NR Beam Management: What Happens When the Beam Misses

Below about 6 GHz, a cell is a place. A UE inside it has a link, and the link degrades gradually as the UE moves away.

At FR2 that stops being true. Beamforming is not an enhancement there — the link budget does not close without it — which means the cell is not a place but a direction, and a UE that is well inside the coverage area can lose its link entirely because the beam is no longer pointed at it. A hand over the antenna, a bus passing, a turn of the wrist: none of these change where the UE is, and all of them can drop the link.

Beam management is the set of procedures for keeping a direction aligned, and beam failure recovery is what happens when alignment is lost faster than the tracking can follow.

---

Three stages, coarse to fine

The specification does not name P1, P2 and P3 — those labels come from the study item and stuck because they are useful. They describe three procedures that narrow a search.

P1 — find any usable direction. The gNB sweeps wide beams, one per SSB, across the whole cell. The UE measures them and reports the best. This is the same sweep a UE uses during initial access, so a UE that has camped on the cell has already done P1 without anything extra being configured.

The beams here are deliberately wide. A wide beam has less gain, but sweeping the cell with narrow ones would take far too long, and P1's job is only to get into the right neighbourhood.

P2 — refine the gNB's beam. Having established roughly where the UE is, the gNB transmits a set of narrower CSI-RS beams around the P1 winner and the UE reports the best of those. The expensive, narrow search runs only in the direction the cheap search already identified.

P3 — refine the UE's beam. The gNB holds its beam fixed and the UE sweeps its own receive beams, measuring the same transmission through each. This is the stage people forget, and it only exists because at FR2 the UE beamforms too — it has an array, and its own pointing is half the alignment problem.

The split is a search-cost argument. One narrow sweep across the whole cell in both dimensions is a product of two large numbers. Coarse-then-fine turns it into a sum of small ones.

None of this is a handover

Worth stating explicitly, because the vocabulary invites the confusion: throughout P1, P2 and P3 the serving cell never changes. The UE keeps its C-RNTI, its RRC connection, its bearers. What changes is which spatial filter the two ends use to talk to each other.

This is why beam management sits in the physical layer and MAC rather than in RRC mobility, and why a beam switch costs microseconds where a handover costs tens of milliseconds. Confusing the two leads to reading beam-switch counters as mobility problems.

TCI states: how the UE knows where to listen

The mechanism connecting all this is the transmission configuration indication. A TCI state tells the UE that a given downlink transmission is quasi-co-located with a particular reference signal — meaning it can use the spatial receive filter it derived from that reference for this transmission too.

In plain terms: receive this the same way you received that. The gNB is telling the UE which of its own beams to point.

This is where beam management meets antenna ports and quasi co-location, and the QCL type matters. Type D is the spatial one — the one that carries "use this receive beam". The others carry Doppler and delay properties. A configuration that gets the QCL type wrong produces a UE listening in the right direction with the wrong assumptions, or the reverse.

Beam failure recovery

Tracking works while the change is gradual. Blockage is not gradual — at millimetre wave, a body between the UE and the gNB can take 20 dB out of the link in the time it takes to walk past.

The procedure has four parts and each is designed to be fast:

Detection. The UE monitors a configured set of reference signals and counts how often the estimated block error rate on the control channel exceeds a threshold. Enough consecutive failures within a timer and the UE declares beam failure. It is deliberately not a single-sample decision — a momentary dip should not trigger a recovery.

Candidate selection. The UE consults a list of candidate beams configured in advance and measures them for one above threshold. Because the list is pre-configured, this is a short check rather than a full sweep. If the network has not configured candidates, the UE has nowhere to go and the failure escalates.

The recovery request. The UE transmits on PRACH using a contention-free preamble associated with the candidate beam it chose. This is the elegant part: the choice of preamble and occasion identifies both the UE and the beam it wants, in a transmission that needs no grant. The random access machinery is reused here precisely because it is the one channel a UE can use without being scheduled.

The response. The gNB replies on the new beam. The link is restored with no RRC re-establishment, no context loss, and no interruption above the MAC layer.

If recovery fails — no candidate above threshold, or no response within the window — then the failure escalates to radio link failure and the UE goes to RRC re-establishment, which is expensive and visible to the user.

Where the configuration goes wrong

No candidate beams configured. Beam failure detection without a candidate list means the UE detects the failure and then has to fall back to full recovery. The fast path exists only if somebody populated the list.

Detection thresholds tuned for FR1 habits. Timers and counters sized for gradual fading are too slow for blockage. The UE spends the interval declaring nothing while the link is already unusable.

Too many candidates. The check is short because the list is short. A long list turns a fast confirmation back into a search.

Reading beam switches as mobility. A high beam-switch rate in a busy indoor FR2 cell is normal — it is people moving between the UE and the gNB, not the UE moving between cells. The counters that matter are recovery failures and escalations to radio link failure, not switches.

---

The mental model

At FR2 a cell is a direction, not a place. A link can fail without the UE moving.

P1 coarse, P2 refines the gNB beam, P3 refines the UE beam. Coarse-then-fine turns a product into a sum.

P3 exists because the UE beamforms too. Alignment is a two-ended problem.

A TCI state says "receive this the way you received that." QCL type D is the spatial one.

Beam failure recovery reuses PRACH because it is the only channel a UE can transmit on without a grant — and the preamble choice names the beam it wants.

None of it is a handover. The cell never changes; only the direction does.

Further reading

• 3GPP TS 38.213 §6 — beam failure detection and recovery procedures
• 3GPP TS 38.214 §5.1.6 — TCI states and quasi co-location assumptions
• 3GPP TS 38.321 — beam failure recovery MAC procedures and timers
• 3GPP TS 38.331 — BeamFailureRecoveryConfig and the candidate beam list
• ShareTechnote's 5G handbook, for signalling traces and configuration examples across these procedures