To improve the precision of the position and speed estimation in ultra-high-speed (UHS) permanent magnet synchronous motors (PMSM) without position sensors, multiple refinements to the traditional extended electromotive force (EEMF) estimation algorithm are proposed in this paper. The key improvements include discretization compensation, high-frequency harmonic filtering, and the real-time adjustment of the phase-locked loop (PLL) bandwidth. Firstly, a discrete model is introduced to address EMF cross-coupling issues. Secondly, an improved proportional resonant (IPR) controller eliminating static errors is utilized in place of the conventional proportional-integral (PI) controller and low-pass filter (LPF) to enable precise electromotive force extraction, effectively filtering high-frequency harmonics that arise in low carrier ratio conditions. Based on a standard PR design, the IPR controller offers a streamlined calculation for target leading angles in delay compensation schemes to effectively mitigate discretization and delay errors. Additionally, an adaptive phase-locked loop (AQPLL) dynamically adjusts its bandwidth during acceleration to balance noise rejection and phase delay, reducing position estimation errors and optimizing torque. Simulations and experimental analyses on a motor (90,000 rpm, 30 kW) validate the effectiveness of the proposed sensorless driving techniques and demonstrate enhanced performance in position and velocity estimation, compared to the conventional EEMF approach.