[Credits: Amit Cherian and Manoj V Mathew
Experiments were conducted at the Central Imaging and Flow Cytometry Facility (CIFF), National Center for Biological Sciences (NCBS), Bangalore, India.]
CMOS camera technology is evolving rapidly. The latest generation of sCMOS cameras performs very similarly to their CCD counterparts in terms of sensitivity and noise levels. They are much better than CCDs in terms of speed of operation as well as chip size. Also taking into account the lower costs of sCMOS cameras, they are increasingly becoming the detector choice for most biological microscopes.
We decided to test few of the parameters of various sCMOS cameras.
Camera:1 Hamamatsu Flash 4 V2 (C11440)
The first camera we got was the Hamamatsu Flash 4 V2 (C11440). See Figure B1.
Figure B1: Hamamatsu Flash 4 V2 sCMOS camera top and back views.
This is a very popular camera in the market and is widely used for biological imaging. This camera features a 2048x2048 chip that operates at max 100fps at full frame.
The camera has two data interfaces:
- Camera Link: 100fps at full frame
- USB 3 Link: 30 fps at full frame
The camera also provides two data transfer modes with different readout times:
- Standard Scan: 10ms read out time (full frame)
- Slow Scan:33ms read out time (full frame)
The slow scan offers lower read noise and is useful for generating high-quality image data trading off speed.
Exp:1 Speed Test using Camera Link Interface and Internal Trigger
We first tested the camera speed using the camera link interface and internal trigger. We used both the transfer modes (standard and slow) and tested the camera speed under various frame sizes without any binning. All experiments were conducted using 16 bit digitization. The Table T1 summarizes our observations.
Table T1: Maximum speed of operation in FPS for various frame sizes using the camera link interface (PDF Format) .
The tests were performed using the Hamatusu HCI software as well as the Micromanager software by operating in the live mode. Figure B2 shows the configuration of the camera in the two modes (standard and slow) as seen from the HCI software control panel.
Figure B2: Configuration of Camera in the two modes standard (left) and slow (right)
Frame rates were read out from the indications on the live window. Both HCI and micromanager software data were consistent.
It can be seen from Table T1 that the maximum speed of 100fps at full frame size, as indicated in the specs sheet was indeed achieved. Figure B3 shows the screen shot of the HCI software indicating 100fps at 2048x2048 at 1.0037ms exposure. (Camera was operated with sensor covered with lid).
Figure B3: Screen shot of HCI software showing 100fps
The Table T1 also indicates the maximum FPS at various cropped frame sizes of the camera. Please note that in the sCMOS architecture the whole row is read out at once. The acquisition speed is hence independent of the column size. So for all the cropped frame sizes, we used the maximum column width of 2048.
Also, note that maximum speed for a given frame size is achieved when the frame is centered on the chip. The reason for this can be understood by looking at the sCMOS architecture as shown in Figure B4. The sCMOS chip is divided logically into two (top half (rows 0 to 1023) and bottom half (rows 1024 to 2047)). Each half has its own data read out circuitry. If you center the frame, the task of reading out the top and bottom halves of the frame is split equally between the two data readout circuitry. This minimizes the time required to read out the frame.
Figure B4: sCMOS internal architecture showing two parallel data read out circuitry. (Source: Hamamatsu)
In the experiments described above the row- offset was set such that the cropped frame is centered. The Table T1 indicates the row offsets required to center the frame.
The table also indicates the maximum and minimum exposure times over which the maximum fps is sustained for a given frame size. For the maximum frame size of 2048x2048, 100 fps is sustained for exposure times between 1.0037ms to 9.99ms. For 2048x8 the camera achieved a speed of 20,543fps at 38.9774us exposure time.
Note that there was no increase in speed when the frame size was decreased from 2048x8 to 2048x4. The acquisition speeds were same for both the frame sizes.
It may also be noted that in the internal trigger mode the camera does not allow exposure times below 1.0037ms for frame sizes 2048x256 and above.
A log-log plot of row size vs speed in FPS returned a straight line as seen in Figure B5. This is as expected theoretically. The acquisition time decreases linearly with the decrease in the number of pixels. Since the acquisition speed is inversely proportional to the acquisition time, the log-log plot of row size vs acquisition speed (in fps) will be linear.
Figure B5. Log-Log plot of frame size vs speed of acquisition in fps for Camera Link Interface using standard and slow scan modes.
We also tested the maximum speed of operation for various bin sizes of the camera. The results are summarized in Table T2. Hamatsu Flash 4 V2 offers 3 bin modes: 1x1, 2x2 and 4x4.
Table T2: Maximum speed of operation for various bin sizes for Camera Link Interface and Internal Trigger
As expected, binning decreases the speed of operation for a given row size compared to the situation where there is no binning.
We tested the variation in speed for change in the row offset for a given frame size. The results are shown in Table T3. Experiments were conducted using a fixed frame size of 2048x256 using the camera link interface.
Table T3: Variation in speed as a function of row offset (PDF Version)
The plot of camera speed in FPS as a function of Row Offset is given in Figure B6.
Figure B6: Speed in FPS as a function of Row Offset
As described in figure B4 the camera chip is divided into two segments of 2048x1024 pixels. Each segment has its own data transfer circuitry. For a frame size of 2048x256, for row offsets below 768 (1024-256=768) the frame lies entirely in the first segment and only one data transfer circuitry is involved. Hence the speed is minimum and constant for row offsets 0 -768 as can be seen from Table T3.
As row offset is increased beyond 768, the second data transfer circuitry gets engaged and the speed increases almost linearly till a row offset of 896 (1024-(256/2)=896). For this row offset the frame is distributed symmetrically between the two segments and the speed is maximum. As the row offset is increased further the speed decreases almost linearly. For a row offset greater than 1024 the frame lies entirely in the second segment. The speed drops to the same minimum value as when the frame was entirely in the first segment. The same behavior was seen for both the scan modes.
Exp:2 Speed Test using USB Link Interface and Internal Trigger
We repeated the speed tests using the USB link interface and Internal Trigger. The results are shown in Table T4.
Table T4: Maximum speed of operation in FPS for various frame sizes using the USB link interface (PDF Format) .
Using the USB interface we achieved the max speed of 30fps at full frame as indicated in the manufacturer specifications. The difference in frame rates for the slow scan and standard scan modes was however minimal (mostly similar). For smaller frame sizes the frame rates increased to a maximum of 800fps at 2048x8 frame size. Like with the camera link interface there is no improvement in the speed while decreasing the frame size from 2048x8 to 2048x4.
The log-log plot of frame size vs speed produced a straight line for frame sizes larger than 2048x64 using the standard scan mode as shown in Figure B7.
Figure B7. Log-Log plot of frame size vs speed of acquisition in fps for USB Interface using standard scan