Skip to content
Documentation Background

5G access network

The radio access network (RAN) is the cellular counterpart of a Wi-Fi WLAN. In 5G it consists of the user equipment (UE, the phone or modem), the New Radio (NR) air interface, and the gNodeB (gNB), the base station. The gNB manages the radio links, schedules transmissions, and connects to the core network. This page covers the radio side. The core is in 5G core network.

Unlike Wi-Fi, which contends for an unlicensed channel, a cellular base station controls the channel centrally. Nothing transmits unless the gNB has granted it radio resources.


5G uses OFDMA: the channel is divided into many narrow subcarriers in frequency and short intervals in time. Each cell of that grid is a resource element carrying one symbol. Resource elements are bundled into resource blocks, the smallest unit the scheduler hands to a device.

Property4G LTE5G NR
Subcarrier spacing15 kHz, fixed15, 30, 60, 120 kHz and up, chosen per band
Symbol timeAbout 66.7 microsecondsShorter as spacing widens
Channel widthUp to 20 MHzUp to 100 MHz in mid-band, wider in millimeter wave
DuplexingFDD or TDDFDD or TDD

Wider subcarrier spacing gives shorter slots and lower latency, at the cost of a shorter range. Low bands use narrow spacing for coverage, and millimeter wave uses wide spacing for speed and latency.

Duplexing divides capacity between downlink and uplink either by frequency (FDD, one band each way) or by time (TDD, the same band in alternating periods). TDD operators set the ratio to fit the traffic, which is heavily downlink-biased.

Downlink and uplink each have separate physical channels:

DirectionChannelCarries
DownlinkPDSCHUser data and control messages
DownlinkPDCCHResource grants: which blocks the device should receive or send on
DownlinkPBCHBasic system information for discovering the cell
UplinkPUSCHUser data
UplinkPUCCHScheduling requests, channel quality reports, and ACKs
UplinkPRACHRandom access, the first contact from a device

PRACH is the one place in cellular where devices contend. It works like slotted ALOHA, described in multiple access protocols: a device transmits a preamble in a random access slot, and if two collide, both retry after a random delay. Once connected, everything is scheduled.


The link layer in the RAN is a stack of sublayers on the UE and on the base station. The user plane carries IP traffic. The control plane configures it.

SublayerPlaneJob
RRC (radio resource control)ControlConnection setup and release, connection states, measurement reports, handover commands
SDAP (service data adaptation)UserMaps IP flows to quality-of-service flows and radio bearers
PDCP (packet data convergence)BothHeader compression, ciphering, and integrity protection
RLC (radio link control)BothSegmentation and reassembly, ARQ retransmission
MACBothScheduling, and mapping data to resource blocks. Also hybrid ARQ, a fast retransmission loop
PHYBothCoding, modulation, OFDMA, antennas

Reliability is layered. The MAC and RLC retransmit lost radio blocks quickly, so TCP sees the radio as a fairly clean, if variable, pipe. The cost is delay variation: a retransmission adds jitter, and TCP may react to the extra delay rather than to real loss.

The connection state matters for battery life and latency. A device is in RRC_CONNECTED while active, RRC_INACTIVE or RRC_IDLE when quiet. Moving from idle back to connected takes a signaling exchange, which is why the first packet after a pause is slower.


Traditionally a base station was one box. 5G splits it into three functions that can sit in different places:

UnitRunsTypical location
Radio unit (RU)RF circuits, analog-to-digital conversion, antennasAt the mast
Distributed unit (DU)PHY, MAC, RLC. Hard real timeNear the mast or at a nearby site
Central unit (CU)RRC, SDAP, PDCP. Less time-criticalRegional data center

Splitting them lets an operator pool processing for many sites, scale it in software, and mix equipment from several vendors through open interfaces (Open RAN). The CU and DU can run as containers on ordinary servers, so this is where cloud-native practice meets cellular. The DU’s tight timing requirements still need careful placement and hardware support.

Software-defined RAN takes the SDN idea into the RAN. The control logic moves to a RAN intelligent controller, which keeps a network-wide view of cells, connections, and channel state, and runs applications (xApps) for tasks such as load balancing, handover, and QoS. It parallels the SDN controller model in the wired world.


Before a device can send data it must find a cell and connect.

  1. Synchronize. The device looks for the primary and secondary synchronization signals, broadcast every few milliseconds. These give it the clock timing and the physical cell ID. Together with the master information block they form the synchronization signal block.
  2. Read system information. SIB1, broadcast on a longer cycle, carries the operator identity (country and network codes), the cell ID, and the minimum signal level needed to camp on the cell.
  3. Choose a cell. The standards do not fix the method. Typical inputs are the operator named on the SIM, a priority list of networks, and signal strength.
  4. Random access. The device sends its RRC connection setup request on PRACH. The gNB replies, and the device now has dedicated signaling resources.

At this point the device is connected to the RAN but is not yet on the network. It has no authentication and no IP address. Those come from registration and session setup in the core.

Terminal window
# Linux with ModemManager: serving cell, technology, and signal metrics
mmcli -m 0 --signal-get
# nr5g: rsrp: -92.00 dBm rsrq: -11.00 dB snr: 14.0 dB
mmcli -m 0 | grep -E 'access tech|operator|signal quality|state'

Cellular signal is reported with different metrics than Wi-Fi. RSRP is the reference signal power, RSRQ its quality relative to the total received, and SINR the signal to interference and noise ratio. Higher is better for each. RSRP above about -90 dBm is good, and below -110 dBm is poor. Indoor coverage problems usually trace to building materials, which a window-mounted antenna, a repeater, or Wi-Fi calling can fix.


Every millisecond the gNB’s MAC scheduler decides which devices send and receive on which resource blocks. Unlike a wired link, where total throughput does not depend on the order of frames, in radio the order changes the total, because each device’s channel quality is different and varies over time. A device with a strong channel gets more bits per block.

The scheduler weighs four things: channel quality (from CQI reports), fairness, traffic class (voice before downloads), and per-flow QoS guarantees. The standards do not define the algorithm, so scheduling is where vendors differentiate.

AlgorithmPicksStrengthWeakness
Round robinEach device in turnSimple, equal turnsIgnores channel quality. Poor-signal devices get fewer bits
Maximum throughputThe device with the best channelHighest total throughputStarves devices at the cell edge
Blind equal throughputThe device with the lowest average rateEqual rates for allLowers total throughput
Proportional fairThe best ratio of current rate to average rateA balance. Serves each device near its own peaksNeeds history per device

Proportional fair is the common choice. A numeric comparison is in the wireless link math deep dive.

For a user this means throughput depends on position and on neighbors. At the cell edge, a device gets low-order modulation and is served less often. Speed tests also vary by the minute as other users come and go.


A cellular radio draws on the order of a watt or more when active and a few milliwatts when asleep, so devices sleep whenever possible. In the connected state, discontinuous reception (DRX) works like this:

  1. After traffic, an inactivity timer runs. If no more data arrives, the device enters short DRX cycles, waking briefly at intervals to see if the gNB has scheduled anything for it.
  2. If it stays quiet, it moves to long DRX cycles with longer sleeps.
  3. The gNB buffers downlink data until the device’s next wake window, so a packet may wait up to one cycle.
  4. After more inactivity the device goes to idle or inactive. The radio connection is released, and the network must page the device to reach it.

The trade is battery life against the delay for the first packet after a quiet period. Push notifications, chat keepalives, and TCP keepalive intervals interact with these timers, and frequent small wakeups are a common cause of mobile battery drain.