Search for SGTIN++ EPC Tag Data Standard 2.3 and you will find two kinds of result: the 448-page standard itself, or nothing much. Trade commentary is thin, which is unusual for a change that lets a brand write its own domain name into a passive UHF tag.
Here is what GS1 actually did, what it does not do, and why it reaches you as a tag-selection question before it is ever a marketing one.
What TDS 2.3 added
GS1’s EPC Tag Data Standard specifies what goes into the memory of a Gen2 tag — the EPC itself, and how it maps to GS1 keys. Release 2.3 is marked Ratified, Oct 2025 on its title page, and the document served at ref.gs1.org/standards/tds/ was still Release 2.3 when we checked it on 3 October 2026.
Its change log describes the addition in one paragraph:
“Introduction of twelve new ’++’ EPC schemes, SGTIN++ and DSGTIN++ that are modelled on existing schemes SGTIN+ and DSGTIN+ but extended to support binary encoding of a custom domain name after the serial number, to better support translation back to non-canonical GS1 Digital Link URIs.”
Twelve is the full count. Alongside sgtin++ and dsgtin++, the release adds sscc++, sgln++, grai++, giai++, gsrn++, gsrnp++, gdti++, cpi++, sgcn++ and itip++. Every one of them is listed as variable-length.
Unpacked: a conventional SGTIN on a tag is a number. To turn it into a web address, something has to map that number to a URL, and the canonical form uses GS1’s own resolver at id.gs1.org. A ’++’ scheme writes your hostname into the tag’s binary EPC, so the binary translates losslessly back to a URL on your own domain. The SGTIN++ coding table gives the target syntax as https://{hostname}/01/{gtin}/21/{serial}, under the header value 11111101. Only https:// is supported, and it is not stored — the standard says it is implied and automatically reinstated on decoding.
Be clear about this before you spend anything
A ’++’ scheme is not what gets your customers onto your own domain. The standard says so itself: for schemes outside the ’++’ set — SGTIN-96, SGTIN-198, SGTIN+ and the rest — it remains possible to construct an equivalent GS1 Digital Link URI using “a hostname that is preferred/recommended within a specific jurisdiction”, or id.gs1.org for the canonical form. Your own domain in front of a shopper is a resolver and middleware decision. A plain 96-bit tag does not stop you.
What ’++’ adds is narrower: the hostname travels in the tag, so binary-to-URI translation is lossless and self-contained, with no external mapping that software has to be told about. That is a real engineering property, and for some programmes it is worth having. It is not the same claim as “you need this for branded URLs” — and anyone selling it to you that way should be corrected.
It is also an encoding standard, not a mandate. It describes how to write something if you choose to. Nothing in it requires anybody to do anything, and as of October 2026 we are not aware of any regulation or retailer programme that calls for a ’++’ scheme. The adjacent digital-passport space contains a good deal of blurring on exactly this point, so be direct with anyone who implies otherwise: ask for the clause and the document reference. Our Digital Product Passport guide covers the carrier question separately, and the honest position there is that it is unsettled — nothing in TDS 2.3 settles it.
For a sense of the installed base any new encoding practice has to move: the RAIN Alliance reported 42.7 billion RAIN tag chips shipped in 2025, in figures published on 2 April 2026 and compiled from member chipmakers EM Microelectronic, Impinj, NXP and Shanghai Quanray. The Alliance reported that as a fall against the year before. Treat the year-on-year gap carefully, though: the 2024 total the Alliance published in February 2025 was compiled from five reporting members, including Shanghai Fudan Microelectronics Group, so the two headline figures are not drawn from the same panel. Our chip shipments briefing sets out what that number does and does not support. Either way, annual volumes at that order of magnitude tell you encoding practice changes slowly.
Why a converter cares: the bit budget
A longer URI consumes EPC memory. That makes this a question about which memory bank, how many bits, and what headroom is left for your serial — a tag and IC selection question, answered long before anyone designs a landing page.
The SGTIN++ coding table sets out the segments. These are the standard’s own figures:
| Segment | Bits | Notes |
|---|---|---|
| EPC header | 8 | Value 11111101 for SGTIN++ |
| +AIDC data toggle | 1 | — |
| Filter | 3 | Shared with SGTIN and DSGTIN+/DSGTIN++ |
| GTIN | 56 | 14 digits, fixed-length numeric |
| Serial number | 8 + up to 140 | 3-bit encoding indicator + 5-bit length indicator + data; up to 20 characters |
| Hostname | 7 + up to 240 | 1-bit encoding indicator + 6-bit length indicator + data |
Add the fixed rows and 68 bits are spent before a single character of serial or hostname. Add both fields’ indicators and it is 90. That figure alone settles one of the buying questions below.
Fill both variable fields to their stated bit maxima and the total reaches 463 bits. The standard footnotes a practical ceiling of 464 bits on tags that support XPC_W1, or 496 bits on tags that do not, because the Length field in the PC bits counts the EPC — plus any appended AIDC data — in 16-bit words. SGTIN++ therefore runs right up against that ceiling. GS1 does not say the two figures were matched deliberately, so read that as an observation rather than a statement of intent.
The word-counting has a second, smaller consequence: because the EPC field is measured in whole 16-bit words, a 125-bit EPC occupies eight words — 128 bits of bank. Bit totals round up.
One caution on the character counts. The coding table gives the hostname as “up to 63 characters” and up to 240 bits. Those are two separate limits, not two ways of saying the same thing, and the bit limit binds first. At 16 bits per three characters, 240 bits holds 45 characters of URN Code 40; at 7 bits per character it holds an optimised length of 34. The 63-character figure is the validity test the standard applies to the input string, not a length you can actually encode.
The comparison that matters: SGTIN-96 spends 96 bits in total. A ’++’ scheme passes that before it has encoded a single character of serial number.
What your domain name costs
The hostname field permits two character encodings, and which one is cheaper depends on your actual domain — so this is arithmetic you can do on the back of an envelope today.
URN Code 40 packs three characters into 16 bits, roughly 5.33 bits per character. The standard’s validity test limits it to digits, uppercase A–Z, dot, colon and hyphen; any lowercase letter or other symbol in the input forces 7-bit encoding instead. Because the host component of a URL is case-insensitive, upper-casing a hostname before encoding is lossless in practice, which is what makes Code 40 usable for an ordinary lowercase domain — but it is a step, not an automatic property. Note too that a final group of one or two characters is padded out to three, so Code 40 bills in whole groups.
7-bit truncated ASCII spends 7 bits per character but carries optimisation tables that compress frequent sequences. Per those tables, .com, .org, .net, .int, .edu, .gov, .mil, .biz, .eco and .med each collapse to a single 7-bit value, as do the subdomains id., www. and qr.; compound country endings including .co.in and .com.in collapse to 14-bit values.
What the field stores in the optimised case is not the raw character count. Per the standard’s own rule, the optimised length is the leftover string after removing the optimised sequences, plus one for each 7-bit sequence and two for each 14-bit sequence; the field then costs 7 bits times that. So you can reproduce the figures below for your own domain.
Totals include the 7 bits of indicators:
| Hostname | Chars | URN Code 40 | 7-bit ASCII | 7-bit, optimised |
|---|---|---|---|---|
brand.com | 9 | 55 | 70 | 49 |
brand.co.in | 11 | 71 | 84 | 56 |
id.brand.com | 12 | 71 | 91 | 56 |
tags.acmeindustries.co.in | 25 | 151 | 182 | 154 |
Two decisions fall straight out of that table. A short domain with a recognised ending is dramatically cheaper — and the optimisation tables cover .co.in and .com.in, so an Indian brand is not penalised for using its national domain. But once the memorable part of the hostname grows long, the optimisations stop paying and plain Code 40 wins. If a ’++’ scheme is genuinely on your roadmap, the cheapest intervention available to you is choosing a shorter hostname, and it is free if you make that choice before the domain is printed on packaging.
So a realistic floor: brand.com, optimised, is 68 + 8 + 49 = 125 bits before the serial digits — eight 16-bit words, so 128 bits of bank. Your serial is then the variable you control, and a long alphanumeric serial is expensive on top of that.
What this means when you buy tags
- A 96-bit EPC bank is not in the running. The fixed segments and the two length indicators consume 90 bits between them, before one character of serial or hostname is written. That is arithmetic, not opinion.
- Ask for the EPC bank size on the exact part, with its variant suffix — not the chip family. Our RFID chip comparison sets out why the variant is the only thing a supplier can be held to, and what else belongs in the memory section of an RFQ.
- Test the read side. Middleware and reader firmware written around a fixed 96-bit EPC do not always handle a long variable-length one gracefully. Nor does every encoder: if you encode in-house, your label printer and its firmware have to write that length.
- Budget air time. More bits per tag means more bits on air per read. In a dense bulk count that shows up as a slower cycle, though by how much is environment-dependent and only a trial on your own goods will tell you.
- Keep it on passive UHF. ’++’ schemes are EPC Gen2 — ISO/IEC 18000-63, RAIN RFID — and apply to passive UHF only. They have no bearing on HF or NFC at 13.56 MHz, or on LF animal tags at 125/134 kHz. Our RFID glossary keeps those families straight.
Encoding is also independent of radio. What you write into memory has nothing to do with which band the tag works in: India’s licence-exempt UHF RFID band is 865–868 MHz under G.S.R. 853(E) of 10 December 2021, the US sits at roughly 902–928 MHz, and that remains true whichever EPC scheme you choose. The frequency allocation page has the detail.
What we can and cannot tell you yet
We manufacture passive UHF inlays and labels, hard tags and on-metal constructions in New Delhi, and units are tested in-house before dispatch.
What we are not going to tell you is that our tags support ’++’ encoding. Nobody here has yet gone through the EPC bank sizes on the specific IC variants we quote against the TDS 2.3 bit budget above, and until someone does, any answer would be a guess dressed as a specification. The arithmetic on this page is from the standard and holds for any supplier’s tags; the part-specific answer comes with the quotation, where it can be held to. If you need it before then, say so in the enquiry and we will check the datasheets rather than reply from memory.
Two further honest limits. We do not issue GS1 company prefixes or operate resolvers — your GS1 member organisation and your own web infrastructure handle those, and the tag is only one end of it. And a branded Digital Link programme is largely a decision about resolver hosting and web analytics, which is not what we sell; we convert and encode the carrier.
Who should care in 2026
Almost nobody, yet — and that is the useful conclusion. If you are tagging assets, tools, files, linen or livestock, the identifier is a database key and a branded URL buys you nothing; a 96-bit EPC remains the right answer, as it is for most projects.
It is worth a look if you are a consumer brand already running serialised item-level tagging, you specifically want the branded hostname carried in the tag itself rather than applied by your resolver or middleware, and you are specifying new tags anyway so the EPC bank size is still an open decision. In that narrow case, the time to do the bit arithmetic is now, during tag selection — not after ten million labels are printed.
And if a scheme or standards version appears in a tender or a quotation, hold it to a document reference. The discipline is the same one we set out for Gen2X and Gen2v3 claims: a claim against a standard nobody has checked is not a claim you can enforce. Tell us the scheme your buyer has specified, in writing, via our RFID manufacturer and exporter page, and we will come back with what the parts can actually hold.
Related reading
- RFID chip comparison: Impinj vs NXP — EPC, TID and user memory, and how to pin a supplier to an exact IC variant.
- RFID glossary of terms and acronyms — EPC, Gen2, RAIN and the frequency families, defined plainly.
- Gen2X and Gen2v3 protocol guide — how to test a standards-version claim before it reaches a contract.