Beam Failure Recovery (MAC) in 5G NR

1. Why Beams Fail, and Why That Is Not a Cell Failure

At FR1 a cell is roughly a place. At FR2 it is a direction. Path loss at 28 GHz is high enough that neither end can afford an omnidirectional pattern, so both concentrate their energy into narrow beams and the link exists only while those two beams point at each other. That arrangement buys the link budget that makes mmWave work at all, and it introduces a failure mode that FR1 barely has: the beam pair can break while everything else — the cell, the security context, the bearers, the C-RNTI — remains completely valid.

Beam Failure Recovery exists because the alternative is absurd. Treating a blocked beam as a radio link failure would mean tearing down and rebuilding a connection, several seconds of work, because someone put a hand over an antenna for eighty milliseconds. BFR is a local repair: it changes which beam the PDCCH arrives on and changes nothing else.

Table 1. Why beam pairs break. The middle two rows are the reason beamFailureInstanceMaxCount is tunable at all: recovery is cheap but not free, and a transient should be allowed to heal.

2. Beam Management in One Page: CORESETs, TCI States, Beam Pairs

To read a BFR log you need to know exactly what "the beam" means to the protocol, because the protocol never uses the word. What it configures is a chain:

• A CORESET is where the UE looks for PDCCH — a set of PRBs and symbols. The companion 08 Scheduling document covers the structure.
• Each CORESET has an active TCI state (Transmission Configuration Indication), which lists one or two reference signals and the type of quasi-colocation the UE may assume with each.
• QCL Type-D is the spatial one: it says "receive this CORESET with the same spatial filter you used to receive that reference signal". The RS named in the Type-D relationship is, operationally, the beam.
• So the PDCCH beam is the QCL Type-D source RS of the TCI state of the CORESET. A beam failure is the statement that those RSs have all become undecodable.

3. Three Nested Timescales: Beam, Cell, Connection

NR monitors the downlink at three levels at once, with three separate counters, three separate timers and three separate recovery procedures. They run concurrently on the same reference signals, and each is designed to be pre-empted by the faster one above it. Almost every confusing BFR log makes sense once you place each line at the right level.

Table 2. The three levels. Note that only the beam level has an active recovery procedure that the UE initiates; the cell level recovers passively or not at all.

4. Beam Failure Detection: Which Reference Signals

Everything downstream depends on the BFD-RS set — the set of reference signals whose collective failure constitutes beam failure. There are two ways the UE ends up with one.

Table 3. Implicit and explicit BFD-RS derivation, TS 38.213 cl. 6. The two failure signatures at the bottom are mirror images and are the fastest way to tell which mode a UE is in when the configuration is not in front of you.

Two exclusions matter. CORESET 0 does not participate, because it is not associated with a UE-specific TCI state. And the CORESET carrying the recovery search space is excluded — including it would make the recovery path depend on the beam whose failure it is meant to repair. The number of RSs the UE must handle is capped, and the cap scales with the maximum number of SSBs in a half frame TS 38.213 cl. 5; maxNrofFailureDetectionResources bounds the ASN.1 list at 10.

5. RLM and BFD: the Same Signals, Different Thresholds

Both radio link monitoring and beam failure detection ask the physical layer the same question — would a PDCCH sent to me right now decode? — and both answer it with a hypothetical PDCCH block error rate rather than a measured power. The UE takes the SINR it estimates from the reference signal, maps it through a reference PDCCH configuration defined in TS 38.133 (a specified DCI size, aggregation level and CCE count), and compares the resulting BLER against a threshold. It is a predicted decoding outcome, not an RSRP.

Table 4. The two mechanisms compared. The single most useful line is the last: BFD does something, RLM only counts.

6. The BFI Counter and beamFailureDetectionTimer

MAC does the counting TS 38.321 cl. 5.17. The rules are four lines long and the third one is the one that gets misremembered:

on receiving a beam failure instance indication from lower layers:
    start or restart beamFailureDetectionTimer
    BFI_COUNTER := BFI_COUNTER + 1
    if BFI_COUNTER >= beamFailureInstanceMaxCount:
        if this Serving Cell is an SCell:  trigger BFR for that SCell
        else:                              initiate Random Access on the SpCell

on beamFailureDetectionTimer expiry:
    BFI_COUNTER := 0

on reconfiguration by upper layers of beamFailureDetectionTimer,
beamFailureInstanceMaxCount, or any BFD reference signal:
    BFI_COUNTER := 0

on successful completion of the Random Access procedure (SpCell):
    BFI_COUNTER := 0
    stop beamFailureRecoveryTimer, if configured
    consider Beam Failure Recovery successfully completed

Sizing is a straight trade. Lowering beamFailureInstanceMaxCount from n4 to n2 halves detection time and doubles the rate of recovery attempts triggered by transients that would have cleared on their own — each of which costs a preamble, a response, and a beam switch that must then be switched back. Shortening the BFD-RS periodicity is the better lever when it is available, because it improves detection time without weakening the filter.

7. Candidate Beam Identification

Detecting failure is useless without somewhere to go. candidateBeamRSList in BeamFailureRecoveryConfig is the pre-agreed answer: a list of SSBs and CSI-RS resources, each already bound to a dedicated preamble index and a set of RACH occasions. That binding is what makes recovery fast and contention-free — the UE does not have to describe the beam it wants, because the preamble it sends says which one.

On beam failure, the UE evaluates L1-RSRP for the entries in the list and picks one whose L1-RSRP exceeds rsrp-ThresholdSSB TS 38.213 cl. 6. If several qualify, the choice is up to the UE. If none does, there is no contention-free option and the UE falls through to an ordinary contention-based Random Access using rach-ConfigCommon.

Table 5. What the candidate search produces. Only the first row is the procedure BFR was designed around; the other three are degraded modes that still work.

8. The Recovery Request: Contention-Free RACH on the Chosen Beam

For an SpCell, the recovery request is a Random Access procedure — the same procedure as initial access, entered from a different trigger and with the ambiguity removed. BeamFailureRecoveryConfig supplies its own rach-ConfigBFR (a RACH-ConfigGeneric: preamble format, power ramping, preambleTransMax, response window), its own rootSequenceIndex-BFR, and optionally its own ra-Prioritization so BFR preambles can ramp faster and retry sooner than ordinary access.

Three details are worth pinning down because they are what distinguishes BFR RACH from any other RACH in a log:

• The preamble is drawn from the CFRA pool, not the contention-based one. It is bound to a specific candidate RS by PRACH-ResourceDedicatedBFR, so detecting it tells the gNB the answer to both questions at once.
• The UE keeps its C-RNTI. There is no TC-RNTI and no promotion; the UE was and remains a known, connected UE. This is the clearest way to tell BFR RACH from re-establishment RACH in a trace.
• The response arrives on a different search space. recoverySearchSpaceId names a search space whose CORESET the gNB will point at the new beam — see §9.

9. The Response, and What the UE Assumes Afterwards

There is no BeamFailureRecoveryResponse message. The response is an ordinary PDCCH whose CRC is scrambled with the UE's existing C-RNTI, received in the search space given by recoverySearchSpaceId, within a window that opens a few slots after the preamble TS 38.213 cl. 6. Receiving it is the confirmation, and MAC then treats the Random Access procedure as successfully completed, clears BFI_COUNTER, and stops beamFailureRecoveryTimer.

The elegant part is what the UE assumes about the beam. Until it is given an explicit TCI state update, the UE assumes the CORESET associated with recoverySearchSpaceId is quasi-colocated with the candidate RS it selected TS 38.213 cl. 6. The gNB knows which candidate that was, because it knows which preamble it detected. So both ends switch to the new beam with no signalling at all — the beam change is carried entirely by the choice of preamble.

BFI_COUNTER

beamFailureRecoveryTimer

candidateBeamRSList

Table 6. The UE's state after recovery. Row seven is the one that produces confused traces: recovery is complete, yet most of the UE's PDCCH configuration is still pointing at a dead beam.

10. When Recovery Itself Fails: the Route to RLF

beamFailureRecoveryTimer is widely described as the timer whose expiry causes radio link failure. It is not. Its documented effect TS 38.331 is narrower and more useful to know: on expiry, the UE stops using the contention-free BFR resources. The dedicated preambles and their beam bindings are no longer available, and any further recovery attempt is an ordinary contention-based Random Access.

The route to RLF runs through the RACH, not through this timer:

1. The UE keeps transmitting preambles, ramping power, up to preambleTransMax from rach-ConfigBFR (or from rach-ConfigCommon once the recovery timer has expired).

2. On reaching preambleTransMax, MAC issues a Random Access problem indication to RRC TS 38.321 cl. 5.1.4. MAC itself never declares a link failure.

3. RRC interprets that indication according to what triggered the RACH. For a BFR-triggered Random Access on the SpCell in RRC_CONNECTED, the interpretation is radio link failure.

4. RLF then runs its own course: T311, cell selection, RRCReestablishmentRequest, T301, and a fall back to RRCSetup if that fails. The companion 16 RLM and RLF document covers all of it.

The distinction matters diagnostically. A log showing beamFailureRecoveryTimer expiry followed by a successful contention-based recovery is a working system whose candidate list was wrong. A log showing the recovery timer expiry followed by preamble exhaustion is a coverage or configuration failure. They look almost identical for the first 200 ms.

11. SCell Beam Failure Recovery (Rel-16): the BFR MAC CEs

Rel-15 defined beam failure recovery only for the SpCell, which left an obvious hole: an FR2 SCell aggregated onto an FR1 anchor is far more likely to lose its beam than the anchor is, and until Rel-16 the only remedies were to let it rot or to release it. Rel-16 fills the hole, and does it by report rather than by procedure — because the UE already has a working uplink on the SpCell, there is nothing a preamble could achieve that a MAC CE cannot.

Table 7. The two procedures. The asymmetry is entirely about whether a working uplink already exists.

SP

C_i

AC

Candidate RS ID

R

Table 8. BFR MAC CE fields, TS 38.321 cl. 6.1.3. One candidate octet follows the bitmap for each bit that is set, in ascending serving cell order.

12. Rel-17: Per-TRP BFR and Multi-CORESET-Pool Operation

Multi-TRP transmission, introduced in Rel-16, lets two geographically separated transmission points serve one UE, distinguished by CORESETPoolIndex values 0 and 1. That immediately breaks the Rel-15 BFD model, which treats the BFD-RS set as a single AND: with two TRPs, one can be completely blocked while the other is fine, and the AND means no beam failure is ever declared. The UE loses half its capacity and reports nothing.

Rel-17 fixes this by making detection and recovery per-TRP:

• Two BFD-RS sets, one per CORESETPoolIndex, each with its own BFI_COUNTER and its own beamFailureDetectionTimer instance. Beam failure is now declared per TRP.
• Two candidate beam lists, so a replacement can be found within the failed TRP's own beams rather than by moving everything to the surviving TRP.
• Enhanced BFR MAC CE and Truncated Enhanced BFR MAC CE, which extend the Rel-16 report with the TRP (CORESET pool) the failure applies to, so one CE can describe a per-TRP failure on the SpCell as well as on SCells.
• Partial recovery as a normal outcome. A UE can be operating with one TRP recovered and one TRP in a failed state, which is a state Rel-15 has no way to express and no way to log.

13. Timers, Counters and Parameter Reference

beamFailureInstanceMaxCount

beamFailureDetectionTimer

failureDetectionResourcesToAddModList

candidateBeamRSList

rsrp-ThresholdSSB

recoverySearchSpaceId

SearchSpaceId

beamFailureRecoveryTimer

rach-ConfigBFR

RACH-ConfigGeneric

rootSequenceIndex-BFR

ra-Prioritization

RA-Prioritization

rlmInSyncOutOfSyncThreshold

Table 9. BFD and BFR parameters. RLM config = RadioLinkMonitoringConfig in ServingCellConfig; BFR config = BeamFailureRecoveryConfig, signalled per uplink BWP inside BWP-UplinkDedicated, which means a BWP switch can change the entire recovery configuration — including whether one exists at all.

14. Failure Modes and What Each One Means

Table 10. BFR failure modes. As with DRX, most of them are detected by nobody: BFR has no failure report of its own, so every row is diagnosed from its side effects.

15. Configuration Reference (ASN.1)

Detection is configured in ServingCellConfig -> radioLinkMonitoringConfig; recovery is configured per uplink BWP in BWP-UplinkDedicated -> beamFailureRecoveryConfig. Abridged, with ... marking omitted members and extension markers:

RadioLinkMonitoringConfig ::= SEQUENCE {
    failureDetectionResourcesToAddModList
        SEQUENCE (SIZE(1..maxNrofFailureDetectionResources)) OF
                                        RadioLinkMonitoringRS   OPTIONAL,
    failureDetectionResourcesToReleaseList
        SEQUENCE (SIZE(1..maxNrofFailureDetectionResources)) OF
                                        RadioLinkMonitoringRS-Id OPTIONAL,
    beamFailureInstanceMaxCount     ENUMERATED {n1,n2,n3,n4,n5,n6,n8,n10}
                                                                OPTIONAL,
    beamFailureDetectionTimer       ENUMERATED {pbfd1,pbfd2,pbfd3,pbfd4,
                                               pbfd5,pbfd6,pbfd8,pbfd10}
                                                                OPTIONAL,
    ...
}

RadioLinkMonitoringRS ::= SEQUENCE {
    radioLinkMonitoringRS-Id        RadioLinkMonitoringRS-Id,
    purpose                         ENUMERATED {beamFailure, rlf, both},
    detectionResource               CHOICE {
        ssb-Index                       SSB-Index,
        csi-RS-Index                    NZP-CSI-RS-ResourceId
    },
    ...
}

Listing 1. Detection. purpose is the field that ties beam failure detection and radio link monitoring to the same reference signal (§5).

BeamFailureRecoveryConfig ::= SEQUENCE {
    rootSequenceIndex-BFR           INTEGER (0..137)            OPTIONAL,
    rach-ConfigBFR                  RACH-ConfigGeneric          OPTIONAL,
    rsrp-ThresholdSSB               RSRP-Range                  OPTIONAL,
    candidateBeamRSList
        SEQUENCE (SIZE(1..maxNrofCandidateBeams)) OF
                                    PRACH-ResourceDedicatedBFR  OPTIONAL,
    ssb-perRACH-Occasion            ENUMERATED {oneEighth, oneFourth,
                                                oneHalf, one, two, four,
                                                eight, sixteen} OPTIONAL,
    ra-ssb-OccasionMaskIndex        INTEGER (0..15)             OPTIONAL,
    recoverySearchSpaceId           SearchSpaceId               OPTIONAL,
    ra-Prioritization               RA-Prioritization           OPTIONAL,
    beamFailureRecoveryTimer        ENUMERATED {ms10, ms20, ms40, ms60,
                                               ms80, ms100, ms150, ms200}
                                                                OPTIONAL,
    ...,
    [[ msg1-SubcarrierSpacing       SubcarrierSpacing           OPTIONAL ]]
}

PRACH-ResourceDedicatedBFR ::= CHOICE {
    ssb                             BFR-SSB-Resource,
    csi-RS                          BFR-CSIRS-Resource
}

BFR-SSB-Resource ::= SEQUENCE {
    ssb                             SSB-Index,
    ra-PreambleIndex                INTEGER (0..63),
    ...
}

BFR-CSIRS-Resource ::= SEQUENCE {
    csi-RS                          NZP-CSI-RS-ResourceId,
    ra-OccasionList                 SEQUENCE (SIZE(1..maxRA-OccasionsPerCSIRS))
                                        OF INTEGER (0..maxRA-Occasions-1)
                                                                OPTIONAL,
    ra-PreambleIndex                INTEGER (0..63)             OPTIONAL,
    ...
}

Listing 2. Recovery, per uplink BWP. BFR-SSB-Resource is the binding that makes CFRA possible: one SSB index, one preamble index, and the gNB can read a beam out of a detected preamble.

16. Illustrative Message Traces

16.1 The configuration the UE is given

RRCReconfiguration
 spCellConfig / spCellConfigDedicated                 -- ServingCellConfig
  radioLinkMonitoringConfig setup
   beamFailureInstanceMaxCount ............ n4
   beamFailureDetectionTimer .............. pbfd2     -- 2 x 20 ms = 40 ms
   failureDetectionResourcesToAddModList   (absent)   -- implicit set
  uplinkConfig / initialUplinkBWP
   beamFailureRecoveryConfig setup
    rootSequenceIndex-BFR ................. 91
    rach-ConfigBFR
     prach-ConfigurationIndex ............. 159
     preambleReceivedTargetPower .......... -104      -- dBm
     preambleTransMax ..................... n8
     powerRampingStep ..................... dB4
     ra-ResponseWindow .................... sl20
    rsrp-ThresholdSSB ..................... 56        -- -101 dBm (S7)
    candidateBeamRSList
     [0] ssb  { ssb 2,  ra-PreambleIndex 52 }
     [1] ssb  { ssb 5,  ra-PreambleIndex 53 }
     [2] ssb  { ssb 9,  ra-PreambleIndex 54 }
     [3] ssb  { ssb 13, ra-PreambleIndex 55 }
    recoverySearchSpaceId ................. 4         -- CORESET 3
    beamFailureRecoveryTimer .............. ms100
    ra-Prioritization
     powerRampingStepHighPriority ......... dB4
     scalingFactorBI ...................... zero

-- derived by the UE, not signalled:
--   BFD-RS set = { SSB 2, CSI-RS res 7 }   (QCL Type-D of CORESET 1 / 2)
--   candidate threshold = -101 dBm
--   detection budget = 4 instances x 20 ms = 80 ms

Listing 3. Detection and recovery configuration as it would appear decoded, with the derived values a good MAC trace prints alongside it.

16.2 A successful SpCell recovery

-- C-RNTI 0x71A4 throughout, FR2, 120 kHz SCS

14:22:08.100  [PHY] BFD eval  SSB2 hyp-BLER 0.7%  CSI-RS7 hyp-BLER 0.9%
              -> both better than Q_out (10%), no indication
14:22:08.118  [PHY] serving beam obstructed (L1-RSRP SSB2 -78 -> -112 dBm)
14:22:08.120  [PHY] BFD eval  SSB2 hyp-BLER 41%  CSI-RS7 hyp-BLER 38%
              -> ALL BFD-RS worse than Q_out
              -> beam failure instance indication to MAC
14:22:08.120  [MAC] BFI_COUNTER = 1   beamFailureDetectionTimer (re)started
14:22:08.140  [PHY] BFD eval  both worse than Q_out -> indication
14:22:08.140  [MAC] BFI_COUNTER = 2   timer restarted
14:22:08.160  [MAC] BFI_COUNTER = 3   timer restarted
14:22:08.180  [MAC] BFI_COUNTER = 4 = beamFailureInstanceMaxCount (n4)
14:22:08.180  [MAC] beam failure declared on SpCell
              -> initiate Random Access, trigger = beamFailureRecovery
              -> start beamFailureRecoveryTimer (ms100)

14:22:08.182  [MAC] candidate beam evaluation, threshold -101 dBm
                    SSB 2  ... -112 dBm   below threshold
                    SSB 5  ...  -84 dBm   ABOVE  -> selected
                    SSB 9  ... -103 dBm   below threshold
                    SSB 13 ... -119 dBm   below threshold
14:22:08.182  [MAC] ra-PreambleIndex 53 (bound to SSB 5), CFRA
14:22:08.185  [PHY] PRACH sent, preamble 53, RO for SSB 5, 18.0 dBm
14:22:08.185  [MAC] monitoring recoverySearchSpaceId 4 (CORESET 3)
14:22:08.189  [PHY] PDCCH found in CORESET 3, DCI 1_1, CRC by C-RNTI 0x71A4
14:22:08.189  [MAC] Random Access successfully completed
              -> BFI_COUNTER = 0
              -> stop beamFailureRecoveryTimer
              -> Beam Failure Recovery successfully completed
              -- CORESET 3 assumed QCL'd to SSB 5 until told otherwise

14:22:08.204  [MAC-DL] TCI State Indication for UE-specific PDCCH MAC CE
                    CORESET 1 -> TCI state 11 (QCL-D source SSB 5)
                    CORESET 2 -> TCI state 12 (QCL-D source CSI-RS res 19)
              -- BFD-RS set now { SSB 5, CSI-RS res 19 }, implicitly

Listing 4. 89 ms from blockage to a decoded PDCCH on the new beam, 80 ms of it detection. The final MAC CE is what moves the rest of the UE's PDCCH configuration off the dead beam.

16.3 A run that recovers without declaring failure

14:31:55.400  [MAC] BFI_COUNTER = 1   timer started, expires 14:31:55.440
14:31:55.420  [MAC] BFI_COUNTER = 2   timer restarted, expires .460
14:31:55.440  [PHY] BFD eval  SSB2 hyp-BLER 6%  CSI-RS7 hyp-BLER 4%
              -> better than Q_out, NO indication
              -- note: BFI_COUNTER stays at 2, timer keeps running
14:31:55.460  [PHY] BFD eval  both better than Q_out, no indication
14:31:55.460  [MAC] beamFailureDetectionTimer EXPIRED -> BFI_COUNTER = 0
14:31:55.480  [PHY] BFD eval  both worse than Q_out -> indication
14:31:55.480  [MAC] BFI_COUNTER = 1   timer restarted
14:31:55.500  [PHY] BFD eval  both better than Q_out, no indication
14:31:55.520  [MAC] beamFailureDetectionTimer EXPIRED -> BFI_COUNTER = 0

-- five bad evaluation periods in 120 ms and no beam failure. Correct:
-- the counter only ever clears on timer expiry, and it cleared twice.

Listing 5. The reset mechanism in action. The line at .440 is the one to internalise: a good measurement does not reset the counter, it merely fails to restart the timer.

16.4 SCell recovery by MAC CE

14:47:31.010  [MAC] SCell 2 (FR2): per-cell BFI_COUNTER = 4 = n4
14:47:31.010  [MAC] beam failure on SCell -> trigger BFR for SCell 2
                    (no Random Access procedure initiated)
14:47:31.012  [MAC] SCell 2 candidate evaluation
                    SSB 11 ... -81 dBm   ABOVE threshold -> offer it
14:47:31.012  [MAC] no UL-SCH resource available -> trigger SR (BFR SR id 1)
14:47:31.014  [PHY-UL] PUCCH SR on the SpCell, SR id 1
14:47:31.020  [PHY-DL] PDCCH DCI 0_1 to C-RNTI 0x71A4, UL grant 12 octets

14:47:31.022  [MAC-UL] MAC PDU on the SpCell, 4 octets
              subheader  R=0 F=0 LCID=<BFR>  L=2
                BFR MAC CE
                  octet 1 = 0x02     -- SP=0, C_2=1, all other C_i=0
                  octet 2 = 0x8B     -- AC=1, R=0, Candidate RS ID = 11

14:47:31.034  [MAC-DL] TCI States Activation/Deactivation MAC CE, SCell 2
                    -> CORESETs on SCell 2 moved to a TCI state sourced
                       from SSB 11
14:47:31.034  [MAC] BFR for SCell 2 cancelled; per-cell BFI_COUNTER = 0

-- SpCell untouched throughout; SCell 2 never deactivated or released.

Listing 6. 24 ms from declaration to repair, most of it waiting for an uplink grant. Octet 1 = 0x02 is bit C_2 alone; octet 2 = 0x8B is AC = 1 with Candidate RS ID 11.

16.5 A failure path: no candidate, then preamble exhaustion

15:02:44.300  [MAC] BFI_COUNTER = 4 = n4 -> beam failure on SpCell
15:02:44.300  [MAC] start beamFailureRecoveryTimer (ms100)
15:02:44.302  [MAC] candidate beam evaluation, threshold -101 dBm
                    SSB 2  ... -118 dBm     SSB 5  ... -121 dBm
                    SSB 9  ... -117 dBm     SSB 13 ... -124 dBm
              -> NO candidate above rsrp-ThresholdSSB
              -> fall back to contention-based Random Access
                 (rach-ConfigCommon, normal SSB selection)
15:02:44.305  [MAC] preamble 19 (CB pool), counter 1, SSB 9, 15.0 dBm
15:02:44.315  [MAC] ra-ResponseWindow expired, no RAR
15:02:44.333  [MAC] preamble 07, counter 2, 19.0 dBm
15:02:44.343  [MAC] ra-ResponseWindow expired, no RAR
15:02:44.400  [MAC] beamFailureRecoveryTimer expired
              -- contention-free BFR resources no longer usable;
              -- this is NOT an RLF and NOT a procedure failure
                       ... attempts 3..7 elided, all at 23.0 dBm (P_CMAX) ...
15:02:44.612  [MAC] preamble 44, counter 8 = preambleTransMax (n8)
15:02:44.622  [MAC] ra-ResponseWindow expired, no RAR
15:02:44.622  [MAC] -> Random Access problem indication to RRC
15:02:44.622  [RRC] RA problem, trigger was beamFailureRecovery, SpCell,
                    state RRC_CONNECTED  ->  RADIO LINK FAILURE

-- meanwhile, on the same bad radio:
15:02:44.240  [RRC] out-of-sync #1 ... #4  -> N310 (n4) reached
15:02:44.240  [RRC] start T310 (ms1000)
15:02:44.622  [RRC] T310 still running (382 ms elapsed) when RLF declared
              -- RLF came from the RACH, not from T310. With T310 = ms200
              -- the same trace would have blamed T310 instead.

Listing 7. The genuine escalation, with both routes to RLF visible. Which one the log records is a function of the relative sizing of T310 and the BFR RACH budget, not of the radio.

17. Release Deltas: Rel-15 to Rel-18

Table 11. BFR-relevant changes by release. Confirm feature support against the UE capability exchange — see the companion 26 UE Capability document.

18. Reading BFR in Logs: A Checklist

1. Establish which BFD-RS set is in force, implicit or explicit (§4). If implicit, find the most recent TCI state change first — it silently redefined what the UE was monitoring, and a beam failure right after one is usually caused by it.

2. Check whether the failure was an AND across the whole set. A single beam's L1-RSRP collapsing produces no beam failure if another BFD-RS is still good. If you expected a failure and did not get one, this is why.

3. Read the BFI instance timestamps, not just the count. Instances spread across seconds mean beamFailureDetectionTimer is too long relative to the RS period; instances 20 ms apart are a real fade (§6).

4. Do not expect the counter to reset on a good measurement. Trace the timer, not the measurements. The counter clears only on timer expiry, on reconfiguration, or on a completed recovery.

5. Convert rsrp-ThresholdSSB from RSRP-Range to dBm before judging whether a candidate should have qualified: -157 + n (§7). The raw integer is not dBm.

6. Classify the recovery request. A dedicated preamble index from candidateBeamRSList means the full contention-free procedure worked. A contention-based preamble means either no candidate qualified, no dedicated preamble was configured, or beamFailureRecoveryTimer had expired — three different problems with three different fixes.

7. Look for the response in the right place. It is a PDCCH to the UE's existing C-RNTI in recoverySearchSpaceId, not a RAR and not a BFR-specific message. There is no TC-RNTI anywhere in a BFR trace.

8. Check what happened to the other CORESETs. Recovery only implies the recovery CORESET moved. Until a TCI State Indication MAC CE arrives, the rest of the UE's PDCCH configuration is still on the dead beam (§9).

9. When BFR fails, read T310 alongside it. Both were racing the same fade and either can produce the RLF. The one the log blames depends on timer sizing, not on the radio (§10, §16.5).

10. For SCells, read the AC bit. AC = 1 means retune; AC = 0 means the UE has nothing to offer and the candidate list or the SCell itself needs attention (§11).

11. On multi-TRP, confirm per-TRP BFD is configured before trusting any beam failure counter. A legacy configuration hides a dead TRP completely (§12).

19. Glossary

BFI_COUNTER

candidateBeamRSList

recoverySearchSpaceId

CORESETPoolIndex

20. References

• 3GPP TS 38.321 — NR MAC protocol specification. Clause 5.17 (beam failure detection and recovery, BFI_COUNTER, the reset rules), 5.1.1 (Random Access initialisation and triggers, including beam failure recovery), 5.1.2 (Random Access resource selection, including the candidateBeamRSList branch), 5.1.4 (Random Access problem indication), 5.4.4 (Scheduling Request, used by SCell BFR), 6.1.3 (BFR and Truncated BFR MAC CE formats), Table 6.2.1-2 (UL-SCH LCID values).
• 3GPP TS 38.213 — Physical layer procedures for control. Clause 6 (link reconfiguration: BFD-RS determination, candidate beam evaluation, the recovery response window and the post-recovery QCL assumption), clause 5 (radio link monitoring and the RLM-RS limits).
• 3GPP TS 38.133 — Requirements for support of radio resource management. The hypothetical PDCCH used to define Q_in and Q_out, the threshold pairs, the beam failure and radio link monitoring evaluation periods, and the RSRP-Range to dBm mapping.
• 3GPP TS 38.331 — RRC protocol specification. RadioLinkMonitoringConfig, RadioLinkMonitoringRS, BeamFailureRecoveryConfig, PRACH-ResourceDedicatedBFR, BFR-SSB-Resource, BFR-CSIRS-Resource, ServingCellConfig, BWP-UplinkDedicated, rlmInSyncOutOfSyncThreshold.
• 3GPP TS 38.300 — NR overall description. The beam management and beam failure recovery overview, and where BFR sits relative to RLM and mobility.
• 3GPP TS 38.214 — Physical layer procedures for data. TCI states, QCL types and L1-RSRP reporting, which supply the measurements BFR consumes.

Companion documents in this set

• 02 Radio Frame Structure — SSB bursts and the half-frame structure the BFD-RS periodicities are built on.
• 03 Random Access — the procedure BFR borrows for its request; the CFRA and preamble-budget material in §8 lives there in full.
• 07 MAC PDU and Control Elements — the subheader and LCID framework the BFR MAC CEs use.
• 08 Scheduling — CORESETs, search spaces and PDCCH monitoring; what a beam failure actually deprives the UE of.
• 10 BSR, PHR, SR — the Scheduling Request mechanism SCell BFR depends on when it has no grant.
• 11 DRX — ps-TransmitPeriodicL1-RSRP and why a power-saving UE can be slower to notice a beam problem.
• 16 RLM and RLF — the cell-level layer of §3, and where a failed BFR ends up.
• 20 Measurements and Events — L1-RSRP and L3 RSRP, and why the two disagree during a beam failure.
• 22 Handover Overview / 25 Conditional HO and DAPS — what happens when the right answer is a different cell rather than a different beam.
• 26 UE Capability — where SCell BFR, per-TRP BFR and unified TCI support is declared.