Resetting the Comparison Target of DAVIS346 or DVXplorer

[Tianwen Fu]

Hi,

I'm a student at Carnegie Mellon and we're working on a project that needs to measure the events when the illumination of a pixel increases from zero in a static scene (such as instantaneously taking the lens cap off). Currently, we already have a DVXplorer and a DAVIS346. Is there any method to "hard reset" either of the two sensors so that the comparator compares the photocurrent with zero instead of the voltage at the previous timestamp and emits ON events according to the static illuminance?

[Tobi Delbruck]

No, sorry! The DVS128 included a global reset switch in each pixel in parallel with the usual event-triggered reset (see the graphic below with highlighted FET). But later cameras leave this out to save space in the pixel, which is always so precious. Also, each switch has parasitic photocurrent junction leakage from its floating source/drain, which results in "leak" events that increase with bright illumination.

If you can get a DVS128 the functionality is built into the jAER DVS128 menu and is exposed via the DVS128 AEChip class as shown in the menu.

Actually the DVS128 had even more stuff, like the global summing of all photocurrent that just this last year was finally used by industry in the latest Propheesee DVS, which uses this idea as a super low power global change detector for suveillance wakeup.

[Tianwen Fu]

Hi Tobi,

I'm sorry I didn't make my question clear. In fact, what we are hoping to do is not "reset" in the context of event cameras, but we're wondering if we can "restart the photodiode voltage/source follower" of the event cameras. As an illustration, is it possible to achieve the effect like the orange-shaded area in the following figure?

I hope this helps explain my question better, and I would be more than happy to further clarify if it still does not make sense to you. Thank you for your kind help!

[Tobi Delbruck]

No, no sensor support this. The diode cap dynamics and the terrible subthreshold mismatch makes this a bad idea in practice. Under low light conditions, the voltage will take a long time to charge (the approach to final value is logarithmic in time). And the voltage that you reach will have the FET threshold mismatch added to it.  You are talking about something like an Octopus sensor here, where the spike rate is somehow proportional to the intensity. Also, the events will come from all pixels at the same time, which would clog the AER readout bus in the worst possible way: Areas of uniform intensity would produce massive amounts of redundant events.