"Mirroring"/Ghosting and CF Peak

[lousyg]

Disclaimer: I'm a complete beginner at SDRs, let alone RF, so what I'm seeing may be completely normal or explainable.

I've managed to get GNURadio and gr-osmocom running on my Fedora 19 machine. However, whenever I look at any spectrum in osmocom_fft, I see two strange features.

1. My center frequency has what I can only describe as a "spike". No matter what frequency I tune to, it appears. If I shift the spectrum by a few Hz to a center frequency with no carrier, the spike shifts with it.

2. When picking up signals that are slightly offset from the center frequency, I see what I can only describe as "mirroring" or "ghosting" at the same, but opposite offset. For example, if I set the FFT CF to 314.96 MHz and hit a button on my car remote (315MHz), I see a signal not only at 315MHz but a VERY similar waveform at 314.92MHz.

To better show these two observables, I've created a screenshot and uploaded it here: http://i.imgur.com/67OCGLD.png

Does anyone know what I'm seeing? Is there a problem with what I'm doing? Is this a bug in the driver or do I perhaps have a faulty board?

Thanks for your help!

[syntroniks]

This seems like completely normal behavior.

[lousyg]

At this point, I'm very inclined to believe that there's either a problem with the driver/software (or my installation of it) or my board.

Here are two companion screenshots with low/high pass filters applied.

Low Pass Filter: http://imgur.com/olgzfvz

High Pass Filter: http://imgur.com/xtqYZSC

Once again, I'm not an RF guy (yet). But based on everything I can find about low and high pass filters, this is not what I would expect to see.

Please help!

[lance]

I don't have my board running (on the todo list...). And I don't have any experience with any of the mainstream SDR software.

But to me it looks like two things: the central spike is simply DC/DC offset -- not much to worry about right now. The "mirroring" looks to be either due to the fact that you are only using I or Q samples (effectively a real dataset rather than complex), or the I/Q balance is way off. I'm not sure how the I/Q (phase and amplitude) balance could be off that far unless one path was broken or turned off-- normally you would see some reduction in the "mirror" (or image) if you had even the hint of a working I/Q demodulator, but those look almost equal.

So using the "I or Q isn't getting sampled properly or used in the computation theory", your FFT display shows what you would expect when you sample the output of a direct conversion receiver that has only one branch (i.e. no quadrature / image rejection).


While this example is for a transmitter, the same structure and mechanism occurs in the receive side -- there is a ton of information out there http://www.dsplog.com/2009/03/08/iq-imb ... ansmitter/

Lance

[lousyg]

So using the "I or Q isn't getting sampled properly or used in the computation theory", your FFT display shows what you would expect when you sample the output of a direct conversion receiver that has only one branch (i.e. no quadrature / image rejection).

Thanks for the response! Would your explanation also hold true for what I'm observing with the low/high pass filters?

[lance]

So using the "I or Q isn't getting sampled properly or used in the computation theory", your FFT display shows what you would expect when you sample the output of a direct conversion receiver that has only one branch (i.e. no quadrature / image rejection).

Thanks for the response! Would your explanation also hold true for what I'm observing with the low/high pass filters?

If I understand your question correctly, yes... At first glance The lowpass/highpass filters seem to be operating the way I would expect (at least they look like LPF/HPF functions, I didn't look at whether they were at the right frequencies for what you were trying to accomplish).


-Lance

[madengr]

To better show these two observables, I've created a screenshot and uploaded it here: http://i.imgur.com/67OCGLD.png

You are sampling at 160 kHz with the filters set to 2.5 MHz. Those may be aliases rather than image due to I/Q imbalance. Set the filters to 1.5 MHz (minimum they go; they are really 750 KHz) and sample at least 1.5 MHz.

Then you can due a frequency translating FIR filter with decimation to get you back down to a reasonable BW for your spectrum, and avoid the DC center. I'll try posting a GRC block later.

There are internal DC offset calibrations in the LMS, but I assume they have to been implemented yet. I was attempting to do so here:

http://nuand.com/forums/viewtopic.php?f=4&t=2858

Ideally it would do a quick DC compensation every time any setting is changed.

[lance]

whoops, missed the sampling rate!

[lousyg]

You guys are great! Thanks for the responses. After I get home from work, I'll look into your suggestions.

[madengr]

That you Lance (RF Dude)? Lou here.

lousyg, do those gains settings work? Last I tried I don't think they were all implemented. I think RXVGA1 tops out at 6 dB. I tried programming them in GRC, then read back the values in the CLI, and they did not match.

[lance]

That you Lance (RF Dude)? Lou here.

Yes sir! Fancy meeting you here.

[madengr]

Likewise. Have not been to IMS since we met last. I'll definetly be going next year.

[lance]

Likewise. Have not been to IMS since we met last. I'll definetly be going next year.

Lou, Yeah -- I went in 2011 and 2012 (both were on east coast and about equal driving distance), but this year decided not to. I'm not sure if it is in the cards for next year. Hope all is well with you.

[LazyDodo]

Back on subject, Tip i got on IRC : try setting lms register 0x5a to 0xa0 to get rid of the mirroring.

[lousyg]

Back on subject, Tip i got on IRC : try setting lms register 0x5a to 0xa0 to get rid of the mirroring.

That appears to have worked. Now, will this setting be persistent across power cycles?

Any recommendations on getting rid of the DC peak at CF?