How do you set up Wi-Fi in a metal-framed factory?

You cannot cover a metal-filled factory with Wi-Fi using office rules. Racks, beams, machines, pipes, sheet metal, and vehicle bodies create reflections, deep signal nulls, and radio shadows that change as production changes. A network can look exemplary on a plan of an empty building while a forklift terminal loses its session at every turn between loaded racks.

A sound design does not start with an access point count or the advertised speed of a standard. First, define the application and client requirements, then measure the spectrum in several site states, place a test access point in the proposed locations, and check the network along the actual vehicle route. Transmit power does not correct poor geometry. Around metal, it often expands the interference zone faster than useful coverage.

Start by defining the connection the factory needs

Factory Wi-Fi requirements must be written as measurable values before the survey. The phrase "cover the entire area" says nothing about latency, packet loss, handoff time, or a specific terminal's ability to transmit data. A barcode scanner making short requests, a video camera, a voice headset, and a controller on a moving cart place different demands on the same radio network.

Make an inventory of devices with the exact radio model, number of spatial streams, supported bands and channels, antenna type, driver version, and mounting method. A terminal installed behind a metal cab post works differently from the same terminal in an operator's hand. The client decides when to roam, so controller features cannot force a poorly configured or old client to switch at the right time.

For every application, record the acceptable connection interruption, average and peak load, main traffic direction, and consequence of failure. If a drop stops a conveyor or an autonomous cart, RSSI alone is not enough. You will need measurements of signal-to-noise ratio, retries, channel utilization, packet loss, latency, and actual roaming time.

A minimum radio service profile fits into one table:

Do not copy a signal threshold from someone else's project. Cisco's Wireless RF Reference Guide gives -66 dBm as a commonly used client coverage reference, while an industrial Cisco and Sandvik design uses -55 dBm and SNR of at least 25 dB. These are not universal limits. The stricter values belong to a specific transport system and its machine requirements. Your threshold must come from testing the weakest client against the application's tolerance.

A reflection is not interference

Reflections and radio interference call for different remedies. Metal reflects the wanted signal, and several copies reach the receiver over paths of different lengths. This is multipath propagation. Interference occupies the same or an adjacent part of the spectrum with another transmission or with radiation that is not Wi-Fi at all. If the two are confused, a team can spend weeks changing channels where an antenna faces a beam, or moving access points while a machine produces constant noise.

Modern OFDM and MIMO systems can use some multipath energy, but that does not make a metal environment harmless. Signal copies may add at one spot and cancel at another. Narrow nulls therefore appear next to strong readings, and the coverage map changes when sheets, containers, or forklifts move. Cisco's Multipath and Diversity document explicitly lists steel mills, manufacturing areas, and distribution centers as environments with a high probability of multipath distortion. Its practical conclusion still holds: correct antenna position and geometry before trying to hide a null with maximum power.

Check the symptom. If RSSI is high but retries rise and modulation falls, investigate multipath, channel congestion, or a hidden transmitter. If RSSI and SNR fall together, the client has entered a radio shadow or moved too far from its serving access point. If RSSI barely changes while SNR drops sharply when a welder, drive, or wireless device starts, you need time-based spectrum analysis.

A single laptop walk does not reveal these dynamics. Record the state of the site with every measurement: which doors are open, how full the racks are, which lines are running, and where rail cars and containers stand. Without this record, two maps of the same zone will contradict each other while the cause remains hidden.

The primary band is usually 5 GHz

For a production network, 5 GHz usually provides the better starting channel plan because it offers more non-overlapping channels and makes cell size easier to control. Cisco's Converged Plantwide Ethernet guide recommends 5 GHz for industrial applications and leaves 2.4 GHz for noncritical, low-load access. I use that as a starting point, not a ban: the clients, channels permitted in the country, and measured spectrum decide the final plan.

The 2.4 GHz band travels farther and bends around some obstacles better, but that range often causes trouble in a factory. Neighboring access points hear each other across several bays, clients cling to a distant cell, and there are few non-overlapping channels. If old scanners only support 2.4 GHz, give them a separate SSID and radio plan. Use only non-overlapping channels permitted by the local regulatory domain, and do not enable 40 MHz width.

At 5 GHz, start with 20 MHz channels. A wide channel gives one client more peak speed, but it consumes several base channels and shortens the frequency reuse distance. In a factory with many access points and moderate traffic, that is a poor trade. Cisco's industrial architecture guide also recommends 20 MHz for IACS control applications. Move to 40 MHz only after a capacity calculation and confirmation that enough channels remain in every site state.

DFS channels can expand a 5 GHz plan, but they should not be enabled blindly. An access point must leave a channel when it detects a signal classified as radar, and some industrial clients support DFS poorly or scan those channels slowly. Check the supported-channel table for every client and the access point event log. The allowed frequencies, power, and indoor or outdoor operating rules must match the equipment's Kazakhstan regulatory domain, not a setting that a supplier happened to leave behind.

6 GHz is worth considering for new compatible devices and local high-capacity zones. It will not repair poor placement: higher frequencies suffer more behind obstacles, while production terminal fleets turn over more slowly than office laptops. First confirm client support for 6 GHz, its security mode, and the permitted local operating conditions. A design that depends on a band unavailable to half the devices will fail acceptance.

Bring access points closer to the route, not the ceiling

A high ceiling is convenient for an installer and almost always tempts a designer to draw large, even cells. For a terminal near the floor, it creates a long path through beams, crane rails, ducts, and moving loads. Cisco's IW6300 industrial access point instructions warn that unnecessary height increases reception of other unlicensed systems and reduces coverage on the ground. In a factory, bring the access point or remote antenna closer to the working height while protecting it from impact and process contamination.

Place radios in relation to aisles and work zones. In a long metal corridor between racks, a directional antenna at the end may provide more predictable coverage than an omnidirectional antenna above the top shelf. Around machines, several smaller cells aimed at work areas usually outperform one powerful access point under the roof ridge. Near a route turn, provide overlap before the vehicle body blocks the current access point.

Do not mount a standard antenna hard against a beam, cabinet, or duct. Metal changes its radiation pattern and input characteristics. An older Cisco voice WLAN guide gives a useful physical reference: keep an antenna one or two wavelengths away from a reflecting surface. A wavelength is about 12.5 cm at 2.4 GHz and about 6 cm at 5 GHz. This does not replace the installation guide for the specific antenna, but it explains why moving one a few centimeters can remove a deep null.

Do not raise access point power until you have compared it with the client. The access point may be audible from far away while the terminal's weak transmitter cannot answer. The result is an attractive one-way RSSI map and unstable data transfer. Set power so that uplink and downlink remain comparable, adjacent cells provide the required overlap, and clients do not hold onto an old access point across half the factory.

Before final installation, check the cable route, power, grounding, temperature range, enclosure rating, vibration, dust, moisture, and service access. A long coaxial cable consumes antenna gain, especially at higher frequencies. It is often better to bring network and power closer to the access point than to run a long radio-frequency path back to a convenient cabinet.

The radiation pattern matters more than the gain label

Choose an antenna for the shape of the service area, not for the largest dBi number. More gain does not create energy; it redistributes energy by angle. A high-gain omnidirectional antenna usually compresses the vertical lobe. Under a high ceiling, its main beam can pass above clients, leaving a weak connection directly below and many reflections farther away.

An omnidirectional antenna with moderate gain works for an open work area when installed in the orientation used in the manufacturer's published pattern. A directional panel or sector suits an aisle, conveyor, or crane path. It limits energy outside the required area, raises signal-to-noise ratio at the client, and makes channel reuse easier. A narrow beam demands precise aiming and tests at every machine position.

Preserve polarization. If access point antennas are vertically polarized while the terminal lies horizontally or turns with a load, the loss changes along the route. MIMO reduces sensitivity to one poor position but does not cancel geometry. Inspect the actual client mount and test the extremes: mast raised and lowered, cab turned, operator's hand covering the enclosure, and load in front of the antenna.

Connect every radio-frequency port on a multi-stream access point exactly as the manufacturer requires. Do not spread one radio's ports among separate aisles in the hope of creating two independent cells. That layout breaks the spatial relationship assumed by MIMO and factory calibration. Antennas in one set need compatible patterns, polarization, and cable lengths.

Record the azimuth, tilt, height, and enclosure orientation of each antenna for the installers. A photograph marked with the intended direction is more useful than "point toward the aisle." Compare the installed position with the design after the work. I have seen networks where a panel ended up facing a wall after the cable was dressed, while engineers spent months trying to correct the effect with power settings.

Survey with a test access point and the real client

A predictive model is useful for initial placement, but it does not approve a design in a metal factory. The model rarely knows every sheet position, load composition, machine shape, and crane movement. An active survey with a temporarily installed access point shows how the selected radio, antenna, channel, and height behave in the actual building.

First make a passive pass through the entire site and record the spectrum separately. Passive collection shows existing BSSIDs, channels, levels, and Wi-Fi utilization. A spectrum analyzer shows energy that a Wi-Fi adapter cannot classify as 802.11 frames. These measurements are not substitutes. Repeat the recording during a working shift, when high-power equipment starts, and during the period when complaints normally occur.

Then install an access point from the same family with the same antenna intended for the project. Set the proposed height, angle, channel width, and power. The access-point-on-a-tripod method is useful only with an honest configuration. A point at two meters and maximum power says nothing about a future installation at eight meters and medium power.

For every test position, perform the same route:

1. Walk or drive it with the survey adapter and record RSSI, SNR, noise, channel, and serving access point.
2. Repeat the route with the weakest production client in its actual mount.
3. Send a test flow similar to production traffic and record packet loss, latency, latency variation, and retries.
4. Mark the exact location of every roam, the interruption length, and the access point selected by the client.
5. Repeat the pass in the opposite direction and with another load or vehicle configuration.

Direction matters. The terminal antenna, forklift body, and load create different radio shadows on the outbound and return runs. An average over the route hides a two-meter null that is long enough to break a control session. Preserve the raw time-and-location log, not only the colored heat map.

Mark measurement points with physical references that will remain after installation: column number, grid axis, door, or rack start. Building plans and the actual site often disagree. If a coordinate exists only in the survey file, the installer cannot reproduce the position and an engineer cannot find the complaint location six months later.

Test the factory both full and empty. A Cisco wireless voice deployment document notes that a warehouse at 50 percent fill has a different radio-frequency footprint from the same warehouse at full capacity. With metal, the difference is even larger: a new row of parts both blocks the direct path and creates a reflecting corridor. If only one state can be surveyed, keep temporary test positions in the plan and repeat validation after production reaches its normal load.

Find an interference source by its timing

Change a channel only after identifying the source and timing of interference. Constant high energy, short broadband pulses, and legitimate traffic from a neighboring network may feel similar to an application, but they require different fixes. A normal Wi-Fi scanner sees access point and client frames but cannot explain unclassified noise. That requires a spectrum analyzer with time recording.

Match the radio log to the production log. Mark the start of welding, variable-frequency drives, wireless bridges, video systems, microwave heating, telemetry, and test transmitters. Do not blame every electric motor near a null. Look for a repeatable connection between an event and a spectrum change, then confirm it with a controlled start if production permits the test.

A neighboring access point on the same channel creates contention, not background noise. Devices hear management frames and wait for their turn, so high utilization reduces available airtime. With adjacent-channel interference, a receiver spends resources on a partially overlapping signal and makes more errors. The remedy starts with channel and power planning, not with adding access points.

After finding the source, choose an action that matches the signal. Move frequency or change the schedule if the source is under your control. Shield, repair, or relocate faulty equipment only with industrial safety and electromagnetic-compatibility specialists. For an external legal source, change the channel, position, and antenna pattern. Document the spectrum before and after, or the improvement will remain subjective.

Test moving equipment while it moves

Coverage on a parked forklift does not prove that a route works. While it moves, antenna orientation, shielding by the load, channel scan rate, and the roaming decision point all change. A client may see a new access point but stay with the old one until the level is too low. It may also select an access point behind a metal partition when a reflected beacon appears stronger than the direct signal.

Run the test at production speed with the normal application. Continuous ping is useful as a basic indicator, but it does not replace the production protocol: a small ICMP packet can pass after the application session has already exceeded its timer. Capture traffic at the client and infrastructure with synchronized clocks. You can then see the last message before the transition, the start of scanning, authentication, access, and the return of data.

Do not enable every fast-roaming feature at once. 802.11k helps a client learn about neighbors, 802.11v gives the network ways to suggest a better access point, and 802.11r shortens part of the transition exchange. Support depends on the client, driver, and security mode. Enable one feature on a test SSID, confirm compatibility across the fleet, and compare traces before and after. An old terminal that stops connecting after 802.11r is enabled matters more than an attractive controller status.

Test a critical route segment as a series, not once. A run without a drop may be a lucky combination of load position and other traffic. Make passes in both directions, stop at process stations, cross two vehicles, and test during peak load. Calculate the worst result and the share of runs that violate the criterion, not just the average.

Use an acceptance log in this form:

The row is a form example, not a standard. Insert the threshold and result from your application's service profile. Also preserve client firmware version, SSID configuration, and radio power. Otherwise, the test cannot be repeated after an update.

Acceptance must survive a working shift

Accept the network against agreed routes, states, and thresholds, not the absence of red spots on a map. Final validation covers every work area, stopping point, door, turn, lift, crane path, and temporary storage area. For stationary clients, check signal margin with doors closed and equipment installed. For mobile clients, check application continuity.

Before handover, record the installed access point plan, antenna models and serial numbers, cable lengths, orientation, channels, width, power, and permitted minimum rates. Attach RSSI and SNR maps, channel utilization, spectrum records, and route-test results. A map without source conditions is useless: it must state the date, shift, warehouse state, and list of running equipment.

After launch, watch metrics by access point and client model. Retries, rate changes, channel load, client count, failed authentications, DFS events, and roaming history are useful. Compare them with the acceptance baseline. A rise in retries near one column after a line is rearranged gives the engineer a specific place to inspect.

Production changes, so a survey has a shelf life. A new rack, metal partition, crane, line, or load type calls for local revalidation. You do not need to redesign the whole network every time. Repeat the preserved route, compare the log, and correct geometry where conditions changed.

As a system integrator, GSE.kz can connect the radio-frequency survey with cabling, compute infrastructure, and ongoing technical support for the site. Even an integrator cannot replace the main acceptance test: the production vehicle must travel its worst route during a working shift and keep the connection within the application's tolerance.