E.LAWSON // PORTFOLIO_SYS
●ONLINE
Ethan Lawson
← ALL PROJECTS

RF SENSING · HARDWARE // SHELLHACKS 2026

PHASE

PHASE uses ESP32 nodes to transmit WiFi across a zone and read the channel state information (CSI) off those signals, picking up motion and sending the data to a live feed. The nodes also carry microphones reporting live audio. Each stream is monitored by a machine learning model watching for signs of activity in the zone, like CSI interruptions and loud sounds, and the outputs feed a heuristic scoring system that rates how much activity is in the field and flags when someone is likely inside. Built in 36 hours with a team of four at ShellHacks 2026.

PHASE feature photo

// OVERVIEW

PHASE is a presence sensing system for first responders: a set of ESP32 nodes you place around a zone to tell whether anyone is inside before you enter. Each node transmits WiFi and reads the channel state information (CSI) back off those signals, so motion in the zone shows up as distortion in the channel, even through walls. The same nodes carry microphones that stream live audio. Every CSI and audio stream is watched by a machine learning model trained to pick out signs of activity, like CSI interruptions and loud sounds, and those outputs feed a heuristic scoring system that rates how much activity is in each zone and flags when someone is likely present. Readings travel over a Heltec LoRa 32 V3 backhaul to a live feed at the command position, and every node is housed in a custom 3D printed enclosure built from scratch for the project. My team of four built it in 36 hours at ShellHacks 2026.

// AUDIO MODEL

Neural network for possible human scream detectionAn audio sample enters a 1D convolution with 16 channels, max pooling by 2, a second 1D convolution with 32 channels, another max pooling by 2, flattening, a dense layer of 64 ReLU units, and one sigmoid output. Connected lines show information passing between layers.⋮MICROPHONEAudio inputShort sound sample16 CHANNELSConv1DPattern featuresPOOL = 2Max poolReduced sequence32 CHANNELSConv1DDeeper featuresPOOL = 2Max poolReduced sequenceVECTORFlattenFeatures in a line64 UNITSDense + ReLUConnected neurons0–1 SCORESigmoidPossible scream

Audio model: a 1D CNN that scores short samples for possible human sounds.

// HIGHLIGHTS

  • ›RF Sensing: ESP32 nodes measure distortion in WiFi channel state information (CSI) to detect human presence and movement through walls.
  • ›Acoustic: INMP441 microphone nodes flag acoustic anomalies inside the structure.
  • ›Scoring: machine learning models monitor each data stream and feed a heuristic scoring system that rates activity per zone.
  • ›Mesh + Dashboard: Heltec LoRa 32 V3 boards form the backhaul, feeding a live occupancy dashboard at the command position.
  • ›Enclosures: every node housed in a custom designed, 3D printed enclosure built from scratch for this project.

// GALLERY

PHASE — hero
PHASE — node front
PHASE — node side
PHASE — gateway