Abstract A comprehensive analysis of the quantization error in spike-frequency-based image sensor is presented. The image sensor generates different frequency spikes under various light. Images can be reconstructed from spikes by counting the spike number in an integration period, however, this method will generate quantization error. Firstly, the quantization errors introduced by nonlinearity, comparator operation delay and synchronous reading mechanism, are studied. Subsequently, a previous designed spike-frequency-based image sensor with a 400(H) × 250(V) pixel array is tested, and the test results match well with the model results. When the illumination increases from 100 lx to 600 lx, the quantization error increases from −2.5% to 8% at the conditions that reset voltage is 3.3 V, reference voltage is 2.2 V, reading period is 25 μs and the integral period is 10 ms. In addition, under the conditions of illumination is 300 lx, reading period is 25 μs and integration period is 20 ms, the minimum quantization error (6.5%) can be obtained by selecting reset voltage is 2.9 V, reference voltage is 2.32 V. Thus, the results can serve design or usage guide for suppressing the quantization error of the spike-frequency-based image sensor.

中文翻译：

---

基于尖峰频率的图像传感器量化误差分析与建模

摘要 综合分析了基于尖峰频率的图像传感器的量化误差。图像传感器在不同光线下会产生不同频率的尖峰信号。可以通过计算积分周期内的尖峰数从尖峰重建图像，但是这种方法会产生量化误差。首先，研究了非线性、比较器操作延迟和同步读取机制引入的量化误差。随后，对先前设计的具有 400(H) × 250(V) 像素阵列的基于尖峰频率的图像传感器进行了测试，测试结果与模型结果吻合良好。当照度从100 lx增加到600 lx时，在复位电压为3.3 V，参考电压为2.2 V的条件下，量化误差从-2.5%增加到8%，读取周期为 25 μs，积分周期为 10 ms。另外，在照度为300 lx，读取周期为25 μs，积分周期为20 ms的条件下，选择复位电压为2.9 V，参考电压为2.32 V可以获得最小量化误差（6.5%）。 因此，结果可以作为抑制基于尖峰频率的图像传感器的量化误差的设计或使用指南。