Post Go back to editing

AD9364 Custom Board vs EVM (ADRV9364-Z7020) — Same API/Clock/QPSK MOD & DE-MOD, but Different Output and Calibration Registers

Category: Hardware
Product Number: AD9364

Hi ADI Team,

We are debugging a custom AD9364 hardware platform and would like guidance on why our custom board behaves differently from the EVM despite using the same intended configuration.

We are comparing two AD9364-based platforms: (1) an ADRV9364-Z7020 EVM using the Zynq PS-based ADI reference API, where the HDL QPSK design output works correctly, and (2) a custom AD9364 board using a MicroBlaze-based API ported from the KC705/reference design, where the same HDL QPSK design shows distortion in the TX spectrum.

Both boards have been matched for reference clock frequency, nominal API initialization, LO settings, and sample rate. Digital loopback checks and FPGA waveform generation appear correct on both platforms.

We can observe some differences in the output spectrum for the EVM board to Custom board

After full register comparison, multiple calibration-related registers differ between EVM and the custom board, including:

TX Quadrature / IQ Correction:

  • 0x08F (QUAD gain correction)
  • 0x092 (TX I offset)
  • 0x093 (TX Q offset)
  • 0x0A7 (TX Quad Cal Status)

RX/TX Calibration:

  • 0x170 / 0x171 (Rx Phase/Gain Correction)
  • 0x174–0x17F (RX IQ DC offsets)
  • 0x19B–0x19D (RX BB DC IQ Offset)

Calibration / Filter:

  • 0x140 / 0x141 (Calibration Table)
  • 0x0C7–0x0C9 (TX BBF)
  • 0x1E8–0x1EC (RX BBF)   
  • Whether the observed difference between the EVM and custom board indicates what kind of issue, either in  PCB layout or signal integrity issue, power supply noise or grounding problem, clock quality or jitter issue, TX output path overload or gain mismatch, RF matching differences, or calibration convergence differences caused by board-level electrical behavior? Below, we are adding the image of the spectrum output. Sampling rate is 7.68MHz on both cards.
  • BW = Rs * (1 + 0.5) = 2.88MHz
  • Rs = Fs/4 
  • EVM Board results
  •             
  • Custom Board Results
  •            
  • Condition 1: We have 2 EVM boards, one as QPSK TX and another one as QPSK RX. It's working fine, we can successfully demodulate the bits which we where transmited. 
  • Condition 2: one EVM board as RX and your custom board as QPSK TX, we cant able to demodulate the data correctly. We suspect there is a spectral difference in EVM vs custome board. what will be the issue could be either in hardware or API configuration. We checked both BIST loopback and digital loopback, and also corrected the data and clock delay settings on the custom board. Since the issue still persists, this suggests the problem may be occurring after RF upconversion, likely in the RF hardware path.
  • HI  

    Thanks for the query.

    The following most common issues in common board

    1) Tx Analog filter or matching network differences - if any filter is used in your setup, pls try removing them and check once again. It could be some delay issue also.

    2) Pls check the Cal status of custom board, if they are not converging then try recalibrations to fix them.

    3) Pls check the clock phase noise, check for reference clock phase noise and clock buffer or spurious on clock signals.

    4) check for Tx linearity on the custom board. Reduce the Tx gain to move the signal in linear and check.

    5) Pls try checking with lower symbol rate or lower modulation data rates to see if any improvements. This could be solved by low phase noise clock in the reference path. 

    Regards,

    Sai

  • Hi SJagini 

    Here I summarize the list of observations after the test.

    1. We didn't have any analog filters from the AD9364 output to the connector. Below, we have attached the schematic of your hardware.

    2. I checked the calibration status as suggested. After running all calibration functions, I read the relevant registers, then forced re-calibration and checked the status again. Register 0x244 consistently reports 0xA9 each time(successfully calibrated ).

    Although I can see that the IQ/DC offset, phase, gain correction, and VCO tuning register values change after each calibration cycle (for both TX and RX), the RF output spectral shape shows no improvement. The calibration values are also still different from the EVM board, even though recalibration updates the correction registers each time.

    4. I tested the internal TX power reduction to check for TX chain nonlinearity, DAC clipping, and BB filter saturation. I internally attenuated the TX output from 0 dB to 30 dB with no change in the spectral shape. I also reduced the transmitted IQ amplitude at the waveform generation stage to check for possible DAC overflow or baseband filter overdrive, but the spectral output still shows no improvement. Since neither attenuation nor reduced digital drive changes the spectrum, the issue does not appear to be caused by TX compression, DAC clipping, or BB filter saturation.

    5. I tested lower symbol/data rates as suggested by running 3.68 MSPS, 7.68 MSPS, and 15.36 MSPS, but the spectral shape remains unchanged across all rates. I also repeated the test with reduced TX attenuation and lower digital IQ amplitude (to rule out TX nonlinearity, DAC clipping, and BB filter saturation), but there is still no improvement. Since changing symbol rate, output power, and transmit gain do not affect the spectral difference, the issue currently does not appear to be data-rate dependent or caused by TX overdrive. After all this, does this test point more toward clock/reference path quality, RF layout, or board-level hardware differences? 

  • HI  

    Please check for reference clock accuracy. And check for power supply noise, try reducing to see the improvement. 

    Regards,

    SJ