Connected hardware
Concept studyInstrumenting a cold chain without stranding the data
A connected sensing device for refrigerated logistics, where the real engineering problem was surviving the environment and the radio budget, not reading a temperature.
- Engagement
- Project-based delivery
- Duration
- Seven months, prototype through manufacturing review
- Year
- 2024
- Client
- Not applicable
The challenge
What made this difficult.
Temperature excursions were being detected after the fact, from shipment records, which meant spoiled product was discovered at delivery. The requirement was continuous monitoring across refrigerated containers — an environment with condensation, vibration, and very limited connectivity. The honest assessment was that the hard constraints were physical and radio-related, and that a consumer device repurposed for the job would fail in the field for reasons unrelated to its software.
Constraints
Non-negotiables we designed around.
Condensation and thermal cycling
Enclosure design, coating, and venting had to survive repeated transitions between freezer and ambient temperatures without trapping moisture against the board.
Intermittent connectivity
Shipments pass through tunnels, depots, and ocean crossings. Data has to survive the gaps without draining the battery.
Battery budget measured in months
Devices are not serviced in transit, so radio and sampling duty cycle dominate the power design.
Calibration over time
A sensor that drifts is worse than no sensor. The measurement chain needed a defensible calibration story.
Part lifecycle and certification
Radio and safety requirements constrain component choice, and every part needs a second source.
Approach
How we would build it.
Sized the power budget before choosing parts
The radio duty cycle sets the architecture, so we modelled energy per transmission, per sample, and per wake against the target service interval, then selected components to fit that budget rather than adjusting the budget to fit the parts.
Designed the data path around disconnection
The device buffers locally with a bounded store, transmits opportunistically, and acknowledges sequence numbers so the backend can detect and resolve gaps. Missing data is an expected state the system is designed around, not an error path.
Made calibration explicit and auditable
Reference points at known temperature are recorded in the field, drift is tracked against them, and readings carry a quality flag the backend acts on. The system degrades visibly rather than silently.
Prototyped on production-intent parts
Early hardware was built on the components intended for the first production revision, so enclosure fit, thermal behaviour, and radio performance were measured rather than estimated.
Treated the device–cloud contract as a versioned interface
Firmware and backend evolve on different schedules, so the protocol carries an explicit version, negotiates capabilities on connect, and can be told to fall back to a minimal mode it can still satisfy.
Architecture
How the pieces fit together.
Device stack. Duty cycle is a design parameter: every layer is sized against the energy budget for one transmission cycle.
Sensing
- Temperature sensor
- Accelerometer
- Door state
- Reference calibration point
Firmware
- RTOS scheduling
- Adaptive sampling
- Local buffer with bounds
- Over-the-air update
Connectivity
- Low-power radio
- Batched transmission
- Sequence acknowledgement
- Capability negotiation
Ingest
- Protocol version handling
- Gap detection and fill
- Quality flag evaluation
- Timeseries store
Application
- Excursion detection
- Alerting
- Shipment timeline reconstruction
- Reporting
Stack
What it would run on.
Hardware
- 32-bit MCU
- Conformal coating and venting
- Primary + secondary sourcing
- Rev A prototype fabrication
Firmware
- C (bare metal and RTOS)
- Low-power radio stack
- Flash wear management
- Signed over-the-air updates
Cloud
- MQTT ingestion
- Timeseries storage
- Streaming excursion detection
- Fleet management console
Expected outcomes
What success would look like.
Power budget
Modelled pre-build
Energy per cycle was computed before component selection so the service interval was a design input, not a hope.
Data integrity
Gap-aware
Sequence acknowledgement means the backend can distinguish a device failure from a connectivity gap — the two had been indistinguishable before.
Calibration
Auditable
Field reference points make drift measurable, and readings carry a quality flag the application can act on.
Hardware revisions
One prototype
Design intent was to reach a manufacturable revision in a single prototype cycle by building early on production-intent parts.
What we learned
The conclusions we would carry forward.
For connected hardware, the power budget determines the architecture — modelling it first prevented a component selection that could not have met the service interval.
Designing for disconnection from the start changed the backend contract, not just the firmware.
A calibration story is a product requirement for anything that makes a measurement claim.
Related