HARQ in 5G NR: Soft Combining, Process IDs, and Flexible Timing

Plain ARQ discards a failed transmission. HARQ (Hybrid ARQ) keeps it in a soft buffer and combines it with the retransmission — and everything else follows from that.

A corrupted transport block still holds most of its information, just not enough to decode.

"Hybrid" means FEC and retransmission working together. Two transmissions that each fail alone often decode once combined.

---

Why soft combining wins

A block received just below the decoding threshold fails. Under plain ARQ the copy arrives at about the same SNR and fails too. HARQ adds the energy of both: two identical copies give roughly 3 dB, often exactly the missing margin.

Cheap recovery lets the scheduler target about 10% first-transmission BLER instead of 0.1%. Higher target BLER means higher MCS and higher average throughput. HARQ is what lets the scheduler run the link hot.

---

Two ways to combine

Chase combining

The retransmission is an identical copy, combined symbol by symbol — maximum ratio combining across time.

The soft buffer holds one block. Each retransmission adds energy but no new information, so returns diminish.

Incremental redundancy

The retransmission carries different coded bits from the same mother code: a different redundancy version.

Each attempt lowers the effective code rate — from 0.9 on the first transmission to about 0.3 after two retransmissions. That is an easier decoding problem, not just a higher SNR.

NR defines RV0 to RV3. RV0 carries the systematic bits and is self-decodable; the others add parity. IR beats chase combining at the cost of a larger soft buffer. NR uses IR; chase combining is the special case of resending RV0.

---

HARQ processes

A single stop-and-wait transmitter would idle for every round trip. NR instead runs several HARQ processes in parallel, each an independent stop-and-wait instance with its own soft buffer.

While process 0 waits for feedback, the others transmit. NR allows up to 16 processes per carrier per direction, double LTE's eight, because shorter slots fit more transmissions into one round trip.

The DCI carries the process ID, the New Data Indicator and the redundancy version, so the receiver knows which buffer to combine into.

HARQ delivers out of order: process 3 may finish after two retransmissions while process 4 succeeded first time. RLC reorders, which is one reason RLC AM sits above a layer that already retransmits.

---

Timing: where NR really differs

LTE fixed HARQ timing: in FDD, data in subframe n was acknowledged in n+4. NR signals timing per transmission in the scheduling DCI.

K1 — downlink feedback timing

K1 is the slot offset from PDSCH to the PUCCH carrying its HARQ feedback.

Latency-critical traffic gets a small K1, even feedback in the same slot if the TDD pattern has uplink symbols. A slower UE gets a larger one.

K2 — uplink scheduling timing

K2 is the slot offset from the uplink grant on PDCCH to the PUSCH it schedules.

Why this matters

Flexible timing enables self-contained slots, with data and acknowledgement in one slot. With mini-slots and wide subcarrier spacing, the round trip falls from LTE's fixed ~4 ms to under a millisecond.

UE processing capability bounds how small K1 and K2 can be: N1 for downlink, N2 for uplink. Capability 2 UEs support aggressive timing; Capability 1 UEs do not. Scheduling below the reported capability is a common field interoperability fault.

---

The soft buffer is the real constraint

The receiver stores LLRs at several bits per coded bit for every block awaiting retransmission. For a high-category UE with carrier aggregation, that is tens of megabits of on-chip memory.

3GPP defines a total soft buffer size per UE category, partitioned across processes. More processes means less memory each, and at high code rates the full block may not fit.

Limited buffer rate matching (LBRM) restricts which coded bits may be transmitted, so the receiver can always store what arrives. Both ends agree the limit in advance.

Soft buffer memory is expensive silicon that scales with peak throughput — a large part of why UE categories exist.

Why redundancy versions are ordered the way they are

RV0 starts at the systematic bits. RV2 is close to self-decodable. RV1 and RV3 are parity-heavy and depend on combining.

Send RV0 or RV2 when the retransmission may be received alone — after a long gap or on a new beam. Send RV1 or RV3 when the receiver still holds the earlier soft values. The usual sequence 0, 2, 3, 1 gets self-decodability early and maximum coding gain later.

---

HARQ and URLLC

Each retransmission costs a round trip. A 1 ms, 99.999% URLLC target allows one retransmission at most, often none.

URLLC therefore relies on other tools:

Conservative link adaptation. Target BLER around 0.001%, so the first transmission usually succeeds, at a deliberate cost in spectral efficiency.

Blind repetition. Send K copies without waiting for feedback. Latency becomes deterministic; some repetitions are wasted.

PDCP duplication. Send the packet over two independent carriers or cells. This is diversity, not retransmission, and it delivers the last nines.

HARQ optimises average throughput; URLLC needs bounded worst-case latency.

---

When HARQ makes things worse

Latency-bound traffic. One retransmission can exceed a 1 ms budget.

Head-of-line blocking. RLC AM restores order, so one block stuck in retransmission holds back blocks that already arrived correctly.

Non-terrestrial networks. A GEO round trip exceeds 500 ms. NTN lets HARQ feedback be disabled per process, falling back to blind repetition and RLC recovery.

Broadcast and multicast. There is no single feedback path, so multicast runs without HARQ or with NACK-only feedback.

Reading HARQ statistics

First-transmission BLER is what the outer loop drives, usually 10% for eMBB. Far below target, the scheduler is too cautious; far above, it is over-reaching.

Residual BLER after all retransmissions is what reaches RLC, typically below 0.1%. If it rises, HARQ is exhausting its budget: the problem is coverage or interference, not link adaptation.

10% initial with 0.01% residual is a healthy link. 40% initial with 1% residual is scheduled far beyond the channel.

---

Where HARQ sits

HARQ runs in MAC; RLC AM adds a second retransmission layer above it.

HARQ is fast, with 1-bit feedback, but that feedback can itself be corrupted: a NACK misread as ACK loses the block silently. RLC ARQ is slower, with detailed status reports, and catches that residue. HARQ brings residual error to ~10⁻¹–10⁻²; RLC takes it to ~10⁻⁵ or better.

Interaction with the scheduler

HARQ feedback drives link adaptation: a run of ACKs pushes MCS up, repeated NACKs pull it down.

This loop runs per UE, continuously. An abnormal retransmission profile in a cell points directly at interference or coverage.

---

LTE to NR at a glance

CBG-based retransmission resends only the failed code block groups. On large transport blocks — 100 MHz carriers with high-order MIMO — that saves significant resources.

---

The mental model

HARQ = FEC + retransmission + soft combining. Failed transmissions are kept, not discarded.

Incremental redundancy lowers the effective code rate with each attempt. Chase combining only adds energy.

Up to 16 parallel processes keep the link busy across the round-trip delay.

K1 and K2 are signalled per transmission — that flexibility is NR's real HARQ innovation.

HARQ enables aggressive link adaptation. ~10% target BLER is a feature, not a fault.

MAC HARQ is fast and imperfect; RLC ARQ catches the residue. Both exist deliberately.

---

Further reading

• 3GPP TS 38.321 — MAC protocol specification, HARQ operation
• 3GPP TS 38.212 — Multiplexing and channel coding, redundancy versions
• 3GPP TS 38.213 — Physical layer procedures for control, K1/K2 and HARQ codebooks
• 3GPP TS 38.214 — Physical layer procedures for data, UE processing time capabilities