sam henri goldhey wanna see something kinda interesting? this was the entire fix to the iPhone Antennagate in 2010. 20 bytes.
(this is going to be a very long thread 🧵)
For context, back in 2010 when the iPhone 4 came out, people noticed you could grip the phone in a certain way and the signal bars would plummet from 5 to, like, 2.
A few weeks later, they published a letter (https://www.apple.com/newsroom/2010/07/02Letter-from-Apple-Regarding-iPhone-4/) admitting fault, blaming a bad formula.
The letter was shared around and clowned upon (https://daringfireball.net/2010/07/translation_iphone_4) but nobody really looked into what the formula between 4.0 and the patch in 4.0.1
I was a stupid eight-year-old at the time, but now I’m a stupid adult with access to a disassembler.
I downloaded both firmwares and started poking around. In the CoreTelephony framework, I found a promising looking binary: CommCenter. Looking at the strings gave me a pretty good sense that this is where the bar formula was.
The actual calculation is dead simple. When converting signal strength to bars, CommCenter loads each threshold from memory and compares until it finds the right range.
This code is not the problem...
...this is. This is the lookup table. When you convert the bytes to actual dBm values, you get:
-115, -111, -107, -103, and -99
(the closer to zero, the better the signal)
For example, here you need -107 or better signal to see 3 bars.
When you plot this onto a chart, you can see how the values are kinda screwed up since the values are really optimistic. Most of the time, you would see 4-5 bars. But when you gripped it, since the falloff is so sharp, you’d see a catastrophic drop from 5 to 2 bars.
In 4.0.1, they changed these values to be way smoother. Mapped onto a chart, you can see that it takes a lot to drop from 5 to 0 bars. It’s harder to see 5 bars, but it’s harder to plummet bars.
yertle@samhenrigold how do they map it now, for 3G signal strength?
@yertle I haven't looked into it but I'm sure modern signal bar calculation is way more complex. You gotta consider RSRP, RSRQ, and SINR, you gotta figure out what you wanna do for different network types (3G/LTE/5G). I'm willing to bet there's a little ML model somewhere that anticipates signal changes too for predictive smoothing.