UHF RFID Anti-Collision and Multi-Tag Read Rates: Q, Sessions, Antennas

Reading one UHF tag is easy. Bulk reading, such as all 200 tagged cartons on a pallet as it passes a dock door or every item on a shelf during a stock count, is where RFID projects succeed or fail. Missed tags almost always have one of three causes: some tags never get enough energy, some lose out in anti-collision, or the host cannot keep up with the reader.

This guide explains how EPC Gen2 anti-collision works, which settings change the result, how the physical layout limits it, and how to run a pilot test that shows which problem you have. It applies to any Gen2 (ISO/IEC 18000-63) reader.

Read rate and read speed are different numbers

Agree on what you are measuring before you change any setting.

A dock door needs every tag within the few seconds the pallet is in the zone; a shelf count needs every tag eventually. Reads per second is the least useful figure, because a reader can report the same strong tags hundreds of times while weak tags stay unread.

Single-tag figures do not predict multi-tag performance either. The U610-M 7 dBi integrated UHF reader lists up to 140 single-tag reads per second, but none of our specifications publishes a multi-tag rate, because it depends on the tags, their layout and the site.

How Gen2 anti-collision works

Gen2 readers separate tags with a slotted random-access scheme, defined in the GS1 EPC UHF Gen2 air interface standard:

1. The reader sends a Query with a Q value from 0 to 15. Each tag picks a random slot number from 0 to 2^Q − 1.
2. Tags whose slot number is 0 reply with a 16-bit random number. If exactly one tag replies, the reader acknowledges it and the tag sends its PC, EPC and CRC.
3. The tag that was read flips its inventoried flag for the current session (from A to B when the reader targets A) and ignores the rest of the round.
4. The reader sends QueryRep to move to the next slot, and every waiting tag counts down by one. QueryAdjust raises or lowers Q by one mid-round when there are too many collisions or too many empty slots.

Each slot ends empty, with a single reply (a read), or with a collision, where two or more tags answer at once and try again later. Efficiency peaks when the number of slots roughly equals the number of unread tags, and even then only about 37% of slots (1/e) hold a single reply. Readers therefore run a dynamic Q algorithm that adjusts Q as tags are read; many start at Q = 4. In a collision the reader sometimes decodes the stronger reply anyway, so weak tags lose out, which is one reason the last few tags take longest.

A Q that is too low wastes time on collisions; one that is too high wastes it on empty slots. Dynamic Q corrects either, but a fixed Q close to a known population saves time at the start of each round.

Sessions and targets decide which tags answer again

Each tag keeps four independent inventoried flags, one per session (S0 to S3). The reader asks for tags whose flag matches its target, A or B. A tag that has been read flips its flag and stays silent until the flag returns to A. How long that takes depends on the session:

The target setting works with the session:

• Single target (A only): each tag is reported once and then keeps quiet, so weaker tags get the free slots. Use it for counts.
• Dual target (A, then B, then A): tags are reported again and again. That suits presence monitoring but hurts dense counts, because strong tags crowd every round.

A handheld that keeps reporting the tags nearest the antenna while tags at the back of the shelf never appear is usually running S0 or dual target; switch to S2 or S3 with single target. The S1 window varies between chip models, so test portal timing with the tags you will deploy. When two readers cover the same tags, put them in different sessions, such as S2 and S3, so one reader does not silence tags for the other.

Before the Query, a reader can also send a Select command with a mask, such as your own EPC prefix, so only matching tags take part. That shortens every round; how to write UHF RFID tags explains how to plan the prefix.

Physical factors come first

No setting can read a tag that never receives enough energy. Check these before you tune the protocol:

A circularly polarized antenna reads tags at any rotation in the antenna plane, but delivers about 3 dB less to a linear tag than a matched linear antenna. More RF power is not a cure: beyond the level that energizes the tags you want, it mostly adds stray reads from outside the zone, and regulations cap radiated power including antenna gain. See UHF RFID read range and frequency bands by country.

Reader settings, in the order to tune them

1. RF power. Start low and raise it until the whole zone reads.
2. Session and target. Large static counts: S2 or S3, single target. Portals: S1 or S2, single target. Presence monitoring: S0, dual target.
3. Q. Start near the value for your expected tag count and leave dynamic Q on unless the population is fixed.
4. Link profile. Tags reply at a backscatter link frequency of 40–640 kHz, with FM0 or Miller 2, 4 or 8 encoding, giving 5 to 640 kbit/s. FM0 at a high link frequency is fastest in a quiet RF environment. Miller 4 or 8 is slower but more robust, and it is the basis of dense reader mode where many readers share a site.
5. Select filters on your EPC prefix.
6. Trigger and dwell. Read only while the pallet, tote or vehicle is in the zone, started by a sensor.
7. Reporting. Filter duplicates and check that the host link keeps up. A 9600 bit/s serial link carries about 960 bytes per second. Sent as 24 hex characters plus CR and LF, each 96-bit EPC takes 26 bytes, so the link tops out near 37 tags per second. Keyboard output and Wiegand deliver one tag at a time and are not bulk channels.

Our integrated long-range readers list RF output of 0–30 dBm set in software, continuous or triggered reading with one trigger input, and an SDK with C#, VC, VB, Java and Delphi sample code. Their specifications do not document session, target or Q settings; which inventory parameters the SDK exposes is confirmed in your quotation.

Run a pilot test that finds the real problem

1. Build the test set from real items, real tags and real packing, such as one full pallet. Record every EPC, or encode sequential EPCs so a missing tag is obvious.
2. Include worst cases: tags in the middle of the pallet, facing away from the antenna, or next to liquids and metal.
3. Mount the reader at its final position and move items at the real speed.
4. Run at least 10 passes per setting. Record unique tags, time to last tag and the missing EPCs.
5. Change one setting at a time, and set the pass criterion before you start.

Choosing hardware for multi-tag reading

• Fixed read points. The U610-M (7 dBi, 0–5 m) keeps the zone short at one door or lane; the U630-M 9 dBi integrated UHF reader (0–6 m) reaches further along the approach, with a narrower beam than a lower-gain antenna. Each has one built-in antenna, so they suit points where tags face the reader, such as lanes, doors and conveyors with consistent tag placement. Portals that must read tags facing every direction usually need a multi-port reader with several antennas, which our range does not include.
• Handhelds. The U550-M Bluetooth UHF handheld reads at 0–3 m and sends each ID as keyboard input to a phone over Bluetooth, or to a PC through its 2.4G receiver or USB; no SDK is documented. The U510-B (0–50 cm) and U520-M (0–35 cm) suit closer checks. The U550-M and U510-B list a 0.5 s read interval and none of the three states a multi-tag rate, so they fit item checks and lookups; test samples before planning dense counts. See handheld UHF and Bluetooth readers.
• Desktop reader/writers read up to 0.5 m by design. Use one to encode the test set and to check tags the pilot keeps missing.

The fixed vs integrated vs handheld guide compares the formats, and inventory and asset tracking shows where each fits.

Multi-tag read checklist

• KPI defined: read rate, time to last tag, or both.
• One tag model, tested on real items and placed in the same spot on each.
• Antenna height, angle and distance fixed; metal shadows avoided.
• RF power at the lowest level that reads the whole zone.
• Session, target and starting Q matched to the job.
• Dense reader mode where several readers share a site.
• Select filter on your EPC prefix; host link capacity checked.
• Pilot of 10 or more passes, with missing EPCs analyzed.

To test with your own tags and layout, request an evaluation unit on the samples page.