Camera Module Phase Detection Auto Focus

1. The realization of PDAF function needs to use the concept of shield pixel, including right-half-shield and left-half-shield, which can imitate the similar function of left and right eyes, refer to the following picture.

2. The shield pixels are mainly distributed within 92% of the center of the chip. The 32×32 pixel array is labeled as a block, and the proportion of shield pixels in this block is 1/64, that is, each There are 16 shield pixels in the block, 8 right-half-shields and 8 left-half-shields, and the organization methods of different manufacturers will be different, which may be due to avoiding patent issues.

3. Using the PDAF function requires module calibration (Calibration), including SPC (shield pixel calibration) and DCC (defocus conversion coefficient). Strictly speaking, the SPC function is a normal correction action, while the DCC function is not a correction but generates the coefficient relationship between DAC and Focus, and then realizes the PDAF function.

4. SPC: In order to realize the PDAF function, in the pixel matrix of the chip, some G pixels are designed with a light-shielding structure, and the light energy signal is weaker than that of normal G pixels. Therefore, in order to achieve the energy efficiency level of a normal G pixel, it is necessary to add an additional gain to this type of G pixel, and this action is called SPC. The effect before and after correction can be referred to the figure below, where the abscissa is the position corresponding to the image height, and the ordinate is the pixel photosensitive performance.

5. As shown in the figure above, two dash lines with different trends represent right-half-shield and left-half-shield respectively. By adjusting the gain of the pixel, its performance level is consistent with other normal G pixel. This calibration process uses a pure white back-transparent target. After shooting the pure white target, the gain of each corresponding position is obtained through linear interpolation, and the gain is written into the OTP position. This OTP data is called SPC Map, and there are two One map, one right-half-shield and one left-half-shield.

6. The figure below shows the organization of the SPC Map. The imaging area is divided into 9×7 blocks in total, each block contains 512×512 pixels, and the center point of the first block in the upper left corner is (280,212). The central points of these 63 blocks are called Knots (nodes where the rope is knotted), and 63 interpolated data are generated. The data is roughly calculated as G-pixel/PD-pixel. Only G pixel has shield pixel, and others do not. .

7. DCC: When discussing the PDAF function, in most cases we will claim that the distance of the current subject can be known through this function module. In fact, we do not physically know the distance of the subject, but map the subject to a certain position in the lens movement distance through the difference between the left and right shield pixels. When the system focuses, it needs to convert the detected phase difference (phase difference) into a defocus rate (Defocus Value). The form data applied to this conversion process is called DCC (defocus conversion coefficient).

8. The DCC test calculation method includes the following actions:

Divide the effective focusing stroke of the module into 10 equal parts, that is, (near focus DAC - far focus DAC)/10, so as to cover the focus range of the motor;

Focus at each focus DAC position, and record the phase difference of the current focus DAC position;

After completing the motor focusing stroke, take a set of 10pcs diagonal DAC to compare with the obtained PD data (PD/DAC);

Generate a similar ratio of 10pcs and record it as k. The two-dimensional data composed of DAC and PD can be used to draw a line through an algorithm to fit a straight line with an approximate slope of k;

9. Due to the individual differences between the modules, DCC needs to be measured separately for each module, and due to the field curvature and shading of the lens, there are also differences between the surrounding and the center of the same module, and the conversion cannot be performed with the same coefficient , generally speaking, the coefficient around the center will be higher than the coefficient in the center. Generally speaking, the area is divided into small M×N blocks, and each block obtains a separate k through the process of ⑧ above. Then, the defocus rate, defocus coefficient and phase difference are calculated according to the following conversion method: Defocus Value = Phase Difference × Coefficient, as shown in the figure below:

10. During the DCC calibration process, it is necessary to use black and white back-transparent targets, and the shooting distance is 15cm. Since the current shield pixel only has two directions of L and R, there is a requirement for the direction of the black and white stripes of the calibration target, and it needs to be parallel to the direction of the LR shield pixel to be recognized and detected; ------------- ------------------ This item is to be confirmed.

11. PDAF detection process (from Truly document):

Obtain the data information of a certain line or several lines of the focus position of the subject;

Read right-half-shield and left-half-shield sequence information;

Use the row information obtained in the sequence ① generated by right-half-shield and left-half-shield to perform convolution integral calculation (convolution integral calculation will be that the original row information only carries the information of the corresponding convolution sequence);

Perform preprocessing such as smooth differential interpolation on the two new series after convolution to obtain two smooth cut relative displacement data;

Import the phase difference of the two data into the system, and use the PDAF library and DCC data to calculate the defocus rate;

Use the defocus rate to give the motor an instruction corresponding to the DAC action and complete the focus;