REAL-TIME RF VISUALIZATION

How WiFi DensePose Works

A real-time 3D WiFi radar that maps every network around you as glowing nodes in a Matrix-style neural space — powered by signal correlation, not guesswork.

Launch 3D Map
The Problem

RSSI Alone Isn't Enough

Every WiFi scanner shows you signal strength (RSSI) — the number of decibels your adapter picks up from each access point. But RSSI is a one-dimensional measurement. It tells you how loud a signal is, not where it's coming from.

📡

The RSSI Problem

Two APs at -65 dBm could be in the same room or on opposite ends of a building. A wall, a microwave, even humidity can swing RSSI by 10-20 dBm without the AP moving an inch. Distance ≠ signal strength.

🧠

The Correlation Fix

Instead of trusting raw dBm, we track how signals co-fluctuate over time. APs that are physically close experience the same environmental interference — their signals rise and fall together. That correlation is the real spatial signal.

Core Algorithm

Pearson Correlation for AP Proximity

The key insight: physically adjacent access points share environmental noise. When a door opens, a crowd moves, or interference spikes on a channel — nearby APs are affected simultaneously. We exploit this with Pearson's correlation coefficient.

// Pearson correlation between two AP signal histories
function pearson(x, y) {
  const n = x.length;
  let sumX = 0, sumY = 0, sumXY = 0;
  let sumX2 = 0, sumY2 = 0;

  for (let i = 0; i < n; i++) {
    sumX += x[i]; sumY += y[i];
    sumXY += x[i] * y[i];
    sumX2 += x[i] ** 2;
    sumY2 += y[i] ** 2;
  }

  const num = n * sumXY - sumX * sumY;
  const den = Math.sqrt(
    (n * sumX2 - sumX ** 2) *
    (n * sumY2 - sumY ** 2)
  );

  return den === 0 ? 0 : num / den;
  // +1 = same location, 0 = unrelated, -1 = inverse
}

For every pair of detected APs, we collect a rolling window of RSSI samples and compute ρ (rho). The resulting correlation matrix tells us which APs are spatially clustered — even through walls, across floors, or in environments where raw signal strength is misleading.

Correlation → Distance Mapping

ρ ≥ 0.8
Same Room / Adjacent
ρ ≈ 0.5
Same Floor / Nearby
ρ ≈ 0.2
Different Area
ρ ≤ 0
Unrelated / Far

High correlation → placed close together in 3D space. Low or negative → pushed apart. This gives us a relative spatial topology without any GPS or floor plans.

How It Works

From Radio Waves to 3D Space

The visualization pipeline transforms raw WiFi scan data into an interactive neural RF map in four stages:

RF SCAN
Sweep 2.4 + 5 GHz
CORRELATION
Pearson ρ matrix
TOPOLOGY
3D position solve
RENDER
Three.js scene
→ → → →
Each scan cycle re-evaluates the entire pipeline in real-time

Stage 1 — RF Spectrum Scan

The system probes all available channels across 2.4 GHz (channels 1-14) and 5 GHz (UNII-1 through UNII-3). For each detected access point, we capture: BSSID, SSID, RSSI, channel, bandwidth, and security protocol. Each scan adds a new data point to the rolling signal history buffer.

Stage 2 — Correlation Matrix

With enough samples (typically 20+ scans), we compute the full Pearson correlation matrix — an N×N grid where each cell represents how closely two APs' signals co-vary over time.

// Build NxN correlation matrix from signal histories
const matrix = APs.map((a, i) =>
  APs.map((b, j) => i === j ? 1 : pearson(a.history, b.history))
);

// Convert correlation to distance: high ρ → small distance
const dist = (rho) => Math.sqrt(2 * (1 - rho));

The distance function √(2(1-ρ)) converts correlation into Euclidean distance: perfectly correlated signals (ρ=1) get distance 0, uncorrelated (ρ=0) get distance √2, and inversely correlated (ρ=-1) get distance 2.

Stage 3 — 3D Topology Solve

The distance matrix feeds into a spring-force layout algorithm that positions nodes in 3D space. Connected APs are pulled together by springs proportional to their correlation; uncorrelated APs repel each other. The system converges on a spatial arrangement that preserves relative distances.

  • Signal strength determines distance from center (router) — stronger signals orbit closer
  • Pearson correlation determines inter-AP distances — co-varying signals cluster together
  • Channel conflict detection draws clash lines between overlapping networks
  • Cross-network paths render as curved bezier arcs between distant, correlated APs

Stage 4 — Neural RF Rendering

The final 3D scene is rendered with Three.js using additive blending for a natural glow effect:

Wireframe Spheres Additive Glow Orbs Beacon Pillars Ground Rings Signal Lines Floating Labels Particle Field

Each network is a wireframe sphere sized by signal strength. Devices orbit their connected AP with emoji icons, glowing halos, and vertical beacon beams on selection. All labels are HTML overlays projected from 3D→2D coordinates every other frame for performance.

Architecture

Technical Stack

Rendering

Three.js r128 with shared geometry instancing. All devices reference pooled BufferGeometries — no per-mesh allocation. Pixel ratio capped at 1.5x. Signal lines update every 3rd frame.

Three.js WebGL AdditiveBlending
📡

Network Data

IP geolocation via ipapi.co (free, no key). Real public IP, ISP, city, region, and coordinates injected into the scan terminal and overview panel. Falls back gracefully if offline.

ipapi.co GeoIP Fetch API
🎯

Interaction

Raycaster-based click detection on 3D meshes. Spherical camera orbit with pointer drag + scroll zoom. Device selection triggers camera rotation, beacon beam, glow pulse, and profile panel slide-in.

Raycaster Spherical Orbit CSS Transitions
🎨

Design System

Glassmorphic panels with animated conic-gradient borders (CSS Houdini @property). Outfit + JetBrains Mono typography. Aurora sky, starfield particles, vertex-colored ground plane.

Glassmorphism Houdini Custom Fonts
Signal Intelligence

What the Lines Mean

The neural RF space renders four types of signal relationships as visible connections between nodes:

Network Links — AP connection to central router
Device → Router — Active device connections, colored by device type
Channel Clash — Same-channel interference between competing APs
Cross-Network — Curved bezier paths showing correlated distant APs
Research Basis

Built on DensePose Research

This visualization is inspired by the WiFi-based dense pose estimation research from Carnegie Mellon University (arXiv:2301.00250), which demonstrated that WiFi signals contain enough spatial information to reconstruct human body poses — proving that RF signals encode rich geometric data about physical space.

We extend this principle: if WiFi can map human bodies, it can certainly map the networks themselves. By treating each AP as a signal source and computing inter-signal correlations, we reconstruct the relative spatial topology of your entire wireless environment — no cameras, no floor plans, no hardware beyond your existing WiFi adapter.

Ready to see your network?

Launch the 3D neural RF space and scan your environment.

Launch 3D Map →