Cell Reselection (Idle / Inactive) in 5G NR

1. Why Reselection Exists, and Who Is Actually in Charge

Most of the time a phone is switched on, it is not connected to anything. It has no radio bearer, no scheduled uplink, no C-RNTI and no ongoing conversation with a base station. It is nevertheless on a cell: it has picked one, it is reading that cell's broadcast information, and it is listening for paging messages at the times that cell told it to listen. That state is called camping, and the UE has to keep doing it correctly while its owner walks, drives or takes a train through a network of cells that come and go.

Somebody has to decide when to move the camp from one cell to another. In RRC_CONNECTED the network decides, because it is already talking to the UE and can afford to. In RRC_IDLE and RRC_INACTIVE it cannot: waking a UE up to ask it what it can hear, every time it can hear something, would cost far more energy and far more air time than the problem is worth. So the decision is delegated. The UE measures, the UE ranks, the UE decides, and the UE moves — and it tells nobody.

That delegation is the single fact this whole document hangs off. The network's only influence over an idle UE's mobility is the set of parameters it broadcast before the UE needed them, plus, in one special case, a short-lived list handed out in the message that released the UE from its last connection (§9). Once the UE is camped and idle, the network is a spectator. It has no way to nudge a particular UE, no way to observe its decisions, and no way to know it made one.

Read the parameter set as four jobs rather than as a list. Almost every field in SIB2 to SIB5 that touches reselection is doing one of these four things, and knowing which one tells you what breaks if it is wrong:

Table 1. The four jobs every reselection parameter is doing. Nothing in TS 38.304 is there for elegance; each field is a defence against a specific way an unsupervised UE misbehaves.

What starts reselection is not an event but a standing obligation. A camped UE evaluates the reselection rules continuously, as often as its measurement duty cycle allows, for as long as it stays in RRC_IDLE or RRC_INACTIVE. There is no trigger message and no request. The relevant question is never "what triggered this reselection" but "which of the three rules fired, and which parameter let it".

2. Reselection and Handover, Side by Side

Reselection and handover are often described as "idle-mode mobility" and "connected-mode mobility", as though they were two settings of one mechanism. They are not. They share only the physical outcome: the UE ends up on a different cell. Everything about how that outcome is reached differs, including who is allowed to decide, what evidence the decision is based on, how long it takes, and what happens when it goes wrong.

Table 2. Reselection against handover, row by row. The two rows that matter most to a test engineer are the last two: one mechanism reports its own failures and the other cannot.

The two mechanisms also differ in something subtler: what a measurement means. In connected mode the UE reports a filtered quantity, smoothed by the layer-3 filter coefficient the network configured, and the network compares it against an event threshold it also configured — see the companion 20 Measurements and Events. In idle mode there is no configurable L3 filter in the same sense; the measurement period and accuracy come from the requirements in TS 38.133, and the smoothing that prevents twitchy behaviour is provided by q-Hyst and t-ReselectionNR instead. Comparing an idle-mode threshold with a connected-mode event threshold is meaningless: they are not measured the same way, and they are not doing the same job.

3. Camping: Selection, Reselection, and What Camped Means

Two words that sound interchangeable are not. Cell selection is what a UE does when it has no cell: at power-on, on return from a coverage hole, or after a PLMN change. It searches, finds candidates, checks them and picks one. That procedure — the raster scan, the SSB detection, the PSS and SSS correlation, the MIB and SIB1 read — belongs to the companion 36 SSB and Cell Search, and is not repeated here. Cell reselection is what the UE does once it already has a cell and keeps it under review. It is a maintenance loop, not a search.

Table 3. Selection and reselection compared. Note the scope row: reselection is deliberately myopic, because a camped UE has been given a list and has no reason to look beyond it.

3.1 What camping actually consists of

"Camped on a cell" is a precise state, not a figure of speech. A UE camped normally on a cell is doing all of the following, and stops doing all of them for the old cell the instant it reselects:

• Monitoring paging on that cell, at the paging occasions derived from its own identity and that cell's pcch-Config — see the companion 19 Paging. This is the reason camping matters at all: it is how the network reaches the UE.
• Reading and re-reading system information from that cell, including the value tag that tells it whether anything changed, and acting on short-message change indications.
• Holding that cell's parameters as the basis for every reselection decision: its q-Hyst, its search thresholds, its neighbour lists, its priorities. Reselecting means adopting a different set of them, which is why a UE's behaviour can change character the moment it crosses a boundary between two differently-configured cells.
• Being able to originate: an access attempt, an emergency call, an RNA update or a resume all start on the camped cell, using its rach-ConfigCommon — see the companion 03 Random Access.
• Evaluating reselection, continuously, per §6.

TS 38.304 distinguishes camped normally, which is the state just described on a suitable cell, from camped on any cell, which is the limited-service state a UE falls into when it can find only an acceptable cell (§4). The difference is not cosmetic: a UE camped on any cell may make emergency calls and receive public warning messages, but it is not registered for normal service and the network will not page it for ordinary traffic.

4. Suitable, Acceptable, Barred: the Three Categories of Cell

Before any comparison between cells can happen, the UE has to sort them into categories. A cell that fails at this stage never enters the ranking at all, which is why a perfectly strong cell can be ignored completely — and why "the UE ignored a 20 dB better cell" is so often a barring or exclusion story rather than a threshold story.

Table 4. Cell categories, TS 38.304 cl. 5.2.1 and 5.3.1. The 300 s figure is a specification constant, not a broadcast parameter — it cannot be tuned.

4.1 cellBarred and intraFreqReselection, the two-bit sledgehammer

The MIB carries exactly two bits that matter here, and between them they can remove an entire carrier from a UE's world. cellBarred (barred / notBarred) bars this cell. intraFreqReselection (allowed / notAllowed) says what the UE may do next if this cell is barred: allowed means it may reselect to another cell on the same frequency, notAllowed means it may not, and must go to a different frequency or a different RAT. The field decode itself belongs to the companion 18 MIB and SIB1 IEs; what matters here is the consequence.

Two adjacent mechanisms are often confused with barring and are worth separating cleanly:

• Unified Access Control (uac-BarringInfo in SIB1) does not bar camping. A UE may camp normally on a cell whose UAC bars its access category; the barring check happens when it tries to access, and it fails the access attempt rather than the camp. UAC therefore does not appear anywhere in the reselection rules — see companions 18 MIB and SIB1 IEs and 36 SSB and Cell Search.
• Cell reservation (cellReservedForOtherUse per PLMN in SIB1) behaves like barring for an ordinary UE, but is per-PLMN rather than per-cell, so the same cell can be reserved for one PLMN of a shared RAN and open for another.

5. The S-Criterion: Is This Cell Even a Candidate?

The S-criterion answers one question: is this cell good enough to be used at all? It is a pass/fail gate, not a comparison. A cell either has enough margin above the minimum the operator declared acceptable, in which case it can be camped on and can compete against other cells, or it does not, in which case it is invisible no matter how it compares with anything else.

Two independent tests must both pass. One is about level — is the signal strong enough — and the other about quality — is the signal clean enough. They are separate because a cell can be loud and useless: in a dense layer, a cell can deliver plenty of RSRP while its signal-to-interference ratio is hopeless, and a level test alone would happily camp a UE on it.

Srxlev = Q_rxlevmeas - (q-RxLevMin + q-RxLevMinOffset)
                     - P_compensation - Qoffset_temp

Squal  = Q_qualmeas  - (q-QualMin  + q-QualMinOffset)
                     - Qoffset_temp

A cell passes the S-criterion when   Srxlev > 0   AND   Squal > 0

  Q_rxlevmeas      measured cell RSRP, derived from SSB beams (S14)
  Q_qualmeas       measured cell RSRQ, derived the same way
  q-RxLevMin       minimum required level, in 2 dB units
  q-QualMin        minimum required quality, in 1 dB units
  q-RxLevMinOffset applied only while evaluating a higher-priority PLMN
                   from a VPLMN; 0 otherwise.  In 2 dB units
  q-QualMinOffset  the same, for quality
  P_compensation   an uplink-limited correction (S5.2)
  Qoffset_temp     a temporary penalty after a failed access (S5.4)

In words: take the level you measured, subtract the minimum level this cell says it needs, subtract anything that makes the uplink harder than the downlink suggests, subtract any penalty you owe this cell from a recent failure, and the answer must be positive. Then do the same thing again with quality instead of level. Srxlev and Squal are therefore margins in dB above the minimum, not measurements — a logged Srxlev of 12 does not mean −12 dBm of anything, it means the cell is 12 dB better than the floor the operator set for it.

5.1 Where each term is broadcast

The same criterion is evaluated for the serving cell and for every neighbour, but the values come from different places depending on which cell is being tested. This is why one UE can compute a healthy margin for its serving cell and a failing one for a neighbour that measures identically:

Table 5. Every carrier carries its own suitability floor. A neighbour carrier with a stricter q-RxLevMin than the serving carrier will look worse than it is, and that is a legitimate steering tool as well as a common misconfiguration.

5.2 P_compensation, and why an uplink term appears in a downlink test

P_compensation is the term that surprises people. Everything else in the S-criterion is measured on the downlink, and then this appears — a quantity derived from transmit powers, subtracted from a received-level margin. The reason is that camping on a cell is only useful if the UE can eventually talk to it. A cell that is comfortably audible but out of the UE's uplink reach is a trap: the UE would camp, be paged, attempt access, and fail. P_compensation shrinks the apparent downlink margin by the amount the uplink is short.

P_compensation = max( P_EMAX - P_PowerClass , 0 )        [dB]

  P_EMAX        the maximum UE output power the cell permits on this
                carrier: p-Max, broadcast per cell / per carrier
  P_PowerClass  the maximum output power of this UE's power class
                (23 dBm for the common power class 3)

For a UE supporting additional maximum power reduction, the higher of
the applicable P_EMAX values is used; for SUL, q-RxLevMinSUL replaces
q-RxLevMin and the SUL carrier's p-Max applies.

In words: if the network allows more uplink power than this UE can produce, the shortfall is deducted from the cell's margin. If p-Max is at or below the UE's own capability, the term is zero and vanishes. The asymmetry is deliberate: the network is describing the link budget it designed for, and a UE that cannot meet it must discount the cell accordingly.

5.3 A worked S-criterion evaluation

Take one cell and walk the whole test, then walk it again at a weaker point so the failure is visible:

Note what the failure does not do. It does not trigger anything by itself. If the UE is camped on that cell and it becomes unsuitable, the UE must find somewhere else to go, and if nothing else qualifies it enters cell selection — back to the procedure in companion 36 SSB and Cell Search — and may end up out of service. The S-criterion is a gate on candidacy, not a reselection trigger.

5.4 Qoffset_temp: the penalty for a recent failure

Qoffset_temp is the only history-dependent term in the criterion. When a UE fails connection establishment on a cell connEstFailCount times, it applies connEstFailOffset as Qoffset_temp to that cell for connEstFailOffsetValidity seconds, which reduces both Srxlev and Squal and can push a cell from suitable to unsuitable on the strength of the UE's own bad experience.

6. The Camped-State Evaluation Loop

Everything from here on is one loop, run by the UE, forever. It is worth seeing the shape of it before the individual rules, because the order of the steps explains several behaviours that look arbitrary in isolation — in particular why a UE can be sitting next to a better cell and not know it, and why the answer to "how quickly does a UE react" is never a single number.

1. Re-evaluate the serving cell's own S-criterion. If the serving cell has become unsuitable, everything else is secondary — the UE needs somewhere to go, and if reselection cannot provide it, cell selection will have to.
2. Apply the measurement rules (§7). Compare the serving cell's Srxlev and Squal against s-IntraSearchP / s-IntraSearchQ and s-NonIntraSearchP / s-NonIntraSearchQ, and decide which classes of neighbour measurement are obligatory right now. Higher-priority frequencies are always obligatory.
3. Measure what is still required, at the rate TS 38.133 demands for the UE's current DRX cycle. This is where the wall-clock time goes: an idle UE with a long DRX cycle measures rarely, and its reselection reaction time is dominated by that, not by t-ReselectionNR.
4. Derive one value per cell from the beams of that cell (§14). A cell is not a single measurement in NR; it is a set of SSB beams that must be consolidated into one number before any comparison is possible.
5. Evaluate the three rules in priority order (§10): higher priority first, then rank among equals, then lower priority. The order is fixed and matters.
6. If a rule is satisfied, check that it has been satisfied for its whole timer, that more than 1 s has elapsed since the UE camped on the current cell, and that the target is a legitimate candidate (§4). Only then reselect.
7. Reselect: retune, acquire the new cell's MIB and SIB1, confirm it is genuinely suitable, adopt its parameters, and start monitoring its paging occasions. In RRC_INACTIVE, additionally check whether the new cell is inside the RAN notification area (§17).

7. Measurement Rules: When the UE Is Allowed Not To Look

This is the section people skip, and then cannot explain why a UE ignored a cell that was plainly better. The rule is simple and its consequences are not: while the serving cell is comfortably good, the UE does not have to measure its neighbours at all. Not "may measure less often" — may not measure them.

The reason is battery. A neighbour measurement on another frequency means retuning the receiver, waiting for an SSB burst, correlating, and coming back; on the serving frequency it is cheaper but still not free. A UE deep inside good coverage has nothing to gain from any of it, and there are a great many such UEs, each of them nominally idle for hours at a time. So TS 38.304 gives the operator two pairs of thresholds that buy silence.

Table 6. Measurement rules, TS 38.304 cl. 5.2.4.2. The last row is not a parameter; it is the absence of one, and it is the most consequential row in the table.

How often "always" is, in practice, comes from TS 38.133 rather than TS 38.304. The requirements are expressed as measurement periods scaled by the UE's DRX cycle, plus a specific obligation on higher-priority layers: a UE must search each higher-priority layer at least once every 60 × N_layers seconds, where N_layers is the number of such layers it has been given TS 38.133 cl. 4.2.2. Two consequences follow. First, steering onto a high-priority carrier can legitimately take a minute or more even in perfect coverage. Second, giving a UE many high-priority layers makes each one slower to be found, because the obligation is per set, not per layer.

7.1 Relaxed measurement, and the exception to the asymmetry

Rel-17 added a second, orthogonal way for a UE to measure less: relaxed measurement, aimed initially at RedCap and other battery-critical devices. Instead of asking "is the serving cell comfortable", it asks "has anything changed". If the UE's own measured level has not moved by more than s-SearchDeltaP over a period t-SearchDeltaP, the UE may treat itself as stationary and relax neighbour measurement; a separate cellEdgeEvaluation criterion lets it relax while it is comfortably inside a cell.

Table 7. Rel-17 relaxed measurement in idle and inactive, from SIB2 relaxedMeasurement. Support is UE-capability-gated; a UE that does not signal the capability measures normally.

8. Frequency Priorities, and the Frequencies With None

A network usually has more than one carrier, and rarely wants idle UEs spread evenly across them. A low band gives coverage and is precious; a mid-band carrier gives capacity and is where the operator would rather have UEs when they wake up; a millimetre-wave carrier is useful only where it is useful. Frequency priority is the tool that expresses that preference to UEs that nobody is supervising.

Each frequency is given an integer priority from 0 to 7, where 7 is the highest and 0 the lowest. The priority attaches to the frequency, not to the cell: every cell on the carrier inherits it. Rel-15 also allows a fractional refinement, cellReselectionSubPriority (oDot2, oDot4, oDot6, oDot8), which adds a fraction to the integer so two carriers can be ordered without either becoming equal to a third.

Table 8. Where priorities come from. Note the asymmetry between SIB2 and SIB4/SIB5: the serving frequency's priority is mandatory, every other one is optional.

Two further rules are worth stating plainly because they are easy to assume the other way round:

• Priorities are not per cell. There is no way to give one cell of a carrier a different priority from its neighbours on the same carrier. Per-cell steering is done with offsets (§11), which act inside the ranking rather than above it.
• Equal priority is a real, distinct case. Two carriers at the same priority are compared by rank, exactly as if they were the same carrier, with q-OffsetFreq available to bias the comparison. This is how an operator makes two carriers behave as one pooled layer.

9. Dedicated Priorities, t320, and Deprioritisation

Broadcast priorities are the same for every UE in the cell, which is occasionally not what the network wants. When it releases a specific UE it has a brief chance to say something to that UE alone, and RRCRelease is where it takes it. Two separate mechanisms live in that message: a full per-UE priority list, and a blunt request to avoid a frequency for a while.

9.1 cellReselectionPriorities and t320

cellReselectionPriorities in RRCRelease carries a list of NR frequencies and E-UTRA frequencies with priorities for this UE, plus the timer t320. While t320 runs, the UE uses the dedicated priorities and ignores the broadcast ones entirely — including for frequencies not mentioned in the dedicated list, which are treated as having no priority at all and therefore drop out of consideration.

Table 9. Dedicated priorities in RRCRelease, TS 38.331 CellReselectionPriorities and RRCRelease-IEs.

Why bother? Because release is the one moment the network knows something about this UE that it cannot express in a broadcast: which slice it was using, which layer served it well, whether it is a high-throughput device that belongs on capacity spectrum or a low-rate device that does not. Dedicated priorities are how load balancing and service-based steering reach an idle UE. They are also how a network hands a UE back to LTE after an NR session, by giving the E-UTRA carrier the higher priority on the way out.

9.2 deprioritisationReq

The second mechanism is cruder and has a narrower purpose. deprioritisationReq in RRCRelease asks the UE to treat either the current frequency (deprioritisationType = frequency) or all NR frequencies (nr) as the lowest priority for the duration of deprioritisationTimer (min5, min10, min15, min30), tracked by T325. It carries no list and no per-frequency detail; it is a way of saying "go somewhere else for a few minutes".

• frequency is used to push a UE off a carrier that is congested or about to be taken out of service, without changing anything for other UEs.
• nr deprioritises NR as a whole, which in practice sends the UE to E-UTRA. It exists for cases where the NR coverage itself is the problem — an early deployment with poor uplink, or a UE that keeps failing access.
• The UE discards the deprioritisation on PLMN selection, on entering RRC_CONNECTED, or when T325 expires.

10. The Three Reselection Rules

There are exactly three ways a camped UE can decide to move, and they are tried in a fixed order. First, is there anything better on a higher-priority frequency — and if so, go, whatever the serving cell is doing. Second, among frequencies of the same priority, including the serving one, is another cell ranked above the serving cell. Third, and only if the serving cell is genuinely poor, is there something usable on a lower-priority frequency to fall back to.

The order is what makes priority steering work. Rank is only consulted among equals, so a strong intra-frequency neighbour can never keep a UE away from a higher-priority carrier — the higher-priority test happened first and did not care.

10.1 Higher priority: go, and never mind the serving cell

The UE reselects to a cell on a higher-priority frequency when, for the whole of t-ReselectionNR (or the inter-RAT equivalent), that cell's Srxlev exceeds threshX-HighP — or, where the quality-based variant is configured, its Squal exceeds threshX-HighQ — and more than one second has passed since the UE camped where it is TS 38.304 cl. 5.2.4.5.

Higher-priority reselection, level-based:

    Srxlev_target > threshX-HighP        held for t-ReselectionNR

Higher-priority reselection, quality-based (used where threshX-HighQ
is signalled for the target frequency):

    Squal_target  > threshX-HighQ        held for t-ReselectionNR

There is NO term for the serving cell in either expression.

In words: if a cell on a preferred frequency is merely good enough on its own account, and stays good enough for a couple of seconds, the UE goes there. The serving cell may be 20 dB stronger. It is not consulted. If several higher-priority frequencies qualify at once, the UE goes to the highest priority among them, and to the best-ranked suitable cell on that frequency.

10.2 Equal priority: rank them

For the serving frequency and for any frequency with the same priority as the serving frequency, there is no absolute threshold at all. The UE builds a ranking of the serving cell and the candidate neighbours using the R-criterion, and reselects to the best-ranked cell if it stays better than the serving cell for t-ReselectionNR. §11 works this through in full, because it is where the offsets live and where most practical tuning happens.

Intra-frequency neighbours are always in this case by definition: the serving frequency's priority is its own. Which is why intra-frequency reselection has no threshX parameters of any kind, and is governed entirely by q-Hyst, q-OffsetCell and t-ReselectionNR from SIB2 and SIB3.

10.3 Lower priority: the double condition

Going down in priority requires two things to be true at once, and this is the only rule with a term for the serving cell in it:

Lower-priority reselection, level-based:

    Srxlev_serving < threshServingLowP     AND
    Srxlev_target  > threshX-LowP          both held for t-Reselection

Quality-based variant, where threshServingLowQ is signalled:

    Squal_serving  < threshServingLowQ     AND
    Squal_target   > threshX-LowQ

and, as always, more than 1 s since the UE camped on the serving cell,
and no higher- or equal-priority cell meeting its own criteria.

In words: the UE must be doing badly where it is, and the fallback must be doing well, before it will step down a layer. The asymmetry against rule 1 is the entire design. Upwards, the UE needs no excuse. Downwards, it needs a reason. Without the first condition a UE would oscillate between layers whenever the lower layer happened to look adequate — which, for a low band under a mid-band carrier, is almost everywhere.

Table 10. The three rules in one table. Read the second column downwards: the presence or absence of a serving-cell term is what distinguishes steering from fallback.

11. The R-Criterion in Full

Among frequencies of equal priority the UE has to choose between cells that are all, by definition, allowed. It does so by computing a single ranked value for each and picking the largest. The value is not the measurement: the serving cell gets a bonus, and each neighbour can be given a handicap or an advantage of its own. Those adjustments are the whole of practical reselection tuning.

For the serving cell:

    Rs = Q_meas,s + Q_hyst - Qoffset_temp

For each neighbour cell n:

    Rn = Q_meas,n - Q_offset - Qoffset_temp

where Q_offset is:

    intra-frequency   Qoffset(s,n)          = q-OffsetCell of that cell
                      (0 if not signalled for the cell)

    inter-frequency   Qoffset(s,n) + Qfreq  = q-OffsetCell of that cell
                      plus q-OffsetFreq of that carrier; Qfreq alone if
                      no per-cell offset is signalled

    Q_meas            RSRP, consolidated from the cell's beams (S14)
    Q_hyst            q-Hyst, plus the speed-dependent q-HystSF term
    Qoffset_temp      the failed-access penalty of S5.4

In words: the serving cell is credited with q-Hyst decibels it has not earned, every neighbour is debited whatever offsets the network chose for it, and the UE then ranks what is left and takes the top one — provided it stays on top for t-ReselectionNR. Note the signs carefully: q-Hyst is added to the serving cell, and Q_offset is subtracted from the neighbour, so a positive q-OffsetCell makes a neighbour less attractive. This sign convention is inverted relative to the connected-mode cellIndividualOffset in a measurement object, which is added to the measurement — see companion 20 Measurements and Events. Confusing the two is how UEs get steered the wrong way.

11.1 Two neighbours, one serving cell, worked through

The figure is the same scenario as the arithmetic below. Three cells, one of them serving, with the offsets a real network would plausibly have configured:

Broadcast configuration in use
  SIB2  q-Hyst .................... dB4
        t-ReselectionNR ........... 2         -- seconds
        cellReselectionPriority ... 5         -- serving frequency f1
  SIB3  intraFreqNeighCellList
          PCI 188  q-OffsetCell ... dB0       -- neighbour N1
  SIB4  carrier f2  priority ...... 5         -- EQUAL priority to f1
          q-OffsetFreq ............ dB2
          interFreqNeighCellList
            PCI 431  q-OffsetCell . dB3       -- neighbour N2

Measurements at t = 0 s
  serving cell S (f1, PCI 12) .... RSRP -92 dBm, falling 0.8 dB/s
  neighbour  N1 (f1, PCI 188) .... RSRP -97 dBm, rising  0.7 dB/s
  neighbour  N2 (f2, PCI 431) .... RSRP -90 dBm, steady

Now change one number to see how sharp the tool is. Set N2's q-OffsetCell to dB-2 instead of dB3 — a legal value, since Q-OffsetRange runs from −24 dB to +24 dB — and its ranked value becomes −90 − (2 − 2) = −90 dBm, which beats the serving cell from t = 0 s. The same UE in the same place now leaves for a different carrier two seconds into the scenario. Nothing else changed.

11.2 The 1-second rule, and what it does and does not forbid

Every reselection case carries the same extra condition: more than one second must have elapsed since the UE camped on the current serving cell TS 38.304 cl. 5.2.4.1. The practical effect is a floor on how fast a UE can move: the shortest possible interval between two reselections is one second plus whatever t-ReselectionNR applies, so no configuration can make a UE hop faster than that.

What it is often described as — a rule forbidding the UE to go back to the cell it just left — is the consequence rather than the wording. There is no cell-specific memory in it; the guard applies to any reselection, including one that happens to return. Combined with q-Hyst, which is applied to whichever cell is currently serving, the pair does make an immediate return unlikely: after reselecting to N1, N1 is the cell that now gets the q-Hyst bonus, so the old serving cell has to become better than N1 plus q-Hyst to win it back.

12. q-Hyst, Ping-Pong, and the One-Second Rule

Two cells of genuinely equal strength are the worst case for any mobility mechanism. The measurements of both will wander by several decibels from fading alone, so whichever is compared first will sometimes look better, and a UE that reacted to every such moment would spend its life reselecting. q-Hyst is the defence: a deliberate, broadcast handicap given to whichever cell the UE is currently on.