Characterization of the iPhone LiDAR[…] https://pmc.ncbi.nlm.nih.gov/articles/PMC10537187/pdf/sensors-23-07832.pdf

So we know it is a 60hz, 940nm infrared signal. We also know that it can be expected to present as a rotating pattern of lattice grid beams of light. Armed with this information, I began brainstorming different approaches to measuring such a signal in a meaningful way. It was at this point that I realized I actually don't really know how to do that, so I looked it up.

LiDAR is a Flashy Light, How Do We Measure a Flashy Light

After a lot of web searching, reading other researchers' existing work, reading a lot of Wikipedia pages, struggling to get good suggestions out of LLMs, and pouring over datasheets, I reached a point where I felt I was beginning to understand the objectives well enough.

1. See a signal as light spread into beams over an area

2. Sense and convert that light to an analog signal

3. Convert the signal from analog to digital

4. Measure it

The iPhone TrueDepth uses a 60hz, 940nm VCSEL DotGrid Lattice LiDAR system. In order to detect this and distinguish it from other signal sources, a device would need to sense IR signals from multiple discrete sources at high speed from which several factors could be measured. Once these factors are measured, the device would need to be able to quickly perform calculations on the measurements and programmatically decide whether the measured signals are the desired target, or noise. The factors we would want to measure are signal frequency, pulse repetition frequency, whether the signal is steady or in bursts, and how many sensors detect the same signal at the same time or not.

Now, armed with even more information I set about looking up what components might suit the needs of the project.

Hardware

This device needs to detect 940nm infrared signals. I tested several ways to accomplish this, including LEDs wired as photodiodes with and without 940nm bandpass filters, pin silicon photodiodes with and without bandpass filters, and 940nm peak pin silicon photodiodes. While LEDs wired as photodiodes were surprisingly effective, the cleanest and clearest signals were obtained using 940nm peak photodiodes.

In addition to just detecting 940nm infrared signals, the device needs to be able to discern a signal's apparent frequency. We know that the iPhone LiDAR is flashing at 60hz, so we need to be able to detect a 60hz signal, and probably harmonics of that same frequency up to some reasonable amount. This aspect of the target is where either having a 940nm peak photodiode, or using a bandpass filter really comes in handy. Most displays around you are going to be at 30, 60, or 120hz and in my testing I found that without filtering for the desired wavelength almost any display would trigger a false positive.

In order to process this signal, we need to perform operations at a high speed. Solutions for the sensors, like pin silicon photodiodes and fast components like 10mhz op-amps or schmitt triggers, work just fine for capturing these signals and making them available. To process them the device needs to strike a fair compromise between energy consumption and processing power. These days, there are countless tiny little chips that fit this bill. Since I already had development boards laying around, I chose the SAMD21 for its 48mhz processor and tiny energy footprint. Using this chip as the platform to build on, I went through several iterations of hardware designs.

Honorable Mention: Photodiode Pixel Grid

Since the LiDAR is projecting a grid lattice and you can kind of see what the pattern its projecting is, I had thought something like a photodiode grid could work. Upon looking into the BOM cost and difficulty of designing a board for it, I did not pursue this design. But wouldn't that be a neat way to solve this?

Final Choice and Design Caveat

Ultimately I decided to progress the Schmitt Trigger version of the hardware since the difference in performance between it and the op-amp version was negligible.

One common factor all designs required was multiple discrete infrared sensors. In order to differentiate the LiDAR dot grid lattice signals from other sources the sensors might pick up (an analysis I will discuss further in the next post), the device would need to be able to detect if some, but not all sensors were detecting the same signal. I made some attempts not much better than eyeballing with a millimeter ruler to measure the distance between centers of the lattice dots at 3, 5, and 15 meters and then chose two distances which best matched for 15 and 5 meters, with matching for 3 meters being coincidentally covered well enough to probably work.

Whats Next