• State Resolved
  • Locked Locked
  • Replies 2 replies
  • Subscribers 31 subscribers
  • Views 481 views
  • Users 0 members are here
Support feedback
Options
Options
Related

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.

AWR1843: GTrack error covariance initial value

liang yu chou
Intellectual 580 points
Part Number: AWR1843


Hi,

   The file "gtrack_unit_start.c" of the gtrack of mmwave SDK 3.2,

I do not understand the initialization of error covariance matrix.

Why P_apriori_hat is diag([0,0,0.5,0.5,1,1])? Does there exist a more

detailed explanation or references?

Thank you

                                                                                 Liang-Yu  

line 111

/* P_apriori_hat = eye(6) */
/* memcpy(inst->P_apriori_hat, eye6x6, sizeof(eye6x6)); */

/* P_apriori_hat = diag([0,0,0.5,0.5,1,1]) */
memset(inst->P_apriori_hat, 0, sizeof(inst->P_apriori_hat));
for(n = 0; n < GTRACK_STATE_VECTOR_SIZE; n++)
inst->P_apriori_hat[n*GTRACK_STATE_VECTOR_SIZE+n] = pInit[n];

  • 0
    TI__Mastermind 22155 points

    Hi Cheng ,

    Let me check with the experts of this application and get back to you.

    Thanks,

    Raghu

  • 0 in reply to Raghunandan Kamath
    TI__Intellectual 830 points

    Hi, Cheng,

    In principle, the tracker will eventually learn the error covariances over the time. The initialization is only helping at the first steps, when tracker has very limited information. I chose this particular parameters due to the following consideration:

    Typically, the first measurement points tracker has are not too reliable. And the question is whether we trust more positional or motion (velocity/acceleration) information. The positional information comes from lossless transformation form Range/Angular measurements. Target velocity in Cartesian space is very unreliable initially. This is because because it is based on radial velocity component only (so we don't yet know how to decompose it to Vx,Vy), and also subject to radial velocity aliasing uncertainty. Therefore, I decided to trust velocity less, and even less to the acceleration - the derivative of uncertain velocity. In practice, this shall translate to Kalman process first following the position, and adjusting the velocity and acceleration at least on initial steps.

    Hope that helps,

    Michael