By Apple Ko, Global Product Sales Business Development, Eelink Technology Detection gets the budget. Absence rarely does.
Failures divide into two kinds. Some produce a visible local effect — a gate that will not open, a turnstile that jams, a vehicle that misses a delivery window — and somebody raises it. Others are only ever visible remotely, and there the monitoring platform is the sole witness. An asset whose entire output is telemetry belongs to the second kind. When it loses its network it raises no error. It goes quiet, and quiet is indistinguishable from a stretch in which nothing happened.
When a network technology is retired, the devices that depended on it do not fail in a way that reaches an operations dashboard. They attach to nothing and stop transmitting. On a platform where alerts are generated by processing incoming readings — temperature out of range, shock event, geofence breach — no incoming reading means no alert is ever evaluated.
The result is a fleet that looks healthier than it is. A device reporting normally on most lanes and silently on one passes every dashboard check it will ever face, and can stay in service far longer than a unit that has simply died — because a unit that has simply died gets replaced.
In its August 2026 switch-off report, the Global mobile Suppliers Association counted 313 completed, planned or in-progress 2G and 3G shutdowns across 89 countries and territories, with 139 operators across 70 markets having shut down or committed to shutting down 2G. Each event removes a fallback radio that some fielded device was specified against.
Those events do not land uniformly. Sweden's regulator PTS records Telia carrying 2G to the end of 2027 while Tele2 and Telenor have already retired theirs; across the border, the joint operator site for Norway's shutdown has Telia finished in 2025 and Telenor's shutdown set for 31 December 2027. The same fleet can go quiet on one route and stay healthy on the next — precisely the pattern a dashboard built around incoming readings will not surface.
The fix is not exotic, and it is mostly not in the hardware.
A monitoring platform needs an expected reporting interval per device and a missed-report threshold that raises an exception on its own, without waiting for a reading to arrive. Liveness monitoring is not a new idea in enterprise IT. What is routinely missing is its application to field assets whose reporting cadence is irregular by design.
A single fleet-wide timeout will either miss real outages or bury the operations team in false positives. Workable thresholds are derived per device class from four inputs.
Expected cadence and sleep schedule. A unit reporting every fifteen minutes and one waking twice a day cannot share a threshold.
Batching and transmission jitter. Store-and-forward designs deliver in bursts, and the grace period has to absorb that without masking a real gap.
Planned suppression. Warehoused stock, maintenance windows and units not yet commissioned should suppress rather than alert. A device that has never reported at all needs its own state, distinct from one that has stopped.
A named escalation owner and an explicit recovery event. An exception nobody owns is a log line, and a reconnection that closes a ticket silently teaches the team to ignore the alert.
The device side is not passive either. Logging to internal memory with automatic forwarding on reconnection turns a coverage gap into a delayed record rather than a missing one — but only if the buffer depth matches the worst plausible outage and original timestamps survive the upload rather than being rewritten to the moment of transmission. Retry policy, buffer-overflow behaviour, clock integrity and watchdog recovery sit on the same side of that line.
Specifications go wrong when they assume a single supplier controls the chain.
Radio capability and market access are separate questions. Capability comes from the modem, RF front end, antenna design, SKU and firmware configuration. Access is jurisdictional: regulatory authorisation, modular versus final-device approval, and carrier acceptance are distinct hurdles, and a module-level approval does not automatically carry to the finished product in every market.
Radio choice follows duty cycle. Telenor IoT's migration guidance positions LTE Cat-1 and Cat-1 bis as the near-term baseline for most global fleets, with LTE-M for battery-powered mobile assets. NB-IoT earns its place on stationary or rarely-moving assets, where deep network penetration matters more than mobility — though network reach is not the same as a position fix. An NB-IoT GPS tracker sited deep inside a building may hold its connection while its GNSS receiver sees too little sky to report where it is, and NB-IoT roaming support has to be confirmed operator by operator rather than read off a coverage map.
Roaming is a contract, not a firmware setting. Which partner networks are reachable is set by the SIM or eSIM and the connectivity provider. Firmware still matters at the margin, through RAT preference and network-selection behaviour, so both sides need checking rather than one.
Silence detection usually belongs to the platform, because it is the only component that knows what should have arrived. Device-side connectivity-loss handling, watchdogs and health telemetry complement it rather than replace it. In practice this is the control that tends to have no owner at all.
Silence monitoring is one of the few controls that can be verified cheaply. Power down a unit in a controlled setting and confirm that the exception fires, reaches the named owner, and closes on reconnection with the buffered records intact and correctly timestamped.
Then look backwards. Pull the reporting history of the fleet and find stretches with no data that were never treated as exceptions. Those gaps are leads rather than findings — a gap can equally be power loss, an antenna fault, a configuration change or an ingestion failure — but each one is a case where the platform had nothing to say. That is the condition worth closing before the next radio is switched off.
Apple Ko is a Global Product Sales Business Development at Eelink Technology, a designer and manufacturer of cellular tracking and monitoring hardware supplying IoT platform providers, logistics operators and cold chain solution builders. The company's design and engineering base is in Shenzhen, with manufacturing in Yibin, China and Haiphong, Vietnam. Its work covers multi-radio asset tracking, temperature-monitored freight and OEM and ODM device programmes built for platform brands.
Failures divide into two kinds. Some produce a visible local effect — a gate that will not open, a turnstile that jams, a vehicle that misses a delivery window — and somebody raises it. Others are only ever visible remotely, and there the monitoring platform is the sole witness. An asset whose entire output is telemetry belongs to the second kind. When it loses its network it raises no error. It goes quiet, and quiet is indistinguishable from a stretch in which nothing happened.
A Retired Radio Does Not Fail Loudly
When a network technology is retired, the devices that depended on it do not fail in a way that reaches an operations dashboard. They attach to nothing and stop transmitting. On a platform where alerts are generated by processing incoming readings — temperature out of range, shock event, geofence breach — no incoming reading means no alert is ever evaluated.
The result is a fleet that looks healthier than it is. A device reporting normally on most lanes and silently on one passes every dashboard check it will ever face, and can stay in service far longer than a unit that has simply died — because a unit that has simply died gets replaced.
Why Silence Is About To Get More Common
In its August 2026 switch-off report, the Global mobile Suppliers Association counted 313 completed, planned or in-progress 2G and 3G shutdowns across 89 countries and territories, with 139 operators across 70 markets having shut down or committed to shutting down 2G. Each event removes a fallback radio that some fielded device was specified against.
Those events do not land uniformly. Sweden's regulator PTS records Telia carrying 2G to the end of 2027 while Tele2 and Telenor have already retired theirs; across the border, the joint operator site for Norway's shutdown has Telia finished in 2025 and Telenor's shutdown set for 31 December 2027. The same fleet can go quiet on one route and stay healthy on the next — precisely the pattern a dashboard built around incoming readings will not surface.
Absence Has To Be Its Own Alert Condition
The fix is not exotic, and it is mostly not in the hardware.
A monitoring platform needs an expected reporting interval per device and a missed-report threshold that raises an exception on its own, without waiting for a reading to arrive. Liveness monitoring is not a new idea in enterprise IT. What is routinely missing is its application to field assets whose reporting cadence is irregular by design.
Setting The Threshold Is The Hard Part
A single fleet-wide timeout will either miss real outages or bury the operations team in false positives. Workable thresholds are derived per device class from four inputs.
Expected cadence and sleep schedule. A unit reporting every fifteen minutes and one waking twice a day cannot share a threshold.
Batching and transmission jitter. Store-and-forward designs deliver in bursts, and the grace period has to absorb that without masking a real gap.
Planned suppression. Warehoused stock, maintenance windows and units not yet commissioned should suppress rather than alert. A device that has never reported at all needs its own state, distinct from one that has stopped.
A named escalation owner and an explicit recovery event. An exception nobody owns is a log line, and a reconnection that closes a ticket silently teaches the team to ignore the alert.
The device side is not passive either. Logging to internal memory with automatic forwarding on reconnection turns a coverage gap into a delayed record rather than a missing one — but only if the buffer depth matches the worst plausible outage and original timestamps survive the upload rather than being rewritten to the moment of transmission. Retry policy, buffer-overflow behaviour, clock integrity and watchdog recovery sit on the same side of that line.
The Controls Have More Than One Owner
Specifications go wrong when they assume a single supplier controls the chain.
Radio capability and market access are separate questions. Capability comes from the modem, RF front end, antenna design, SKU and firmware configuration. Access is jurisdictional: regulatory authorisation, modular versus final-device approval, and carrier acceptance are distinct hurdles, and a module-level approval does not automatically carry to the finished product in every market.
Radio choice follows duty cycle. Telenor IoT's migration guidance positions LTE Cat-1 and Cat-1 bis as the near-term baseline for most global fleets, with LTE-M for battery-powered mobile assets. NB-IoT earns its place on stationary or rarely-moving assets, where deep network penetration matters more than mobility — though network reach is not the same as a position fix. An NB-IoT GPS tracker sited deep inside a building may hold its connection while its GNSS receiver sees too little sky to report where it is, and NB-IoT roaming support has to be confirmed operator by operator rather than read off a coverage map.
Roaming is a contract, not a firmware setting. Which partner networks are reachable is set by the SIM or eSIM and the connectivity provider. Firmware still matters at the margin, through RAT preference and network-selection behaviour, so both sides need checking rather than one.
Silence detection usually belongs to the platform, because it is the only component that knows what should have arrived. Device-side connectivity-loss handling, watchdogs and health telemetry complement it rather than replace it. In practice this is the control that tends to have no owner at all.
Test The Detection Before Trusting It
Silence monitoring is one of the few controls that can be verified cheaply. Power down a unit in a controlled setting and confirm that the exception fires, reaches the named owner, and closes on reconnection with the buffered records intact and correctly timestamped.
Then look backwards. Pull the reporting history of the fleet and find stretches with no data that were never treated as exceptions. Those gaps are leads rather than findings — a gap can equally be power loss, an antenna fault, a configuration change or an ingestion failure — but each one is a case where the platform had nothing to say. That is the condition worth closing before the next radio is switched off.
Apple Ko is a Global Product Sales Business Development at Eelink Technology, a designer and manufacturer of cellular tracking and monitoring hardware supplying IoT platform providers, logistics operators and cold chain solution builders. The company's design and engineering base is in Shenzhen, with manufacturing in Yibin, China and Haiphong, Vietnam. Its work covers multi-radio asset tracking, temperature-monitored freight and OEM and ODM device programmes built for platform brands.