MAC Layer Overview in 5G NR

Introduction

The MAC sublayer (TS 38.321) is the lowest layer of the radio protocol stack that still reasons about which data goes out and when. It sits between RLC above and the physical layer below, and it is where the abstract idea of "a UE with data to send" becomes the concrete reality of "a transport block on a shared, error-prone medium, delivered to the right UE at the right time."

It is active the entire time a UE is attached, in every direction and in every state that carries traffic. Every scheduled uplink and downlink transmission passes through it: the MAC receives a grant, fills a transport block, runs HARQ, and reports its status back to the scheduler — slot after slot. It also owns several housekeeping procedures that keep the link alive: random access, timing-advance maintenance, DRX, BWP switching and SCell control.

It matters because the air interface is one shared medium that many UEs and many logical flows must contend for, and it is fast and lossy. Higher layers cannot react quickly enough to schedule slot-by-slot or to recover from air errors in a few milliseconds. The MAC is the layer built for exactly that speed — multiplexing competing flows, correcting errors with soft-combining before RLC ever sees them, and keeping the uplink time-aligned so dozens of UEs land inside the same slot boundary.

What the MAC layer actually does

Everything above the MAC thinks in terms of logical channels — streams of a particular kind of information (signalling, user data, paging). Everything below the MAC thinks in terms of transport channels — chunks of bits the physical layer knows how to code and modulate. The MAC's core job is to bridge those two worlds and to arbitrate the one thing every UE in a cell must share: the radio. It is the lowest layer that still reasons about which data goes out, in what order, and how much of it.

The MAC provides data-transfer and radio-resource-allocation services to RLC above it, and uses the transport-channel services of the physical layer below. TS 38.321 enumerates its functions as a checklist, and it is worth being able to name every one of them — each is a distinct piece of machinery the layer runs.

Why the MAC layer is needed

Higher layers deliver neat, orderly streams and assume they will arrive. The radio makes no such promise: it is shared among every UE in the cell, its capacity changes slot by slot, and it corrupts blocks constantly. Three hard problems fall out of that, and the MAC exists to solve all three at once.

Everything the MAC does is in service of doing these fast and in-band — carrying control as MAC CEs inside the same transport block as the data, so a slot-timescale decision never has to wait for the slow, reliable RRC path.

Logical channels above, transport channels below

The two interfaces the MAC bridges are named and fixed. Above it, RLC delivers data on logical channels, classified by what kind of information they carry — control-plane signalling versus user-plane traffic, dedicated versus common. Below it, the physical layer offers transport channels, classified by how the information is transported over the air. The MAC's channel-mapping function is exactly the wiring between these two sets.

On the transport side, the MAC deals with the DL-SCH (Downlink Shared Channel) and UL-SCH (Uplink Shared Channel) for the vast majority of traffic, the PCH (Paging Channel) for paging, the BCH (Broadcast Channel) for the fixed-format MIB, and the RACH (Random Access Channel) for preambles. The shared channels (DL-SCH/UL-SCH) are the workhorses: they support dynamic transport-block size, HARQ, link adaptation (dynamic modulation and coding) and beam-specific transmission. "Dynamic TB size" is the crucial property — every grant tells the UE exactly how many bits fit in this transmission, computed from the allocated resource blocks, the modulation-and-coding scheme and the number of layers, so the transport block is sized fresh for each slot rather than fixed in advance.

The mapping is many-to-one where it needs to be: BCCH maps to BCH (for the MIB) and to DL-SCH (for SIBs); PCCH maps to PCH; CCCH, DCCH and DTCH all map to DL-SCH/UL-SCH. That collapsing of many logical channels onto the shared transport channel is precisely why multiplexing and prioritization exist.

One MAC entity per cell group, addressed by C-RNTI

A UE has a separate MAC entity for each cell group: one for the Master Cell Group (MCG) and, when dual connectivity is configured, one for the Secondary Cell Group (SCG). There is one MAC per UE per cell group — each entity independently runs its own HARQ, DRX, BSR, timing-advance and random-access machinery for the cells in its group, because the two groups are served by different nodes on different carriers. Within a cell group, one cell is the special cell (SpCell — the PCell for MCG or PSCell for SCG) and the rest are SCells added by carrier aggregation.

On the shared channel there is no dedicated pipe per UE, so the network needs a way to say "this grant is for you." That is the job of the RNTI — a Radio Network Temporary Identifier that scrambles the CRC of the PDCCH that carries the grant. The UE's persistent identity in a cell is its C-RNTI (Cell RNTI, a 16-bit value in the range 0x0001–0xFFEF); a PDCCH whose CRC unmasks with the UE's C-RNTI is a grant addressed to that UE. Other RNTIs address shared or transient procedures.

Because each MAC entity is bound to one cell group, dual-connectivity operation is cleanly partitioned: a NACK on an SCG carrier, a BSR for MCG data, or a random-access attempt on the PSCell are all handled by their own entity without touching the other. This is what lets a UE be scheduled by two schedulers at once without the two stepping on each other.

Where the MAC sits and what flows through it

The MAC lives directly below RLC and directly above the physical layer. RLC hands it PDUs (which become MAC SDUs) over logical channels; the MAC multiplexes them with any control elements into a transport block and passes it to PHY over a transport channel. Read the figure top to bottom: logical channels enter, the entity's functions act on them, and one transport block leaves.

Multiplexing, MAC PDU structure, LCP, padding and HARQ

When the scheduler grants the UE resources, the MAC must fill exactly that many bits — no more, no less. It runs LCP to decide the order in which logical channels contribute their SDUs, then assembles a MAC PDU: a sequence of MAC subPDUs, each a subheader plus either a MAC SDU, a MAC CE (control element), or padding. MAC CEs carry in-band control — a BSR, a PHR, a C-RNTI, DRX commands, SCell activation — and are placed at defined positions relative to SDUs (CEs first in the UL PDU, near the end in the DL PDU). See MAC PDU & CEs for the byte-level layout.

LCP (logical channel prioritization) gives each logical channel a priority and a Prioritized Bit Rate (PBR) with an associated bucket. In the first round each channel is served up to its PBR in strict priority order, so a high-priority channel cannot completely starve a lower one; any remaining grant is then filled by strict priority until either the data or the grant runs out. NR adds LCP restrictions — a logical channel can be mapped only to permitted numerologies, cells, or PUSCH durations (allowedSCS-List, maxPUSCH-Duration, allowedServingCells) — so latency-critical traffic is confined to fast numerologies. This is how the UE splits a single grant fairly across, say, a signalling DCCH and a data DTCH. If, after all SDUs and CEs are placed, bits remain, the MAC inserts padding (a padding subheader plus padding bits) so the PDU exactly matches the transport-block size; a padding BSR may opportunistically occupy that space instead.

Each HARQ process holds one transport block awaiting acknowledgement, identified by a HARQ process ID carried in the DCI. Because several processes run in parallel, the UE can keep transmitting new blocks while earlier ones await feedback — stop-and-wait per process, pipelined across processes. A New Data Indicator (NDI) toggle in the DCI tells the UE whether a grant carries fresh data or a retransmission of the block already buffered in that process, and the Redundancy Version selects which coded bits are sent so incremental-redundancy combining can improve the odds on each try.

Reporting to the scheduler: BSR, SR, PHR, timing advance and DRX

The gNB scheduler cannot allocate uplink well unless it knows two things about the UE: how much data is waiting, and how much transmit power is left. The MAC reports both as MAC CEs, and asks for resources with a SR when it has none.

The BSR reports per Logical Channel Group (LCG — up to 8 groups) rather than per channel, keeping the report compact; a regular BSR triggered by newly arrived higher-priority data is what usually kicks off an SR when the UE has nothing to send on. If even PUCCH resources for the SR are missing, the UE falls back to random access to get a grant.

Timing advance and uplink time alignment are the MAC's job of keeping uplink transmissions from different UEs arriving aligned at the gNB despite differing propagation delays. The gNB measures the UE's timing and sends a Timing Advance Command MAC CE; the UE applies the offset and (re)starts its timeAlignmentTimer for that timing-advance group (TAG). While the timer runs the uplink is considered synchronised; if it expires, the UE releases PUCCH/SRS resources and must re-acquire timing through random access before transmitting again. This is why a UE that has been silent for a while may have to do random access even though it is still in RRC_CONNECTED.

Once the UE is connected, blindly monitoring PDCCH every slot would drain the battery. DRX lets the MAC switch the receiver off on a schedule. During the On Duration the UE monitors PDCCH; if nothing is scheduled it may sleep until the next cycle. Any activity restarts the drx-InactivityTimer, keeping the UE awake while traffic flows, and a drx-RetransmissionTimer keeps it listening for pending HARQ retransmissions. See DRX for the timer details.

BWP operation, SCell (de)activation and the MAC CE catalogue

Two more control functions round out the MAC's remit, and both are driven by MAC CEs. BWP operation concerns the active Bandwidth Part — the slice of carrier bandwidth the UE is currently tuned to. A UE may be configured with several BWPs (up to four per direction) but only one is active per direction at a time. A DCI or a MAC-visible timer can switch it: the bwp-InactivityTimer runs while the UE is on a non-default BWP, and on expiry the MAC switches back to the default BWP (typically narrower) to save power. When the active BWP changes, the MAC suspends transmission on the old BWP and resumes on the new one, keeping HARQ buffers intact.

SCell (de)activation lets carrier aggregation add bandwidth only when it is needed. Configured SCells start deactivated; the gNB sends an SCell Activation/Deactivation MAC CE (a bitmap, one bit per SCell) to bring them up, and each active SCell runs an sCellDeactivationTimer that deactivates it again after a period of inactivity. Deactivating an SCell stops its PDCCH monitoring, CSI reporting and SRS, which is a large power saving when the extra carriers are idle.

All of these control functions travel in-band as MAC control elements, each identified by a reserved LCID in its subheader. The catalogue below is the working set you should recognise.

Summary

The MAC (TS 38.321) is the sublayer that turns many logical streams into one transport block on shared radio, delivered reliably and on time. It maps logical channels (BCCH/PCCH/CCCH/DCCH/DTCH) to transport channels (BCH/DL-SCH/UL-SCH/PCH/RACH), multiplexes SDUs and in-band MAC CEs into a dynamically-sized transport block via LCP and padding, and runs HARQ with soft combining to catch air errors in a few slots.

It is organised as one entity per cell group (MCG and, under dual connectivity, SCG), each with independent HARQ, DRX, BSR and timing-advance state, and it addresses grants on the shared channel by scrambling PDCCH CRC with the UE's C-RNTI. Around that core it runs the machinery that keeps the link alive and efficient: BSR/SR/PHR reporting to the scheduler, timing-advance maintenance via the timeAlignmentTimer, DRX for battery, and BWP and SCell control — all carried as fast, in-band MAC CEs rather than slow RRC signalling. Master those pieces and the deeper topics — channel mapping, MAC PDU layout, HARQ, random access, DRX — are each just one function of this layer examined up close.