5G: The Radio, the Core, and How They Fit Together

5G is usually explained as a list of superlatives — faster, lower latency, more devices. That framing is useless the moment you have to work on one. It is more accurate, and far more useful, to see 5G as two systems designed to different briefs that then had to be made to agree.

On one side is the radio: an air interface deliberately built with almost every parameter left configurable, so a single specification can serve a rural macro cell at 700 MHz and a factory small cell at 28 GHz. On the other is the core: the parts of a mobile network that were once purpose-built boxes, rewritten as software services that call each other over HTTP.

Nearly every "why does 5G do it that way" question resolves into one of those two stories.

The air interface: flexibility as the organising idea

LTE fixed its subcarrier spacing at 15 kHz and built everything on top of that one decision. It worked, because LTE was aimed at one kind of traffic in one kind of spectrum. NR could not make the same simplification, so it made the spacing itself a parameter — the choice that numerology describes.

That single decision propagates outward. Double the subcarrier spacing and the symbol halves, so the slot halves, so latency halves — but the cyclic prefix halves too, and with it the delay spread the cell can tolerate. A wide spacing suits a small cell with short echoes and a device that needs a fast answer. A narrow one suits a large cell whose echoes arrive microseconds apart. There is no universally correct setting, which is exactly why it is configurable.

The time-frequency grid

Everything the radio does is an allocation on a grid of subcarriers and symbols. Time is divided into frames, subframes and slots, with the mini-slot available when a transmission cannot wait for a slot boundary. Frequency is divided into resource blocks of twelve subcarriers, and a device is told to operate within a bandwidth part rather than the whole carrier — so a modest device never has to tune across 100 MHz it will never use.

The transmitted signal is CP-OFDM in the downlink, with DFT-s-OFDM available in the uplink where a device's power amplifier cares more about peak-to-average ratio than about spectral shape. The cyclic prefix in front of every symbol is what turns a multipath channel into one a receiver can undo with a single multiplication per subcarrier.

Finding the cell, then keeping the link

Before any of that grid matters, a device has to find the cell at all. It searches for the synchronisation signals, decodes the PBCH to read the MIB, and from there learns enough to receive system information properly. In FR2 that search is complicated by beams: the cell sweeps its SSB across directions in turn, and the device finds whichever one reaches it.

Once connected, the link is maintained by scheduling — the base station deciding, every slot, which devices get which resources — and repaired by HARQ, which combines failed transmissions with their retransmissions rather than discarding them.

Antennas: where the capacity actually comes from

The largest real-world gain in 5G is spatial. Massive MIMO puts enough antenna elements at the base station to serve several devices on the same time-frequency resource, separated only by their spatial signatures. That is the capacity multiplier; the modulation improvements are rounding errors by comparison.

Beamforming is the related but distinct idea of concentrating energy toward one receiver instead of radiating it everywhere. At FR2 it is not an enhancement — without it the link budget does not close. This is why beam management, beam failure detection and recovery occupy so much of the FR2 specification.

The core: from boxes to services

The 5G core is a genuine architectural break, not an upgrade. Where the 4G core had named network elements with fixed interfaces between them, the 5G core has network functions that register themselves and are discovered at runtime, talking over a service-based interface.

The practical consequence is separation of concerns. The AMF handles the device's connection and mobility signalling. The SMF decides what a data session looks like. The UPF actually forwards packets, and can be placed near the edge without moving the control logic with it — which is what makes low-latency edge deployments tractable.

A device's data connectivity is carried in a PDU session, and within that session traffic is sorted into QoS flows that carry the network's treatment policy. Network slicing sits above all of it: several logical networks over one physical one, each with its own selection, isolation and policy.

RAN architecture and how it gets deployed

The base station is the gNB, and unlike its predecessors it is not necessarily one box. The central unit and distributed unit can be separated, and the split point determines what runs where and how much transport capacity the connection between them needs. That split is the hinge the entire Open RAN discussion turns on.

Deployment adds another axis. Non-standalone reuses the LTE anchor and the 4G core; standalone runs the 5G core natively. Dual connectivity is the mechanism underneath — a device maintaining links to two nodes at once — and it appears in both.

Beyond the mainstream case, 5G stretches in two directions worth knowing about: RedCap trims the device requirements for sensors and wearables that do not need full 5G capability, and non-terrestrial networks adapt the timing and mobility machinery for satellites whose propagation delay dwarfs anything terrestrial.

How to read this section

If you are new to the subject, the air-interface articles repay being read in order — numerology, then frames and slots, then resource blocks and bandwidth parts. Each builds on the vocabulary of the last, and the scheduling and MIMO material assumes all three.

If you already work in the field and arrived looking for one specific answer, the articles are written to stand alone; each opens with the problem it solves rather than with background you already have.