For the vehicle-mounted phased-array antenna in the mobile satellite receiver, a novel beamforming algorithm based on improved simultaneous perturbation stochastic approximation (SPSA) is developed. It estimates the gradient of the received signal power in the assured direction rather than fully isotropic by adding a fit searching direction. Besides the SPSA beamforming algorithm's merit, the proposed method can track the target satellite quickly, accurately and steadily in the single-channel receiver. Computer simulations verify that the proposed algorithm is indeed robust and effective.Introduction: Recently, there has been a greatly increased user demand for satellite communication in vehicular stations such as cars, buses, trains and ships beyond a fixed station [1][2][3][4]. In the satellite communication on-the-move (SOTM) system at Ku-band, due to the small intervals of the adjacent geostationary satellites, the radiation pattern of the vehicle-mounted antenna is required to be narrowed and strictly aligned to the desired satellite, to prevent interference from other adjacent satellites and not to miss the target satellite during motion [3]. This Letter is focused on the tracking method to carry out the accurate alignment for SOTM.In [3], a kind of hybrid tracking scheme composed of a mechanical stabilisation loop and an electronic beamforming loop for the vehiclemounted phased-array antenna is developed. The mechanical stabilisation loop orients the beam approximately to the target and then an electronic beamforming algorithm for the single-channel receiver is utilised to realise the accurate beam direction. In [3], a DOA estimation method which perturbs all phase shifters sequentially to estimate the gradient of the received signal power is proposed, which is similar to the finite difference stochastic approximation (FDSA) algorithm in [4]. Nonetheless, the sequential perturbation is time-consuming and the algorithm may converge slowly with the increasing number of phase shifters according to [4]. Therefore a stochastic algorithm by perturbing all phase shifters simultaneously is developed in [5], which converges faster than ZKBA. However, it is likely to diverge in some cases.In this Letter, a new beamforming algorithm based on an improved simultaneous perturbation stochastic approximation (SPSA) is proposed. It converges quickly, accurately and steadily by adding a fit searching direction. Consequently the probability of algorithm divergence is significantly lower than that in [5].