Official bladeRF forum

I am debugging a system on the bladeRF, the system is something like ATSC. The receiver performance is suffering. I tried lots of configuration of LMS6002D, but cannot find the solution, can you please help to find the problem, or tell me how to debug to improve the performance.
Thanks.

My working environment is as following:
1. the source signal is from RF signal generator, output level -10dbm, the source signal is good.
2. the receiver is bladeRF 115x, with LMS6002D
3. the central frequency is 786MHz
4. the signal bandwidth is 8MHz
5. the RX samplerate is 30.375M
6. the RX part configuration is: RXLNAGain=0db / RXVGA1=5db / RXVGA2=0db
7. the modulation mode is 4QAM
8. We get signal from ADC output of LMS6002D, and probe them by logic analyzer.

After analyzing the probed data, we see there is problem of angle error in receiving signal, but the receiving signal amplitude seems good. As following:
1, The constellation of received signal is shown as below.
From the constellation, it is obvious that there is severe phase noise or phase error. 2, The amplitude of received signal seems good. 3, Observing the phase error, we find that there is large phase variation on the receiving signal and the error signal is a periodic signal. The symbol rate is about 7.5MBaud, so the frequency of the error signal is about 15KHz.
This phase error may degrade the overall C/N performance about 1.5dB. So, our question is where comes from the periodic phase error signal? How can we eliminate this phase error?

What frequency is the TX side tuned to while you're doing this?

Do you see better performance if you ensure TX is tuned further than +/- 1 MHz away from RX?

Take a look at this post. We've found that this is a symptom of the LMS6002D PLLs interfering with one another, hence Lime's recommendation to keep them 1 MHz apart. Upon learning this, I've tried to make this recommendation in the libbladeRF docs.

Let us know whether or not this helps.

- Jon

thanks Jon,
We only use LMS6002 as RX receiver, the TX part is not in use, so we set the LMS6002 TX frequency at 2.4G, as far from RX as I can...
But we do find something interesting by adding frequency offset. The source signal is 786M, when we add an offset at the 6002 RX frequency (set RX frequency=782M), the system sensitivity turn better, ie. 1db better than no offset. is this also from PLL interfering?

Was the original QPSK plot after residual frequency offset recovery? I see there is no DC offset as well, so I am guessing you removed that with post processing?

The offset-tuned extra sensitivity might come from the IQ imbalance going to the opposite side of the spectrum and you filtering it out instead of having it land on top of your spectrum under your actual signal.

You shouldn't drive the device at -10dBm. Try to keep it lower - maybe around -40dBm or lower? Can you link to your received samples?