0
\$\begingroup\$

What happens if I overclock an image sensor? For instance the OV7670 (datasheet) which has a rated maximum system clock input (XCLK) of 24 MHz and maximum pixel clock (PCLK) of again 24 MHz. The maximum standard frames per second is 30. Is it possible to completely "speed up" the sensor to get more FPS?

Will the overclocking affect the automatic image control functions like auto exposure control, gain control, white balance etc? Maybe resulting in the AEC running faster and producing darker images?

\$\endgroup\$
  • \$\begingroup\$ The datasheet says that Fclk is from 10-48 MHz, so it appears that the chip may function at 60 fps ...do you want to go higher? \$\endgroup\$ – Jack Creasey Sep 15 '18 at 23:09
  • \$\begingroup\$ Oh, thanks for finding that out. But yes, I look for much higher FPS, maybe hundreds of frames per second... \$\endgroup\$ – Kozuch Sep 15 '18 at 23:43
  • \$\begingroup\$ I'd suggest it is not going to work without cooling. Just read about overclocking CPU's. It seems very unlikely that the DSP would run much over 48MHz and the A/D definitely not. There was a kickstarter for a high fps camera, but I'm not sure any have been delivered yet: kickstarter.com/projects/1623255426/… (up to 800 fps) There are plenty of high fps sensors, but they are expensive. \$\endgroup\$ – Jack Creasey Sep 15 '18 at 23:57
  • \$\begingroup\$ Often the power-supply-pin ringing will be setting the acquisition of pixel-charge by the ADC. Thus implementing well-dampened VDD voltages is your task. \$\endgroup\$ – analogsystemsrf Sep 16 '18 at 2:24
4
\$\begingroup\$

In general, image quality will drop rapidly. The sensor circuitry is carefully designed to minimize internal noise and crosstalk, but it is usually optimized for the specified master clock frequency range.

The usual method to get high frame rates on sensors that allow it, is to read out only a tiny portion of the sensor area for each frame. In other words, you can generally directly trade off the number of active pixels for frame rate.

For example, I have recently been working with an IR sensor that can do 1280x1024 frames (1.2 Mpix) at 180 FPS, but it can also do a 64x4 sub-frame (256 pix) at >6 kFPS. Note that the speedup in this case is less than linear, because sensor integration time (for adequate exposure) begins to dominate the frame period (instead of the readout time).

\$\endgroup\$
  • \$\begingroup\$ Well this is exactly my case - I can sacrifice resolution for FPS. Low-cost is a benefit. Would you mind disclosing the type of your sensor? Or can you suggest an affordable sensor capable doing high FPS in low resolution? I prefer parallel camera interface for easy integration. \$\endgroup\$ – Kozuch Sep 16 '18 at 0:14
  • \$\begingroup\$ Sorry, no, we don't do product recommendations here. \$\endgroup\$ – Dave Tweed Sep 16 '18 at 0:57
1
\$\begingroup\$

The simple answer is you can’t tell until you try.

Some devices might ignore your attempts to provide an invalid clock, others might return broken data or spectacularly crash. Maybe you’ll shorten the expected lifespan of the part through overheating or another fault condition.

Some oscilloscope manufacturers are rumoured to do this with fast analogue to digital converters (ADCs). These are a significant portion of the unit cost, perhaps >£50 in a £500 device. The scope manufacturers test thoroughly and sort them into “bins” based on performance. This often works because the original chip manufacturer conduct their own testing and often label a part as a lower spec to ensure it works throughout the operating range (temperature, voltage, expected life etc.).

You might be able time get a generic answer about the viability of overclocking a particular type of device, e.g. SDRAM or a DAC or an SD card or an image sensor.

In the data sheet you posted are a few useful quotes, for example:

this A/D converter operates at speeds up to 12Mhz and is fully synchronous to the pixel rate

There is also a video timing generator built in. This suggests the capture device has sophisticated internal control circuitry that you can’t just fool with a faster external clock.

Regardless you’ll be using the device outside the specified operating conditions, so you shouldn’t rely on it without significant testing and error checking.

\$\endgroup\$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.