星链下行信号实时高精度频率估计算法设计与实现是LEO卫星动态机会导航工程应用的关键技术。针对传统极大似然估计、频域滑窗估计及卡尔曼滤波等算法在低信噪比星链信号捕获中鲁棒性差、实时性不足的问题，本文提出多子载波联合频偏估计（MC-JFE）算法，通过深度挖掘信号多子载波结构特征，联合优化载波频率与频率间隔参数，提升了频率估计精度与实时性。为突破MC-JFE算法工程应用中密集计算瓶颈，创新构建了一种CPU+GPU异构并行的加速处理架构，通过协同调度CPU逻辑控制与GPU大规模并行计算能力，算法执行效率实现超一个数量级提升。为验证设计算法的理论与技术实现有效性，基于半实物仿真平台生成的星链下行信标数据，开展了5978颗星链卫星信号实时频率估计试验，并结合我国边境地区实测信号进行多普勒估计算法对比研究。结果表明：所提MC-JFE算法在-10dB至10dB全信噪比范围内保持最低估计误差边界，估计精度提升50%以上（0dB）；通过相位信息融合机制，在部分子载波中断时维持稳定输出；基于CUDA最优线程块配置的CPU+GPU异构架构，加速比峰值达47倍，较传统CPU方案提升2.8倍，且精度与加速比呈正相关特性，为LEO卫星动态机会导航提供了高可靠、强实时的频率估计技术支撑，具有重要工程应用价值。

The design and implementation of a real-time high-precision frequency estimation algorithm for Starlink downlink signals is a critical technology for the engineering application of LEO satellite dynamic opportunistic navigation. Traditional algorithms such as maxi-mum likelihood estimation, frequency-domain sliding window estimation, and Kalman filtering suffer from poor robustness and insufficient real-time performance in capturing low signal-to-noise ratio (SNR) Starlink signals. To address these issues, this paper proposes a Multi-Carrier Joint Frequency Estimation (MC-JFE) algorithm, which enhances frequency estimation accuracy and real-time performance by deeply exploiting the multi-subcarrier structural characteristics of signals and jointly optimizing carrier frequen-cy and frequency interval parameters. To overcome the intensive computational bottleneck in the engineering application of the MC-JFE algorithm, an innovative CPU+GPU heterogeneous parallel acceleration architecture is constructed, achieving over an order of magnitude improvement in execution efficiency through coordinated scheduling of CPU logic control and GPU large-scale parallel computing capabilities. To validate the theoretical and technical effectiveness of the proposed algorithm, real-time frequency estima-tion experiments were conducted using Starlink downlink beacon data generated by a hardware-in-the-loop simulation platform, involving 5,978 Starlink satellite signals, and a comparative study of Doppler estimation algorithms was performed using measured signals from China’s border regions. Results show that the MC-JFE algorithm maintains the lowest estimation error boundary across the full SNR range (-10 dB to 10 dB), with accuracy significantly improving as the number of subcarriers increases, and stable out-put is maintained during partial subcarrier interruptions through a phase information fusion mechanism. The CUDA-optimized CPU+GPU heterogeneous architecture achieves 0.1 Hz-level high-precision frequency estimation, with a peak speedup ratio of 47× (2.8× faster than traditional CPU solutions) and a positive correlation between accuracy and acceleration, providing highly reliable and real-time frequency estimation technical support for LEO satellite dynamic opportunistic navigation, demonstrating significant engineering application value.