UHF RFID Deep Dive: From Reader Carrier to ERP Record

UHF RFID is often described as "radio barcodes", which is accurate about the use case and misleading about the engineering. A passive UHF system is a bidirectional radio link with a power budget measured in microwatts, an arbitration protocol that must resolve hundreds of simultaneous talkers, and a physical layer that changes behaviour the moment a tag touches metal.

Understanding it at that level is what separates a deployment that reaches 99 % read rate from one that stalls at 90 % and blames the hardware.

What happens in a single read cycle?

The reader never stops transmitting. In a passive UHF system the carrier is continuous, and the protocol is built on top of it.

The reader sends a Select command to filter the tag population, then a Query to start an inventory round and announce which session it is using. Tags respond with a 16-bit random number (RN16). The reader acknowledges that specific tag, the tag backscatters its EPC, the reader acknowledges again, and may optionally read TID or user memory.

• Select: filters by EPC mask or flag. Done once per round, not per tag.
• Query: opens a slot, defines the session and the Q parameter.
• RN16: the tag's temporary handle; the reader uses it to avoid collisions.
• EPC read: 96 bits typical, up to 496 bits, transmitted at 40–640 kbit/s.
• Access: optional read or write of TID and user memory, password protected if locked.

The whole exchange for one tag takes roughly 2–6 milliseconds at typical settings, which is why a well-tuned reader reports 700–1,200 tags per second on the air interface.

Where does the power actually go?

A passive tag has no battery. It runs on the fraction of the reader's carrier that reaches its antenna and survives the rectifier.

Start with reader power. In Europe the limit is 2 W ERP (about 3.3 W EIRP); in the United States 4 W EIRP; in China 2 W EIRP in the 920–925 MHz band. That power spreads over a sphere, so free-space path loss at 3 metres on 915 MHz is about 41 dB before any real-world loss.

• Cable and connector loss: 1–3 dB, more if the run is long or the connector is cheap.
• Antenna gain: 6–9 dBi typical for a fixed circular antenna.
• Tag antenna gain: 1.5–2.2 dBi for a dipole inlay.
• Polarization mismatch: 3 dB on average, 20 dB or more if orthogonal.
• Tag efficiency: the rectifier converts perhaps 20–40 % of received RF into DC.

The result is that a tag at 6 metres receives tens of microwatts and must wake at roughly 15–30 microwatts. There is no margin to waste, which is why a 1 dB improvement in the antenna or a 2 dB loss in a bad connector changes the read zone visibly.

How does Gen2 arbitrate hundreds of tags?

Pure random access would collapse. Gen2 uses slotted ALOHA with an adaptive frame size: the reader announces a Q value, tags pick one of 2^Q slots, and only the one in a given slot replies. Q=4 gives 16 slots; Q is raised when collisions dominate and lowered when slots go empty.

Q is not the only lever. Sessions matter just as much. Gen2 defines four sessions, and a tag maintains an independent inventoried flag in each. Reading in session 2 leaves session 0 untouched, so a portal can read a pallet while a handheld, using a different session, still sees those tags as new.

• Session 0: typical for handheld cycle counts; the flag resets quickly.
• Session 1: longer persistence, useful for conveyor reads.
• Session 2: the standard choice for portals and overhead readers.
• Session 3: rarely used, long persistence.

Getting sessions wrong is one of the most common causes of "the reader only sees the tags once" complaints. The tags are fine; the session flag says they have already been counted.

Why do dense-reader mode and frequency hopping matter?

In a warehouse with 40 readers, interference is not a hypothetical. Dense-reader mode (DRM) narrows the tag's backscattered signal so adjacent channels interfere less, at the cost of roughly 3 dB of tag sensitivity. If your tags are marginal, DRM makes them worse; if your readers are crowded, DRM makes the site work.

Frequency hopping is regulatory as much as technical. In the EU, readers must hop across 865–868 MHz with a defined dwell time; in the US, across 902–928 MHz. A reader that transmits on a fixed channel is either illegal or interfering, and both show up as mysterious read-rate drops that move around the site.

• Plan channel reuse between adjacent portals; do not stack them on the same channel.
• Use DRM where reader density is high and tag sensitivity allows.
• Log reader transmit power and channel per read event — it explains most intermittent failures.

What limits real-world performance?

Four physical effects account for nearly all shortfalls, and none of them appears in a datasheet.

Multipath creates nulls: a tag can sit in a spot where two reflected paths cancel, and no amount of reader power fixes it. That is why antennas are mounted at different heights and angles, and why read zones are measured, not assumed.

Tag detuning from nearby material shifts the resonant frequency, sometimes outside the band entirely. A tag specified for 915 MHz on a plastic tote may resonate at 890 MHz once it is on a metal shelf.

Mutual coupling: tags packed tightly shadow each other, and the ones in the middle of a pallet are read last or not at all. And cable and connector degradation is progressive — a site that worked in month one and drifts by month six is usually a connector, not firmware.

• Measure read rate per antenna, not per reader.
• Test at the worst orientation, at the far edge of the zone, with a full pallet.
• Compare month-one and month-six baselines to catch degradation.

Where does UHF sit against barcode, BLE and UWB?

UHF is the only technology that reads hundreds of passive identifiers per second at several metres for a few cents per tag. That combination is why it owns inventory, apparel and logistics.

Barcode is cheaper per label and needs no RF design, but it is line-of-sight, one at a time, and labour-bound. BLE gives you location and battery-powered sensors, but at dollars per tag and with no bulk identification. UWB gives centimetre accuracy, but at a hardware cost that only makes sense for high-value assets and people.

• Bulk identification at metres: UHF.
• Exact position to 10–30 cm: UWB.
• Temperature or motion data over months: BLE.
• Zero-cost label on a carton: barcode.

Bottom line

The UHF read cycle is a short, tightly budgeted conversation: the reader keeps talking, the tag answers with a random number, and the whole system depends on enough microjoules arriving at the right moment. Design for the link budget, get the sessions right, plan channels between readers, and measure per antenna. Everything else is tuning.