Physical Layer Overview in 5G NR
What the physical layer delivers to the MAC, and the map of the 38.2xx specification series.
The physical layer is the factory floor of 5G NR: everything the higher layers decide — schedule this user, send this transport block, retransmit that one — eventually becomes bits that get coded, modulated and painted onto radio resources. The PHY's job is to move those bits across the air as reliably as the channel allows, and to hand the MAC a clean, measured, error-checked interface built entirely out of transport channels.
Introduction
The NR physical layer (layer 1) is the lowest layer of the radio stack, sitting directly beneath MAC. It owns everything between a transport block handed down by MAC and the radio waveform that leaves the antenna — channel coding, modulation, MIMO mapping, OFDM generation — and the reverse on receive. It is specified across the 38.2xx series (TS 38.201 through TS 38.215), with the general description in TS 38.201 and the upward service list in TS 38.202.
The PHY runs on every single transmission, in both directions, for the whole life of a connection: initial access (PSS/SSS/PBCH, PRACH), every scheduled slot of data (PDSCH/PUSCH), and every piece of L1 control (PDCCH/PUCCH). It is invoked continuously and on the tightest timescale in the system — per OFDM symbol and per slot — which is exactly why its interface to MAC is kept deliberately narrow and abstract.
Understanding the PHY overview means understanding one contract and one pipeline: the transport-channel service the PHY offers MAC (send, recover, measure), and the transmit chain that turns a transport block into a waveform. Everything more detailed — the resource grid, the coding, the procedures, the measurements — hangs off those two ideas and lives in the specific 38.2xx documents mapped later on this page.
On this page
Why the PHY is built this way
In plain words: think of MAC as someone placing an order with a warehouse's shipping department. MAC says "send this package (transport block), this big, by this deadline (this slot), at this service level (MCS)." The PHY is the shipping department: it packs the box, adds a tracking label (CRC), chooses trucks and routes (coding, modulation, antennas), drives it across the air, and sends back a delivered/failed receipt (ACK/NACK). MAC never has to know how forklifts, trucks or roads work — it just gets a clean receipt.
The whole layer is framed by two documents. TS 38.201 is the general description: it says what the PHY is, how it sits directly under MAC, and how the rest of the 38.2xx series fits together. TS 38.202 then lists the services and functions the PHY offers upward. Read together they answer one question — what does MAC get to assume the PHY will do for it?
The PHY carries MAC's data over transport channels and provides the control machinery around them — channel coding, HARQ soft-combining, and physical-layer measurements. MAC never touches subcarriers directly; it hands down transport blocks and gets back decoded transport blocks plus indications.
Separating "what to send" (MAC) from "how to get it across the air" (PHY) lets the scheduler stay channel-agnostic. MAC picks a transport-block size and a modulation-and-coding scheme; the PHY turns that into a robust waveform and reports whether it worked.
Through a defined set of transport channels (DL-SCH, UL-SCH, PCH, BCH, RACH), a set of indications (HARQ ACK/NACK, CSI/CQI feedback), and measurement quantities (RSRP, RSRQ, SINR) defined across the 38.2xx series.
The service interface is deliberately narrow, and that narrowness is the point. MAC sees only transport channels; the PHY maps each transport channel onto one or more physical channels that actually occupy resource elements in the grid. For example, DL-SCH is carried on PDSCH, UL-SCH on PUSCH, and BCH on PBCH. Above that line the MAC scheduler drives everything, but it does so purely in the currency of transport blocks and HARQ processes — it never reasons about QAM constellations, MIMO layers, or IFFT sizes. Below the line the PHY is free to change coding rate, modulation order, precoder, and antenna mapping on a slot-by-slot basis without MAC having to understand any of it. That clean contract is what lets NR support wildly different waveforms, band ranges, and antenna configurations behind one stable interface.
A transport channel is characterised not by what information it carries but by how and with what characteristics it is transferred over the radio. Each transport channel comes with an associated transport format that fixes, per Transmission Time Interval, how the data is sent: the transport-block size, the modulation scheme, the antenna mapping, and the redundancy applied. That is the abstraction MAC programs against.
Transport channels: the services offered to MAC
NR defines a small, fixed set of transport channels — three in the downlink and two in the uplink. Learn what each one is for and you have essentially learned the PHY-to-MAC contract.
| Transport channel | Direction | Carries | Notes |
|---|---|---|---|
DL-SCH (Downlink Shared Channel) | Downlink | User data and most downlink signalling | The workhorse. Supports HARQ, dynamic and semi-static link adaptation, dynamic resource allocation, and DRX for power saving. Carried on PDSCH. |
PCH (Paging Channel) | Downlink | Paging messages | Supports DRX so the UE can sleep between paging occasions. Mapped onto PDSCH resources. |
BCH (Broadcast Channel) | Downlink | The MIB (Master Information Block) | Fixed, pre-defined transport format. Carried on PBCH inside the SS/PBCH block. |
UL-SCH (Uplink Shared Channel) | Uplink | User data and most uplink signalling | The uplink workhorse. Supports HARQ, link adaptation, and dynamic/configured grants. Carried on PUSCH. |
RACH (Random Access Channel) | Uplink | Random-access preambles | Not a data channel — it carries the preamble that lets an unsynchronised UE get on the network. Carried on PRACH. |
Notice what is not here. There is no dedicated per-UE data channel and no separate control transport channel: NR is a shared-channel design, so nearly all user and signalling traffic rides the shared channels (DL-SCH/UL-SCH) and is multiplexed dynamically by the scheduler. Layer-1/2 control such as DCI and UCI is generated inside the PHY itself and does not correspond to any transport channel — that is why you will not find "control" in the list above.
Around these channels the PHY delivers three concrete services to MAC. First, data transport: each transport block is delivered per TTI with the requested transport format. Second, HARQ with soft combining: when a transport block fails its CRC, the PHY retains the soft bits and combines them with the retransmission (chase combining or incremental redundancy), reporting ACK/NACK per HARQ process so MAC can decide whether to schedule another attempt — NR supports up to 16 parallel downlink HARQ processes per cell, which is what lets many transport blocks be "in flight" while feedback catches up. Third, measurements: physical-layer quantities such as RSRP, RSRQ and SINR, plus CSI/CQI reporting, that feed link adaptation, scheduling, and mobility decisions above.
In plain terms: the PHY gives MAC three things — a way to send (transport channels), a way to recover (HARQ soft-combining), and a way to see (measurements). Everything else in the 38.2xx series is detail underneath those three.
Mapping the 38.2xx series
The physical-layer specifications split cleanly by function. Learn this map once and you can always find the clause you need without hunting.
| Spec | Owns | Think of it as… |
|---|---|---|
TS 38.201 | General description | The table of contents: what the PHY is, where it sits under MAC, and how the other 38.2xx documents relate. |
TS 38.202 | Services provided to higher layers | The service contract: the transport channels, indications and measurements the PHY exposes upward to MAC. |
TS 38.211 | Physical channels & signals | The grid and the symbols: frame structure, the resource grid, physical signals (DMRS, PSS/SSS, CSI-RS, PT-RS, SRS), and how physical channels (PDSCH, PUSCH, PDCCH, PUCCH, PBCH) map to resource elements. Defines CP-OFDM and DFT-s-OFDM. |
TS 38.212 | Multiplexing & channel coding | The error-protection factory: CRC attachment, code-block segmentation, LDPC for data, Polar for control, rate matching, concatenation, and how DCI/UCI bit fields are built. |
TS 38.213 | Physical-layer procedures for control | The control choreography: cell search, random access, power control, PDCCH monitoring, and how the UE finds and reacts to DCI. |
TS 38.214 | Physical-layer procedures for data | The data rules: how a transport-block size is computed, MCS tables, CSI reporting, and PDSCH/PUSCH resource-allocation procedures. |
TS 38.215 | Physical-layer measurements | The definitions of the numbers: exactly what RSRP, RSRQ and SINR mean and how they are computed. |
Rule of thumb: "how do the documents fit?" → 38.201; "what does the PHY promise MAC?" → 38.202; "what does a signal look like?" → 38.211; "how are the bits protected?" → 38.212; "what does the UE do for control?" → 38.213; "…for data?" → 38.214; "what is RSRP?" → 38.215.
The downlink transmit processing chain
Follow one downlink transport block from MAC all the way to the antenna. Each stage is defined in TS 38.211 or TS 38.212, and every stage has a clear, single purpose. The order matters: this is a pipeline, and the receiver runs it in reverse.
| # | Stage | What happens | Why |
|---|---|---|---|
| 1 | CRC attachment | A CRC is appended to the whole transport block — 24 bits (CRC24A) for a transport block larger than 3824 bits, otherwise 16 bits (CRC16). | So the receiver can tell whether decoding of the transport block succeeded — this is what ultimately drives HARQ ACK/NACK. |
| 2 | Code-block segmentation | If the block exceeds the maximum LDPC code-block size (8448 bits for base graph 1, 3840 bits for base graph 2), it is split into several code blocks, and a 24-bit per-code-block CRC (CRC24B) is added to each. | LDPC decoders work on bounded block lengths; segmentation keeps each piece decodable and lets the receiver check code blocks individually. |
| 3 | LDPC channel coding | Each code block is encoded with a Quasi-Cyclic LDPC code — base graph 1 (large blocks, code rates up to ~8/9) or base graph 2 (small blocks / low rate). | Adds structured redundancy so channel errors can be corrected. (Control on PDCCH uses Polar coding instead.) |
| 4 | Rate matching | Bits are written to a circular buffer and the exact number needed for the allocation is selected, per redundancy version (RV 0/1/2/3). | Matches the code output to the granted resources and enables incremental-redundancy HARQ via the different redundancy versions. |
| 5 | Code-block concatenation | The rate-matched code blocks are joined back into one bit stream. | Reassembles the transport block's coded bits into a single sequence for transmission. |
| 6 | Scrambling | Bits are XOR-ed with a UE/cell-specific pseudo-random sequence (initialised from the C-RNTI and cell ID). | Randomises inter-cell and inter-user interference so it averages out rather than correlating. |
| 7 | Modulation | Scrambled bits become complex symbols: QPSK, 16-QAM, 64-QAM, or 256-QAM (modulation order Qm = 2, 4, 6 or 8 bits/symbol). | Packs more bits per symbol when the channel is good; drops back to robust QPSK when it is not. |
| 8 | Layer mapping | The symbol stream is mapped onto one or more MIMO transmission layers (up to 8 layers on PDSCH). | Enables spatial multiplexing — parallel streams on the same time and frequency. |
| 9 | Antenna-port mapping / precoding | Layers are mapped to antenna ports through a precoding operation. | Steers energy (beamforming) and adapts the transmission to the spatial channel. |
| 10 | RE mapping | Symbols are placed on the resource elements of the allocated PDSCH, around reference signals. | Puts data exactly where the scheduler granted it in the time-frequency grid. |
| 11 | OFDM modulation | An IFFT builds the time-domain waveform and a cyclic prefix is inserted, per antenna port. | Produces the CP-OFDM signal that is actually radiated over the air. |
LTE ↔ NR: the shape of the chain is familiar from LTE, but the coding factory was rebuilt. LTE used Turbo coding for data and tail-biting convolutional coding (TBCC) for control; NR replaced these with LDPC for data (better at high throughput and easier to parallelise) and Polar coding for control. LTE also topped out at 64-QAM in the base releases (256-QAM added later), whereas 256-QAM is native to NR. The CRC-then-segment-then-code-then-rate-match skeleton, though, is the same idea in both.
The uplink chain and transform precoding
The uplink UL-SCH → PUSCH chain is structurally the same pipeline as the downlink: CRC attachment, code-block segmentation, LDPC coding, rate matching, concatenation, scrambling, modulation, layer mapping, precoding onto antenna ports, RE mapping, and OFDM modulation. MAC hands the PHY a transport block; the PHY returns a waveform. Everything you learned about the downlink chain transfers directly.
There is one important extra option in the uplink: transform precoding. When the network configures it (via the msg3-transformPrecoder / transformPrecoder parameters), a DFT is applied to the block of modulation symbols before they are mapped to resource elements. The result is DFT-s-OFDM (also called single-carrier FDM), whose signal has a markedly lower peak-to-average power ratio than plain CP-OFDM. That matters at the cell edge, where the UE's power amplifier is the limiting factor: a lower PAPR lets the amplifier run closer to saturation without distortion, extending uplink coverage. When transform precoding is disabled, the uplink uses the same CP-OFDM waveform as the downlink, which is friendlier to spatial multiplexing. The network chooses per situation — coverage vs. capacity.
LTE ↔ NR: in LTE the uplink was always SC-FDMA (i.e. DFT-spread OFDM) — there was no CP-OFDM uplink option. NR makes it a choice: CP-OFDM uplink (for MIMO and capacity) or DFT-s-OFDM (for cell-edge PAPR/coverage), selectable per configuration. This flexibility is new to NR and is one reason NR uplink can both spatial-multiplex aggressively and reach further than LTE at the edge.
The other asymmetry is the receiver. In the downlink the gNB transmits and the UE decodes; in the uplink the roles swap, and the gNB — with far more compute and antennas — performs the heavy equalisation and MIMO detection. That is why NR can afford richer uplink receiver processing than a handset could ever run in the downlink.
CRC check fail → NACK → PHY keeps soft bits, MAC schedules a retransmission (incremental redundancy)
That closed loop — enabled by the very first CRC stage of the chain — is what makes the PHY robust without MAC ever understanding the channel. MAC only sees ACK or NACK per HARQ process and reacts; the PHY quietly combines soft information across attempts to claw back marginal transport blocks. The same mechanism runs in both directions, which is why HARQ is listed as a core transport-channel service for both DL-SCH and UL-SCH.
Physical channels vs physical signals, and the mapping
Everything the PHY radiates is either a physical channel or a physical signal, and the distinction is simple: a physical channel carries information that originates in the higher layers (a transport channel, or L1/L2 control such as DCI/UCI); a physical signal is generated inside the PHY itself and carries no higher-layer payload — it exists purely to make the channel measurable, estimable, or synchronisable.
| Downlink | Uplink | |
|---|---|---|
| Physical channels (carry higher-layer / control info) | PDSCH (data), PDCCH (DCI), PBCH (MIB) | PUSCH (data), PUCCH (UCI), PRACH (preamble) |
| Physical signals (generated in the PHY, no payload) | PSS/SSS (synchronisation), DMRS, CSI-RS, PT-RS | DMRS, SRS, PT-RS |
With both halves defined, the transport-channel-to-physical-channel mapping is the seam where MAC's abstraction meets the grid. It is almost one-to-one:
| Transport channel | Maps to physical channel | Also on the grid alongside it |
|---|---|---|
DL-SCH | PDSCH | Scheduled by DCI on PDCCH (CRC scrambled by C-RNTI); demodulated using DMRS |
PCH | PDSCH | Paging DCI (P-RNTI) on PDCCH points to it |
BCH | PBCH | Sits in the SS/PBCH block with PSS/SSS |
UL-SCH | PUSCH | Granted by DCI (format 0_0/0_1); may multiplex UCI; uses DMRS/SRS |
RACH | PRACH | Preamble only; the response (RAR) comes back on PDSCH |
Two entries have no transport channel at all. PDCCH carries DCI (downlink assignments, uplink grants, power-control commands) and PUCCH carries UCI (HARQ feedback, CSI reports, scheduling requests). These are L1/L2 control generated within the PHY, which is exactly why they are absent from the transport-channel list: the scheduler produces them, but they never pass through the MAC transport-channel interface.
How numerology, BWP and HARQ tie the PHY to the higher layers
Three PHY concepts leak upward and shape how the whole stack behaves. They are worth understanding as the glue between "the factory floor" and everything above it.
Numerology. NR does not have one fixed subcarrier spacing; it has a family, indexed by μ = 0…4, giving spacings of 15·2μ kHz — 15, 30, 60, 120 (and 240) kHz — each a power-of-two scaling of the LTE-style 15 kHz. A slot is always 14 OFDM symbols, but higher numerology means wider subcarriers and shorter slots (a 15 kHz slot is 1 ms; a 30 kHz slot 0.5 ms; a 60 kHz slot 0.25 ms), so the scheduling and HARQ timing cadence all scale with μ. When MAC schedules "in a slot," how long that slot lasts is a PHY choice. Numerology is what lets one radio design serve both low-band coverage (15/30 kHz) and mmWave (120 kHz), and it is defined in TS 38.211.
Bandwidth Part (BWP). A UE is not required to receive across the full carrier at all times. A BWP is a contiguous subset of the carrier's resource blocks, with its own numerology and configuration, that the UE is told to operate within. Up to four BWPs can be configured per direction, but only one downlink and one uplink BWP is active at a time. BWPs let a UE run in a narrow, power-efficient slice when idle-ish and switch to a wide slice when it needs throughput — a direct PHY-configured knob that MAC scheduling and RRC configuration both live inside. Every CORESET, every PDSCH allocation, every measurement is scoped to the active BWP.
HARQ. As covered above, HARQ is where the PHY and MAC most tightly interlock. The PHY provides the soft-combining engine and the per-process ACK/NACK; MAC owns the HARQ processes, decides retransmissions, and manages the process identifiers. The number of parallel HARQ processes (up to 16) and their timing depend on numerology and on the round-trip processing budget — another place where a PHY parameter (slot length) sets an upper-layer behaviour (how many transport blocks can be in flight before feedback returns).
The through-line: MAC schedules in transport blocks, slots, and HARQ processes — but the size of a slot (numerology), the span it can address (BWP), and the depth of retransmission pipelining (HARQ) are all PHY-defined. The interface is narrow, but these three parameters are how PHY reality shapes higher-layer decisions.
LTE ↔ NR: LTE had a single fixed numerology (15 kHz, 1 ms subframe) and no bandwidth-part concept — a UE always operated across the whole configured bandwidth. NR adds scalable numerology (μ) and the BWP so one air interface spans sub-1 GHz to mmWave and UEs can adapt their bandwidth for power. LTE also used a fixed 8 HARQ processes (FDD); NR raises this to up to 16 to keep the pipe full at short, high-numerology slot lengths.
Summary
The NR physical layer is best remembered as one contract plus one pipeline. The contract (TS 38.202) is that the PHY gives MAC exactly three services over a fixed set of transport channels — a way to send (DL-SCH/UL-SCH/PCH/BCH/RACH onto PDSCH/PUSCH/PBCH/PRACH), a way to recover (HARQ soft-combining with per-process ACK/NACK, up to 16 processes), and a way to see (RSRP/RSRQ/SINR and CSI). MAC never touches subcarriers; it trades in transport blocks, MCS and HARQ processes.
The pipeline turns a transport block into a waveform in a fixed order: CRC → code-block segmentation → LDPC → rate matching (RV) → concatenation → scrambling → modulation (up to 256-QAM) → layer mapping → precoding → RE mapping → OFDM (IFFT+CP). The uplink is the same chain with an optional DFT (transform precoding) for lower-PAPR DFT-s-OFDM at the cell edge. Physical channels carry higher-layer/control payload; physical signals (PSS/SSS, DMRS, CSI-RS, PT-RS, SRS) are generated in the PHY to synchronise and measure.
Finally, three PHY parameters reach up and shape MAC: numerology (μ, 15·2μ kHz) sets slot length; the active BWP scopes every allocation; and HARQ depth sets how much can be in flight. To find any detail, use the spec map — 38.201 (overview), 38.202 (services), 38.211 (grid/signals), 38.212 (coding), 38.213/214 (control/data procedures), 38.215 (measurements).
Quick Q&A
Q. Name the NR transport channels and which physical channels carry them.
A. Downlink: DL-SCH→PDSCH, PCH→PDSCH, BCH→PBCH. Uplink: UL-SCH→PUSCH, RACH→PRACH. PDCCH/PUCCH have no transport channel — they carry L1/L2 control (DCI/UCI).
Q. Which spec defines LDPC/Polar coding, and which defines RSRP?
A. CRC, code-block segmentation, LDPC (data) and Polar (control), and rate matching are in TS 38.212. The measurement definitions — RSRP, RSRQ, SINR — are in TS 38.215. The general map lives in TS 38.201; the service list in TS 38.202.
Q. Give the downlink transmit chain in order.
A. CRC attach → code-block segmentation (+CB CRC) → LDPC coding → rate matching → code-block concatenation → scrambling → modulation → layer mapping → antenna-port/precoding → RE mapping → OFDM modulation (IFFT + CP).
Q. What is the difference between a physical channel and a physical signal?
A. A physical channel carries higher-layer or L1/L2 control information (e.g. PDSCH, PDCCH, PBCH). A physical signal is generated inside the PHY and carries no payload — it exists to synchronise, estimate, or measure (e.g. PSS/SSS, DMRS, CSI-RS, SRS).
Q. What does transform precoding do, and when is it used?
A. It applies a DFT to the modulation symbols before RE mapping, turning CP-OFDM into DFT-s-OFDM with lower PAPR. It is an uplink option used mainly to extend coverage at the cell edge, where the UE's power amplifier is the bottleneck.
Q. How do numerology and BWP reach up into MAC's world?
A. Numerology (μ = 0…4, spacing 15·2μ kHz) sets the subcarrier spacing and slot length, so it scales scheduling and HARQ timing. The active BWP scopes every allocation and measurement to a subset of the carrier. Both are PHY-defined but constrain how MAC schedules.
Where this fits
The PHY overview is the hub; the detail lives in the pages around it. Next, drop into the time domain and the resource grid, then into the waveform, and into the coding factory that protects every transport block.