RFID Read Accuracy: Missed Reads, Metal & Liquids

What problem does this solve?

Almost every RFID project meets the same wall. The pilot in a clear area reads everything; the production environment reads most things. The gap is usually a small, stubborn set — a carton at the bottom of the pallet, the tag on the metal side of a tool, the bottle in the middle of the case, the item pushed to the back of the shelf. Because the misses are not random, the system produces confident, wrong answers, and the operations team goes back to manual verification while still paying for the RFID infrastructure.

The usual first response makes it worse. Transmit power is raised, which extends the field into the next bay and adds phantom reads without recovering the shadowed tags. More antennas are added at the same height and angle, duplicating coverage that already exists. Reader firmware and middleware settings are changed one after another with no controlled test in between, so nobody can say which change helped. After a few weeks the installation has drifted far from its original configuration and there is no baseline to return to.

The underlying issue is that read accuracy is a property of the whole system — tag design, what the tag is attached to, where it sits, the antenna geometry, the RF environment, the pass speed and the tag population — and it is measurable. Treated as a physics and measurement problem it is usually solvable. Treated as a software setting it is not.

How the solution works

Start from the energy budget. A passive UHF tag has no battery: it must harvest enough energy from the reader's field to power its chip, and then modulate a reflection strong enough for the reader receiver to decode. Anything that reduces the energy reaching the tag, detunes its antenna, or absorbs the reflected signal produces a miss. Metal placed directly against a standard tag detunes and effectively shorts its antenna; water and water-based liquids absorb energy at UHF frequencies; a tag oriented at right angles to a linearly polarised antenna couples poorly; and several layers of dense goods attenuate and reflect the field before it reaches the tag behind them.

Each cause has a specific fix. Metal needs an on-metal tag — a design with a ground plane and spacer that uses the metal surface as part of its antenna rather than fighting it. Liquids need the tag on the outer face of the packaging, a spacer, or a decision to identify at case or pallet level instead of item level. Orientation needs circularly polarised antennas or a second antenna on a different axis. Shadowing needs a placement standard, an additional antenna angle, or a controlled second pass. The engineering discipline that makes this work is testing: a fixed test load, a fixed pass speed, at least twenty passes, per-tag logging, and a map of which tags were missed — because the pattern of misses tells you the cause far more reliably than any setting screen.

1. 1
   Input — characterise the population List what is being read: materials, packaging, item size, how items are stacked, how many tags pass at once, tag model and where it sits on each item. Photograph the worst cases; they define the achievable read rate.
2. 2
   Capture — build a repeatable test Assemble a fixed test load with a known tag list, define a fixed pass speed and path, and run at least twenty passes. Log every read with tag ID, antenna, timestamp and RSSI so results can be compared change to change.
3. 3
   Processing — analyse the misses Classify: is the same tag always missed (tag or placement), is the same physical position always missed (antenna geometry or null), or are misses random (power, speed, density or collision)? Compare the RSSI distribution of successful reads against the read threshold.
4. 4
   Processing — change one variable at a time Apply a single change — tag model, placement, antenna angle, polarisation, power, sensitivity, session or Q — re-run the same test, and record the delta. Undo anything that does not help.
5. 5
   Integration — set the operational rule Where a residual set cannot be read reliably, define the handling rule instead of pretending: a second pass, a turn of the pallet, a handheld sweep, or a barcode fallback for that item family, implemented in the WMS or asset workflow.
6. 6
   Action — accept against a stated target Sign off the read point against a measured read rate and a separately measured false-read rate, both written into the acceptance document with the test procedure that produced them.
7. 7
   Reporting — monitor for drift Track read rate, reads per tag per pass and mean RSSI per read point over time. A falling baseline points to a moved antenna, a damaged cable or a new tag batch long before users complain.

Reference architecture

Read accuracy is designed at five points. Weakness at any one of them caps the whole system, and no amount of tuning downstream recovers it.

Deployment options: Diagnostics run on site with the customer's own goods. Logging and analysis tooling can run locally on a laptop or on the on-premise middleware; no read data needs to leave the facility, which matters for government and defence environments.

Key capabilities

Integrations

Accuracy work touches the operational systems in two ways: read statistics have to be visible, and unreadable items need a defined fallback that the business system understands.

Industry use cases

Implementation approach

1. 1
   Define the target and the fallback Agree, in writing, the read rate the process needs and what happens to items that miss. A system without a defined fallback will always look broken.
2. 2
   Build the reference test load Assemble a physical test load representing the real worst case, with a documented tag list. This load stays on site and is used for every future test.
3. 3
   Baseline measurement Twenty or more passes with full per-tag logging and RSSI capture, before any change is made. Without this baseline nothing later can be proven.
4. 4
   Tag and placement trial Compare candidate tags and placements on the same test load; keep the winner and document it with photographs.
5. 5
   Antenna and configuration tuning One variable at a time — geometry first, then power and sensitivity, then session/Q and dwell — re-testing after each change.
6. 6
   Environment and process adjustments Where RF cannot solve it, change the process: pallet build rule, pass speed, clearance around the read point, or turning the load. These changes are often free and effective.
7. 7
   Acceptance and documentation Sign off against the measured read rate and false-read rate, with the test procedure, the commissioned settings and the placement standard handed over.
8. 8
   Monitoring and periodic re-test Automated read-rate statistics plus a quarterly re-run of the reference test, and a re-test whenever packaging, racking or tag supplier changes.

Security & deployment

Read-rate diagnostics work on tag identifiers, timestamps and signal strength, not on business data, so the logs are low sensitivity — but for government and defence sites we still run the tooling locally and keep all captured data inside the facility, with nothing exported to a vendor cloud. Where reads are used as audit evidence, the raw per-tag logs behind an accepted count are retained alongside the count itself so a disputed figure can be reconstructed. Monitoring and analysis components deploy on-premise or air-gapped alongside the rest of the RFID stack, using the organisation's own directory for access control.

Limitations & prerequisites

• Some items cannot be read reliably at item level at any reasonable cost — dense metal assemblies, liquid-surrounded units in the middle of a case, and items inside sealed metal enclosures. The correct answer is a different identification level or a fallback process, not a bigger reader.
• Read rate is environment-specific. A figure measured in one warehouse, on one pallet build, at one pass speed, does not transfer to another site — including another door in the same building.
• Tags fail over time. Adhesive labels peel in heat and humidity, hard tags are knocked off, and a bad manufacturing batch can produce a cluster of dead tags; a retagging rate must be budgeted and monitored.
• Higher transmit power is not an accuracy strategy. It is capped by regulation, it increases stray reads and reflections, and it rarely recovers shadowed or detuned tags.
• Very large tag populations read simultaneously introduce collision effects; sessions, Q value and read-window length must be tuned, and beyond a point the process must slow down.
• Chip sensitivity varies between tag models and generations, so results from one tag do not predict another; every material change needs a fresh trial.
• Handheld read rates depend heavily on operator technique — sweep speed, distance and angle — which means training and periodic re-checks are part of maintaining accuracy, not a one-off.

Missed-read symptoms, likely causes and the real fix

Use the pattern of the misses to find the cause before changing any setting.

The pattern of misses is diagnostic. Consistent misses point to tags and placement; positional misses point to antennas; random misses point to timing, density and the energy budget.

FAQ

Related topics & solutions

• RFID Gate System Design & Tuning
• RFID Inventory Management & Warehouse Systems
• RFID Asset Tracking for Enterprise & Government
• RFID Site Survey & POC Checklist
• RFID vs Barcode vs BLE vs UWB: Choosing a Tracking Technology
• RFID Integration with SAP, Oracle, Odoo & IBM Maximo
• Octopus WMS
• MOWQIE ÔÇô RTLS Tracking