The channel corrupts bits; coding is how the link survives it. A CRC detects errors, forward error correction fixes them, and HARQ combines coding with retransmission — and 5G picks LDPC for data and Polar for control for concrete reasons.
A radio channel corrupts bits — noise, interference and fading flip them. Two tools handle that. A CRC (cyclic redundancy check) detects whether a block arrived intact; it cannot fix anything. Forward error correction (FEC) adds structured redundancy — parity — so the receiver can correct a certain number of errors without asking for a retransmission. The ratio of information to total bits is the code rate, and lowering it (more parity) buys robustness at the cost of throughput.
FEC is what lets a link run close to the channel's theoretical capacity. Get within a fraction of a decibel of the Shannon limit and you squeeze the most data through the spectrum you paid for — which is exactly what a modern cellular system is trying to do.
5G NR uses two FEC families, split by job. LDPC (low-density parity-check) codes carry the data channels (PDSCH, PUSCH). They excel at long blocks and very high throughput, decode with parallel message-passing, and support flexible rate matching through a circular buffer — ideal for gigabit data. Polar codes carry the control channels (PDCCH, PBCH, PUCCH). They are provably capacity-achieving and strong at the short block lengths control messages use, where LDPC is less efficient.
This is a change from LTE, which used Turbo codes for data and convolutional codes for control. Turbo served well but LDPC decodes faster at NR's data rates, which is why NR switched. Reed–Solomon is a different animal — a block code that corrects symbol errors, dominant in broadcast, optical and storage systems and in older standards, but not used for NR data. It is worth knowing because it still appears across adjacent systems a radio designer meets.
FEC corrects what it can; when a block is still wrong, the system retransmits. Doing that naively (plain ARQ) throws away the failed copy. Hybrid ARQ (HARQ) is smarter: it keeps the failed receive and combines it with the retransmission.
Two flavours. Chase combining resends the same coded bits and the receiver adds the two soft copies, improving the signal-to-noise ratio. Incremental redundancy resends different parity, so each retransmission lowers the effective code rate and adds correcting power — usually the stronger scheme. The receiver soft-combines across attempts in a HARQ buffer, and only a CRC pass releases the block upward. NR runs several HARQ processes in parallel (stop-and-wait per process) so the link does not stall waiting for one acknowledgement.
The decoder is one of the heaviest blocks in the physical layer. An LDPC decoder at NR data rates is a large, parallel engine — layered min-sum message passing across a parity-check matrix, iterated until it converges — and it has to finish inside the slot's latency budget every time. Polar decoding (successive-cancellation list) is lighter per block but runs on the latency-critical control path.
So channel coding is a major part of the high-PHY compute that lives in the DU, and a core reason the baseband needs the silicon it does. The engineering is in hitting throughput and latency together: enough decoder parallelism for gigabit data, flexible rate matching for every code rate the scheduler picks, and a HARQ buffer and combining path that keep up. Get it right and the link runs near capacity; get it wrong and either throughput or latency gives way.
Coding that adds structured redundancy (parity) to data so the receiver can correct a number of errors without a retransmission. A CRC detects whether a block is corrupt; FEC actually fixes errors. It lets a link run close to the channel's theoretical capacity.
5G NR uses LDPC codes for the data channels (PDSCH, PUSCH) and Polar codes for the control channels (PDCCH, PBCH, PUCCH). LDPC suits long, high-throughput blocks; Polar suits the short blocks of control signalling.
LDPC (low-density parity-check) decodes with parallel message-passing and excels at high throughput, so NR uses it for data. Polar is capacity-achieving and strong at short blocks, so NR uses it for control. Turbo was LTE's data code, superseded by LDPC in NR because LDPC decodes faster at NR's rates.
Not for NR data. Reed–Solomon is a block code that corrects symbol errors and dominates broadcast, optical and storage systems and older standards. NR data uses LDPC and control uses Polar. Reed–Solomon is still worth knowing because it appears in adjacent systems.
Hybrid ARQ combines forward error correction with retransmission. Instead of discarding a failed block, the receiver keeps it and combines it with the retransmission, so each attempt adds correcting power. NR runs several HARQ processes in parallel so the link does not stall.
Chase combining resends the same coded bits and the receiver adds the soft copies to raise the SNR. Incremental redundancy resends different parity, lowering the effective code rate on each retransmission and adding more correcting power — usually the stronger scheme.
At NR data rates the LDPC decoder is a large, parallel engine that must converge within the slot's latency budget every time. It is one of the heaviest high-PHY blocks, a major part of the DU's compute, and getting throughput and latency together is the core challenge.
The other high-PHY heavy lifting.
KNOWLEDGEWhere high-PHY sits.
KNOWLEDGEThe layers FEC protects.
TECHNOLOGYWhere coding lives in the PHY.
KNOWLEDGEMapping a decoder to silicon.
KNOWLEDGEWhat a gigabit decoder costs.
An LDPC decoder that hits gigabit throughput inside the slot budget, flexible rate matching, and a HARQ combining path that keeps up are the real work. Tell us the rate and the latency target.