Multiple signals - messy spectrum

[Zortz]

Hi!

I am trying to use bladeRF to process multiple signals in one bandwidth, i ran into problems related to the I/Q calibration.
As can be seen from the image, there are significant peaks in addition to 2 sine signals. I have tried to to compensate for phase mismatch and gain mismatch, but when i compensate for spur 1 then spur 2 gets stronger and vice-versa.
It seems that the optimal setting is when both spurs are about 20 dB from the noise level.

Is there a way to clean up this spectrum?

Thank you!

[bpadalino]

This is the output of the TX spectrum from a bladeRF?

What is txvga1 and txvga2 set at? How much dynamic range are you using on the sample stream going down? How much SSB suppression do you expect to get? What is the baseband signal supposed to be?

[Zortz]

Thank you for the reply bpadalino.
I am using bladeRF as a relay device at this moment, so the previous image shows a situation where test signal passes trough bladeRF RX and then TX.
This time i am going to use only RX module, for simplicity.

Here is the starting point for my measurements (i dont care for the DC component at the moment):


Now i will input a single sinusoid to bladeRF:
Frequency: 1.01G
Level: -50dBm
Notice that i corrected the IQ imbalance and managed to completely remove the spur.


Question 1: What is up with those side-lobes?
I constructed following GNURadio flowgraph for further inspection:

Spectrum:

And zoomed in to sine frequency:


So what is going on with this sine wave?

Question 2: Spur removal for multiple input signals?
So for one input i was able to fully compensate for the spur. Now for two signals i am not.
This seems to be the best result:

I can remove the lower spur, but then the higher spur increases even more (same story if i try to remove the higher spur):


So it seems that the system can only provide ~ -35dBc attenuation. To meet standards i would need as high as -55dBc, is there any way to achieve that?

All help appreciated

[bpadalino]

OK - there's a lot of stuff here.

First lets try doing a quick hardware mod if that's OK with you. Can you remove and/or short FB12 (upper right corner of the board) and tell me if the spectrum looks better.

For an experiment with the two tones, can you try something for me and manually calibrate out the single tone image, and then slowly move the tone to the other side and see if/when the image pops up again?

[Zortz]

OK - there's a lot of stuff here.

First lets try doing a quick hardware mod if that's OK with you. Can you remove and/or short FB12 (upper right corner of the board) and tell me if the spectrum looks better.

Shorted FB12 spectrum:

Zoomed:

Looks a bit better, but still problematic...

For an experiment with the two tones, can you try something for me and manually calibrate out the single tone image, and then slowly move the tone to the other side and see if/when the image pops up again?

Calibrated single tone:


Spur reappears when +500kHz added to tone:


Thank you for the support

[Zortz]

I found a resource about one guy having similar problems with I/Q balance:
https://groups.google.com/forum/#!topic ... EduAeZJcjM

He solves the problem and refers the solution to be at umtrx mailing list - that sadly seems to be unavailable today... So thats that.

In the same thread there is a good document about LMS "SSB calibration vs CW offset":
https://docs.google.com/viewer?a=v&pid= ... MS4xAQF2Mg
I can see that my results of -35dbC could be realistic (of course they cut the graphs before the SSB goes below 50dbc ).

Do you have any additional ideas bpadalino?

[bpadalino]

Try setting your TX frequency more than 10MHz away from your RX frequency so the 2 PLL's are not that close to each other.

See if that helps to improve the spectrum.

[Zortz]

I am only using the RX module for these tests, the input tone is generated with R&S Vector signal generator.
Nevertheless i set TX to 900M while RX is 1G - as expected, nothing changed.
TX spectrum looks the same as RX - there are +-(N * 550kHz) components in addition to the main lobe.

[Zortz]

I have now tested the bladeRF TX module and i can see similar problems.

Here is a spectrum picture of an tone generated with R&S generator - this could serve as an reference:

refSignal.png

Now, to compare this to bladeRF output i used osmocom_siggen to generate single tone:

Code: Select all

osmocom_siggen --samp-rate=40M --tx-freq=1G --waveform-freq=10M --sine --gains=VGA1:-12 --amplitude=1

gen.png

Resulting spectrum:

bladeTX.png

There is a +-(N*78.1 kHz) spur problem. Note that i have removed and shorted FB12 component from my board, as suggested by bpadalino - that had an positive effect on RX spectrum.

Am i the only one with this kind of spectrum?

[Zortz]

I have verified that gain settings nor DC offset or IQ balance contribute to those spurs.

Sampling rate seems to affect the results, here are results for 20M and 10M sampling rates:

[bpadalino]

Interesting it wasn't the PLL issue - I've seen that cause this type of issue. The RX and TX PLL's can internally pull on each other since they are both on all the time.

If you just use a constant to transmit and output a tone using just DC, does that exhibit the same spurs? If so, if you short out FB11, does it appear that the spectrum changes? You should be able to see results just by using tweezers while it's running and short the sides.

With TX, I've seen the PLL fighting issue cause lots of problems. Otherwise I've just seen tones with the phase noise skirt out to around +/-50kHz or +/-100kHz, depending on where you think the skirt ends.

I know there are large harmonic sample rate spurs that can show up as well. I am not sure what's best with the sample rates. When doing receive only applications, I've tried setting the TX samplerate to 160kHz just so it would be extremely slow and it would be many more harmonics before it would get into the RF.

[Zortz]

Interesting it wasn't the PLL issue - I've seen that cause this type of issue. The RX and TX PLL's can internally pull on each other since they are both on all the time.

If you just use a constant to transmit and output a tone using just DC, does that exhibit the same spurs? If so, if you short out FB11, does it appear that the spectrum changes? You should be able to see results just by using tweezers while it's running and short the sides.

With TX, I've seen the PLL fighting issue cause lots of problems. Otherwise I've just seen tones with the phase noise skirt out to around +/-50kHz or +/-100kHz, depending on where you think the skirt ends.

I know there are large harmonic sample rate spurs that can show up as well. I am not sure what's best with the sample rates. When doing receive only applications, I've tried setting the TX samplerate to 160kHz just so it would be extremely slow and it would be many more harmonics before it would get into the RF.

Here is constant tone spectrum:

constant.png

Shorting FB11 had no effect at all. Lobes are not stable, jumping about 3db.
This spectrum is similar to the figures presented in http://www.analog.com/media/en/technica ... AN_927.pdf . Maybe some kind of power supply noise?

Waiting for further instructions, master

[Zortz]

Okay now i get different results.

Spectrum changes when i tune the TX frequency to X and then back to 1G. I used 300M or 3.5G for X atm.
Here are some results, all measured at 1G:

[bpadalino]

The big spikes that come up when you output DC only is due to the PLL's fighting. Change your RX PLL to something like 433MHz or something not close to the 1GHz you are transmitting. Or change your transmit to some other area. You should see it go away.

The spikes to the left and right of the tone you're transmitting we've been able to recreate and are looking into it.

Can you let us know what versions of the library, FX3 firmware and FPGA image you're using?

[Zortz]

It seems that i am not using the latest firmware:

Code: Select all

  bladeRF-cli version:        1.2.1-git-19c7f7c
  libbladeRF version:         1.4.3-git-19c7f7c

  Firmware version:           1.8.0
  FPGA version:               0.3.3

I used the apt-get method for installing just a few weeks ago, it seems that those repositories are not updated.

After i updated to latest firmware:

Code: Select all

  bladeRF-cli version:        1.2.1-git-19c7f7c
  libbladeRF version:         1.4.3-git-19c7f7c

  Firmware version:           1.8.1
  FPGA version:               0.4.1

The lobes around DC output was due to RX PLL. I once again changed the RX to lower frequency and they went away, but the 78.1kHZ spurs for tone signal remained. As i understand you were able to reproduce it?

Dual tone spurious emission retests:
I have retested the TX and RX module for dual tone(20M separation) spurs, due to I/Q imbalance:

1gTxDualSpur.png

As can be seen i am already getting close to my -55dBc goal. There are some other spurs present ...maybe related to this 78.1kHz spur?

Also retested dual tone for RX(TX set to 300M) and sadly results are the same - wont even make it to -40dBc

1gRxDualSpur.png

We are currently evaluating this SDR development path and while it seems promising we are afraid that the quality is just not good enough. So if you have any other ideas/solutions to these problems please let me know
Thanks for all the support.