A chilled meat case is, from a radio’s point of view, close to the worst thing you can ask a passive UHF tag to work inside: trays of high-water product stacked two and three deep, on a metal shelf, with the label on film touching the product. A tag that reads across a shop floor on an apparel hangtag can fail at arm’s length here. That is the difficulty with RFID tags for fresh food — a physics problem before it is a procurement one.
This post is about what water does to a tag’s antenna and to your power budget, not about temperature — for thermal history, see our cold-chain guide.
Two failure mechanisms, not one
“Metal and liquid” usually gets written as a single obstacle. On liquid there are two mechanisms, and they fail differently.
Detuning is an impedance problem. A passive inlay is a small dipole conjugate-matched to a chip whose input is largely capacitive: Impinj publishes 3.50 kΩ parallel resistance and 0.930 pF chip capacitance for its M830 at AutoTune mid-range, 1.045 pF once mount parasitics are counted (M830/M850 datasheet v2.1, 2025). Bring a high-permittivity medium within millimetres of that antenna and the effective permittivity it sees rises sharply: the guided wavelength shortens, resonance slides downwards, the feed impedance walks off the match, and power harvesting collapses. That part is partly recoverable by design.
Absorption is an efficiency problem. The same food has a large loss factor, so energy crossing into the product is dissipated as heat. No retuning recovers it. Bianco and Marrocco separate the two in the IEEE Journal of Radio Frequency Identification (2024): a liquid load both absorbs and detunes, and self-tuning silicon compensates only limited impedance mismatches — limited is their word.
Stacked trays add a third effect, usually misread as collision: Floerkemeier and Lampe (ETH Zurich) measured stacks reading worse than an anti-collision model predicts, isolating mutual detuning between neighbouring tags as the cause. Their work was 13.56 MHz HF — the mechanism transfers, the numbers do not.
The numbers, measured on food rather than on water
Water sits near a relative permittivity (ε′) of 78.2 at 25 °C (Hasted 1972); RFID Journal’s glossary puts frozen water near 3.2. Food is neither. Wang, Wig, Tang and Hallberg measured real products at 915 MHz (J. Food Engineering 57, 2003) — the US band rather than India’s, and these properties shift with frequency, so read the table as the regime you are in, not as India-band values.
| Material (20 °C, 915 MHz) | ε′ | Loss factor ε″ | Power penetration depth |
|---|---|---|---|
| Liquid whey protein mixture | 61.7 | 33.6 | 12.6 mm |
| Whey protein gel | 59.0 | 34.8 | 12.0 mm |
| Cheese sauce (salted) | 42.9 | 46.2 | 8.2 mm |
| Cooked macaroni (56% moisture) | 29.2 | 4.2 | 67.3 mm |
Penetration depth is where power falls to 1/e, about 36.8%. In salted cheese sauce that is 8.2 mm — roughly two-thirds of the power entering the product is gone within a centimetre of the surface. A loaded tray is far thicker than that, so the position worth testing is the one where the field must cross product rather than reach the tag around it.
The salt subtlety. Cheese sauce is less permittive than the whey gel (42.9 against 59.0) yet more lossy (46.2 against 34.8): loss factor splits into a dipole-rotation and an ionic-conduction term, and cheese sauce is salted. Brined, cured or marinated meat is therefore RF-harder than lean meat at the same water content — moisture and ionic content are two variables, not one.
Dielectrically, frozen is the easier case. At ε′ near 3.2, a hard-frozen pack is closer to dry plastic than to water. Walmart, announcing work with Avery Dennison on 22 October 2025, framed the challenge as “high-moisture, cold environments, like meat cases” — chilled, not frozen. If you assumed the freezer was the hard problem, reverse it — then test the transition anyway: a pack out of cold storage carries a condensation film over the label.
What actually works, in order of leverage
Antenna geometry. Because loading pulls resonance downwards, a tag meant for wet product can be resonated deliberately high in free air and brought onto band by the load. A physically small antenna is also less sensitive to its surroundings, at the cost of gain. The six Avery Dennison IdentiFresh inlays announced on 25 February 2026, during EuroShop, on Impinj M830 silicon are all small — published sizes from 28 × 28 mm up to 68 × 26.4 mm — though Avery Dennison publishes neither its tuning approach nor any performance data.
Standoff and ground plane, from the on-metal toolkit. Space the radiating element off the lossy medium, or put a thin ground plane behind it so the antenna radiates away from the food. The costs are the ones on-metal tags always carry: thickness, rigidity, price, and a label that may not run through your scale-printer.
Self-tuning silicon — real, and bounded. Impinj’s AutoTune trims the chip’s own input capacitance at power-up across ±100 fF in 40 fF steps against that ~0.93 pF baseline: roughly ±11%, enabled by default and disableable by the reader. It widens the window in which one inlay works across different materials (the M830 is specified over 860–960 MHz), but it cannot rescue an antenna pushed far off resonance, and does nothing about absorption.
Link budget on the reader side. Gen2X, released December 2024 and extended to further Impinj parts on 12 November 2025, improves the tag-to-reader waveform; Impinj’s datasheet claims up to 7.5 dB of Eb/N0 gain, modelled from bit error probability rather than measured. It needs support at both ends, reader IC and tag IC — our own fixed readers are R2000-generation, so we do not offer it. A few decibels decides a marginal read, but antenna choice, polarisation and geometry buy more than tag heroics.
One buying trap: on IdentiFresh parts the MR suffix means microwave resistance, not moisture resistance — Avery Dennison’s own AD Saucer page says so.
The India point no global datasheet makes
India’s licence-exempt allocation is 865–868 MHz, set by G.S.R. 853(E) of 10 December 2021 — the Use of Low Power Equipment in the Frequency Band 865–868 MHz for Short Range Devices (Exemption from Licence) Rules, 2021, made by the WPC Wing expressly in supersession of the 2005 RFID rules that had set 865–867 MHz. Two details from Table IV of those rules matter on a wet product. Reader transmissions at the full 2 W e.r.p. are permitted only inside four 200 kHz channels, centred at 865.7, 866.3, 866.9 and 867.5 MHz — so you have four channels to hop across, not a continuous 3 MHz to spread over, and in a chilled cabinet full of absorbing product that channel budget is part of your read-rate maths. The tag’s reply is specified at around −20 dBm e.r.p., which is the number worth holding on to here: the return signal is tiny before water ever enters the picture. Note also that the 867.5 MHz channel sits outside the old 865–867 window, so the 2021 rules genuinely widened the band rather than relabelling it — a detail most Indian RFID pages, including ours until recently, still had wrong. ETSI uses the same 865–868 MHz range and the FCC roughly 902–928 MHz. Allocations are revised, so confirm the current position with WPC before you design around it, and see our BIS and WPC guide
Now put that beside the direction of detuning. A tag qualified on meat in an American store was qualified in a band starting 35 MHz above India’s top edge, with 26 MHz of channels to hop across; here it has to work inside a 2 MHz slice, with no second channel to rescue a bad one, and the loaded resonance sits where the product puts it. That result does not transfer — a checkable reason to insist on an India-band test at legal power rather than a vendor’s US trial data.
Read rates and ranges you can defend
Claucherty and colleagues measured the collapse cleanly at 915 MHz (Foods, 2024, open access). One tag read 16.34 m on an empty PET bottle at −20.78 dBm sensitivity; on a water-filled bottle the same tag needed +8.37 dBm — about 29 dB worse, which on the usual r ∝ 10^(dB/20) relation is roughly a 28-fold range reduction. A different, liquid-tolerant design in the same study measured 1.76 m empty, 1.58 m on deionised water and 1.88 m on apple juice.
The long-range design loses almost everything; the tolerant one starts low and barely moves. Choose the second, and plan read points inside the 0.5–2 m region that measured data actually supports rather than around the free-air ceilings vendors quote for apparel tags. None of this is a guaranteed range: on wet product it is environment-dependent by construction, so a specification should name the distance, orientation and case load it applies to.
Read rate is a different quantity, and there is one independent anchor worth holding a supplier to. ECR Retail Loss reported on 29 June 2024 that two grocery retailers ran fresh-meat trials and both demonstrated read rates of 95% or better, against a prior assumption that the category was untestable — though the retailers are unnamed and the data sits behind a retailer-only recording, so treat it as an existence proof rather than a benchmark. Avery Dennison’s 99%-plus claim for IdentiFresh, by contrast, carries no methodology, tag population, read distance or third-party lab that we can find. Nor is there a standard to fall back on: as of 26 September 2026 Auburn’s ARC programme lists no specification covering food, produce or high-moisture goods, and its retired Spec M has no replacement. Your pilot is the certification.
Where to tag, by product class
| Product class | Sensible starting point | Prove in the pilot |
|---|---|---|
| Chilled meat, poultry, fish trays | Small liquid-tolerant inlay on lid film, clear of contact | Loaded case; worst tray position alone |
| Brined, cured, marinated meat | As above, with standoff or a moved label | The worst SKU, not the average |
| Fluid dairy, curd, paneer | Shoulder or lid, never the fluid face | Orientation sweep; crate-level fallback |
| Reusable crates and roll cages | On-metal or rugged sealed tag on the container | Read after wash cycles, stacked high |
Note the last row: tagging the container rather than the product sidesteps the hard RF problem — pallet and crate identity first, item-level later. Slaughterhouse lines work the same way.
What Identium can and cannot do here
We manufacture passive UHF inlays and labels, hard tags, on-metal, cable-tie and rugged sealed tags at our New Delhi factory, and supply readers and antennas that are WPC-approved for India’s licence-exempt UHF RFID band. Every reader ships with an SDK, demo app and source code, so you can instrument the pilot yourself. Custom tags typically turn around in 5–10 days once the design is frozen, and units are tested in-house.
What we do not have: no sensor or temperature-logging tags, and no dedicated fresh-food inlay range with published performance data on meat or dairy. If anyone offers you one, ask for the test methodology before the price. We are BIS certified and WPC-approved for India’s band; we do not hold CE, FCC, RoHS, REACH or ISO 9001, so if an export customer needs those marks, weigh it as a real gap and ask every supplier the same question.
The pilot is genuinely non-negotiable
No datasheet predicts your combination of water content, salt, pack geometry, stacking depth, shelf metal and antenna position inside a 2 MHz band, and nobody honest will quote a read range sight unseen.
Four things separate a pilot that proves something from one that flatters the tag: three or more candidate inlays, one deliberately liquid-tolerant; India-band readers at the power that allocation permits, not an imported 902–928 MHz setup; read rate logged per tag position rather than averaged; and a condensation test on stock coming out of chilled storage. Our guide to scoping read range and environment covers the surrounding discipline.
This is one of the harder passive UHF applications there is, and anyone telling you it is solved is selling something. Tell us the product, its packaging, the stacking depth and where the reader has to sit, and we will scope a test around it — including when the honest answer is that item-level tagging is not yet worth it and the crate is.
Related reading
- On-metal RFID tags — the standoff and ground-plane techniques fresh-food tags borrow.
- RFID cold chain temperature monitoring — thermal history, and where sensing happens.
- RFID tag not reading: troubleshooting — the diagnostic sequence when a pilot underperforms.