Figure 1. Active learning example. The estimated state vectors of the communication link (green crosses) label the data from other sensors as a valid MPC. Therby active learning of possible beam direction from other sensors is rendered possible.

Figure 2. Geometric relationship of all variables used for the algorithm. The MPC to track is marked as squqre. The employed array geometry (uniform circular array) is sketched as well.

Figure 3. Beampattern of the proposed uniform circular array with N=64 elements and 4 bit quantization of all steering vectors. The desired array pattern is marked in green; the array factor for this direction (0°) is marked with a green dot; the neighbouring codebook pattern are drawn in black. The uniform circular array (UCA) pattern is inherently symmetric for all directions; illustrated by the red pattern pointing in opposite direction.

Figure 4. Blocked manoeuvre: a) sketch of the scenario. The green car block a direct communications between the red TX and black RX car; b) the Doppler shift estimate for this scenario obtained from a real world experiment in [9] and [23].

Figure 5. a) Manoeuvre sketch of the scenarios; b) exemplary relative position estimates (x,y) of the proposed estimator with 20 ms update rate; c) range normalized mean squared error comparisons with 20 ms update rate; d) (x,y) estimates with 50 ms update rate; e) NMSE with 50 ms update rate.