Carrier Aggregation in 5G NR: PCell, SCells, and Why Throughput Doesn't Just Add Up

Spectrum is allocated in fragments. An operator might hold 20 MHz at 700 MHz, 40 MHz at 1800 MHz, and 100 MHz at 3.5 GHz — bought at different auctions, in different bands, years apart.

Carrier aggregation lets a single UE use several of those fragments simultaneously as one logical connection.

Each aggregated carrier is a component carrier (CC). NR supports up to 16, though real deployments and real devices use far fewer. The UE sees one connection; the MAC layer sees several carriers to schedule across.

---

PCell and SCell

The carriers aren't equals, and the asymmetry matters.

The PCell (Primary Cell) is the anchor. It carries the RRC connection, handles the initial random access, maintains security context, and carries essential control signalling. There is exactly one, and losing it means losing the connection — a radio link failure on the PCell triggers re-establishment regardless of how healthy the SCells are.

SCells (Secondary Cells) add capacity. They're added and removed by RRC reconfiguration, carry user data, and can be activated or deactivated by MAC control element — a much faster operation than RRC signalling.

That two-level distinction between configured and activated is worth internalising:

Configured means the UE knows the SCell's parameters but isn't monitoring it. Activated means the UE is measuring CSI and can be scheduled on it.

Activation costs power — the UE must run additional RF chains and report channel state. So the network keeps SCells configured but deactivated during light traffic and activates them when there's enough data to justify it. Activation takes a few milliseconds; configuration takes an RRC round trip. Keeping carriers configured-but-dormant is what makes fast capacity ramp-up possible.

---

The three layouts

Intra-band contiguous

Component carriers sit adjacent within the same band.

The easiest case by a wide margin. One RF chain, one local oscillator, one power amplifier can cover the whole aggregate. Device complexity barely increases over a single wide carrier.

Intra-band non-contiguous

Same band, but with a gap between the carriers — usually because another operator holds the spectrum in between.

Harder. The RF front end must handle a wider span, and filtering gets more demanding because there's a neighbour's transmission sitting in the middle.

Inter-band

Component carriers in different bands — the combination that matters most in practice.

This is the one worth understanding properly, because it does something the others can't: it combines coverage with capacity.

A low-band carrier at 700 MHz propagates well and penetrates buildings but offers little bandwidth. A mid-band carrier at 3.5 GHz offers 100 MHz but has poor edge coverage. Aggregating them means a UE at the cell edge keeps a reliable low-band anchor while opportunistically using mid-band capacity whenever conditions allow.

The cost is RF complexity: separate chains, separate filters, separate antennas, plus intermodulation and harmonic interference between bands that must be designed out. Which combinations a device supports is a hardware fact, not a software one.

---

Where the throughput actually goes

The naive expectation is that aggregating 100 MHz and 40 MHz gives you the throughput of 140 MHz. It rarely does, and the reasons are worth knowing.

Carriers experience different channel conditions. A mid-band SCell may be at the edge of its usable range while the low-band PCell has a strong signal. The SCell contributes at a low MCS, not at its theoretical peak.

Each carrier has independent link adaptation and HARQ. The scheduler runs separate MCS selection and separate HARQ processes per carrier — a good PCell doesn't rescue a marginal SCell.

Uplink is asymmetric. Most devices support far fewer uplink CCs than downlink, often just one. Uplink aggregation costs power amplifiers and battery, and the traffic asymmetry rarely justifies it. Downlink 4CC with uplink 1CC is a completely normal configuration.

Higher layers may be the bottleneck. TCP congestion control, server capacity, or transport backhaul frequently limit throughput before the radio does.

SCell activation is not instant. A short transfer may complete before the SCells are usefully contributing.

The realistic framing: carrier aggregation raises the ceiling. Whether you approach it depends on radio conditions, device capability, traffic duration, and everything above the radio layer.

---

What the scheduler does with it

Scheduling happens per carrier, coordinated by one MAC entity.

Each carrier gets its own resource allocation, MCS selection, and HARQ processes. What's shared is the logical channel prioritisation and the overall view of the UE's buffer.

Two configuration details matter operationally:

Cross-carrier scheduling allows the PDCCH on one carrier to schedule data on another. Useful when one carrier suffers control-channel interference, or for scheduling a carrier that doesn't carry its own control region.

PUCCH SCell. By default all HARQ feedback for all carriers returns on the PCell, which can overload its uplink control channel with many aggregated carriers. Configuring a secondary PUCCH group spreads that load — a real consideration once you're past three or four carriers.

---

Activation, deactivation and the timer nobody tunes

Configuring an SCell and using an SCell are different things, and the gap between them is where most of the practical behaviour lives.

An SCell is first configured by RRC — a relatively slow procedure that tells the UE the carrier's parameters. A configured SCell starts deactivated: the UE does not monitor its PDCCH, does not report CQI for it, and does not transmit on it. It costs almost nothing.

Activation happens through a MAC control element, which is fast — a few milliseconds. The scheduler activates an SCell when there is enough buffered data to justify it, and deactivates it when there is not.

Deactivation happens two ways. The network can send an explicit MAC CE, or the sCellDeactivationTimer can expire after a period without scheduling on that SCell. The timer exists because leaving an SCell active costs the UE real power: it must monitor control channels and report channel state on a carrier carrying no data.

The tuning tension is straightforward. A short timer saves battery but means bursty traffic repeatedly pays activation delay. A long timer keeps carriers ready but drains the device. Operators tend to set this once and never revisit it, which is why activation behaviour is a productive place to look when a device shows good peak throughput but poor throughput on short transfers.

Cross-carrier scheduling

By default each carrier schedules itself: the PDCCH on a carrier grants resources on that same carrier. Cross-carrier scheduling breaks that link, allowing one carrier's control channel to grant resources on another.

The reason is interference. In a heterogeneous deployment a small cell's control channel may be swamped by a macro cell on the same frequency, while its data region remains usable. Moving the control channel to a cleaner carrier keeps the data capacity available.

The cost is that the scheduling carrier's PDCCH now carries the control load for several carriers, and the carrier indicator field must be configured so the UE knows which carrier each grant refers to.

---

Carrier aggregation vs dual connectivity

These get confused constantly. The distinguishing question is simple: how many schedulers are involved?

Carrier aggregation — one gNB, one MAC entity, one RRC connection. Carriers must be tightly synchronised because a single scheduler coordinates them all within slot timing.

Dual connectivity — two nodes, two MAC entities, split at PDCP. The nodes need only loose coordination over Xn, so they can be geographically separated with real backhaul latency between them.

CA gives tighter coordination and better aggregate efficiency. DC tolerates non-ideal backhaul and works across separated sites — including across radio technologies, which is what EN-DC does in NSA deployments.

They also compose: a UE can run CA within each of the two DC legs.

---

NR vs LTE-Advanced

The concept is inherited from LTE-Advanced. What changed is scale and flexibility:

The bandwidth jump is the substantial one. Five LTE carriers gave you 100 MHz total. A single NR FR1 carrier gives you that, and aggregating a handful of FR2 carriers reaches into the gigahertz.

Mixed numerology across carriers is genuinely new. A low-band carrier can run 15 kHz subcarrier spacing while a mid-band carrier runs 30 kHz within the same aggregation — the different propagation environments get the numerology each actually needs.

---

Uplink aggregation is rarer than you would expect

Downlink carrier aggregation is ubiquitous. Uplink carrier aggregation is deployed far less often, and the reasons are physical rather than architectural.

Transmitting on two carriers at once means two simultaneous transmit chains in a device with one battery and a strict thermal envelope. It also means the power amplifier must handle a higher peak-to-average ratio, which pushes it away from its efficient operating point.

Worse, simultaneous transmission on two bands generates intermodulation products. Two carriers at frequencies f1 and f2 produce spurious energy at combinations of them, and for certain band pairs those products land squarely in the device's own receive band. The device desensitises its own receiver. 3GPP documents these as specific band-combination restrictions, and they are why an operator's uplink CA support looks arbitrary until you check which combinations are physically viable.

Most deployments therefore aggregate heavily in the downlink and run a single uplink carrier, choosing it for coverage — usually the lowest band available. This asymmetry is why a device may report 1 Gbps down and 50 Mbps up on the same connection.

Aggregating across duplex modes and bands

The permitted combinations are constrained beyond simple physics. A band combination must be explicitly defined by 3GPP, and each device declares which it supports through its capability signalling. A network offering a combination the device has not declared simply will not use it.

This is why capability exchange is worth checking when aggregation is not engaging as expected. The carriers may be present, the coverage adequate, and the scheduler willing — while the device never declared the combination and the network correctly declined to configure it.

---

The mental model

One PCell anchors the connection. Multiple SCells add capacity.

Configured ≠ activated. Activation is a fast MAC operation; configuration is an RRC round trip.

Inter-band CA is the valuable case: it combines low-band coverage with mid-band capacity.

Aggregate throughput ≠ sum of carriers. Each carrier has its own channel, MCS, and HARQ.

Uplink aggregation lags downlink in almost every device.

CA = one scheduler. DC = two. That's the whole distinction.

Carrier aggregation is fundamentally a spectrum-utilisation tool. It exists because spectrum arrives in fragments and the radio has to make one coherent connection out of them — and understanding why the pieces don't simply add up is most of what makes it useful to reason about.

---

Further reading

• 3GPP TS 38.300 — NR overall description, carrier aggregation
• 3GPP TS 38.321 — MAC specification, SCell activation and deactivation
• 3GPP TS 38.331 — RRC specification, SCell configuration
• 3GPP TS 38.101-1 and 38.101-3 — UE radio transmission, supported CA band combinations