TL;DR:

  • Cost of commercial-grade air quality sensors has dropped enough to make dense sensor networks viable for building and campus environments — PM2.5, CO2, VOC, temperature, and humidity sensors in a single node now cost £50-200 at component level
  • Edge inference on the sensor node itself (using TinyML models on Cortex-M microcontrollers) enables local anomaly detection, data compression, and reduced LoRaWAN transmission overhead compared to sending raw readings
  • UK Building Regulations Part F (ventilation) and WELL certification are creating commercial pull for indoor air quality monitoring in commercial buildings, driving more standardised deployment patterns

Air quality monitoring used to mean expensive fixed stations run by local authorities, or single-point sensors in meeting rooms. The combination of cheap sensors, low-power connectivity (particularly LoRaWAN), and edge AI inference has changed the economics entirely. You can now deploy a building-wide air quality monitoring network for a few thousand pounds, with real-time data and local anomaly detection.

This guide covers the architecture, sensor selection, connectivity, and edge inference approaches for 2026 deployments.

What You’re Measuring and Why

A typical air quality node measures some combination of:

CO2 (carbon dioxide): The most practical proxy for indoor ventilation adequacy in occupied spaces. NDIR (non-dispersive infrared) sensors are the gold standard — they are accurate, stable, and not prone to cross-sensitivity from other gases. Target: below 800 ppm for well-ventilated spaces, WELL Building Standard recommends below 1000 ppm. Sensor cost: £15-40 at component level (Sensirion SCD40, SCD41).

PM2.5 and PM10 (particulate matter): Laser particle counters measure fine and coarse particulates. Critical for near-road locations, industrial environments, and during outdoor pollution events. Sensor cost: £8-25 (PMS5003, Sensirion SPS30 — the SPS30 adds mass concentration estimates and autoclean).

VOC (volatile organic compounds): Metal oxide sensors measure total VOC as a proxy for indoor chemical pollution from cleaning products, furniture off-gassing, cooking. SGP30 and SGP41 from Sensirion are widely used; they require a humidity/temperature input to compensate readings. Sensor cost: £3-8.

NO2 and O3 (outdoor deployments): Electrochemical sensors for nitrogen dioxide and ozone, primarily for outdoor urban monitoring. These require more careful calibration and are sensitive to temperature, humidity, and cross-sensitivity between gases. Alphasense OX-A431 and NO2-A43F are common choices. Sensor cost: £20-50.

Temperature and relative humidity: Required for compensation of most gas sensors. SHT40 (Sensirion), BME280 (Bosch) are reliable and cheap. Sensor cost: £2-5.

A complete indoor air quality node (CO2, VOC, PM2.5, temperature, humidity) can be built for £50-80 in component costs. Add enclosure, PCB, microcontroller, and connectivity module, and a complete deployable node lands at £150-250 at small quantities.

Edge Inference: Why Process on the Node

Sending raw sensor readings to a cloud backend every 30 seconds works, but it creates problems at scale: LoRaWAN has limited bandwidth (typically 250 bytes per transmission in the EU868 band), battery life on wireless nodes is reduced by frequent transmissions, and cloud processing adds latency for real-time alerting.

Edge AI inference on the sensor node addresses all three. A small TinyML model running on the microcontroller (typically a Cortex-M33 or similar) can:

Anomaly detection: A model trained on baseline air quality patterns for a specific location can identify anomalous events (a cooking event in a school corridor, a cleaning product spill, unusual PM2.5 during normal hours) and send an alert immediately, rather than relying on a cloud system to detect the anomaly from received readings.

Data compression: Instead of transmitting every 30-second reading, the edge model transmits only when readings change significantly from a predicted trajectory, plus a summary of the interval. This can reduce transmission frequency by 5-10x while preserving the information content.

Sensor fusion: Combining CO2, VOC, and PM readings with temperature and humidity to produce a single IAQ (Indoor Air Quality) index computed locally, rather than sending five separate values and computing the index in the cloud.

For Cortex-M class microcontrollers (STM32L5, nRF9160), TinyML frameworks like TensorFlow Lite for Microcontrollers or Edge Impulse’s C++ SDK can run sub-10ms inference cycles on small anomaly detection models, consuming milliwatts.

Connectivity: LoRaWAN for Building and Campus Scale

For building-scale deployments, LoRaWAN is the dominant choice for wireless air quality nodes. The advantages:

  • Range: LoRaWAN signals penetrate building materials well. A single gateway covers a multi-storey office building or a campus of 50,000 square metres.
  • Power: Nodes can operate for 2-5 years on a pair of AA batteries at typical air quality transmission intervals.
  • Cost: LoRaWAN gateways start at £150-300 for indoor models. The Things Network provides free community network coverage in many UK cities.
  • Capacity: A single gateway handles thousands of nodes.

For urban-scale environmental monitoring, public LoRaWAN networks (The Things Network, Helium) mean you don’t need your own gateway infrastructure.

Alternatives worth knowing:

  • Narrowband IoT (NB-IoT): Better for outdoor deployments across areas not covered by LoRaWAN, uses cellular infrastructure, but higher power consumption and running costs.
  • Wi-Fi: For mains-powered indoor nodes where power is not a constraint and you want simpler integration with existing IT infrastructure.
  • Thread/Matter: For smart home scale deployments where integration with building management systems via Matter is a requirement.

Smart Building Integration

The growing commercial deployment of air quality monitoring in UK commercial buildings is being driven by two regulatory angles:

Building Regulations Part F: The 2021 update to Part F (Ventilation) requires CO2 monitoring in certain new and renovated commercial buildings. The technical specifications are increasingly driving building owners toward networked monitoring rather than standalone units.

WELL Building Standard: Version 2 of the WELL standard includes specific air quality requirements (WELL Air feature set) with measurement requirements that networked IoT monitoring systems satisfy more cost-effectively than periodic manual checks.

For integration with building management systems, the key protocols are BACnet (the BMS standard in commercial buildings) and MQTT/REST APIs for cloud-connected platforms. Sensirion, Netatmo Business, and IQAir AirVisual Pro all offer commercial-grade devices with BACnet or API integration. For custom deployments, Home Assistant’s BACnet integration or a dedicated IoT gateway (such as an AWS Greengrass device) can bridge LoRaWAN data into existing BMS infrastructure.

A Basic Deployment Architecture

A starting point for a building deployment:

  1. Node: ESP32-based or Nordic nRF52840 board with SCD41 (CO2), SPS30 (PM), SGP41 (VOC), SHT40 (temperature/humidity). Battery powered, LoRaWAN radio module (Semtech SX1262).
  2. Gateway: RAK7268 indoor LoRaWAN gateway, connected via Ethernet to building network.
  3. Network server: ChirpStack (self-hosted on a Raspberry Pi or cloud VM) or The Things Stack Cloud.
  4. Data pipeline: MQTT from network server to InfluxDB time-series database.
  5. Visualisation: Grafana dashboard with per-room panels, alert rules for CO2 threshold breaches.

This architecture costs around £500-1,000 in hardware for a 10-node pilot, plus ongoing hosting costs of under £20/month. It scales linearly: add nodes, the gateway and backend handle the load without modification up to several hundred nodes on a single gateway.

Edge AI anomaly detection can be added by training a model in Edge Impulse on a few weeks of baseline data, deploying it as a firmware update to the nodes via LoRaWAN OTA update.