Measurement Gaps & SMTC in 5G NR

1. Why a Gap Has to Exist

A receiver is tuned to a centre frequency and a bandwidth. A UE camped on a carrier at 3.55 GHz with a 100 MHz bandwidth part is, in hardware terms, a filter and a set of ADCs pointed at that band. Ask it to measure a neighbour at 1.85 GHz and there is no clever protocol trick available: the local oscillator has to move, and while it is moving and while it is pointed elsewhere, the serving carrier is not being received at all. Uplink is worse -- transmitting on the serving carrier while receiving on another is a filtering problem no handset solves for an arbitrary pair of bands.

So NR does the honest thing and schedules the outage. A measurement gap is a periodic, network-configured interval during which the UE is not expected to transmit or receive anything on the serving cells, and is free to retune, measure a different frequency or a different RAT, and retune back TS 38.133 cl. 9.1.2. The scheduler knows the gap pattern, so it simply does not schedule into it. Nothing is lost to collision; the cost is paid in advance, as capacity.

Table 1. What needs a gap and what does not. The second row is the one that catches people, because the frequency looks identical in the configuration.

2. When No Gap Is Needed, and the Exception

Intra-frequency measurement is free, and that single fact shapes NR network design: intra-frequency mobility can be configured aggressively with fast filters and short timeToTrigger, while inter-frequency mobility is rationed. The reason it is free is that the neighbour's SSB arrives in the same receive chain the UE is already using for the serving cell -- the UE only has to know which symbols to demodulate, which is exactly what the SMTC tells it.

The exception is bandwidth parts. The active DL BWP is not the whole carrier; it can be a narrow slice placed anywhere in it, with its own subcarrier spacing. If the SSB the UE has to measure does not fall inside the active DL BWP, or the SSB's subcarrier spacing differs from the active BWP's, then the receiver is not configured to demodulate it and a gap becomes necessary for what the configuration calls an intra-frequency measurement TS 38.133 cl. 9.1.2.

Two further cases produce gap-free measurement in practice. A UE with genuinely independent receive chains per frequency range can measure FR2 while receiving FR1, which is what per-FR gaps formalise (§5). And from Rel-17, network-controlled small gaps let the UE take an interruption measured in symbols rather than milliseconds, for cases where the retuning is cheap (§14).

3. Anatomy of a Gap Pattern

Table 2. The four fields of a gap pattern plus the reference indicator TS 38.331 `GapConfig`. mgl and mgrp together select a standardised pattern (§4); gapOffset and mgta position it.

3.1 Where a gap actually falls

The gap position is derived from the SFN and subframe of the reference serving cell, not from any timer the UE starts TS 38.133 cl. 9.1.2:

T = MGRP / 10                     (repetition period, in radio frames)

gap occurs in frames where:  SFN mod T = FLOOR(gapOffset / 10)
gap starts at subframe:      gapOffset mod 10
gap actually begins:         mgta earlier than that subframe boundary
gap ends:                    MGL later than the nominal start

4. The Standardised Gap Patterns

mgl and mgrp are not freely combinable in practice: TS 38.133 tabulates the legal combinations as numbered gap patterns, and UE capability signalling is expressed in terms of those numbers. The pattern id is what appears in vendor documentation and in most log formats, so it is worth being able to convert in both directions.

Table 3. Gap patterns 0-23 TS 38.133 Table 9.1.2-1, printed as two halves side by side. Duty cycle is MGL/MGRP and is computed here, not tabulated in the spec. Patterns 0-11 are the general set; the shorter-MGL patterns 12-23 are specified for FR2 measurement, where the SSB burst is short and retuning is cheaper.

5. Per-UE and Per-FR Gaps

The original model was one gap pattern for the whole UE: when the gap arrives, everything stops. That is correct for a UE with one receive chain, and needlessly expensive for a UE with separate FR1 and FR2 front ends, where retuning the FR2 chain to measure another FR2 carrier need not disturb the FR1 serving cell at all.

Table 4. The three gap scopes. The rule that matters operationally: gapUE is exclusive of the other two, and per-FR gaps need the UE to have said it can do them.

For a UE in NR-DC or EN-DC the picture gains a second dimension: the gap may be configured by the master node and has to be honoured by both cell groups, and refServCellIndicator says whose frame timing the gapOffset refers to. The companion 26 UE Capability document covers where independentGapConfig and the per-band-combination gap capabilities are reported.

6. MeasGapConfig and the Coexistence Rules

Gaps are configured inside measConfig, alongside the measurement objects that need them -- which is convenient, because the two have to be designed together, and dangerous, because a delta reconfiguration that touches one can disturb the other. Each of the three gap fields is a SetupRelease, so the network sets up, modifies or releases them independently.

• gapUE must not be configured together with gapFR1 or gapFR2. Per-UE and per-FR gaps are alternatives, not layers.
• gapFR1 and gapFR2 may be configured together, and then apply independently to their own frequency ranges.
• Per-FR gaps require the UE to have reported independentGapConfig. Configuring them regardless is a specification violation whose observable effect is UE-dependent.
• Releasing a gap does not release the measIds that needed it. Those measIds simply stop producing results, silently -- exactly the failure described in the companion 20 Measurements and Events document.

7. What the UE Gives Up During a Gap

Inside a measurement gap the UE is not expected to transmit or receive anything on the serving cells the gap applies to. Not deprioritised -- absent. Everything in the following table is simply not there for the duration.

Table 5. The full cost of a gap. The configured-grant row is the one that produces user-visible complaints fastest.

8. Gap Sharing and the Scaling Factor

One gap pattern usually has to serve several purposes at once: gap-assisted intra-frequency measurement, two or three inter-frequency NR carriers, an inter-RAT carrier, sometimes positioning. They cannot all use the same 6 ms. measGapSharingConfig tells the UE how to divide the gap occasions TS 38.133 cl. 9.1.9.

Table 6. MeasGapSharingScheme values. The exact percentage for each scheme is tabulated in TS 38.133 cl. 9.1.9.2 and depends on whether gap-assisted intra-frequency measurement is needed at all; the design point is that the network fixes the split rather than leaving it to the UE. Like the gaps themselves, sharing can be per-UE or per-FR.

8.1 The carrier-specific scaling factor

Sharing has a formal consequence in the RRM requirements. The time the UE is allowed to take to identify and to measure a cell on a given carrier is scaled by a carrier-specific scaling factor, which grows with the number of carriers competing for the same gaps TS 38.133 cl. 9.1.5. Add a fourth inter-frequency carrier to a configuration and you have not just diluted the gap allocation, you have relaxed what the UE is required to achieve on the other three.

9. Gaps and DRX

DRX and gaps are two independent mechanisms that both remove the UE from the air, configured by different parts of the network for different reasons, and they do not coordinate themselves. DRX timers keep running during a gap: drx-onDurationTimer and drx-InactivityTimer are unaffected, so a PDCCH occasion that falls inside a gap is not postponed, it is lost. See the companion 11 DRX document for the timer machinery.

Table 7. Gap and DRX interactions. Only the first row costs latency, and it is entirely avoidable through gapOffset.

10. SMTC: Telling the UE When to Look

SSBs are not continuous. A cell transmits a burst of SS/PBCH blocks -- up to 8 in FR1, up to 64 in FR2, all inside a 5 ms half-frame -- and then transmits nothing on those resources until the burst repeats, typically 20 ms later. A UE searching blindly for a neighbour's SSB would spend most of its receive time looking at silence. The SSB Measurement Timing Configuration is the network telling the UE where the silence is not.

Table 8. SSB-MTC TS 38.331. Two fields, and between them they decide whether a neighbour is measurable at all.

The window recurs on the same arithmetic as the gap:

window occurs in frames where:  SFN mod (periodicity / 10) = 
                                    FLOOR(offset / 10)
window starts at subframe:      offset mod 10
window length:                  duration subframes

  -- for sf5, the window recurs every 5 subframes and the offset is
     0..4 within that, so the frame arithmetic degenerates

Table 9. SMTC periodicity against the neighbour's actual SSB periodicity. Note that only the actual periodicity matters, and it is not visible anywhere in the UE's configuration -- the network has to know it.

11. smtc1, smtc2 and Per-Cell Overrides

MeasObjectNR carries more than one SMTC because one carrier can hold cells with different SSB periodicities. smtc1 is the primary window and applies to every cell on the carrier. smtc2 is a periodicity override for a listed set of PCIs.

Table 10. The SMTC family. smtc2 inheriting smtc1's offset and duration is the detail that produces surprises.

Two further fields in MeasObjectNR shape what happens inside the window. ssb-ToMeasure is a bitmap of the SS/PBCH block indices worth measuring -- useful when a neighbour uses only some of its beams, because the UE can stop processing the rest. ssb-PositionQCL-Common declares which SSB positions are quasi-co-located, and deriveSSB-IndexFromCell says whether the UE may assume the neighbour's SSB timing is aligned with the serving cell's, which is what lets it derive an SSB index without decoding the PBCH.

12. The Three-Way Overlap

This is the section to read first if a neighbour is missing from every report. Three independent periodic things must overlap before a single sample exists:

1. The neighbour's SSB burst -- owned by the neighbour cell's configuration, invisible to the UE, and known to the network only through planning data.
2. The SMTC window -- owned by the measObject in this UE's measConfig.
3. The measurement gap -- owned by measGapConfig, and only required when the measurement needs retuning (§2).

One further subtlety: the usable part of the gap is shorter than MGL. The UE spends mgta retuning at the start and a comparable interval retuning back at the end, so a 6 ms gap yields roughly 5 ms of measurement, and a 1.5 ms gap yields well under 1 ms. An SMTC of duration sf5 cannot be captured inside a 1.5 ms gap at all, however perfectly the offsets are aligned -- which makes mgl and duration a pair that must be chosen together.

13. CSI-RS Based Measurement and Its Timing

SSB-based measurement is coarse in time -- one burst every 20 ms -- and coarse in space, since an SSB beam is deliberately wide. CSI-RS-based mobility measurement exists for the cases where that is not enough: finer beams, more frequent samples, and a measurement that lives on the same resources used for beam management. The price is configuration coupling, because the UE must be told the neighbour's CSI-RS configuration in detail, which means the network has to know it.