Understanding the Error state (indirect) kalman filter

I have read about the error state kalman filter several places, but I have a hard time understanding the concepts. Can someone please give a step by step explanation of the concept of this filter? I have seen the equations of course, but what are each equation doing and why? Matlab code below is for trying to estimate orientation with IMU measurements.

    %%Update
    y           = [acc_x(i); acc_y(i); acc_z(i)]; % IMU accelerometer measurements
    K           = P * H' * inv(H * P * H' + R);
    delta_x_hat = K * (y - y_hat);
    P           = (I - K * H) * P * (I - K * H)' + K * R * K';
    
    %%Prediction
    % delta_x_hat = F*delta_x_hat; // not necessary according to link below
    P          = F * P * F' + G * Q * G';
  
    
    %%Error injection
    phi         = phi      + delta_x_hat(1); // roll
    theta       = theta    + delta_x_hat(2); // pitch
    b_b_ars     = b_b_ars  + delta_x_hat(3:5); // angular rate sensor biases

    %%ESKF reset
    delta_x_hat = zeros(5,1);

<http://www.iri.upc.edu/people/jsola/JoanSola/objectes/notes/kinematics.pdf>.

Thanks for any comments!