Scheduling & HARQ Timing (n+4) in LTE 4G

Introduction

Time-domain scheduling in LTE is the set of rules that decide when each piece of a transaction happens: when the data follows its grant, when the acknowledgement follows the data, and when a retransmission follows a NACK. Unlike the frequency-domain allocation that names which resource blocks a UE gets, the timing is almost never signalled — it is baked into the standard. These relationships live in TS 36.213 (procedures) and TS 36.211 (physical channels).

You lean on this timing on every scheduled transmission a connected UE makes. The moment the eNB sends a downlink assignment or an uplink grant, both sides already know — without any extra field — the exact subframe where the data, the acknowledgement, and any retransmission will land. On FDD that knowledge is a single constant, n+4; on TDD it becomes a lookup keyed by the UL/DL configuration.

It matters because fixed timing is what makes LTE's HARQ deterministic and cheap: the uplink loop closes on a fixed 8 ms round trip that fills exactly eight HARQ processes, and the UE can implement its timing with a counter rather than by decoding signalling. Reading LTE's rigidity against NR's flexibility is the clearest way to see why NR spends signalling bits on K0/K1/K2 and SLIV — and why LTE never had to.

Why LTE fixes the clock

Concretely, LTE was designed for a single, fixed numerology: 15 kHz subcarrier spacing, a 1 ms subframe of 14 (normal cyclic prefix) OFDM symbols, and a 10 ms radio frame. When there is only one timescale, you can hard-code the relationships between channels and never pay the signalling cost NR pays to stay flexible. Four subframes (4 ms) is the processing budget the UE and eNB were designed around — time to decode, prepare a response, and turn it around — so fixing it removes signalling and makes HARQ timing deterministic, which is the whole point of a synchronous HARQ design. NR reverses that trade because it must serve numerologies from 15 kHz to 120 kHz and use cases from broadband to ultra-low-latency, so it makes timing a signalled parameter instead of a constant.

The Big Idea: LTE Time-Domain Timing Is Fixed

If you have already met 5G NR, the first thing to unlearn is flexibility. NR lets the scheduler pick K0 (PDCCH-to-PDSCH delay), K1 (PDSCH-to-HARQ-ACK delay) and K2 (grant-to-PUSCH delay) per transmission, and even choose the exact start symbol and length inside a slot via the SLIV (Start and Length Indicator Value). LTE does none of that. For FDD the timing relationships are baked into the standard as constants, and the smallest unit anyone gets scheduled on is a whole 1 ms subframe — one TTI (Transmission Time Interval).

Why is a fixed clock even acceptable? Because LTE was designed for a single, fixed numerology: 15 kHz subcarrier spacing, a 1 ms subframe of 14 (normal cyclic prefix) OFDM symbols, and a 10 ms radio frame of ten such subframes. There is exactly one slot pair per subframe and no scalable numerology to accommodate. When there is only one timescale, you can hard-code the relationships between channels and never pay the signalling cost NR pays to stay flexible. NR reverses that trade because it must serve numerologies from 15 kHz to 120 kHz and use cases from broadband to ultra-low-latency, so it makes the timing a signalled parameter instead of a constant.

FDD Downlink HARQ Timing (n+4)

On the downlink the eNB already holds your data, so scheduling is direct. The eNB sends a DL assignment — DCI format 1/1A/2/2A etc. on the PDCCH, CRC-scrambled with your C-RNTI — in the control region of subframe n, and the PDSCH carrying the data sits in the same subframe n (this is the LTE equivalent of NR's K0 = 0, and it is fixed, not signalled). You decode it, and exactly four subframes later, in subframe n+4, you report the HARQ-ACK.

That feedback rides on PUCCH if you have no uplink data scheduled at n+4, or is multiplexed onto PUSCH if you happen to be transmitting user data at n+4. Either way the timing is the same fixed +4. Because the eNB can retransmit whenever it likes — it just sends a new assignment for the same HARQ process in some later subframe — downlink HARQ in LTE is asynchronous. The retransmission is not pinned to a fixed later subframe, so the DL DCI carries an explicit 3-bit HARQ process number plus the NDI (New Data Indicator) so the UE knows whether the transport block is new or a retransmission to soft-combine.

How many DL HARQ processes are there? For FDD the maximum is 8 processes, the same count as the uplink, but for a completely different reason: on the downlink 8 is a configured ceiling that gives the scheduler enough outstanding transport blocks to keep the pipe full across the 4 ms feedback delay, not a value forced by a fixed round trip. The scheduler is free to service them in any order.

FDD Uplink Timing and the 8 ms HARQ RTT

Uplink is where the fixed timing really shows its shape, because the eNB cannot see your buffer and must grant permission first. The chain on FDD is:

1. The eNB sends an UL grant — DCI format 0 on PDCCH — in subframe n. 2. You transmit the PUSCH four subframes later, in subframe n+4. 3. The eNB decodes it and returns its HARQ acknowledgement on the PHICH (Physical HARQ Indicator Channel) four subframes after that — at n+8 relative to the original grant, i.e. n+4 relative to your PUSCH. 4. A NACK on the PHICH triggers a retransmission, which lands another four subframes on, at n+8 relative to the first PUSCH.

Because uplink HARQ is synchronous, every step of this loop happens at a fixed time — the retransmission occupies the same HARQ process in a predictable subframe, so the process ID is derived from the subframe number and is not carried in the DCI. Walk the clock around: grant/PUSCH (subframe m), PHICH and possible retransmission (m+4 from the PUSCH), next PHICH (m+8)... the loop closes on an 8 ms round-trip time. Fit eight 1 ms subframes into that RTT and you get exactly 8 uplink HARQ processes running back to back, so a single UE can keep one transmission in flight in every subframe of the RTT window.

Retransmissions come in two flavours. A non-adaptive retransmission is triggered purely by a PHICH NACK and reuses the same resources and MCS as before — no new DCI needed, which is the cheapest possible retransmission signal. An adaptive retransmission is signalled by a fresh DCI format 0 with the NDI not toggled, letting the eNB move the retransmission to different resources or change the MCS (a new grant on the PDCCH overrides the PHICH). Both stay on the same synchronous 8 ms timeline; adaptivity changes the where and how, never the when.

One subtlety worth carrying: the PHICH is a tiny channel that carries essentially a single ACK/NACK bit per uplink transmission, and the specific PHICH resource a UE listens to is derived from the lowest resource block index of its PUSCH and its uplink DMRS cyclic shift — again, no explicit pointer, the mapping is implicit and deterministic. This implicit, timing-driven design is precisely what NR discarded: NR has no PHICH at all, because its uplink HARQ is asynchronous and every (re)transmission is scheduled explicitly by DCI.

The Complete FDD Timing Picture

It helps to see every fixed relationship at once. On FDD the entire scheduling loop — control, data, feedback, and retransmission — is a small set of constants. There are no tables to read and nothing to signal; you count subframes.

Read the figure as a loop, not a line. On the uplink, once the PHICH at n+8 says NACK, the next transmission of that process starts at n+8 and its PHICH lands at n+16 — the process reappears every 8 ms. Eight processes offset by 1 ms each tile the timeline with no gaps. That perfect tiling is the direct arithmetic consequence of a 4 ms one-way processing budget applied symmetrically in both directions.

Why There Is No SLIV in LTE

NR's SLIV packs a start symbol and a length into one field so the scheduler can hand a UE a fraction of a slot — a few OFDM symbols — for low latency. LTE simply does not work at that granularity in the time domain. The unit of allocation is the subframe: when the scheduler picks you, you get the data region of a whole 1 ms subframe, all 12 or 14 symbols of it minus the control region.

The only "start symbol" question in LTE is answered globally, not per UE. The first 1–3 OFDM symbols (up to 4 in a narrow bandwidth of ≤ 10 RBs) of every subframe form the control region that carries PDCCH, PCFICH and PHICH. The number of symbols the control region actually uses is broadcast to everyone in the cell by the PCFICH (Physical Control Format Indicator Channel), which carries the 2-bit CFI (Control Format Indicator). Your PDSCH begins right after that control region and runs to the end of the subframe. There is nothing per-UE to encode as a start-symbol-plus-length, because the start is fixed cell-wide by the PCFICH and the length is always "the rest of the subframe."

Contrast this with NR, where each UE's assignment carries its own SLIV and two UEs in the same slot can legitimately start at different symbols and occupy different numbers of symbols. NR added that machinery precisely to schedule at symbol granularity for lower latency and for flexible coexistence of different traffic types in one slot — a need LTE's single-numerology, whole-subframe model never had.

Shortened TTI (sTTI): The One Crack in the 1 ms Model

The whole-subframe TTI is comfortable but slow: even in the best case, the fixed 4 ms feedback delays put a floor under LTE's user-plane latency. Rel-15 attacked this with sTTI (shortened Transmission Time Interval), the one feature that breaks LTE's rigid subframe granularity in the time domain.

Instead of a 1 ms subframe, sTTI schedules on much shorter units — a slot sTTI of 7 symbols (0.5 ms) or a subslot sTTI of 2–3 symbols (roughly 0.14 ms) — carried by shortened physical channels sPDCCH, sPDSCH and sPUSCH. Crucially, sTTI also shortens the processing timeline: the n+4 relationship is replaced by much tighter, capability-dependent turnarounds so HARQ feedback and grants follow the data within a fraction of a subframe. The result is dramatically lower round-trip latency for delay-sensitive traffic, at the cost of extra control overhead — LTE reaching toward the sub-slot scheduling NR made native from day one.

TDD: Same Idea, Table-Driven

On TDD, downlink and uplink share one carrier and take turns, following one of seven UL/DL configurations (patterns of D, U and special S subframes across the 10 ms frame). A plain n+4 rule cannot hold — the subframe four ahead might be the wrong direction. So TDD keeps the spirit of fixed timing but replaces the single constant with lookup tables in TS 36.213, keyed by the configuration in force.

The DL HARQ-ACK timing uses a set of k values (the DL HARQ-ACK association set, often written as the set K or by index M) that maps each DL subframe to the uplink subframe where its ACK is reported; the UL grant-to-PUSCH and PUSCH-to-PHICH timings use their own separate k tables. All of these offsets depend entirely on which UL/DL configuration is in force, and — because UL subframes are scarcer than DL subframes in DL-heavy configs — several DL subframes may fold their acknowledgements into a single UL subframe. That is why TDD needs HARQ-ACK bundling (logical-AND the ACKs) or multiplexing (report them separately), which FDD never needs. The number of HARQ processes also varies by configuration rather than being a fixed 8. Read the exact values from the tables — do not memorise or invent them.

Worked example — the shape of it, config 1 vs config 2. Take config 1 (D S U U D D S U U D). Its uplink subframes are 2, 3, 7 and 8; its downlink subframes are 0, 4, 5 and 9. The DL HARQ-ACK association set assigns each of those DL subframes to one of the UL subframes at least four subframes later, and with four UL subframes spread through the frame the loading is modest — roughly one or two DL subframes report into each UL subframe. Now take config 2 (D S U D D D S U D D): it has only two UL subframes, 2 and 7. Six downlink subframes must funnel their acknowledgements into just those two uplink opportunities, so each UL subframe carries the bundled/multiplexed ACKs for several DL subframes at once (the association set for each UL subframe therefore holds multiple k entries). Same principle in both cases — every DL subframe's ACK is deferred to the first suitable UL subframe at least four subframes on — but the fan-in is small in config 1 and large in the DL-heavy config 2. The precise k entries for each UL subframe come straight from the TS 36.213 association-set table for that configuration.

LTE vs NR Time-Domain Scheduling

Side by side, the philosophy difference is stark: LTE fixes the clock and the granularity; NR makes both flexible and signals them per transmission. Everything on this page is a symptom of that one root decision.

The takeaway is not that NR is "better" and LTE is "primitive." LTE's fixed clock buys determinism and near-zero timing overhead for a network that only ever served one numerology — you can implement its HARQ timing with a counter. NR pays a real signalling cost — K0, K1, K2, SLIV, per-transmission HARQ IDs — to earn the flexibility a multi-numerology, latency-diverse air interface demands. LTE's sTTI and TDD's table-driven timing are where LTE itself began to bend toward that flexibility.

Summary

LTE time-domain scheduling runs on a rigid clock. On FDD the whole loop is a handful of constants: the DL assignment and its PDSCH share subframe n, the HARQ-ACK follows at n+4, an uplink grant schedules PUSCH at n+4, and the PHICH answers that PUSCH at n+8 — an 8 ms round trip that tiles exactly into 8 uplink HARQ processes. Uplink HARQ is synchronous with the process ID derived from the subframe; downlink HARQ is asynchronous with the process ID carried in the DCI. The allocation unit is the whole 1 ms subframe, so there is no per-UE SLIV — the start is set cell-wide by PCFICH and the length is the rest of the subframe.

TDD keeps the same fixed-timing spirit but replaces the single n+4 constant with TS 36.213 lookup tables keyed by the UL/DL configuration, and DL-heavy configs force several DL subframes to fold their ACKs into one UL subframe via bundling or multiplexing. Rel-15 sTTI is the one crack in the 1 ms model. Read against NR — flexible K0/K1/K2, symbol-level SLIV, asynchronous HARQ with no PHICH — every LTE rigidity turns out to be a deliberate consequence of serving a single numerology cheaply.

Quick Q&A