Ferkans — Interactive Telecom Tutor

Why Beams Must Be Managed at mmWave

At FR1, a UE without a beam can still close the link — an omnidirectional reference signal at 3.5 GHz reaches several hundred meters indoors and several kilometres outdoors. At FR2, the link budget is completely different: 28 GHz has 20 dB of extra Friis path loss relative to 3.5 GHz, and the only way to recover it is directional gain at both the BS and the UE. A connected mmWave UE is effectively never operating omnidirectionally. This means the BS and UE must first discover the right beam pair before any useful data transfer, then track that beam pair as the UE moves, and recover gracefully when the beam is lost (by pedestrian blockage, rotation, or hand-over).

5G NR codifies these three tasks as procedures P1, P2, P3 (for initial acquisition, BS-side refinement, and UE-side refinement), plus a beam failure recovery procedure. They run on top of the SSB (Synchronization Signal Block) as the broadcast beacon and CSI-RS as the refinement pilot.

Definition:
Synchronization Signal Block (SSB)

The SSB is a 4-symbol, 240-subcarrier block containing the primary synchronization signal (PSS), secondary synchronization signal (SSS), and physical broadcast channel (PBCH). It is used by UEs to detect the cell, acquire timing and frequency, and decode the master information block (MIB). Critically for beam management, the SSB is the beacon: in FR2 the BS transmits multiple SSBs in different directions, and the UE measures each to determine the best beam direction.

The number of SSBs per SS burst set, $N_{\text{SSB}}$ , is

up to 4 below 3 GHz,
up to 8 between 3 and 6 GHz,
up to 64 above 6 GHz (FR2).

Each SSB occupies 4 consecutive OFDM symbols, and consecutive SSBs are time-multiplexed (beam-swept) within a 5-ms SS burst set window. The SS burst set period is configurable at $\{5, 10, 20, 40, 80, 160\}$ ms, with 20 ms as the default for connected-mode beam tracking.

,

Definition:
Beam Management Procedures P1, P2, P3

5G NR specifies three hierarchical beam-management procedures, identified by their use cases rather than by explicit names in the standard:

P1: Initial beam acquisition. The BS sweeps through $N_{\text{SSB}}$ wide beams covering its angular sector. The UE measures each SSB's received power (RSRP) and selects the strongest. P1 runs at RRC-connection setup and again whenever the UE re-enters RRC- CONNECTED mode. Sweep duration: $T_{\text{sweep}} = N_{\text{SSB}} \cdot 4 / \Delta f$ .
P2: BS-side beam refinement. Once P1 has selected a wide beam, the BS transmits a sequence of CSI-RS resources, each on a narrower beam within the wide-beam angular range. The UE reports which narrow beam is best. P2 runs periodically in connected mode at 10-40 ms cadence, allowing the BS to track slow UE movement.
P3: UE-side beam refinement. The BS fixes its narrow beam and sends multiple repeated CSI-RS, while the UE sweeps its own receive beam across the refined angular range. The UE selects its best receive beam without reporting back. P3 handles UE rotation (handheld devices) and is the only procedure where the UE's receive codebook matters.

,

Theorem: Beam-Sweep Latency

For an FR2 cell with $N_{\text{SSB}}$ SSBs transmitted sequentially, each occupying 4 OFDM symbols at numerology $\mu$ , the P1 beam discovery latency (ignoring inter-SSB gaps) is $T_{\text{P1}} = \frac{4 N_{\text{SSB}}}{\Delta f(\mu)} = \frac{4 N_{\text{SSB}}}{15\,\text{kHz}\,\cdot 2^{\mu}}.$ For $N_{\text{SSB}} = 64$ at $\mu = 3$ (typical FR2 configuration), $T_{\text{P1}} = 256 / 120\,\text{kHz} \approx 2.13$ ms. Adding gaps and the 5-ms SS burst window, a complete P1 in practice takes about 5 ms once per SS burst period (e.g., 20 ms).

The beam sweep is inherently serial: the BS can transmit only one beam at a time, so $N_{\text{SSB}}$ beams take $N_{\text{SSB}}$ SSBs worth of time. Numerology shortens each SSB but cannot parallelize the sweep — so P1 latency is fundamentally limited by the product of beam count and symbol duration.

Show Hint

Compute OFDM symbol duration from $\Delta f = 15\,\text{kHz}\cdot 2^{\mu}$ .

Multiply by 4 (symbols per SSB) and $N_{\text{SSB}}$ .

Remember SSBs are time-multiplexed, not frequency-multiplexed.

Proof

Symbol duration

$T_{\text{OFDM}}(\mu) = 1/\Delta f(\mu)$ (ignoring cyclic prefix). At $\mu = 3$ , $T_{\text{OFDM}} = 1/(120\,\text{kHz}) \approx 8.33$ µs.

One SSB

Each SSB is 4 symbols: $T_{\text{SSB}} = 4 T_{\text{OFDM}} \approx 33.3$ µs at $\mu = 3$ .

Full sweep

$T_{\text{P1}} = N_{\text{SSB}} \cdot T_{\text{SSB}} = 64 \cdot 33.3\,\mu s \approx 2.13$ ms at $\mu = 3$ . The actual SS burst window is padded to 5 ms to match the frame structure. $\blacksquare$

Beam Sweep Latency: P1/P2/P3 Budget

Beam sweep latency for P1, P2, and P3 procedures as functions of the number of beams and the NR numerology. Total end-to-end acquisition time (P1+P2+P3) is highlighted against a typical handover latency budget of 10-20 ms.

Parameters

N_{\text{SSB}}

32

P2 refinement beams8

P3 UE-side beams4

Numerology

\mu

3

Definition:
Beam Correspondence

Beam correspondence is the property that a UE's best receive beam for a given BS transmit beam equals its best transmit beam for the same channel — i.e., the UE's beam is the same in both directions up to a reciprocity assumption. When beam correspondence holds, the UE can use the beam it discovered during P3 (a downlink measurement) for its own uplink transmissions, and the BS is guaranteed that the uplink SRS arrives from the same spatial direction as the downlink signal.

Beam correspondence is assumed for simple UEs in Rel-15, but it is not mandatory — a UE is allowed to declare its beam-correspondence capability. When the UE lacks it, the BS must run additional P3-style refinement on the uplink to find the right beam pair, doubling the beam-management overhead. Beam correspondence is a UE-implementation quality metric.

Example: FR2 Beam Management at 28 GHz

An FR2 gNB at $f_0 = 28$ GHz uses numerology $\mu = 3$ , $N_{\text{SSB}} = 32$ wide beams (coverage), $N_{\text{P2}} = 8$ narrow beams per wide beam, and SS burst period $20$ ms. Compute the P1 and P2 latency and the total beam-management overhead as a fraction of the 20 ms cycle.

Solution

P1 latency

At $\mu = 3$ , $T_{\text{OFDM}} \approx 8.3$ µs. One SSB is $4 T_{\text{OFDM}} \approx 33.3$ µs. A complete P1 sweep is $32 \times 33.3\,\mu s \approx 1.07$ ms, which fits in the 5-ms SS burst window.

P2 latency

P2 uses CSI-RS resources, not SSB. With 8 narrow beams at 4 symbols each: $T_{\text{P2}} = 8 \cdot 33.3\,\mu s \approx 266\,\mu s$ at $\mu = 3$ . Performed once per 10-20 ms in connected mode.

Total fraction

P1 fraction of 20 ms cycle: $1.07/20 \approx 5.35\%$ . P2 fraction (if once per 20 ms): $0.27/20 \approx 1.3\%$ . Together, beam management costs about $7\%$ of the slot budget — a substantial but manageable overhead. In practice, P1 runs less frequently than once per SS burst period when the UE is stationary, so the average overhead is 2-3%. $\blacksquare$

Definition:
Beam Failure Recovery

Beam failure is declared by the UE when the RSRP of all configured beam-reference resources falls below a threshold for $N$ consecutive measurement instances. This typically indicates a pedestrian blockage, a hand-over, or a rapid UE rotation.

On detecting failure, the UE initiates the beam failure recovery (BFR) procedure: it transmits a dedicated PRACH (physical random access channel) preamble on a contention-free resource associated with a candidate beam, signaling the gNB that beam reacquisition is needed. The gNB responds with a new beam assignment. Total BFR latency is 10-30 ms for FR2, which determines the minimum blockage duration the system can gracefully handle without dropping the call.

Beam Tracking in Mobility: Combined P1-P2-Recovery Loop

Complexity:

O(1)

per measurement instance;

O(N_{\text{SSB}} + N_{\text{P2}})

per BFR event.

Input: Cell beam codebook

\mathcal{B}_{\text{wide}}

(for P1, size

N_{\text{SSB}}

), narrow codebook

\mathcal{B}_{\text{narrow}}

(for P2, size

N_{\text{P2}}

),

failure threshold

\theta

, counter limit

N_{\max}

.

State: current best wide beam

b_w^\star

, current best narrow

beam

b_n^\star

, failure counter

n_{\text{fail}}

.

1. Initial: Perform P1 sweep over

\mathcal{B}_{\text{wide}}

, select

b_w^\star

.

2. Perform P2 sweep within

b_w^\star

over

\mathcal{B}_{\text{narrow}}

, select

b_n^\star

.

3. Repeat every SS burst period:

4.

\quad

Measure RSRP on

b_n^\star

and 2 nearest narrow neighbors.

5.

\quad

If RSRP

> \theta

and best neighbor is current then

6.

\qquad

n_{\text{fail}} \leftarrow 0

; (no change needed).

7.

\quad

else if neighbor beam RSRP is best then

8.

\qquad

b_n^\star \leftarrow

best neighbor;

n_{\text{fail}} \leftarrow 0

.

9.

\quad

else (RSRP

\leq \theta

)

10.

\qquad

n_{\text{fail}} \leftarrow n_{\text{fail}} + 1

.

11.

\qquad

If

n_{\text{fail}} \geq N_{\max}

then

12.

\quad\qquad

Declare beam failure, initiate BFR PRACH.

13.

\quad\qquad

Re-run P1 sweep, set

b_w^\star

to winner.

14.

\quad\qquad

Re-run P2 to set

b_n^\star

.

15.

\quad\qquad

n_{\text{fail}} \leftarrow 0

.

16. end repeat

The tracking loop trades latency against robustness: a small $N_{\max}$ recovers quickly but triggers BFR on transient shadowing; a large $N_{\max}$ is robust to transient dips but loses time on true blockages. Commercial UEs use $N_{\max} \in \{2, 3, 4\}$ with $\theta$ set 3 dB below the nominal link margin.

,

🚨Critical Engineering Note

Beam Tracking in High Mobility

At vehicular speeds, the beam coherence time can drop below the SS burst period. For a UE moving at 60 km/h past a gNB at 50 m range with 5-degree beamwidth, the angular rate is $\dot\theta = v/r \approx 0.33$ rad/s $\approx 19$ deg/s, which traverses a 5-degree beam in $\approx 260$ ms. If the SS burst period is 20 ms, the system sees the beam slide by roughly 2% per measurement, which is well within tracking limits. At 300 km/h (high-speed rail) past a 30-m gNB with 3-degree beamwidth, the beam traverses in $\approx 17$ ms — marginal. The practical design rule is that $T_{\text{SSB}} < 0.2 \cdot T_{\text{beam}}$ , where $T_{\text{beam}}$ is the time for the UE to cross one beam.

Practical Constraints

•
Beam tracking requires $T_{\text{SSB}} < 0.2 \cdot T_{\text{beam}}$
•
High-speed rail (300 km/h) at FR2 pushes P1 + P2 cadence to 10 ms
•
Handover between gNBs must complete within one beam dwell time

📋 Ref: 3GPP TS 38.213 Section 5.1

P1, P2, P3 Beam Sweep Procedures — Hierarchical beam refinement in 5G NR: P1 sweeps the BS wide beams (broad angular coverage); P2 refines within the selected wide beam to a narrow BS beam; P3 holds the BS beam fixed and lets the UE sweep its own receive beam. Each stage narrows the angular uncertainty by roughly an order of magnitude.

Common Mistake: Omnidirectional Coverage Assumption at FR2

Mistake:

A common mistake when designing FR2 protocols is to assume that a UE can momentarily operate in omnidirectional (no-beam) mode for control channel reception, the way LTE UEs can.

Correction:

At mmWave, an unbeamformed reception pattern is 15-25 dB worse than a beamformed one. All FR2 control channels (SSB, PDCCH, PUCCH, PRACH for BFR) must be received through a specific beam. The UE must always have a best-guess receive beam, even during initial acquisition — P1 is essentially the UE trying its default (typically broadside) beam first and then rotating as needed. The notion of a "no-beam" state does not exist in operational FR2.

Historical Note: Beam Management: From 802.11ad to 5G NR

2012-2018

Hierarchical beam sweep procedures predate 5G NR: the 802.11ad standard (2012) already defined a sector-level-sweep (SLS) followed by a beam refinement phase (BRP) for 60 GHz Wi-Fi. The 3GPP design borrowed the two-level structure but adapted it to the SSB/CSI-RS framework and added the P3 UE-side refinement as a response to handheld-device rotation. The SLS-BRP-P3 correspondence was made explicit in the 2017 RAN1 "beam management for NR" study item, which cited the 802.11ad experience as evidence that sector-level sweeping was operationally feasible.

Why This Matters: Beams as a Sensing Resource

The same SSB sweep that acquires a UE beam also produces a coarse angular scan of the cell. Rel-18 and beyond propose reusing this scan for joint communication and sensing (ISAC — Chapter 24): the RSRP measurements gathered during P1 can map not only the UE location but also fixed scatterers, moving objects, and blockages. This turns the beam management overhead into dual-purpose sensing data, closing a link between Chapter 22 and Chapter 24.

SSB (Synchronization Signal Block)

A 4-symbol, 240-subcarrier block containing PSS, SSS, and PBCH. Used for cell detection and, at FR2, as the beacon for P1 beam sweep. Multiple SSBs per SS burst set are beam-swept by the BS.

Beam Failure Recovery (BFR)

Procedure invoked by the UE when all configured beams fall below a threshold RSRP. The UE sends a contention-free PRACH preamble on a beam-associated resource to request beam reassignment.

Related: Beam Management, Prach

RSRP (Reference Signal Received Power)

The linear average of the power of resource elements carrying the reference signal (SSB or CSI-RS). Used for beam selection, handover decisions, and beam-failure detection.

Quick Check

At $\mu = 3$ ( $\Delta f = 120$ kHz), how long does a P1 sweep of $N_{\text{SSB}} = 32$ beams take, ignoring SS burst gaps?

$\approx 1.07$ ms

$\approx 2.13$ ms

$\approx 5$ ms

$\approx 0.5$ ms