To address the problem of high common-mode voltage in deadbeat predictive control for permanent magnet synchronous motor (PMSM), this paper proposed the use of opposite-phases basic voltage vectors within one control cycle to generate a virtual zero voltage vector, which suppressed the common-mode voltage. A dynamic selection method for generating virtual voltage vectors was adopted to decrease switching frequency. To further improve the performances of deadbeat predictive control, 12 virtual non-zero voltage vectors were introduced to expand candidate voltage vector set. Simulation results showed that, compared to using 7 basic candidate voltage vectors, the deadbeat predictive control based on 19 candidate voltage vectors reduced torque ripple by 6.89%, decreased flux linkage ripple by 13.33% and effectively suppressed common-mode voltage. The dynamic selection method for virtual voltage vector generation reduced the switching frequency by 25.89% compared to the fixed generation method. Although this method significantly improved system performance, it increased the number of iterations, leading to a substantial increase in computation burden. To enhance the real-time performance of the algorithm, a simplified method was proposed, which determined the optimal voltage vector by identifying the region where the ideal voltage vector resided, eliminating the need for exhaustive calculations and reducing the computation burden. Real-time experiments based on the STM32H743IIT6 microcontroller showed that compared with using 7 voltage vectors, deadbeat predictive control using 19 voltage vectors increased execution time by 273.40%, while the optimal voltage vector simplified determination method reduced execution time by 88.09%.