AD9371 receive path spikes

Hello,

I have started to see spikes in the data from the AD9371 receiver and would like to understand the cause and eliminate them.

Initially I tested the AD9371 with the default centre frequency of 2.55GHz. The data was captured from JESD into RAM and analysed over of a range of test tone frequencies and amplitudes. The amplitudes were measured in -10dbm intervals down to -100dbm and no-signal. At the lower end the output was noise in the range of +/-10 counts.

Altering the centre frequency to 1.3GHz and re-measuring now shows single sample noise spikes in the input signal when significantly lower than 8-bits (eg/ +/-40 counts). Spikes are in the range -256 to -249 counts, and +248 to +255 appear over about 1% of the samples (example time domain plot of captured test data attached showing problem).

I am surprised that just changing the centre frequency would cause this, however now I am wondering about the LO/PLL and problems with it, and things like decoupling. The hardware is a AD9371 evaluation board.

Any ideas as to the cause of this and how to eliminate the source of the noise?

AD9371-receive-path-spikes.pptx
    •  Analog Employees 
    on Jul 19, 2018 1:02 PM

    Are you using TES GUI or IIO?

    We did the same with TES GUI and could not reproduce the issue . Can you share the raw I and Q file which you are plotting?

  • Hello,

    I am using TES to generate a profile (header/source C files) to build into a headless (no OS) version of the Mykonos SW (version 1.5.2-3566) which produces my own embedded control SW based on the ADI supplied SW driver. I have my own Xilinx FPGA which hosts the JESD interface.

    I have attached the raw I and Q file, and the excel file which imported this data and has the graph of the I and Q data. Interestingly I noticed all the spikes appear to be on the Q samples...

    IQ-data-2K-samples.zip
  • Does the spike problem resolve if you disable calibration: Rx QEC (quadrature error correction) and/or Rx LOL (LO leakage correction).

  • No it does not resolve. I ran a before (with existing cals) and after test (removing the initial calibrations for RX_LO_DELAY, RX_QEC_INIT). I can see same characteristic spikes on both outputs as before.

    •  Analog Employees 
    on Jul 27, 2018 1:16 PM

    Is the LO locked at 1.3GHz ? Have you tried any other frequency ?

    Is your TX working at 1.3GHz.or Tx as well has some problem. ?