Imaging Sensor Column Fixed Pattern Noise Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional CMOS image sensors suffer from column fixed pattern noise (FPN) due to device and interconnect mismatches, leading to visible vertical lines in images, especially under varying conditions or during mass production, which limits image quality and yield.
Innovation Solution
The implementation of channel randomization in the readout circuitry, where the sign of the offset is randomized for each pixel readout, causing some pixels to have a positive and others a negative offset, thereby reducing the visibility of vertical lines in the resultant image.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If conventional column sampling circuits are used with fixed offset, then the circuit structure is simple, but visible vertical lines appear in the image due to column FPN
Solution Approach 1:
The patent applies dynamics by making the sampling channel selection variable rather than fixed. A random number generator dynamically selects different sampling channels for different pixels within the same column, causing the offset sign to change randomly. This dynamic approach transforms the static, visible column FPN pattern into a dynamic, randomized noise pattern that is much less noticeable to the human eye.
Solution Approach 2:
The patent changes the parameter of offset sign from fixed to randomized. By introducing randomness in the selection of sampling channels, the offset parameter fluctuates between positive and negative values for different pixels. This parameter change effectively converts systematic column FPN into random noise, significantly reducing its visual impact while maintaining the same hardware structure.
2Measurement precision
If correlated double sampling or delta double sampling is applied, then some offset noise is reduced, but residual column FPN remains due to non-linearity and circuit variation
Solution Approach 1:
The patent uses multiple sampling channels (copies) to read the same pixel signal. Instead of relying on a single sampling path, the invention provides several parallel sampling channels and randomly selects among them. This copying approach ensures that even if one channel has non-linearity or variation, other channels can compensate, thereby reducing residual column FPN more effectively than conventional single-path sampling methods.
Solution Approach 2:
The patent converts the harmful effect of offset variations into a beneficial randomization effect. Rather than trying to eliminate offset variations through complex compensation circuits, the invention embraces them by randomly distributing positive and negative offsets across pixels. This transforms the harmful systematic error into a beneficial random noise pattern that is visually imperceptible.
3Illumination intensity
If signal gain is increased to boost images in low lighting, then image visibility improves, but residual column FPN becomes more visible
Solution Approach 1:
The patent applies dynamics by making the offset sign variable through random channel selection. When signal gain is increased, the randomized offset pattern remains dynamic and unpredictable, preventing the formation of visible vertical lines even at high gain settings. This dynamic approach allows aggressive signal boosting without amplifying systematic column FPN artifacts.
Data Source
Figure 1A
Figure 1B
Figure 2
AI summary
An imaging sensor having reduced column fixed pattern noise includes a plurality of imaging pixels and a column sampling circuit. The plurality of imaging pixels are arranged in a column the column sampling circuit is coupled to the column. A plurality of sampling channels are included in the column sampling circuit, where the column sampling circuit randomly selects a first sampling channel from among the plurality of sampling channels to sample a first data signal from a pixel included in the plurality of imaging pixels and where the column sampling circuit randomly selects a second sampling channel from among the plurality of sampling channels to sample a second data signal from the pixel.