• State
  • Locked Locked
  • Replies 7 replies
  • Subscribers 33 subscribers
  • Views 157 views
  • Users 0 members are here
Support feedback
Options
Options
Similar topics

This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

Original question:

IWR6843: Variability in performance

IWR6843: Variability in performance

Matthew Woolfe
Prodigy 35 points
Part Number: IWR6843

Continuing on from: IWR6843: Variability in performance - Sensors forum - Sensors - TI E2E support forums

We have implemented calibration in our sensor. However as we have based our firmware on the people counting demo which does not support calibration.
Everything I read about calibration also only indicates that it should improve accuracy, rather than extend the range/capabilities of the sensor.

Nevertheless, we adapted the OOB demo to obtain the compRangeBiasAndRxChanPhase coefficients, however after testing several samples with compRangeBiasAndRxChanPhasecoefficients (unique to each board) the performance in general went down significantly, most sensor could no longer identify a target within its FOV, except nearest to the sensor.

Is there any specific testing that could suggest why there is such variability in performance as it seems very likely that this is a hardware issue? 
The primary though I have to the variability is with the manufactured antenna geometry, PCB materials or the reflow activity.

Would a VNA be useful in diagnosis of antenna performance? as it could potentially diagnose the antenna geometry and PCB materials and remove question about the reflow activities.

Thanks in advance
Matthew

  • 0
    TI__Mastermind 27660 points

    Hi Matthew,

    Are you confident that your coefficients were measured correctly? Can you show the range-profile during the measurement to be sure it's picking up the correct peak?

    Best,

    Nate

  • 0 in reply to Nathan Block
    Prodigy 35 points

    Hi Nathan,

    I am not aware of how to obtain the range-profile, could you please explain?

    I did test the compRangeBiasAndRxChanPhase calibration over multiple sensors, at multiple distances ranging from 0.5m to 3m. But I do not believe this calibration stage will improve the sensor performance as it has been highlighted that these coefficients are not supported by the people counting demo, of which our code is based. Please see: https://e2e.ti.com/support/sensors-group/sensors/f/sensors-forum/966360/iwr6843isk-does-68xx_3d_people_counting-support-cli-comprangebiasandrxchanphase?tisearch=e2e-sitesearch&keymatch=compRangeBiasAndRxChanPhase

    Nevertheless, we analyzed the coefficients that were collected, the standard deviation of each calibration procedure (25 samples were collected over a short period of time, 2 minutes) was small, with a maximum of 0.0052. For example (3 of 25 calibrations. Calibration set at 1m, with 0.2m search area):

    compRangeBiasAndRxChanPhase 0.164448 0.74982 0.00726 0.83185 -0.0054 0.77472 0.2359 0.73807 0.3147 0.60248 -0.23462 0.63251 -0.27802 0.66571 -0.04437 0.68057 0.05917 -0.08374 -0.91217 -0.16302 -0.98663 0.11548 -0.99249 0.2424 -0.90671
    compRangeBiasAndRxChanPhase 0.164177 0.75345 0.00122 0.82965 -0.00531 0.78952 0.22574 0.72971 0.30634 0.59933 -0.24033 0.6239 -0.27405 0.6683 -0.04858 0.67386 0.04861 -0.09564 -0.91504 -0.16898 -0.97757 0.11185 -0.99371 0.23178 -0.90555
    compRangeBiasAndRxChanPhase 0.164082 0.74771 0.00281 0.83084 -0.00922 0.77649 0.22421 0.7298 0.3096 0.59769 -0.23703 0.62524 -0.27887 0.66214 -0.04895 0.67233 0.05267 -0.09225 -0.91391 -0.17023 -0.97443 0.10922 -0.99402 0.22742 -0.90247

    This suggested to me that the corner reflector was effectively located.

    In a trial where a corner reflector was omitted the standard deviation over 25 trials was a maximum of 0.0947. Also, in one trial at close distance, 0.5m, the sensor clearly identified an incorrect peak as noted by the rangebias term which toggled between ~0.003 to ~0.115. The other coefficients equally jumped around and the maximum standard deviation rose to 0.40. This calibration was deemed unsuccessful as it could not reliably find the corner reflector.

    As I queried before, is there any specific testing that could suggest why there is such variability in performance as it seems very likely that this is a hardware issue. My primary thought is with the manufactured antenna geometry, PCB materials or the reflow activity. But the reflow activity is the least likely.

    Can you indicate if a 60GHz VNA analysis of the bare PCB antennas could reveal which would be more successful than others.

    I theorize that in high performing sensors (PCB + IWR6843) the Antenna tuning is more effective than in the lower performing sensors. This would mainly be because of manufacturing differences, especially considering the tolerance requirements of the antenna geometry and PCB stack up. From this, analyzing the S11 at 60-64GHz of each antenna should identify the less tuned antennas and hence suggest PCBs not suitable for population. Has TI had any quality issues similar to what I am seeing, and how were they identified/corrected for?

    If possible, would we be able to discuss this further in a video chat to go through the issue in more detail.

    Thanks in advance
    Matthew

  • 0 in reply to Matthew Woolfe
    TI__Expert 6932 points

    Hi Matthew,

    Nate is OoO through the U.S. holiday on Monday. He should have a response for you by EOD Tuesday.

    Regards,

    Tim

  • 0 in reply to Tim Henderson
    TI__Expert 5335 points

    Hi Matthew,

    We are looking into your question and will get back to you soon.

    Regards

    Ankit

  • 0 in reply to Ankit Mohanty
    TI__Expert 5335 points

    Hi Matthew,

    Kindly give the schematic and assembly files for your design, so that we can review it.

    Regards

    Ankit

  • 0 in reply to Ankit Mohanty
    Prodigy 35 points

    Hi Ankit,

    Please provide a confidential method for me to upload.

    Thanks in advance
    Matthew

  • 0 in reply to Matthew Woolfe
    TI__Expert 5335 points

    Closing this thread as it is taking place in a private conversation.

    Regards

    Ankit