37 Sensing with Existing Infrastructure
The mathematical gist. Free-space loss in MHz is , so solving for distance gives . At 2,400 MHz and −65 dBm this ideal model gives 17.7 m; replacing the logarithm with gives 169 m, a 9.55-fold error even before walls and multipath.
Start with Structural Health Monitoring (SHM): Continuous or periodic measurement of structural response — strain, vibration, displacement, crack propagation — to detect damage, assess remaining service life, and prevent catastrophic failure. Then Fiber Bragg Grating (FBG) Sensor: An optical fiber sensor where periodic refractive index variations reflect specific wavelengths proportional to strain and temperature; immune to electromagnetic interference, suitable for embedded concrete installations. Next Acoustic Emission (AE) Sensor: Detects high-frequency stress waves (100 kHz - 1 MHz) generated when cracks form or grow in materials; provides early warning of fatigue crack initiation before visible damage appears. After that Tiltmeter: A precision inclinometer measuring angular deviation from vertical; MEMS tiltmeters achieve sub-millidegree resolution; used for monitoring slope stability, building settlement, and dam deformation. Continue by Corrosion Sensor: Electrochemical probes measuring instantaneous corrosion rate of metal structures; deployed in bridges, offshore platforms, and pipelines to schedule maintenance before structural compromise. Continue by Distributed Sensing: Fiber optic sensing systems measuring temperature or strain continuously along the entire cable length — thousands of virtual sensors in a single optical fiber installation. Continue by Wireless Sensor Node for Infrastructure: Battery or energy-harvesting powered nodes with sensing, processing, and wireless communication; must operate 5-10 years with minimal maintenance in outdoor temperature extremes and moisture. Finally Infrastructure Sensing: The use of sensors to monitor physical infrastructure condition — bridges, roads, buildings, pipelines, power grids — enabling predictive maintenance and safety monitoring at scale.
Learning Objectives
After completing this chapter, you will be able to:
- Evaluate existing infrastructure for repurposing as sensing platforms
- Apply Wi-Fi CSI sensing techniques for presence and gesture detection
- Analyze power meter waveforms to identify individual appliances via NILM
- Design hybrid sensing systems that combine infrastructure signals with targeted dedicated sensors
37.2 Prerequisites
- Sensor Classification: Types of sensors
- Braitenberg Model: Sensor-behavior relationships
37.3 Use Existing Infrastructure
37.3.1 Use Existing Sensors First
Traditional IoT thinking: “We need temperature data -> Deploy temperature sensors.”
Infrastructure-leveraging thinking: “We need occupancy data -> Use existing Wi-Fi routers.”
This paradigm shift—leveraging existing infrastructure instead of deploying dedicated sensors—can reduce costs by 10-100x while providing area-wide coverage instead of point measurements.
Imagine you want to detect if someone is in a room. Traditional approach:
Deploy Dedicated Sensors ($$$):
Start by Buy PIR motion sensors ($15-150 each depending on installation and housing). Then Install wiring and power. Next Calibrate and maintain batteries. Finally Get point measurements (one spot per sensor).
Leverage Infrastructure ($0):
Start by Use existing Wi-Fi router (already there!). Then Detect phone/device connections. Next Get room-wide coverage. Finally No installation, no batteries, no maintenance.
The Wi-Fi router becomes a “free” presence sensor. This is infrastructure-leveraging sensing—using existing devices as sensors.
37.4 Indirect Infrastructure Sensing
Instead of deploying new sensors, extract sensing information from infrastructure already in place:
| Infrastructure | Traditional Use | Sensing Capability | What It Measures |
|---|---|---|---|
| Wi-Fi Router | Internet connectivity | RSSI, CSI, device count | Presence, gestures, breathing, occupancy |
| Power Meter | Energy billing | Current waveform analysis | Appliance identification (NILM) |
| Cell Tower | Mobile calls/data | Handoff patterns, signal strength | Traffic density, crowd size, movement |
| Street Light | Illumination | Current draw, vibration | Pedestrian count, vehicle detection |
| Water Pipe | Water delivery | Acoustic vibrations | Leak detection, flow rate |
| HVAC Ducts | Climate control | Airflow patterns | Occupancy, room-level activity |
Key Insight: Every piece of infrastructure emits signals. We can sense the world by observing how the environment disturbs these signals.
37.5 Comparison: Deploy vs Leverage
| Dimension | Deploy Dedicated Sensors | Infrastructure-Leveraging |
|---|---|---|
| Upfront Cost | $10-100 per sensor point | $0 marginal cost (already installed) |
| Installation | Drilling, wiring, positioning | Software-only deployment |
| Coverage | Point measurements (discrete) | Area-wide coverage (continuous) |
| Maintenance | Battery replacement, calibration | Minimal (infrastructure already maintained) |
| Privacy | Obvious sensors (cameras visible) | Less intrusive (passive RF sensing) |
| Accuracy | High (purpose-built) | Moderate (indirect inference) |
| Latency | Immediate (direct sensing) | May require processing (feature extraction) |
When to Choose What:
- Deploy sensors when: Accuracy critical, point measurement needed, safety-critical
- Leverage infrastructure when: Large-scale deployment, cost-sensitive, retrofit scenario, quick pilot
37.6 Wi-Fi Breathing/Gestures
Infrastructure: Standard Wi-Fi router (802.11n/ac/ax)
Sensing Capability: Channel State Information (CSI) captures how Wi-Fi signals propagate through space. Human movement, breathing, even heartbeat disturbs these signals.
What It Measures:
Start with Occupancy estimation: Device count gives headcount estimates (+/-15% accuracy at zone level using devices-per-person divisor). Then Presence detection: CSI-based binary detection (occupied vs. empty room, no device needed). Next Gesture recognition: Hand waves, swipes (85-95% accuracy). After that Breathing rate: Chest movement modulates signal (+/-2 breaths/min under controlled single-occupant, line-of-sight conditions; accuracy degrades with multiple occupants or movement). Finally Fall detection: Sudden signal disruption pattern.
Real Deployment:
Start with University dorms: Detect occupancy without cameras (privacy-preserving). Then Elderly care: Fall detection without wearables. Finally Smart homes: Gesture control of lights and appliances.
37.7 Non-Intrusive Load Monitoring
Infrastructure: Smart electricity meter (already installed for billing)
Sensing Capability: Analyze current and voltage waveforms to identify unique “signatures” of individual appliances.
What It Measures:
Start with Appliance state: On/off, standby, active. Then Energy breakdown: Per-appliance consumption. Finally Usage patterns: When devices are used.
How It Works:
A NILM algorithm watches changes in total home power:
Start with 150 W baseline: Only always-on devices are active. Then 1,650 W: A +1,500 W jump strongly suggests a hair dryer or heater. Next 1,750 W: A further +100 W step suggests a refrigerator compressor. Finally 250 W: A -1,500 W drop suggests the hair dryer turned off.
Each appliance has unique electrical signatures: Start with Resistive loads (heaters, toasters): Clean on/off transitions. Then Motors (refrigerators, fans): Inrush current spike, steady-state hum. Finally Switched-mode power supplies (computers, chargers): High-frequency harmonics.
37.8 Example 3: Cellular Signal Analysis
Infrastructure: Cell towers (already everywhere)
Sensing Capability: Analyze signal strength patterns and handoff events to infer crowd density and movement.
What It Measures:
- Crowd density: Number of active devices in area
- Traffic flow: Movement patterns between cell zones
- Event detection: Large gatherings, unusual patterns
37.9 The Three-Layer Model
Start with Infrastructure Layer: Existing devices emitting signals (Wi-Fi, power, cellular). Then Disturbance Layer: Environment modulates these signals (human movement, appliance usage). Finally Analytics Layer: Software extracts sensing information from signal perturbations.
37.10 Design Guidelines
Good candidates for infrastructure sensing:
Start by Large buildings (retrofit without new wiring). Then Privacy-sensitive environments (no cameras needed). Next Budget-constrained projects. After that Quick proof-of-concept deployments. Finally Aggregate measurements (counts, not identity).
Still need dedicated sensors for:
Start by Safety-critical applications (fire, gas). Then High-precision measurements (temperature +/-0.1C). Next Point-specific data (this exact location). After that Real-time response (<100 ms). Finally Regulatory compliance (certified sensors).
the microcontroller had a brilliant idea: “What if we do not need to BUILD new sensors? What if sensors are already ALL AROUND US?”
“What do you mean?” asked Temperature Terry.
“Think about your Wi-Fi router at home,” Max explained. “It sends out radio waves all the time. When a person walks through the room, they disturb those radio waves — like making ripples in a pond. A smart computer can analyze those ripples and figure out that someone walked by, WITHOUT any cameras or motion sensors!”
the LED was amazed: “So the Wi-Fi router IS a sensor?”
“Exactly! And your electricity meter is a sensor too,” Max continued. “Every appliance in your house uses electricity differently. A hair dryer uses 1500 watts, a fridge uses 100 watts, a phone charger uses 5-25 watts. By watching the power meter, a computer can figure out which devices are turned on — like identifying people by their footsteps!”
the battery loved this: “No new sensors to install, no batteries to replace, no wires to run. The infrastructure does double duty!”
“But,” Sammy reminded everyone, “for really important measurements — like fire detection or medical monitoring — you still need dedicated sensors. Infrastructure sensing is clever but not as precise.”
37.11 Campus Occupancy Monitoring
Scenario: The University of Southern California (USC) wants to monitor real-time occupancy across 12 campus buildings to optimize HVAC scheduling. Installing dedicated occupancy sensors in every room (2,400 rooms) would cost over $360,000. Instead, they propose leveraging 480 existing Cisco Aironet Wi-Fi access points.
Given:
Start by 12 buildings, 200 rooms average per building = 2,400 rooms. Then 480 Wi-Fi APs already deployed (Cisco Aironet 3802i), 40 per building. Next Average 18,000 daily connected devices across campus (phones, laptops, tablets). After that Each device associates with one AP at a time; AP coverage radius ~25m indoors. Finally Current HVAC bill: $2.8 million/year across all 12 buildings.
Step 1 — Extract occupancy signal from AP data:
Each Cisco AP reports associated client count every 60 seconds via SNMP. We map AP location to room/zone:
| Building Zone | APs in Zone | Avg Devices (9am) | Avg Devices (3am) | Occupancy Estimate |
|---|---|---|---|---|
| Engineering Lecture Halls | 8 | 340 | 2 | ~227 students + staff (340 ÷ 1.5) |
| Library Main Floor | 6 | 280 | 5 | ~187 occupants (280 ÷ 1.5) |
| Admin Offices | 12 | 95 | 0 | ~63 staff (95 ÷ 1.5) |
| Gym / Recreation | 4 | 120 | 0 | ~80 people (1.5 devices/person) |
Key correction factor: Students carry 1.2-1.8 devices on average (phone + laptop). Apply divisor of 1.5 for headcount estimate.
Step 2 — Compare costs: Dedicated sensors vs. Wi-Fi infrastructure:
| Approach | Hardware Cost | Installation | Annual Maintenance | Accuracy |
|---|---|---|---|---|
| PIR sensors (2,400 rooms) | $360,000 | $120,000 | $48,000/year | +/-1 person per room |
| CO2-based (2,400 rooms) | $840,000 | $200,000 | $96,000/year | +/-3 people per room |
| Wi-Fi AP analysis (existing) | $0 | $12,000 (software) | $3,600/year | +/-15% per zone |
| Wi-Fi + 50 PIR (high-traffic) | $7,500 | $14,000 | $4,200/year | +/-10% per zone |
Step 3 — Calculate HVAC savings:
Start by Baseline: HVAC runs fixed schedule (6am-10pm) regardless of occupancy. Then With occupancy-responsive HVAC: Reduce conditioning in unoccupied zones. Next Measured savings from pilot (2 buildings, 6 months): 22% energy reduction. After that Projected annual savings: $2.8M x 22% = $616,000/year. Continue by Implementation cost: $33,500 (software $12,000 + 50 PIR sensors $7,500 + installation $14,000). Finally Payback period: ~20 days.
ROI for infrastructure-based sensing is compelling. The hybrid Wi-Fi + PIR approach costs:
- Software: $12,000
- 50 supplementary PIR sensors: $7,500
- Installation: $14,000
- Total implementation cost: $33,500
Annual savings from 22% HVAC reduction:
- Annual HVAC bill: $2,800,000/year
- Expected reduction: 22%
- Annual savings: $616,000/year
Payback period: about 20 days. Compare that with a dedicated PIR deployment costing $480,000 ($360,000 sensors + $120,000 installation), which takes about 284 days to pay back. The infrastructure approach delivers the same HVAC savings at 93% lower capital cost.
Result: By leveraging existing Wi-Fi infrastructure plus 50 supplementary PIR sensors in high-traffic areas, USC achieved zone-level occupancy estimates accurate to +/-10% at 93% lower cost than a dedicated PIR sensor deployment ($33,500 vs. $480,000). The system identified that 35% of HVAC runtime was wasted on empty zones, saving $616,000/year.
Key Insight: Infrastructure-leveraging sensing excels for “good enough” aggregate measurements. Zone-level accuracy (+/-10%) is sufficient for HVAC optimization — you do not need to know exactly who is in which room. The hybrid approach (Wi-Fi infrastructure + targeted dedicated sensors) gives the best cost-accuracy tradeoff.
37.12 Wi-Fi Occupancy Estimator
Use this calculator to estimate room occupancy from Wi-Fi access point device counts. Adjust the number of connected devices and the average devices per person to see the estimated headcount range.
37.13 Infrastructure Sensing ROI
Estimate the payback period and cost savings when using infrastructure-based sensing instead of dedicated sensors for building occupancy monitoring.
37.14 Activity: Wi-Fi CSI Presence Detection
This compact example demonstrates the core idea behind Wi-Fi Channel State Information (CSI) occupancy sensing. When a person moves through a Wi-Fi signal path, the CSI amplitude fluctuates more than in an empty room. Real deployments process many subcarriers over time, but the student-facing decision is simple:
Start by Observe CSI amplitude for an empty room and treat it as the baseline. Then Observe CSI amplitude while a person walks through the signal path. Next Compare how much the values fluctuate in each case. Finally Mark the room as occupied only when the current fluctuation is much larger than the empty-room baseline.
def variance(values):
mean = sum(values) / len(values)
errors = [
(x - mean) ** 2
for x in values
]
return sum(errors) / len(values)
empty_room = [
0.50, 0.51,
0.49, 0.50,
]
person_moving = [
0.20, 0.82,
0.35, 0.75,
]
baseline = variance(empty_room)
current = variance(person_moving)
ratio = current / baseline
occupied = ratio > 2.0
print(f"variance ratio = {ratio:.1f}")
print(f"occupied = {occupied}")Key design decisions:
| Decision | Choice | Rationale |
|---|---|---|
| Window size | 5 seconds | Smooths transient noise while detecting walking pace movement |
| Variance ratio threshold | 2.0x baseline | Empirically validated; 1.5x causes false positives from HVAC airflow |
| Subcarrier count | 52 | Standard for 802.11n 20 MHz channels |
| Calibration | Empty-room baseline | Accounts for room geometry and furniture reflections |
37.15 Concept Relationships
| Concept | Related To | Connection Type |
|---|---|---|
| Wi-Fi CSI | Signal Variance | Human movement increases channel state variance |
| NILM | Power Signatures | Each appliance has unique current draw pattern |
| Cell Tower Density | Crowd Estimation | Device count per tower estimates population |
| Infrastructure Cost | Marginal vs Fixed | Leveraging existing = $0 marginal cost |
| Privacy | RF Sensing | Wi-Fi CSI preserves privacy better than cameras |
37.16 Label the Diagram
37.17 Code Challenge
37.18 Proxy Sensing and Loop Limits
The body above explains how existing Wi-Fi, power, cellular, and building infrastructure can act as indirect sensors. Continue to Infrastructure Proxy Sensing and Inductive Loop Limits when you need the deeper physics and evidence boundaries behind those proxies. That page separates the L2 material: Wi-Fi fingerprint reuse, inductive-loop oscillator shifts, eddy-current blind spots, threshold trade-offs, auto-baselining, and proxy drift.
37.19 Summary
Infrastructure-leveraging sensing turns existing equipment into sensors through software analysis, providing area-wide coverage at near-zero marginal cost while trading some accuracy for convenience. Key takeaways:
Start with Existing infrastructure can serve as sensors — Wi-Fi routers (CSI for presence/gestures), power meters (NILM for appliance identification), and cell towers (crowd density estimation). Then Software deployment replaces hardware installation — Faster, cheaper, and easier to maintain. Next Area-wide vs point measurements — Zone-level accuracy (+/-10-15%) is sufficient for many applications like HVAC optimization. After that Privacy-preserving options — RF sensing avoids cameras while still detecting presence and activity. Continue by Combine approaches for best results — Hybrid deployments (infrastructure + targeted dedicated sensors) deliver the best cost-accuracy tradeoff. Finally Reserve dedicated sensors for safety-critical, high-precision, or real-time applications.
37.20 See Also
Start with Sensor Classification - Active vs passive sensing principles apply to infrastructure. Then Signal Processing - CSI data requires filtering and feature extraction. Next Common Mistakes - Infrastructure sensing still needs validation logic. Finally Selection Guide - When to leverage infrastructure vs deploy dedicated sensors.
37.21 Try It Yourself
Challenge: Estimate room occupancy from Wi-Fi access point data.
Given data: A conference room’s access point reports these device counts over 1 hour:
09:00 → 2 devices
09:30 → 8 devices
10:00 → 8 devices
10:30 → 12 devices
11:00 → 3 devicesYour task:
Start by estimate occupancy at 10:30. Then account for the fact that people carry 1.2-1.8 devices on average. Finally determine when the meeting likely started and ended.
Click for solution
1. Occupancy at 10:30:
Use the average device ratio: 12 devices ÷ 1.5 devices/person = 8 people.
2. Accounting for device ratio:
Start by conservative (1.8 devices/person): 12 ÷ 1.8 = 6.67 ≈ 7 people. Then liberal (1.2 devices/person): 12 ÷ 1.2 = 10 people. Finally Best estimate: 7-10 people at 10:30.
3. Meeting timeline:
Start with Started: Between 09:00 and 09:30 (jump from 2 → 8 devices). Then Ended: Between 10:30 and 11:00 (drop from 12 → 3 devices). Finally Duration: Approximately 1-1.5 hours.
Key insight: This ±2-person accuracy is sufficient for HVAC optimization but inadequate for safety-critical occupancy limits (fire code compliance).
Common Pitfalls
A single strain gauge only measures strain at its exact location. A fatigue crack forming 20 cm away produces no signal. Infrastructure monitoring requires sensor arrays with spacing matched to the expected damage scale. Work with structural engineers to determine appropriate sensor placement density.
Infrastructure sensors are often installed for 10-20 year lifespans. A strain gauge with 0.01% annual drift accumulates significant error over 10-20 years — enough to mask meaningful structural changes. Select sensors with certified long-term stability or include redundant sensors to detect drift through comparison.
High-frequency vibration monitoring (1 kHz sampling on 10 channels) generates 1.4 GB per day. Design the processing pipeline to extract features (RMS, spectral peaks, damping ratios) at the node and transmit only compressed summaries, not raw time series.
IP67 ratings are tested in clean water but real deployments involve condensation cycles and chemical exposure. Use hermetically sealed enclosures with desiccant for sensors in harsh environments. Specify higher IP ratings (IP68, IP69K) for washdown applications.