Case study · Telecom site monitoring
Cell Towers That Report Their Own Health
ESP32 + Raspberry Pi CM controller: Modbus RTU sensors, MQTT telemetry, ML tower-health regression.
An AI-enabled remote monitoring system for telecom tower sites, built at Geocell Canada in Islamabad, Pakistan. Firmware for a custom controller pairing an ESP32 with a Raspberry Pi Compute Module reads industrial sensors over Modbus RTU, controls relays and alarms, and reports to the cloud over MQTT. A machine-learning regression model turns the site data into a predictive tower-health assessment.
- Org
- Geocell
- Year
- 2025–26
- Status
- Live
- Channels
- CH1CH3CH4CH5CH6
The problem
Remote telecom towers need constant checks on structure, power, fuel and security, and site visits are slow and expensive.
What I built
- Firmware for a custom ESP32 + Raspberry Pi Compute Module controller with Modbus RTU, MQTT, relay control, alarms, cameras and industrial sensor interfaces
- Cloud monitoring of tower tilt, vibration, 3-phase power, batteries, solar, fuel, environment and security
- An ML regression model for predictive tower-health assessment
Stack
ESP32 · Raspberry Pi CM · Modbus RTU · MQTT · ML regression
Results
- 8 site signals monitored from one controller
- Built at Geocell Canada (2025–26)
One controller watches eight kinds of site signal: tower tilt, vibration, three-phase power, batteries, solar, fuel, environment and security, with cameras on site. Relay control and alarms run locally on the controller.
I developed the firmware and worked across the delivery: custom controller hardware, industrial sensor interfacing, IoT connectivity, bench testing and field deployment in outdoor enclosures, with a touchscreen dashboard on site and cloud monitoring for the operator.
Gallery
Straight from the archive: every photo I could dig up for this build.
