec / 32 = 1.25MS/sec as shown on the GUI, and therefore the bandwidth of the spectrum shown is 1.25 MHz. You’ve applied a filter and Blackman windowed the data, which results in the rolloff we see in the spectrum window.
So: decimation controls the sampling rate, and the “IF Bandwidth” control in zero span mode controls the filter cutoff points, if I’m understanding things correctly. Question 1: Is this same filtering applied to the raw I and Q data recorded in the .iq files?
In the case above, we’ve chosen IF bandwidth of 1 MHz, so this is controlling the 6 dB down points of said filter… if we are only controlling the 6 dB down points, the effective noise bandwidth (ENBW) of this filter must be somewhat greater than 1 MHz, correct? Question 2: how much greater, and how can I determine/estimate the filter ENBW based on the description of the windowed filter you’re using?
I’ve looked a lot of places trying to find the ENBW for the combination of window and filter you’re using, but typically I find references to the resolution bandwidth of spectrum analyzers and so on, and I’m not sure these are really applicable to the ENBW for an IQ recording. An example is this page.
We’re using a calibrated noise source to assess our system gain and noise figure, but until we have a handle on ENBW for the BB60C, we’re floundering a bit. Any help/discussion would be appreciated.
AndrewModeratorHello Dbreton,
1) It sounds like your understanding is correct. When you select an IF bandwidth of 1MHz, we use a filter with cutoff points (6dB down) that have 1MHz banwdidth. This filter is also applied for the I/Q recordings.
2) (edit) See Justins response below. Also see this paper that lists various window functions and their ENBW/3/6dB bandwidths. See Eq. 11 for calculating ENBW.
https://www.utdallas.edu/~cpb021000/EE%204361/Great%20DSP%20Papers/Harris%20on%20Windows.pdf
We use the “blackman” window from this paper with the 1.73 bin ENBW.Let me know if you have follow up questions.
Regards
Justin CrooksModeratorThe ENBW is going to be very close to the selected bandwidth. My guess would be at most a 5% difference, or maybe ~0.2 dB error.
Dan0ParticipantGentlemen, thank you for your replies & links, this is very helpful; I think we are on the right track now.
- AuthorPosts
You must be logged in to reply to this topic.