A wireless network can look sound on a coverage map but fail when a scanner crosses an aisle or machinery changes the radio environment. The useful question is which design carries a defined workload through the actual site, within an acceptable failure limit and operating budget. Private 5G and industrial Wi-Fi both need to answer that question in a trial.
NIST studies industrial wireless systems in terms of propagation, interference, reliability and timing. Its NIST’s Industrial Wireless Systems project is a useful starting point: describe the application and its environment before judging a radio technology.
Define the work before choosing the network
Separate the proposed deployment into workloads. A scanner sends small transactions but may need prompt confirmation while its operator moves. A maintenance tablet may tolerate a slow document download but still need coverage along the whole route. A sensor reporting once an hour has different needs from one feeding a time-sensitive process. A control command has a deadline and a consequence if it arrives late. Calling all of these “industrial IoT” conceals the differences that decide the design.
For each workload, record endpoint models, traffic direction and size, simultaneous device count, movement path, acceptable interruption, recovery behaviour and consequence of failure. Decide whether a delayed message is still useful. An inventory update may be recoverable; a late control command may not be. Include the application server and wired path in the timing boundary. A good radio-link result alone cannot show what an operator experiences.
NIST’s industrial wireless problem-space research frames technology selection around operating problems and use cases. That is a firmer basis for a purchase than a claimed peak speed.
Check the constraints that can rule out a design
Endpoint compatibility. Inventory the radios already in scanners, tablets, sensors and machines. An existing Wi-Fi fleet makes Wi-Fi a practical first candidate. Private 5G may require different endpoints, modems or gateways, affecting power, mounting and device management. A planned equipment refresh may change the calculation. Test exact device and software combinations: support for a technology family does not prove that a particular endpoint works in the proposed configuration.
Coverage and movement. Survey aisles, loading areas, metal structures, enclosed rooms and outdoor paths at the heights where devices operate. Walk or drive the real route while an application transaction runs. Record interruption, retries and recovery at each handoff, as well as signal strength. A stationary reading cannot prove that a moving device keeps a usable session. Repeat measurements after layout or stock changes that materially affect the radio path.
Spectrum and interference. Wi-Fi uses shared spectrum, so channel planning and coexistence tests matter. A private 5G proposal must identify its spectrum, the conditions for using it and who manages interference. Neither label guarantees an uncontested channel. NIST’s industrial 5G testbed evaluates performance under path-loss conditions and with wireless “aggressors”. A site trial should likewise include likely sources of degradation, not only a quiet floor.
Operational ownership. Name who designs the network, changes settings, monitors alarms, replaces equipment, manages identities and restores service. For private 5G, include the radio network and core. For Wi-Fi, include access points, controllers or management software, switching and authentication. Ask the team that will run the plant to trace an application complaint through those components. A design that passes a demonstration still needs a workable fault and support process.
Check access control with the same care as coverage. List how each class of device receives credentials, how a lost unit is removed, and who can change network policy. Walk through software updates and a failed authentication service during a maintenance window. These checks do not favour either radio technology; they expose dependencies that can stop a usable endpoint from reaching its application.
Match each workload to a test
Mobile scanners: Start with the installed devices and map a complete working route. Count completed transactions, retries and the longest interruption at each handoff. Trial private 5G if the tested Wi-Fi design misses the agreed limit and compatible endpoints are available. A wider coverage claim alone does not establish that scanner transactions survive movement.
Worker tablets: Test document access and other real tasks at weak-signal locations, including the transitions between indoor and outdoor areas. Record session continuity and the effort required to support the devices. A separate network is justified only if it solves a measured requirement that the proposed Wi-Fi design cannot meet at an acceptable total cost.
Monitoring sensors: Compare delivered readings, the age of data after disruption, recovery after an outage and endpoint power use. The reporting interval and tolerance for stale data determine whether a missed message matters. Check gateway needs and installation access alongside radio coverage.
Time-sensitive control: Measure the complete application path under load and interference. Count deadline misses and test the process response to lost communication. Research on NIST’s wireless time-sensitive networking testbed addresses latency and reliability with channel disturbances; it does not certify every Wi-Fi installation. The same limit applies to a private 5G claim. Approve only the tested configuration for the defined function and failure behaviour.
Write failure limits into the trial
Agree on a scorecard before either candidate is installed. For every workload, define the measurement boundary, route, production load, observation period and failure threshold. Measure application transaction completion, end-to-end latency where relevant, deadline misses, consecutive losses, interruption during movement and recovery after equipment failure. Record conditions with each result so a quiet-shift run is not compared with a busy one as though they were equivalent.
Set thresholds with the process owner. A scanner limit might concern transactions requiring manual retry on a route. A control limit should describe the missed-deadline behaviour the process can safely tolerate and what happens when communication fails. These are types of limits, not universal pass marks. If a process cannot tolerate the tested failure mode, a favourable average latency does not make the network suitable.
Run both candidates with the same workload scripts and floor conditions. Include simultaneous devices, shift changes, moving equipment, obstructed paths and known interference. Repeat movement and fault-recovery tests. Keep raw logs, timestamps, configurations and test-position maps. Without those records, a result is hard to reproduce or attribute to a particular design.
Make the acceptance test observable by the people who own the process. Pair network logs with application timestamps and a record of operator actions, using clocks that can be compared. When a task fails, identify whether the cause lies in the endpoint, radio path, wired network or application. Record unresolved causes as untested rather than assigning them to whichever technology happens to be under review.
Compare the cost of meeting the same accepted workload. Include endpoint changes, survey and installation work, backhaul, power, spectrum obligations where applicable, software or service contracts, monitoring, spares, staff time and planned expansion. Account for any area that needs a second access method. A pilot price that omits device replacement or ongoing support is not a useful comparison.
Record a narrow decision
Different workloads may need different answers. Existing Wi-Fi may serve tablets and scanners while a separate network earns its place for a measured mobility or timing requirement. A well-designed Wi-Fi deployment may also pass a workload initially earmarked for private 5G. State the endpoints, route, configuration, load and failure limits behind each decision.
Close procurement with a record of what passed, failed and remains untested. Require a remedy and retest for fixable failures. The defensible choice is the design the site can operate and whose measured behaviour stays within the workload’s limits under difficult conditions.