MATLAB: Efficiency of code: function call with structure vs script

efficienykalman filter

Hello, I have implemented a piece of code two ways and the times of computation are drastically different. I am currently not able to understand why. Here are the two methods: Method 1:
      for k=2:1:N   
          U = uKal(:,k-1);
          Y_k = Y(:,k);
          P_k_m = sysEST.A*P_k_p*Atrans + Q_k;
          K_k = P_k_m*Ctrans/(sysEST.C*P_k_m*Ctrans+R_k);
          x_hat_m = sysEST.A*x_hat_p + sysEST.B*U;
  %         r(k) = Y_k - sysEST.C*x_hat_m;

          x_hat_p = x_hat_m + K_k*(Y_k - sysEST.C*x_hat_m);
          P_k_p = (I - K_k*C)*P_k_m;
          xHatHist(:,k) = x_hat_p;
      end
Method 2:
     for k=2:1:N
          %extract measurements:
          s.U = uKal(:,k-1);
          s.Y_k = Y(:,k);
          %advance kalman filter:
          s = kalmanfilt(s);
          %save the state:
          xHatHistTwo(:,k) = s.x;
      end
Here is the kalmanfilt function:
function s = kalmanfilt(s)
      %a prior estimates:
          P_k_m = s.A*s.P*s.Atrans + s.Q;
          x_hat_m = s.A*s.x + s.B*s.U;
      %compute kalman gain:
          K_k = P_k_m*s.Ctrans/(s.C*P_k_m*s.Ctrans + s.R);
%         r(k) = Y_k - sysEST.C*x_hat_m;
    %a posteriori estimates:    
        xHat = x_hat_m + K_k*(s.Y_k - s.C*x_hat_m);
        P_k_p = (s.I - K_k*s.C)*P_k_m;
      %output structure:
          s.x = xHat;
          s.P = P_k_p;
end
The computation time for method 1 is 2 seconds and the computation time for method 2 is 0.01 seconds. Does anyone have an explanation of this. Also, any comments on how to further improve the efficiency are welcome too. Thank you.

Best Answer

scripts are less Just In Time compiled than functions are, but how much less has been changing over the releases, with more and more JIT being done for scripts (which is affecting the interpretation of scripts that "poof" variables into existence.)
Related Question