If someone is using WiFi signals to observe your space, can you detect them? The answer depends heavily on what kind of attacker — active emitters are tractable; pure passive receivers may be undetectable. This post maps five detection layers, assesses each against published research, and identifies what is shippable on commodity hardware today.

For the attacks these layers defend against, see WiFi sensing attacks . For the obfuscation-based alternative, see defenses and limits .

Layer 1: RF emitter inventory

Detect RF devices that transmit.

The most concrete, near-term layer. Scan for:

Unknown BSSIDs and hidden SSIDs
Probe requests and probe responses from unrecognized MACs
Beacon anomalies (unexpected intervals, vendor OUIs associated with sensing hardware)
New MAC addresses appearing in the environment
Channel hopping patterns inconsistent with normal WiFi clients
Espressif devices operating in mesh or CSI-extraction modes

Hardware: ESP32 nodes in promiscuous mode, RTL-SDR for sub-GHz protocols, existing WIDS infrastructure for WiFi management frames.

What it catches: Rogue APs, rogue ESP32 sensing nodes, SDR transmitters, active radar emitters. Any attacker who must transmit to collect CSI (e.g., injecting sounding frames, operating a rogue hotspot for WindTalker-style attacks).

What it misses: Pure passive BFI sniffing (LeakyBeam). The attacker’s NIC receives but never transmits. WIDS sees nothing because there is nothing to see at the management-frame level.

Feasibility: High. Shippable on ESP32 mesh with existing promiscuous-mode APIs. This is the foundation of an RF timeline — “Device X appeared on channel 6 at 14:32.”

Layer 2: RF fingerprinting

Identify hardware, not just MAC addresses.

MAC addresses are trivially spoofed. RF fingerprinting extracts hardware-intrinsic characteristics from the transmitted waveform:

IQ imbalance
Carrier frequency offset
Phase noise profile
Power amplifier nonlinearity
Clock skew

These features persist even when the attacker rotates MAC addresses, allowing detection of returning hardware that presents as a new device.

Research foundation: Mature literature in IEEE TIFS and IEEE Communications Magazine. See the RF fingerprinting survey and Junqing Zhang’s RFFI work for WiFi-specific methods including channel-robust extraction (Channel2Channel, IEEE JSAC 2025).

What it catches: Spoofed MACs on known surveillance hardware. Rogue ESP32 boards that reappear with different addresses. SDR platforms with distinctive transmitter impairments.

What it misses: Passive receivers (no waveform to fingerprint). Novel hardware with no baseline entry.

Feasibility: Medium. Requires labeled hardware datasets and per-environment calibration. Useful as a Layer 1 enhancement — when a “new” MAC exhibits a known RF fingerprint, confidence increases. Not standalone.

Layer 3: Environmental anomaly detection

Detect the effects of sensing, not the sensor.

Instead of finding the attacker, monitor whether the RF environment itself behaves abnormally:

Sudden CSI variance spikes without corresponding occupancy
Abnormal channel occupancy patterns
Unexpected RSSI distribution shifts
Multipath structure changes inconsistent with known activity
Persistent low-amplitude RF perturbations (attacker’s body/equipment in the channel)

Methods: Isolation Forest, autoencoders, temporal transformers, Bayesian changepoint detection — applied to CSI statistics, spectrum occupancy, and cross-node correlation from a distributed mesh.

What it catches: Scenarios where the attacker’s presence or equipment changes the channel even if the attacker is receive-only. A person sitting in a car with a monitor-mode laptop still perturbs local multipath. Sustained CSI polling at unusual rates (active sensing nodes) creates detectable traffic patterns.

What it misses: A well-positioned passive sniffer that adds no multipath perturbation and collects BFI from traffic the victim’s own devices generate. LeakyBeam’s attacker at 20 meters may be invisible to in-room anomaly detectors.

Feasibility: Medium. Leverages the same baseline and change-point expertise Wavey uses for occupancy detection — but inverted. Instead of detecting human perturbation, detect perturbation that does not match known occupancy. Promising research direction; not yet validated against BFI-specific passive attacks.

Layer 4: Distributed sensing mesh

Multiple nodes for triangulation, localization, and confidence.

A single ESP32 sees one channel geometry. A mesh of nodes sees:

Spatial correlation of anomalies across locations
Direction-of-arrival estimates from phase differences
Confidence scores that increase with node agreement
Source localization for active emitters

What it catches: Active emitters (Layers 1+4 combined enable triangulation). Environmental anomalies visible from multiple vantage points. RF heatmaps showing unexpected signal sources.

What it misses: Passive sniffers outside the mesh coverage area (e.g., LeakyBeam attacker across the street). Mesh nodes inside the building cannot detect a receiver that never transmits and sits beyond the building envelope.

Feasibility: Medium. Requires multi-node deployment — which Wavey’s architecture already supports. High value for enterprise (offices, data centers, secure facilities) where node density is feasible.

Layer 5: Behavioral baseline

Build normal RF behavior; alert on deviations.

Long-term RF timeline for a space:

Expected WiFi clients and their traffic patterns
Normal BLE, Zigbee, and IoT device inventory
Typical beacon frequencies and channel usage
Baseline CSI polling rates from legitimate sensing systems

Alert when:

A new RF source appears that was never seen during baseline period
Beacon frequency changes on a known device (firmware update vs. compromise)
CSI extraction signatures appear (Espressif mesh coordination, rapid channel scans)
Hidden network detected where none existed before
Unusual timing patterns in management frame sequences

What it catches: Everything Layers 1 and 3 catch, plus slow changes — an attacker who appears gradually, a sensing node deployed over weeks, a neighbor’s new ESP32 project that starts polling your channel.

What it misses: An attacker who mimics normal traffic patterns from the start (unlikely for CSI collection, which has distinctive signatures). Pure passive BFI sniffing with zero local footprint.

Feasibility: High. This is an RF SIEM — inventory plus timeline plus alerting. The most shippable component of an RF privacy platform and the closest to what goop-veil implements today.

The passive sniffer problem

The hardest case: attacker with a monitor-mode NIC, receive-only, well-shielded, positioned outside the building.

Can they be detected directly? Current research says probably not.

LO leakage exists but is microvolts at meters — indistinguishable from ambient noise in a WiFi-noisy office
No management frames to capture — WIDS blind
No RF fingerprint — nothing transmitted
No channel perturbation if the receiver is outside the sensing mesh

Can they be detected indirectly? Partially.

If they piggyback on your infrastructure (BFI from your clients), the attack leaves no local trace — the victim’s own devices generate the leaked signal
If they deploy active sensing (ESP32 transmitter, rogue AP), Layers 1, 2, and 5 catch them
If their body or equipment perturbs the local channel, Layer 3 may flag an anomaly

Honest product framing: “We detect active and semi-active RF sensing. Passive collection from outside the building remains an open research problem with fundamental physical limits.”

Recommended detection stack

For a practical RF privacy deployment, prioritize:

Priority	Layer	Deliverable
Now	1 + 5	RF inventory + timeline with alerts
Next	3	Anomaly detection on CSI/multipath statistics
Then	2	RF fingerprinting for spoofed-device detection
Later	4	Multi-node localization and confidence scoring
Research	Passive sniffer	Publish or perish; do not productize

The fastest path to a demo: “RF timeline for a room” — every emitter that appeared, every anomaly in CSI polling or beacon behavior, with alerts like “Unknown WiFi observer signature detected on channel 6 for 47 minutes.”

Detecting Unauthorized RF Sensing: A Layered Approach