A generation is a contract, not a magic year
You already know the shape of a cellular network from the last guide: the land is carved into cells, each served by a base station, and behind them sits a core network that authenticates you, tracks where you are, and stitches you onto the Internet. A generation — 3G, 4G, 5G — is not a date or a faster phone. It is a fresh edition of the rulebook that the base stations and the core agree to speak, standardised so that any vendor's tower can talk to any vendor's phone. When the rulebook changes enough, the marketers paint a bigger number on it.
Each leap really only pulls on a few physical levers, so it helps to name them up front. First, spectrum: how much radio bandwidth you are allowed to use, and at what frequencies. Second, the air interface: how cleverly the radio packs bits into that spectrum, fighting the attenuation, interference, and multipath you met in the first guide of this rung. Third, the core: whether it is built from rigid telephone-era boxes or from flexible software. Watch those three levers and the whole 3G-to-5G story stops being mysterious — it becomes a sequence of sensible trade-offs.
3G to 4G: the network becomes the Internet
3G (around the 2000s) was the generation that made mobile data genuinely usable — email, maps, the early mobile web — but it grew out of a voice phone system. Picture the network as a telephone exchange that bolted on a data side-channel. Speeds were a few megabits per second at best, and the architecture still thought in terms of phone calls. It was good enough for browsing, painful for video.
4G, whose dominant flavour is LTE (Long-Term Evolution), made one decisive philosophical move: it threw out the circuit-switched voice machinery and declared the whole network to be all-IP. Everything — your video, your web pages, and now even your phone calls — is just packets. On top of that it pulled the first two levers hard. A smarter air interface (OFDMA, which slices the spectrum into many narrow sub-carriers that dodge multipath gracefully) and multiple antennas (MIMO, sending several data streams in parallel through the same air) pushed real-world speeds to tens or even hundreds of megabits per second. This is the generation that made on-the-go video streaming feel ordinary.
Two honest cautions. First, those headline speeds are peak, lab-flavoured numbers shared across everyone in the cell — your actual throughput is whatever is left after the tower divides its capacity among all the phones it is serving, just like a single home pipe split among roommates. Second, all this extra speed does almost nothing for latency. Adding lanes to a motorway lets more cars through, but each car still takes the same time to drive across town. We will come back to this, because it is exactly where 5G's marketing gets slippery.
What 5G actually changes
5G is best understood not as one thing but as three promises aimed at three very different customers — the standards bodies literally gave them three names. Keep them separate and the hype dissolves into engineering.
5G's three faces (one network, three service profiles):
eMBB Enhanced Mobile Broadband
-> the speed story: more spectrum + better radios
-> who cares: you, watching 4K on the train
URLLC Ultra-Reliable Low-Latency Communication
-> the latency story: aim for a few milliseconds, rarely dropped
-> who cares: factory robots, remote control, cars
mMTC Massive Machine-Type Communication
-> the density story: ~1 million sleepy devices per km^2
-> who cares: sensors, meters, the Internet of ThingsThe speed face (eMBB) is the one you feel, and it is mostly more of the same levers turned harder: much wider spectrum (including high-frequency millimetre-wave bands), denser antenna arrays (massive MIMO with dozens of elements), and beamforming, where the tower aims a focused beam of radio straight at your phone instead of flooding the whole cell. Millimetre wave can deliver gigabit-per-second speeds, but honestly: those high frequencies barely pass through walls and fade over a block or two, so they only work in dense hotspots with towers on nearly every corner. Most 5G you actually use runs on lower bands that feel like a brisk 4G, not a revolution.
URLLC and slicing: the genuinely new parts
The genuinely new face is URLLC: ultra-reliable, low-latency communication. This is the part that is not just a faster 4G, because it targets a dimension 4G barely touched — latency, the time for one round trip, which you learned cannot be cheated past the speed of light. 5G attacks the parts of delay it CAN control: shorter radio time-slots, scheduling that does not make a message queue, and — crucially — moving the servers physically closer. That last move is edge computing: instead of every request travelling to a distant data centre, a small server sits right at the base station, so the round trip is metres instead of thousands of kilometres. This is what makes a few-millisecond response plausible for a robot arm or a self-driving car.
Holding all three faces on one network needs the third lever: a software-built core. 5G's core is virtualised — its functions run as software on ordinary servers rather than as fixed telephone boxes — which unlocks network slicing. Slicing carves the one physical network into several logical networks, each tuned to a different promise: a low-latency slice for the factory, a high-throughput slice for video, a low-power slice for a million sensors, all sharing the same towers. Think of one motorway repainted into a fast lane, a freight lane, and a bicycle lane — same tarmac, different guarantees.
Cutting through the hype, honestly
Now the honest accounting, because this subject is drowning in salesmanship. The most common confusion is treating the speed number as a latency number. They are different things, as the very first networking guides hammered home: bandwidth is how wide the pipe is, throughput is how much you actually push through it, and latency is how long one trip takes. A 5G phone showing a huge download speed can still feel sluggish when tapping a button, because that delay is dominated by distance and round trips, not by pipe width. More bandwidth does not, and physically cannot, cut the speed-of-light floor.
Two more myths worth puncturing. 5G is not a single switch you flip; for years it has run in non-standalone mode, leaning on existing 4G infrastructure for the control work while only the data ride is upgraded — so an early 5G phone was often a 4G network wearing a 5G badge. And the millimetre-wave coverage problem is real and permanent physics: those high frequencies trade reach for capacity, which is why your phone so often falls back to a lower 5G band or to plain LTE the moment you step indoors or round a corner.
Where this is heading
Step back and the pattern across generations is steady, not magical. 3G made data real; 4G made the whole network an IP network and made data fast; 5G widened the pipe again, but its truly new ambition is reliability and low latency for machines, delivered by slicing one software core into many tailored networks and by pushing computation out to the edge. Each generation pulled the same handful of levers — spectrum, air interface, core architecture — a little further.
The same lens already frames what comes next. Research toward 6G talks about even higher frequencies, tighter integration of sensing and communication, and weaving in satellite links so coverage reaches places no tower ever will. Treat those claims the way you now treat 5G: ask which lever moves, which face it serves, and whether the physics — especially the unbeatable speed-of-light latency floor — actually allows the promise. You are no longer a spectator to the hype; you can audit it.
One thread we have quietly set aside: all of this assumes you sit still in one cell. The moment you move — between cells, between towers, even between 5G and 4G as you round that corner — the network must hand you off without dropping your connection, and the loss and delay swings of a wireless link play havoc with TCP, which (as you will recall) reads any packet loss as congestion. Mobility, handoff, and why TCP gets confused on the move are exactly the subject of the final guide in this rung.