在LED照明应用中，为了能够获得、记录或重现特定的照明模式，要求检测装置可以测量出每个LED灯点照明空间中某一位置时的独立"贡献量"，这可以归结为对驱动LED工作的脉宽调制波形(PWM)的参数(振幅、频率偏移量、相位延迟)估计问题。为了达到参数估计的目的，首先将频率偏移空间和相位延迟空间离散化成二维网状格点，然后根据检测装置测量得到的数据在格点空间具有稀疏性的特点建立稀疏模型。接着，采用正交匹配追踪算法(OMP)，用很少的采样点快速有效地重建未知参数。最后，为了有效抑制估计误差，本文使用了一种逐级迭代细分网格的技术作为前面稀疏模型的补充。实验结果表明，本文方法仅使用相当于奈奎斯特采样定理要求的27.5%的少量采样点就完成了快速估计的任务，同时，不同噪声条件下的对比试验说明算法在信噪比大于20dB时鲁棒性较好。

In order to acquire, record or reappear some special lighting pattern in LED illumination application, detection devices are required to measure the "contribution" of each LED point illuminating the scene of interest and target locations, which can come down to estimate the parameters (amplitude, frequency offset and phase delay) of pulse-width modulation (PWM) which drives LEDs illumining. To this end, the frequency offset-phase delay space is discrete into two-dimensional reticulated grids firstly, and then a sparsity model is built based on the character that the measured data from detection devices is sparsity. Next, employing orthogonal matching pursuit (OMP) algorithm, we only use very few samples to reconstruct the unknown parameters. Finally, to suppress estimation error efficiently, a technique of iterative refinement of grids is also presented as a complement of previous sparse model. Experimental results indicate that the method presented in this paper finishes the estimation task rapidly, using only 27.5% of samples which are needed in traditional Nyquist Sampling theorem. Simultaneously, we find that this algorithm is robustness when SNR is higher than 20dB through lots of contrast tests in different noise conditions.