my esp32 reads my phone at -83 dbm from the desk, -95 through a wall, and somehow stronger outside the apartment than inside the room. the log-distance path loss model says that shouldn't happen. reality doesn't care about your model. i've been trying to build a three-zone presence system — room, home, outside — and bluetooth is fighting me every step of the way.
tuning thresholds is just guessing with a feedback loop. you set -70 for "in the room" and then your phone reads -85 while you're literally sitting at the desk. so you lower it to -75. now the bathroom triggers "in the room" because the signal bounces off the tiles. you raise it back to -70. now you're "away" while walking to the kitchen. the cycle continues. at some point you just accept that bluetooth is vibes-based technology.
the real problem is that rssi isn't distance. it's a combination of distance, orientation, wall materials, other devices on the same frequency, whether your phone is in your pocket or on the table, and probably the phase of the moon. i have a smoothed rssi sensor with a 10-sample sliding window and exponential moving average and it still jumps 10dbm between readings while i'm sitting perfectly still.
so now i'm combining ble with the ld2410 radar for a multi-sensor fusion approach. the radar says "someone is in this room" and the ble says "that someone is probably nikhil" and home assistant templates combine them into a single state. it works about 80% of the time. the other 20% my phone decides it's in a different room or doesn't exist at all. the esp32-c3 couldn't handle both ble and radar simultaneously — ram fills up, framework conflicts, builds explode. needed the full esp32 with its extra headroom. the c3 is now the closet radar. the middle child of microcontrollers.