Correct is not enough.
Packets must be punctual.
Sync says what time it is; delay management says when each packet must arrive. The O-RU has a tiny reception window — outside it, correct samples are dropped. Here we label the six T-parameters (T1a · T12 · T2a downlink, Ta3 · T34 · Ta4 uplink), compute a window, read the early/on-time/late counters, pair O-DU/O-RU latency categories, and meet CTI for PON. Verbatim from O-RAN WG4 CUS.0 v21 §4.4–4.6 + Annex B/L.
Transport & delay — the T-windows that keep packets punctual
The player screen is a live animation stage — the downlink timeline builds arrow by arrow (T1a → T12 → T2a), the packet lands inside or outside the reception window, the counters tick early/on-time/late, and latency categories pair up, exactly as the narration reaches each idea. Karaoke subtitles EN + Persian, fullscreen.
Will this packet land in time?
The rule: T1a − T12 must land inside the O-RU reception window (T2a). Grow the fibre or add switches and the one-way delay T12 rises — until the packet arrives too late and is dropped.
budget = O-RU window depth (this category)
Downlink and uplink, six arrows
Learn these six and you can read any fronthaul delay diagram. Downlink flows O-DU → O-RU; uplink is the mirror, O-RU → O-DU.
| Param | Direction | What it is |
|---|---|---|
| T1a | DL | O-DU transmission window — how early the O-DU sends |
| T12 | DL | one-way fronthaul delay, O-DU → O-RU |
| T2a | DL | O-RU reception window — data must land inside it |
| Ta3 | UL | O-RU transmission window — how early the O-RU sends UL |
| T34 | UL | one-way fronthaul delay, O-RU → O-DU |
| Ta4 | UL | O-DU reception window — UL must land inside it |
Downlink closes when T1a − T12 lands inside T2a. You learn T12 two ways (§4.4.4): the defined transport method (configured from design) or the measured transport method (eCPRI one-way delay measurement, O-DU initiates). Measurement is truer because it includes the real switches.
The O-RU counts exactly this
| Counter | Meaning | Action |
|---|---|---|
| on time | received within the reception window | no action needed |
| early | arrived before the window opened | reduce T1a advance / shorten path |
| late | arrived after the window closed — dropped | too much delay — fix T12 |
Before you blame a flaky radio, read the early and late counters. Rising late = the path grew (new switch, longer fibre, congestion) and packets miss the window. Rising early = the O-DU advance is too large. Fronthaul engineers graph these counters like a heartbeat.
Shaping the window, and sharing the fibre
| Item | What it does | Clause |
|---|---|---|
| Normal | packets sent as scheduled in the window | §4.6.1 |
| Uniformly distributed | packets spread evenly across the window (smooths switch load) | §4.6.2 |
| Ordered | packets sent in a defined order | §4.6.3 |
| Scheduled | O-RU shifts/resizes its UL window (via M-plane or SE18) | §4.6.4 |
| CTI-TMP | coordinates transmission on shared transport (PON) — report/allocate cycle | WG4 CTI |
When delay exceeds the ideal, two things break first: the HARQ round trip stretches (acks return later) and RACH access becomes fragile. Annex L analyses exactly how they degrade. On shared media like PON, CTI (Cooperative Transport) runs a report-and-allocate cycle so devices do not collide on the fibre.
🛠 Engineer detail block — what you'll actually meet in the field
- The rule is one line: T1a − T12 must land inside T2a. Everything else — fibre length, switch count, congestion — is just what pushes T12 until that stops being true.
- Measure T12, don't assume it. The eCPRI measured-transport method reflects the real switches; a "defined" value from a design doc lies the moment someone adds a hop or the network congests.
- Late counter first, radio last. A cell losing throughput with a climbing reception-window late counter is a transport problem, not an RF one. Swapping the O-RU wastes a truck roll.
- Category pairing is a purchase gate. Two boxes that each pass conformance can still fail together if a strict O-RU sits behind long fibre from a tolerant O-DU. Check Annex-B pairing before you buy.
- PON needs CTI. Open Fronthaul over shared access (PON/DOCSIS) does not work by dropping packets on a switch — it needs CTI-TMP to grant bursts, or the timing collapses under contention.