HARQ — Hybrid ARQ
HARQ combines FEC with ARQ retransmissions. Failed decodes are soft-combined with retransmissions for higher decode success. Covers Chase Combining, Incremental Redundancy, 16 parallel processes, CBG retransmission, and 6G evolution (predictive HARQ, HARQ-free).
#1Why "Hybrid" ARQ?
HARQ is hybrid because it combines FEC (forward error correction) with ARQ (retransmission requests):
Pure ARQ
CRC check → if fail → discard + retransmit. Simple but wastes previous transmission energy.
Pure FEC
Add heavy parity bits → correct errors without reTX. High overhead for worst case.
HARQ = FEC + ARQ (Hybrid)
TX#1: FEC encode → transmit → decode fails → STORE soft bits (LLRs)
TX#2: Retransmit → UE combines old + new LLRs → decode succeeds!
Coding gain: 1 reTX = ~3 dB, 2 reTX = ~4.8 dB, 3 reTX = ~6 dB
HARQ vs RLC ARQ — Two Layers
| Aspect | HARQ (MAC) | RLC ARQ |
|---|---|---|
| Speed | 0.5–1 ms (per-slot) | 10–100 ms |
| Mechanism | Soft-combining (CC/IR) | Full SDU retransmission |
| Max reTX | 4 (typical) | 8–32 |
| Residual BLER | ~1% | ~0.001% |
#2HARQ MAC Protocol — Process States
Per-Process State Variables
HARQ_Process_ID (0-15), NDI (toggle for new data), RV (0→2→3→1 cycling), Retransmission_Count, Max_Retransmissions (4), Soft_Buffer_Pointer
State Machine:
IDLE ──New data──► WAIT_ACK ──ACK──► IDLE (reuse process)
│
NACK
▼
RETRANSMIT ──max reTX?──► IDLE (RLC ARQ)
│ │
└──NACK, more──┘
TX (next RV)
#3Chase Combining (CC)
Retransmission is an exact copy. UE adds LLRs from both transmissions:
Soft-combining — watch LLRs combine and effective SNR increase with each retransmission
CC Properties
LLR_combined = LLR₁ + LLR₂ → SNR_combined = SNR₁ + SNR₂
Each copy is self-decodable. Simple to implement. ~3 dB gain per reTX.
#4Incremental Redundancy (IR)
Each retransmission carries different parity bits. Effective code rate decreases:
| RV | Self-Decodable? | Content | Effective Rate |
|---|---|---|---|
| 0 | YES | Systematic + Parity A | ~2/3 (high) |
| 2 | NO | Parity B (new bits!) | ~1/2 |
| 3 | NO | Parity C (new bits!) | ~1/3 (low) |
| 1 | NO | Parity D (overlap) | ~1/4 |
Chase Combining
Same bits retransmitted. Energy gain only. ~3 dB per reTX. Lower complexity.
Incremental Redundancy
New parity bits each reTX. Energy + info gain. ~3-5 dB per reTX. Higher complexity.
RV Sequence in 5G NR: 0 → 2 → 3 → 1 → 0 → 2 → ...
Circular Buffer: [S S S S | P_A P_A P_A | P_B P_B P_B | P_C P_C]
RV=0 start RV=2 start RV=3 start
#516 Parallel HARQ Processes
16 parallel processes enable continuous pipeline — no idle slots waiting for ACK:
16 HARQ processes pipeline — watch processes cycle through TX, wait-ACK, reTX, and reuse
Pipeline Utilization
Utilization = min(1, N_HARQ / RTT)
RTT ≤ 16 → full pipeline. RTT > 16 → pipeline stall → throughput drops proportionally.
#6Code Block Group (CBG) Retransmission
CBG retransmits only failed code block groups instead of the entire TB:
TB vs CBG retransmission — see how CBG saves bandwidth by retransmitting only failed groups
#7HARQ-ACK Feedback
| PUCCH Format | Max Bits | Use Case |
|---|---|---|
| 0 | 1–2 | 1–2 HARQ-ACK (no SR/CSI) |
| 1 | 1–2 | With Scheduling Request |
| 2 | > 2 | Multiple ACK + CSI + SR |
| 3 | > 2 | Large payload (long PUCCH) |
Multiplexing
Per-process ACK/NACK bits: [1,0,1]. Precise but higher PUCCH overhead.
Bundling (AND)
ACK AND NACK = NACK. Lower overhead but triggers unnecessary reTX.
#8HARQ Throughput Analysis
HARQ throughput vs first-TX BLER — compare with-HARQ vs without-HARQ performance
#9Practical Field Considerations
| Pitfall | Impact | Best Practice |
|---|---|---|
| Max reTX too low (1) | RLC ARQ overloaded | Use max reTX = 4 |
| Soft buffer overflow | Oldest LLRs flushed | Monitor buffer usage < 80% |
| NDI misalignment | reTX treated as new | Verify NDI toggling in DCI |
| RV stuck at 0 | No IR benefit | Ensure RV cycles: 0→2→3→1 |
| K1 too small | ACK timing violation | Verify K1 ≥ N1 |
| Pipeline stall | 20-40% throughput loss | Reduce K1 or add processes |
LTE vs 5G NR HARQ
| Aspect | LTE | 5G NR |
|---|---|---|
| Processes | 8 | 16 |
| CBG reTX | Not supported | Supported |
| RV sequence | Configurable | Fixed: 0→2→3→1 |
| Min RTT | 8 ms | 2–5 slots (0.5–2.5 ms) |
#106G HARQ Evolution
HARQ evolution — compare capabilities across 4G, 5G, and 6G
| Dimension | 5G NR | 6G Target |
|---|---|---|
| Retransmit decision | Reactive (wait for NACK) | Predictive (AI pre-reTX) |
| HARQ-free mode | CG K-repeat (UL only) | DL+UL HARQ-free |
| Adaptive strategy | Fixed CC/IR | AI selects per-packet |
| Cross-layer | None | PHY-MAC-RLC joint |
| NTN HARQ | Extended K1 | Predictive + FEC |
| Processes | 16 | 16–64 |
6G Key HARQ Innovations
- Predictive HARQ: AI predicts failure → pre-schedules reTX before NACK → zero delay
- HARQ-Free: Ultra-low code rate + K-repetition → BLER < 10⁻⁷, no reTX needed
- Adaptive: AI selects optimal strategy per-packet (CC/IR/K-repeat/HARQ-free)
- Cross-layer: PHY-MAC-RLC joint optimization → 10-15% latency reduction
#11Further Reading
3GPP Specifications
- TS 38.321 — NR MAC Protocol §5.3.2 (HARQ entity, processes, retransmissions)
- TS 38.214 — NR Physical Layer Procedures for Data (HARQ process, RV, CBG)
- TS 38.213 — NR Physical Layer Procedures (HARQ-ACK timing, K1, PUCCH)
- TS 38.212 — NR Multiplexing and Channel Coding (circular buffer, RV, LDPC)
Companion Resources
Channel Coding → Link Adaptation → ★ CA (Multi-Cell HARQ) → ★ DC (Dual HARQ) → ★ Sidelink HARQ → ☆ NTN HARQ →