Problems about the operations of DAVIS240 with different bais setting

[Vicki Shen]

Hello,

I'm doing experiments on testing DAVIS240's behavior at a blinking LED. During my investigations, I came up with some problems. 

1. LED blinking at 1 kHz. As shown in the attached figures, when I adjust the pixel bandwidth from slow to fast, there starts to be a hole with no event in the middle of where the LED is imaged.  However, from my experience,  the center of the LED should have the strongest emitting power. I wonder if this hole is because of the saturation of the pixels.

2.   I want to check the operation of DAVIS240 when generating an event. According to a recent paper you published[1], the photocurrent Iph(t0) used to be compared with is the one triggering the last event instead of the photocurrent detected right after the refractory period. Am I correct about this?

3. When I set the bias using DV, there are only coarse and fine control ticks, whereas, in jaer, I can see the configured current. Is there a way to map DV's coarse and fine ticks to direct configured current?

[1] Experimental methods to predict dynamic vision sensor event camera performance. 

Thank you in advance for your help!

[Tobi Delbruck (UZH-ETH)]

Hi Vicki thanks for detailed forum post.

The bright LED can create parasitic photocurrent in the refractory period pulse circuit such that the pixel cannot reset; the current can be so large that the pixel is simply held in reset because the pullup current set by the bias is smaller than the parasitic photocurrent. This is particularly true in Davis240 when we did not realize it might be a problem. Did you check to make sure to set a short refractory period (largish bias current RefrBp?)

I don't have time to reproduce the behavior now - we are working extremely hard on a chip design. But I can take a look after mid November to see if I can reproduce the effects with Davis346 and maybe Davis240.

Be warned that the computation of bias current in jAER is quite approximate, but it as shows the order of magnitude of the bias currents.

Please also read the v2e paper and if you are a circuits person then the original DVS128 paper has the most detailed description of the operation of the DVS pixel. The paper from me and Yuji [3] has detailed measurements of the effect of parasitic photocurrents in Davis346.

Good luck and look forward to more questions

Tobi

1. Hu, Yuhuang, Shih-Chii Liu, and Tobi Delbruck. 2021. “v2e: From Video Frames to Realistic DVS Events.” In 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), 1312–21. https://doi.org/10.1109/CVPRW53098.2021.00144.
2. Lichtsteiner, Patrick, Christoph Posch, and Tobi Delbruck. 2008. “A 128×128 120 dB 15 μs Latency Asynchronous Temporal Contrast Vision Sensor.” IEEE Journal of Solid-State Circuits 43 (2): 566–76. https://doi.org/10.1109/jssc.2007.914337.
3. Nozaki, Y., and T. Delbruck. 2017. “Temperature and Parasitic Photocurrent Effects in Dynamic Vision Sensors.” IEEE Transactions on Electron Devices 64 (8): 3239–45. https://doi.org/10.1109/TED.2017.2717848.

[Vicki Shen]

Hello Tobi, 

Thanks for your reply!

Currently, the RefrBp value is about 50n. With this value, the detectable LED frequency is about 11 kHz (with some missed events occasionally).

I've tried to enlarge the RefrBp value to about 1u. With this value, there start to be some points with a lot of events, however, with a double frequency of the LED frequency.

I will check the papers you recommended to investigate this phenomenon. Thanks!

Separately, about the bias setting,  right now, I'm using jaer. But if I want to switch to DV API to control the DAVIS240, is there a way I can apply the bias setting in jaer to DV?

Thanks!

Vicki

to...@ini.phys.ethz.ch 在 2022年11月3日 星期四清晨6:08:34 [UTC+8] 的信中寫道：

[Luca Longinotti]

Hello,

in the jAER bias control panel there should be an Advanced tab that lists all the biases and their precise values, you can look at the values for coarse and fine sliders and then apply the same values in DV.

In DV the full configuration of the Capture module does contain the same sliders and toggles to configure the biases.

Hope this helps, have a nice day!

-- 

Luca Longinotti (llongi)

Head of Embedded Platforms

iniVation AG - https://inivation.com/

Zurich, Switzerland

Office: +41 44 500 32 14

[Vicki Shen]

Thanks for your detailed explanation.

I will give it a try.

Luca Longinotti 在 2022年11月4日 星期五下午4:07:20 [UTC+8] 的信中寫道：