Scheduling: Dynamic, SPS & Configured Grant in 5G NR

1. What Scheduling Is, and What the Scheduler Is Optimising

Every downlink and uplink transmission in NR except the synchronisation signal, the broadcast of SIB1, and a small set of pre-configured periodic resources happens because the gNB said so, in a message called a grant. The UE has no autonomy at all in the uplink beyond a scheduling request; it cannot transmit data on its own initiative, and it cannot receive data it has not been told to look for. Scheduling is the process that produces those grants, slot by slot, and it is the single largest determinant of what a cell actually delivers.

What makes it hard is that the scheduler is solving several problems at once with one instrument. In any given slot it must decide which UEs get resources, how much each gets, where in frequency, with what modulation and coding, and when the corresponding data appears — and the objectives conflict:

Table 1. What a scheduler is trading. None of these objectives, and no weighting between them, appears anywhere in a 3GPP specification — see §13.

2. The Scheduling Loop, End to End

One turn of the loop is: collect measurements and requests from the UE, decide, transmit a DCI on PDCCH, transmit or receive the data at the offset the DCI specified, collect the HARQ answer, and feed it back into the next decision. The period is one slot — 0.5 ms at 30 kHz — and the loop runs independently for the downlink and the uplink.

Two properties of this loop cause most of the confusion when reading logs. First, everything is offset in time: the DCI, the data and the feedback are in three different slots, related by K0, K1 and K2, and a log sorted by timestamp interleaves three different transactions. Second, the inputs are stale. A CSI report describes the channel as it was when the reference signal was measured, several milliseconds before the grant it influences is transmitted, which at vehicular speeds is long enough for the channel to have changed completely. Link adaptation is therefore always a prediction, and the outer loop in §9 exists precisely to correct for how wrong that prediction turns out to be.

3. The Scheduler's Inputs

Table 2. Everything the scheduler knows. Notice that none of it describes the future, and that two of the most important inputs — CQI and BSR — are the UE's own reports about conditions that have already changed by the time they are acted on.

4. CORESETs: Where a PDCCH Is Allowed to Live

A control resource set is a rectangle of the resource grid — a set of PRBs, over one, two or three consecutive OFDM symbols — inside which PDCCH may be transmitted. It exists because the UE cannot search the whole carrier for control information: the search has to be bounded, and the CORESET is the bound.

4.1 The unit hierarchy

• A REG (resource element group) is one PRB in one OFDM symbol: 12 resource elements, of which 3 carry PDCCH DMRS and 9 carry control information TS 38.211 cl. 7.3.2.2.
• A CCE (control channel element) is 6 REGs — so 72 resource elements, 54 of them usable. This is the allocation quantum for PDCCH.
• A CORESET's size in CCEs is (number of PRBs × duration) / 6. A CORESET of 48 PRBs over 2 symbols is 96 REGs, hence 16 CCEs.
• A PDCCH occupies 1, 2, 4, 8 or 16 CCEs — its aggregation level. More CCEs means a lower effective code rate for the same DCI payload, which is how the network trades control-channel capacity for control-channel reliability.

controlResourceSetId

frequencyDomainResources

duration

cce-REG-MappingType

reg-BundleSize

interleaverSize

shiftIndex

precoderGranularity

pdcch-DMRS-ScramblingID

Table 3. ControlResourceSet fields, TS 38.331 and TS 38.211 cl. 7.3.2. Up to three dedicated CORESETs may be configured per bandwidth part, plus CORESET 0 and a common CORESET.

5. Search Spaces, Aggregation Levels and Blind Decoding

A CORESET says where PDCCH may appear in the grid. A search space says when the UE looks, how hard it looks, and what kind of DCI it expects. The two are separate objects because several search spaces with different periodicities and different DCI expectations can share one CORESET.

5.1 Common and UE-specific search spaces

searchSpaceOtherSystemInformation

pagingSearchSpace

commonSearchSpaceList

Table 4. Search space types, TS 38.213 cl. 10.1. The type names are procedural labels, not ASN.1 fields — the encoding is just common or ue-Specific plus which DCI formats are enabled.

searchSpaceId

monitoringSlotPeriodicityAndOffset

duration

monitoringSymbolsWithinSlot

nrofCandidates

Table 5. SearchSpace fields, TS 38.331. monitoringSymbolsWithinSlot is the one most people forget exists, and it is what makes mini-slot scheduling reachable.

5.2 The hashing function

Candidate m at aggregation level L does not sit at a fixed CCE offset. Its position is computed from a per-UE, per-slot hash, so that two UEs whose candidates collide in one slot will not collide in the next TS 38.213 cl. 10.1:

CCEs of candidate m at aggregation level L, in CORESET p, slot n:

  L * { ( Y_p,n + floor( m * N_CCE,p / (L * M_p,L) ) + n_CI )
        mod floor( N_CCE,p / L ) }  +  i        for i = 0 .. L-1

  N_CCE,p   total CCEs in CORESET p
  M_p,L     number of candidates configured at level L
  n_CI      carrier indicator, 0 unless cross-carrier scheduling

  Common search space:        Y_p,n = 0     (fixed -- every UE agrees)
  UE-specific search space:   Y_p,n = ( A_p * Y_p,n-1 ) mod D
                              D = 65537,  Y_p,-1 = n_RNTI  (non-zero)
                              A_p = 39827, 39829 or 39839
                                    for p mod 3 = 0, 1, 2

Three things follow. A common search space has Y = 0, so every UE computes the same candidate positions — necessary, because the DCI is addressed to a group. A UE-specific search space is seeded with the C-RNTI, so positions differ per UE and advance every slot. And because the recursion is deterministic, both the UE and the gNB can compute the same positions without any signalling — which also means a wrong C-RNTI produces a UE that searches entirely the wrong CCEs and finds nothing, with no error anywhere.

5.3 The blind-decode budget

The UE does not know whether anything was sent to it. It demodulates every candidate at every DCI size it is configured for and finds out from the CRC. That is expensive, so TS 38.213 caps it per slot per serving cell:

Table 6. Per-slot, per-serving-cell PDCCH monitoring limits, TS 38.213 Tables 10.1-2 and 10.1-3. With carrier aggregation the limits scale with the number of configured cells, up to a UE capability.

6. DCI Formats, Grant Contents, and Cross-Carrier Scheduling

6.1 The catalogue

Table 7. Every DCI format in NR, TS 38.212 cl. 7.3.1. Payload sizes exclude the 24-bit CRC and are approximate except for 1_0, which is exactly computable. Formats 2_x carry no grant at all — they are group-addressed indications.

6.2 The one DCI you can compute exactly

Table 8. DCI 1_0 field by field, total 41 bits at a 106-PRB bandwidth part. Add the 24-bit CRC and the PDCCH is carrying 65 bits, which at aggregation level 2 (108 usable resource elements, QPSK) is a code rate of about 0.30.

6.3 What the full-featured formats add

DCI 1_1 and 0_1 add everything the fallback formats deliberately leave out: a second transport block for two-codeword MIMO, antenna port and TCI state selection, SRS resource indication and precoding information for the uplink, rate-matching and ZP-CSI-RS indicators, code block group fields, the bandwidth part indicator, the carrier indicator, and the Rel-16 priority indicator. The cost is that the size becomes a function of about a dozen configuration choices, which is why they are only allowed in a UE-specific search space where the UE's configuration is unambiguous.

6.4 Cross-carrier and cross-bandwidth-part scheduling

Table 9. Cross-carrier and cross-BWP scheduling. Both change where the grant applies without changing the format.

7. Resource Allocation in Frequency: Type 0, Type 1 and the RIV

A grant has to name a set of PRBs in a field of fixed width. NR offers two encodings with different strengths, selected by resourceAllocation in PDSCH-Config / PUSCH-Config — or, with dynamicSwitch, by a bit in the DCI.

7.1 Type 0: the resource block group bitmap

PRBs are grouped into resource block groups of P consecutive PRBs, and the DCI carries one bit per group. The group size depends on the bandwidth part size and on rbg-Size TS 38.214 Table 5.1.2.2.1-1:

Table 10. Resource block group sizes. The number of groups is ⌈(N_BWP_size + (N_BWP_start mod P)) / P⌉, so a bandwidth part that does not start on a P boundary gets one extra, partly unusable group.

Type 0's strength is discontiguity: a scheduler can hand a UE groups 2–6 and 11–12 in one grant, which matters when the rest of the band is already committed. Its weakness is granularity — with P = 16 the smallest allocation is 16 PRBs, so a 20-byte VoIP packet on a 100 MHz carrier wastes most of what it is given.

7.2 Type 1: the resource indication value

Type 1 encodes a contiguous interval — a start and a length — as a single integer, chosen so that every legal (start, length) pair maps to a distinct value and the field is exactly ⌈log₂(N(N+1)/2)⌉ bits wide TS 38.214 cl. 5.1.2.2.2:

Encoding (N = N_BWP_size, L = number of PRBs, S = starting PRB):

  if (L - 1) <= floor(N / 2):
        RIV = N * (L - 1) + S
  else:
        RIV = N * (N - L + 1) + (N - 1 - S)

Decoding:

  L' = floor(RIV / N)          S' = RIV - L' * N

  if (L' + 1) + S' <= N:       L = L' + 1        S = S'
  else:                        L = N - L' + 1    S = N - 1 - S'

Field width = ceil( log2( N * (N + 1) / 2 ) )  bits

An RIV decoded with the wrong bandwidth part size is not merely off by a little; it lands somewhere unrelated, because N appears as the multiplier. If a decoded allocation is implausible — negative, off the end, or suspiciously round — check N first. Common causes: decoding a DCI 1_0 in a common search space against the dedicated bandwidth part instead of the initial one, or against the carrier bandwidth instead of the BWP.

8. Resource Allocation in Time: K0, K2, SLIV and Mapping Type

The time-domain allocation is not carried field by field in the DCI. Instead the DCI carries a 4-bit (or 0–4 bit) row index into a table the network configured in advance, and each row holds the slot offset, the mapping type and a start and length indicator value that packs the starting symbol and the symbol count into one integer.

k0

PDSCH-TimeDomainResourceAllocation

k2

PUSCH-TimeDomainResourceAllocation

mappingType

startSymbolAndLength

Table 11. Time-domain allocation, TS 38.214 cl. 5.1.2.1 (downlink) and 6.1.2.1 (uplink). The row-index indirection is what keeps the DCI small.

SLIV encoding (normal cyclic prefix):

  if (L - 1) <= 7:   SLIV = 14 * (L - 1) + S
  else:              SLIV = 14 * (14 - L + 1) + (14 - 1 - S)

with S + L <= 14, L >= 1

Decoding:
  if floor(SLIV / 14) + 1 <= 14 - (SLIV mod 14):
        L = floor(SLIV / 14) + 1      S = SLIV mod 14
  else:
        L = 14 - floor(SLIV / 14) + 1 S = 14 - 1 - (SLIV mod 14)

9. Link Adaptation: MCS Tables, TBS and the Outer Loop

Link adaptation is the choice of modulation order, code rate and layer count for a given allocation. It runs in two nested loops. The inner loop maps the UE's CQI report to an MCS index; the outer loop corrects that mapping using the HARQ feedback that came back, and it exists because the inner loop is systematically wrong.

9.1 The three MCS tables

Table 12. The three PDSCH MCS tables. The low-SE table is not "a table for bad channels" — it is a table designed for a 10⁻⁵ residual error target, which is why it starts at a code rate of 30/1024 and why URLLC uses it.

MCS-C-RNTI is a second C-RNTI, assigned alongside the ordinary one, whose only function is to select the low-SE MCS table for the DCIs it scrambles. It lets a scheduler switch a single UE between a 10⁻¹ target and a 10⁻⁵ target per grant, with no reconfiguration — which is how a UE can carry eMBB and URLLC traffic simultaneously. In a log, two different RNTIs scheduling the same UE is not an error; it is the reliability switch.

9.2 From MCS to transport block size

Transport block size is derived, never signalled. Both ends compute it from the allocation, the MCS and the configured overhead, and quantise the result so they always agree TS 38.214 cl. 5.1.3.2. The full procedure is in the companion 05 HARQ document; here is a smaller one, worked end to end.

9.3 The outer loop

Nothing in any 3GPP specification describes outer-loop link adaptation. It is universal anyway, because the inner loop cannot work without it. The CQI a UE reports is a prediction made from a reference signal measured milliseconds earlier, using a UE-internal receiver model the network cannot see, and reported against a nominal 10 % BLER that different UEs interpret differently. The result is a systematic bias, different per UE and per environment.

The standard fix is a per-UE offset applied to the CQI-derived MCS, adjusted from the HARQ feedback: decrease it on a NACK by a step Δ_down, increase it on an ACK by a much smaller Δ_up, with the ratio chosen so the loop settles at the target BLER. For a 10 % target, Δ_up ≈ Δ_down/9. The offset is commonly clipped to a few dB in each direction and reset on handover.

Table 13. Reading the outer loop from its symptoms. Because it is not specified, its behaviour is the clearest fingerprint of a particular scheduler implementation.

10. Semi-Persistent Scheduling and Configured Grants

Dynamic scheduling costs one PDCCH per transport block. For traffic that is small, periodic and predictable — a voice codec producing 20 bytes every 20 ms, an industrial sensor, a video conference's audio stream — the PDCCH can cost more resource than the data. NR therefore allows a resource to be granted once and used repeatedly: semi-persistent scheduling in the downlink, configured grants in the uplink.

Table 14. The three flavours. Type 1 is the only one that needs no DCI at all, which makes it the choice where even a PDCCH is too much latency or too much overhead.

10.1 The validation pattern

An activation DCI carries no data and schedules nothing immediately. Its HARQ, redundancy-version and MCS fields are repurposed as a fixed pattern the UE checks before treating the DCI as an activation rather than an ordinary grant TS 38.213 cl. 10.2:

Table 15. Abridged from TS 38.213 Tables 10.2-1 to 10.2-4. A DCI addressed to cs-RNTI that matches neither pattern is discarded, silently — which is the hardest SPS failure to diagnose.

10.2 What it costs

The saving is real but bounded. A configured grant cannot follow the channel: its MCS and PRB count are fixed until the network reactivates it, so it must be sized for the worst case the UE is expected to see. For a 20-byte voice packet that is cheap insurance; for anything with a variable rate it is not. configuredGrantTimer bounds how long a configured-grant HARQ process is considered occupied, and Rel-16's cg-RetransmissionTimer (introduced for NR-U) lets the UE retransmit on a later configured occasion without waiting for a grant — the one place in NR where uplink retransmission happens without the gNB asking.

11. Slot Aggregation and Repetition

When one transmission cannot close the link budget and waiting for HARQ feedback costs too much latency, the answer is to send the same transport block several times without waiting. NR has three mechanisms for this and they are frequently confused with HARQ retransmission, which they are not — repetition happens whether or not the first copy would have decoded.

Table 16. Repetition mechanisms. All three trade spectral efficiency for reliability without adding a round trip — the opposite trade from HARQ.

12. Scheduling for Latency, and Intra-UE Priority

A slot at 30 kHz is 0.5 ms, and a full dynamic scheduling round trip is several slots. For a 1 ms end-to-end budget that is already too slow, so NR provides a set of mechanisms whose common theme is not waiting for the next slot boundary.

12.1 Mini-slots and sub-slot monitoring

Mapping type B allocations can start at any symbol and last as few as two symbols for PDSCH, or one for PUSCH — a mini-slot. To use them the UE has to be able to find a grant mid-slot, which is what monitoringSymbolsWithinSlot provides: setting several bits creates several PDCCH monitoring occasions inside one slot. See the companion 02 Radio Frame Structure document for the symbol-level picture.

12.2 Downlink pre-emption

When a URLLC transmission has to go out now and the resources are already carrying an eMBB transport block, the scheduler simply overwrites them. The victim UE would then soft-combine corrupted symbols into its HARQ buffer and poison every subsequent retransmission of that block — so it has to be told. DCI 2_1, scrambled by INT-RNTI, carries a bitmap of pre-empted resources.

Table 17. The two pre-emption mechanisms, and the direction of each. DCI 2_1 is retrospective — it repairs damage. DCI 2_4 is prospective — it prevents it.

12.3 Intra-UE prioritisation

Rel-16 introduced the problem of one UE having two things to send at once with different reliability requirements — and no way to choose. Three separate mechanisms now exist:

• Physical-layer priority: a 1-bit priority indicator in DCI 0_1 and 1_1 (priorityIndicatorDCI-0-1, priorityIndicatorDCI-1-1), splitting uplink transmissions into two priority classes. When a low-priority PUSCH and a high-priority PUCCH collide, the high-priority one wins and the other is dropped.
• Logical channel priority: priority (1 … 16) in LogicalChannelConfig, which governs how the UE fills a granted transport block. This is the logical channel prioritisation procedure — see the companion 09 LCP document.
• Uplink control information multiplexing: whether HARQ-ACK, SR and CSI of different priorities may share one PUCCH at all, or must be dropped in favour of the highest.

13. Scheduler Policy: Where the Specification Stops

Everything up to here is specified. What follows is not, and it is worth being explicit about the boundary, because a great deal of scheduling folklore is presented as though it were standardised. 3GPP does not define a scheduling algorithm. It does not define fairness, it does not define how QoS weights are combined, and it does not define outer-loop link adaptation. Two conforming gNBs can behave completely differently in the same conditions.

Table 18. Scheduler policies. None of these names appears in a 3GPP specification. They are engineering vocabulary, and the mapping from a vendor's tunable parameters to one of these is a question for that vendor's documentation.

14. Parameter Reference: ASN.1, Ranges and Effects

controlResourceSetToAddModList

PDCCH-Config

frequencyDomainResources

ControlResourceSet

duration

ControlResourceSet

searchSpacesToAddModList

PDCCH-Config

monitoringSlotPeriodicityAndOffset

SearchSpace

nrofCandidates

SearchSpace

monitoringSymbolsWithinSlot

SearchSpace

resourceAllocation

rbg-Size

PDSCH-Config

pdsch-TimeDomainAllocationList

PDSCH-Config

mcs-Table

pdsch-AggregationFactor

PDSCH-Config

BWP-DownlinkDedicated

BWP-UplinkDedicated

configuredGrantTimer

ConfiguredGrantConfig

PDCCH-Config

CrossCarrierSchedulingConfig

ServingCellConfig

Table 19. Scheduling-relevant configuration. The three with the largest practical blast radius are nrofCandidates (silent search-space drops), pdsch-TimeDomainAllocationList (allocations the scheduler cannot express) and SPS-Config periodicity (a saving that only exists if it matches the traffic).

15. Failure Modes and What Each One Means

Table 20. Scheduling failure modes. The recurring theme is that almost nothing here is reported: PDCCH has no failure indication, the blind-decode budget is enforced silently, and an invalid activation DCI is simply dropped.

16. ASN.1 Extracts

Scheduling configuration lives in PDCCH-Config and PDCCH-ConfigCommon (where PDCCH may be and when to look), PDSCH-Config / PUSCH-Config (how allocations are encoded), and the SPS and configured-grant structures. Abridged from TS 38.331; ... marks omitted fields and extension markers.

ControlResourceSet ::= SEQUENCE {
    controlResourceSetId       ControlResourceSetId,   -- INTEGER (0..11)
    frequencyDomainResources   BIT STRING (SIZE (45)),
    duration                   INTEGER (1..3),
    cce-REG-MappingType        CHOICE {
        interleaved  SEQUENCE {
            reg-BundleSize     ENUMERATED { n2, n3, n6 },
            interleaverSize    ENUMERATED { n2, n3, n6 },
            shiftIndex         INTEGER (0..274)             OPTIONAL },
        nonInterleaved NULL
    },
    precoderGranularity        ENUMERATED { sameAsREG-bundle,
                                            allContiguousRBs },
    tci-StatesPDCCH-ToAddList  SEQUENCE (SIZE (1..64)) OF TCI-StateId
                                                            OPTIONAL,
    tci-PresentInDCI           ENUMERATED { enabled }       OPTIONAL,
    pdcch-DMRS-ScramblingID    INTEGER (0..65535)           OPTIONAL,
    ...
}

SearchSpace ::= SEQUENCE {
    searchSpaceId              SearchSpaceId,          -- INTEGER (0..39)
    controlResourceSetId       ControlResourceSetId         OPTIONAL,
    monitoringSlotPeriodicityAndOffset CHOICE {
        sl1 NULL, sl2 INTEGER (0..1), sl4 INTEGER (0..3),
        sl5 INTEGER (0..4), sl8 INTEGER (0..7), sl10 INTEGER (0..9),
        sl16 INTEGER (0..15), sl20 INTEGER (0..19),
        sl40 INTEGER (0..39), sl80 INTEGER (0..79),
        sl160 INTEGER (0..159), sl320 INTEGER (0..319),
        sl640 INTEGER (0..639), sl1280 INTEGER (0..1279),
        sl2560 INTEGER (0..2559)
    }                                                       OPTIONAL,
    duration                   INTEGER (2..2559)            OPTIONAL,
    monitoringSymbolsWithinSlot BIT STRING (SIZE (14))       OPTIONAL,
    nrofCandidates SEQUENCE {
        aggregationLevel1      ENUMERATED { n0,n1,n2,n3,n4,n5,n6,n8 },
        aggregationLevel2      ENUMERATED { n0,n1,n2,n3,n4,n5,n6,n8 },
        aggregationLevel4      ENUMERATED { n0,n1,n2,n3,n4,n5,n6,n8 },
        aggregationLevel8      ENUMERATED { n0,n1,n2,n3,n4,n5,n6,n8 },
        aggregationLevel16     ENUMERATED { n0,n1,n2,n3,n4,n5,n6,n8 }
    }                                                       OPTIONAL,
    searchSpaceType CHOICE {
        common SEQUENCE {
            dci-Format0-0-AndFormat1-0 SEQUENCE { ... }      OPTIONAL,
            dci-Format2-0              SEQUENCE { ... }      OPTIONAL,
            dci-Format2-1              SEQUENCE { ... }      OPTIONAL,
            dci-Format2-2              SEQUENCE { ... }      OPTIONAL,
            dci-Format2-3              SEQUENCE { ... }      OPTIONAL },
        ue-Specific SEQUENCE {
            dci-Formats ENUMERATED { formats0-0-And-1-0,
                                     formats0-1-And-1-1 } }
    }                                                       OPTIONAL
}

Listing 1. Where PDCCH may be, and when the UE looks. nrofCandidates is the field that decides the blind-decode load, and it is the one most often set without checking the budget in §5.3.

PDSCH-Config ::= SEQUENCE {
    dataScramblingIdentityPDSCH  INTEGER (0..1023)           OPTIONAL,
    dmrs-DownlinkForPDSCH-MappingTypeA SetupRelease { DMRS-DownlinkConfig }
                                                             OPTIONAL,
    tci-StatesToAddModList       SEQUENCE (SIZE (1..128)) OF TCI-State
                                                             OPTIONAL,
    vrb-ToPRB-Interleaver        ENUMERATED { n2, n4 }        OPTIONAL,
    resourceAllocation           ENUMERATED { resourceAllocationType0,
                                              resourceAllocationType1,
                                              dynamicSwitch },
    pdsch-TimeDomainAllocationList SetupRelease {
                                     PDSCH-TimeDomainResourceAllocationList }
                                                             OPTIONAL,
    rbg-Size                     ENUMERATED { config1, config2 },
    mcs-Table                    ENUMERATED { qam256, qam64LowSE } OPTIONAL,
    maxNrofCodeWordsScheduledByDCI ENUMERATED { n1, n2 }      OPTIONAL,
    prb-BundlingType             CHOICE { ... },
    pdsch-AggregationFactor      ENUMERATED { n2, n4, n8 }    OPTIONAL,
    ...
}

SPS-Config ::= SEQUENCE {
    periodicity        ENUMERATED { ms10, ms20, ms32, ms40, ms64, ms80,
                                    ms128, ms160, ms320, ms640, ... },
    nrofHARQ-Processes INTEGER (1..8),
    n1PUCCH-AN         PUCCH-ResourceId                       OPTIONAL,
    mcs-Table          ENUMERATED { qam64LowSE }              OPTIONAL,
    ...,
    [[ sps-ConfigIndex-r16   SPS-ConfigIndex-r16              OPTIONAL,
       harq-ProcID-Offset-r16 INTEGER (0..15)                 OPTIONAL,
       periodicityExt-r16   INTEGER (1..5120)                 OPTIONAL ]]
}

ConfiguredGrantConfig ::= SEQUENCE {
    frequencyHopping    ENUMERATED { intraSlot, interSlot }   OPTIONAL,
    cg-DMRS-Configuration DMRS-UplinkConfig,
    mcs-Table           ENUMERATED { qam256, qam64LowSE }     OPTIONAL,
    resourceAllocation  ENUMERATED { resourceAllocationType0,
                                     resourceAllocationType1,
                                     dynamicSwitch },
    nrofHARQ-Processes  INTEGER (1..16),
    repK                ENUMERATED { n1, n2, n4, n8 },
    repK-RV             ENUMERATED { s1-0231, s2-0303, s3-0000 } OPTIONAL,
    periodicity         ENUMERATED { sym2, sym7, sym1x14, sym2x14,
                                     sym4x14, ... , sym5120x14 },
    configuredGrantTimer INTEGER (1..64)                      OPTIONAL,
    rrc-ConfiguredUplinkGrant SEQUENCE {
        timeDomainOffset          INTEGER (0..5119),
        timeDomainAllocation      INTEGER (0..15),
        frequencyDomainAllocation BIT STRING (SIZE (18)),
        mcsAndTBS                 INTEGER (0..31),
        ...
    }                                                         OPTIONAL,
    ...
}

Listing 2. How allocations are encoded, and the two recurring-resource structures. The presence or absence of rrc-ConfiguredUplinkGrant is the only thing that distinguishes a Type 1 configured grant from a Type 2 one.

17. Illustrative Message Traces

17.1 The configuration the scheduler is working within

RRCReconfiguration
 cellGroupConfig
  spCellConfig / spCellConfigDedicated
   initialDownlinkBWP
    pdcch-Config
     controlResourceSetToAddModList
      ControlResourceSet [1]
       controlResourceSetId ........ 1
       frequencyDomainResources .... 111111110000...0   -- 8 bits set
                                                        -- = 48 PRB
       duration .................... 2                  -- 16 CCEs
       cce-REG-MappingType
        interleaved
         reg-BundleSize ............ n6
         interleaverSize ........... n2
         shiftIndex ................ 214                -- = the PCI
       precoderGranularity ......... sameAsREG-bundle
       pdcch-DMRS-ScramblingID ..... 214
     searchSpacesToAddModList
      SearchSpace [2]
       searchSpaceId ............... 2
       controlResourceSetId ........ 1
       monitoringSlotPeriodicityAndOffset  sl1
       monitoringSymbolsWithinSlot . 11000000000000     -- symbols 0-1
       nrofCandidates
        aggregationLevel1 ......... n6
        aggregationLevel2 ......... n6
        aggregationLevel4 ......... n4
        aggregationLevel8 ......... n2
        aggregationLevel16 ........ n1                  -- 19 candidates
       searchSpaceType
        ue-Specific
         dci-Formats .............. formats0-1-And-1-1  -- 2 DCI sizes
    pdsch-Config
     resourceAllocation ............ resourceAllocationType1
     rbg-Size ...................... config1
     mcs-Table ..................... qam256
     pdsch-TimeDomainAllocationList
      [0] k0 0  mappingType typeA  startSymbolAndLength 53  -- S=2 L=12
      [1] k0 0  mappingType typeB  startSymbolAndLength 20  -- S=6 L=2
      [2] k0 1  mappingType typeA  startSymbolAndLength 53
      [3] k0 0  mappingType typeA  startSymbolAndLength 27  -- S=0  L=14
     maxNrofCodeWordsScheduledByDCI  n1

Listing 3. The CORESET, search space and PDSCH configuration used throughout §5 and §17. Row [0] is the SLIV worked out in §8; row [1] is the mini-slot.

17.2 A downlink grant, decoded field by field

10:14:22.0000  [PHY-DL] PDCCH search, SFN 412 slot 7, CORESET 1
                candidates tried .......... 19  (AL1 x6, AL2 x6, AL4 x4,
                                                AL8 x2, AL16 x1)
                DCI sizes tried ........... 2   (1_1 = 57, 0_1 = 49 bits)
                decode attempts ........... 38
                hit: AL 2, candidate m=1, CCEs 4-5

10:14:22.0000  [PHY-DL] DCI 1_1, CRC descrambled with C-RNTI 0x4601
                Identifier for DCI formats ......... 1     -- downlink
                Carrier indicator .................. (absent)
                Bandwidth part indicator ........... 0     -- no switch
                Frequency domain resource assign ... 1585
                  -- FDRA type 1, N_BWP = 106, 13-bit field
                  -- L' = 1585/106 = 14, S' = 101
                  -- (14+1)+101 = 116 > 106 -> wrapped branch
                  -- L = 106-14+1 = 93 PRB, S = 106-1-101 = PRB 4
                Time domain resource assignment .... 0
                  -- row 0: k0 = 0, typeA, SLIV 53 -> S = 2, L = 12
                VRB-to-PRB mapping ................. 0    -- nonInterleaved
                Modulation and coding scheme ....... 15
                  -- mcs-Table qam256, DCI 1_1, UE-specific SS
                  -- -> Table 5.1.3.1-2 row 15: Qm 6, R 666/1024
                New data indicator ................. 1
                Redundancy version ................. 0
                HARQ process number ................ 2
                Downlink assignment index .......... 1
                TPC command for scheduled PUCCH .... 1    -- 0 dB
                PUCCH resource indicator ........... 1
                PDSCH-to-HARQ timing indicator ..... 1
                  -- dl-DataToUL-ACK[1] = 2 -> HARQ-ACK in slot 9
                Antenna port(s) .................... 0
                Transmission configuration indication  2
                DMRS sequence initialisation ....... 0

10:14:22.0000  [MAC-DL] PDSCH scheduled
                PRB 4-96 (93 PRB), symbols 2-13, 1 layer
                N_RE ....................... 132 x 93 = 12276
                N_info ..................... 12276 x 0.6504 x 6 = 47915
                TBS computed ............... 48168 bits  (6021 bytes)
                effective code rate ........ 0.656
                LDPC base graph ............ 1,  6 code blocks

Listing 4. One DCI 1_1, with the RIV decode carried out inline. Note that the wrapped RIV branch is used here — a naive decode would give "15 PRBs at PRB 101", which is off the end of a 106-PRB bandwidth part.

17.3 An uplink grant driven by BSR and PHR

10:14:23.5000  [MAC-UL] Long BSR MAC CE received  (LCID 62)
                LCG bitmap ................. 0x0A   -- LCG 1 and LCG 3
                Buffer Size LCG 1 .......... index 8   -> 40 .. 47 B
                Buffer Size LCG 3 .......... index 45  -> ~11-12 kB
10:14:23.5000  [MAC-UL] Single Entry PHR MAC CE received  (LCID 57)
                PH ......................... index 46  -> 8 .. 9 dB
                P_CMAX,f,c ................. index 39  -> 23 dBm

10:14:23.5020  [SCHED] uplink decision for C-RNTI 0x4601
                buffer to serve ............ 11.8 kB
                SRS-based MCS estimate ..... 13
                outer-loop offset (UL) ..... -1.0 dB  -> MCS 11
                PH allows widening to ...... 48 PRB at MCS 11
                free PRB in slot 4 ......... 44
                -> 41 PRB, MCS 11, 1 layer

10:14:23.5030  [PHY-DL] DCI 0_1, C-RNTI 0x4601, SFN 412 slot 1
                Identifier for DCI formats ......... 0    -- uplink
                Frequency domain resource assign ... 4265
                  -- L' = 4265/106 = 40, S' = 25
                  -- (40+1)+25 = 66 <= 106 -> first branch
                  -- L = 41 PRB, S = PRB 25
                Time domain resource assignment .... 2
                  -- row 2: k2 = 3, typeA, SLIV 27 -> S = 0, L = 14
                Frequency hopping flag ............. 0
                Modulation and coding scheme ....... 11
                New data indicator ................. 1
                Redundancy version ................. 0
                HARQ process number ................ 5
                TPC command for scheduled PUSCH .... 1   -- 0 dB
                SRS resource indicator ............. 0
                Precoding info and no. of layers ... 0
                Antenna ports ...................... 0
                DMRS sequence initialisation ....... 0

10:14:23.5045  [MAC-UL] PUSCH  SFN 412 slot 4  (slot 1 + k2 = 4)
                41 PRB, 14 symbols, MCS 11 (16QAM, R 378/1024)
                N_RE = min(156, 156) x 41 = 6396 -> TBS 9480 bits
                                                 = 1185 bytes
                LCP fills: LCG 1 (44 B), then LCG 3 (1129 B), then
                           a padding BSR reporting 10.6 kB still queued

Listing 5. The BSR and PHR arriving, the decision they drive, and the grant that results. The PHR is what bounds the allocation: without it the scheduler would have to assume the worst case.

17.4 SPS activation and release

10:20:00.0000  [RRC] SPS-Config applied
                periodicity ................ ms20
                nrofHARQ-Processes ......... 4
                n1PUCCH-AN ................. 3
                cs-RNTI .................... 0x3F02
                -- configured, but NOT active

10:20:01.2000  [PHY-DL] DCI 1_1, CRC descrambled with CS-RNTI 0x3F02
                New data indicator ......... 0
                HARQ process number ........ 0000    -- validation
                Redundancy version ........ 00       -- validation
                Modulation and coding scheme  9       -- real value
                Frequency domain resource ... RIV 742 -> 8 PRB @ PRB 0
                Time domain resource ....... row 0 -> k0 0, S 2, L 12
10:20:01.2000  [MAC-DL] validation pattern matched -> SPS ACTIVATION
                stored assignment: 8 PRB, MCS 9, S=2 L=12
                N_info = 1056 x 0.6631 x 2 = 1400, quantised to 1392,
                  then up to the next Table 5.1.3.2-1 entry:
                TBS ........................ 1416 bits (177 bytes)
                first occasion ............. SFN 500 slot 4

10:20:01.2400  [MAC-DL] SPS PDSCH occasion 1, no PDCCH
                HARQ pid derived from the slot number .... 0
10:20:01.2450  [PHY-UL] PUCCH HARQ-ACK on resource 3 -> ACK
10:20:01.2600  [MAC-DL] SPS PDSCH occasion 2, +20 ms, HARQ pid 1
                ... 148 occasions, 0 PDCCH transmissions ...

10:20:04.2000  [PHY-DL] DCI 1_1, CS-RNTI 0x3F02
                HARQ process number ........ 0000
                Redundancy version ........ 00
                Modulation and coding scheme  11111   -- all ones
                Frequency domain resource ... all ones
10:20:04.2000  [MAC-DL] release pattern matched -> SPS RELEASED
                remaining SPS occasions cleared

-- PDCCH cost over 3 seconds: 2 DCIs for 149 transport blocks.
-- Dynamically scheduled, the same traffic would have cost 149.

Listing 6. Two DCIs for 149 transport blocks. The last two lines are the whole argument for SPS — and also the whole argument against it, since none of those 149 transmissions could adapt to the channel.

17.5 A grant that never arrives

11:02:17.0000  [RRC] RRCReconfiguration applied
                searchSpacesToAddModList: SearchSpace [4] added
                 searchSpaceId ............ 4
                 controlResourceSetId ..... 1
                 monitoringSlotPeriodicityAndOffset  sl1
                 monitoringSymbolsWithinSlot  11111110000000
                                              -- 7 occasions per slot!
                 nrofCandidates  AL1 n8, AL2 n4
                 searchSpaceType  ue-Specific formats0-1-And-1-1

11:02:17.0000  [PHY-DL] PDCCH budget check, mu = 1: M_max = 36
                SearchSpace 2 ..... 19 candidates
                SearchSpace 4 ..... 12 candidates x 7 occasions = 84
                total ............. 103  >  36
11:02:17.0000  [PHY-DL] dropping UE-specific search spaces in increasing
                        searchSpaceId order until the budget is met
                        -> SearchSpace 4 retained (lower id first),
                           SearchSpace 2 DROPPED
                -- no signalling, no error, no counter

11:02:17.0010  [SCHED] (gNB) DCI 1_1 for C-RNTI 0x4601 placed in
                        SearchSpace 2, AL 2, CCE 4-5
11:02:17.0010  [PHY-DL] (UE)  SearchSpace 2 not monitored -> nothing found
11:02:17.0013  [MAC-DL] (gNB) PDSCH transmitted, no HARQ-ACK expected
                        until slot 9
11:02:17.0020  [PHY-UL] (UE)  PUCCH: nothing to report for this occasion
11:02:17.0020  [MAC-DL] (gNB) HARQ-ACK absent -> treated as NACK
                        retransmit pid 2, RV 2, same search space
                -- and it will be missed again, indefinitely

-- Downlink throughput for this UE: ~0. Radio conditions: excellent.
-- CQI 13, no PDCCH BLER, no HARQ failures reported by either side.

Listing 7. A search space added for mini-slot monitoring pushes the candidate count past M_max, and the UE silently stops monitoring the search space that carried its ordinary grants. Every radio metric looks perfect. This is the failure mode §5.3 and §15 both warn about.

The trace above is the reason to add up nrofCandidates by hand after any change to PDCCH-Config. There is no counter for "search space dropped", no RRC failure, and no radio symptom — the UE is behaving exactly as specified. The only visible evidence is that the gNB places DCIs in a search space the UE is not monitoring, which requires you to have both sides of the log and to know the budget for the numerology.

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

Table 21. Scheduling changes by release. Confirm feature presence against the UE capability exchange — see the companion 26 UE Capability document — before assuming a DCI format or field exists.

19. Reading Scheduling in Logs: A Checklist

1. Establish the bandwidth part before decoding anything. N_BWP_size is the multiplier in the RIV formula and the divisor in the RBG calculation. A DCI 1_0 in a common search space is decoded against the initial bandwidth part, not the active one (§7.2).

2. Identify the DCI format from the RNTI and the search space, not from the size. Two formats of the same size are told apart by the "identifier for DCI formats" bit; two formats in different search spaces may share a size by design (§6.1).

3. Add up `nrofCandidates` across every search space that shares a monitoring occasion, multiply by the occasions per slot, and compare against M_max for the numerology. This one calculation explains a whole class of "missing grant" problems and nothing reports it (§5.3).

4. Count CCE occupancy, not PRB occupancy, when throughput saturates. PDCCH blocking and PRB exhaustion look identical from a throughput graph and have completely different fixes (§4).

5. Recompute the RIV in both branches. If ⌊RIV/N⌋ + 1 gives an allocation that runs off the end of the bandwidth part, you are in the wrapped branch — that impossibility is the branch test (§7.2).

6. Resolve the TDRA row index against the configured list, and check the list is wide enough. Allocations clustering on two or three shapes usually means the list is short, not that the scheduler is unimaginative (§8).

7. Compare granted MCS against CQI-implied MCS. A persistent gap is the outer loop, and whether that is correct depends on the first-transmission BLER, not on the gap (§9.3).

8. For SPS and configured grants, check the validation pattern field by field. An activation that fails validation produces no error and no counter, so "SPS never started" is almost always this (§10.1).

9. Distinguish repetition from retransmission by counting DCIs. One DCI and four transmissions is aggregation and one HARQ attempt; four DCIs and four transmissions is four HARQ attempts (§11).

10. Before blaming radio, check the three silent failures: blind-decode budget exceeded, activation validation mismatch, and RIV decoded against the wrong N. None of them produces a log line, and all three are pure arithmetic (§15).

20. Glossary

cs-RNTI

MCS-C-RNTI

21. References

• 3GPP TS 38.213 — Physical layer procedures for control. Clause 10.1 (UE procedure for determining PDCCH candidates: the search space definitions, the hashing function, and Tables 10.1-2 and 10.1-3 giving the per-slot monitoring limits), clause 10.2 (validation of DCIs activating and releasing SPS and configured grants), clause 11.1 (slot format indication and DCI 2_0), clause 11.2 (pre-emption indication and DCI 2_1), clause 11.2A (uplink cancellation and DCI 2_4), clause 9 (HARQ-ACK reporting and PUCCH resources).
• 3GPP TS 38.212 — Multiplexing and channel coding. Clause 7.3.1 (every DCI format, field by field, including 0_0 to 0_2, 1_0 to 1_2 and 2_0 to 2_7), clause 7.3.1.0 (the DCI size budget), clause 7.3.2 (PDCCH coding and CRC scrambling by RNTI).
• 3GPP TS 38.214 — Physical layer procedures for data. Clause 5.1.2 (PDSCH resource allocation in time and frequency: K0, mapping type, SLIV, FDRA types 0 and 1, Table 5.1.2.2.1-1 for RBG sizes), clause 5.1.3 (modulation order, target code rate and transport block size determination; Tables 5.1.3.1-1 to 5.1.3.1-3 for the MCS tables), clause 5.1.7 (code block group based transmission), clause 5.2.2 (CSI reporting, and Tables 5.2.2.1-2 to 5.2.2.1-4 for the CQI tables), clause 6.1.2 (PUSCH resource allocation, K2, repetition Types A and B).
• 3GPP TS 38.211 — Physical channels and modulation. Clause 7.3.2 (PDCCH: REG and CCE definitions, the CCE-to-REG mapping and the interleaver), clause 7.4.1.3 (PDCCH DMRS).
• 3GPP TS 38.321 — MAC protocol specification. Clause 5.4.1 (configured uplink grants and the HARQ process ID derivation), clause 5.8.1 (reception of a downlink assignment on the configured downlink assignment), clause 5.13 (activation and deactivation handling), clause 6.1.3.7 (Configured Grant Confirmation MAC CE).
• 3GPP TS 38.331 — RRC protocol specification. PDCCH-Config, PDCCH-ConfigCommon, ControlResourceSet, SearchSpace, PDSCH-Config, PUSCH-Config, PDSCH-TimeDomainResourceAllocationList, SPS-Config, ConfiguredGrantConfig, DownlinkPreemption, UplinkCancellation, CrossCarrierSchedulingConfig, BWP-Downlink, BWP-Uplink.
• 3GPP TS 38.300 — NR overall description. Clause 10 (scheduling and rate control, including the explicit statement that the scheduling algorithm is implementation-specific).
• 3GPP TS 23.501 — System architecture. Clause 5.7 (5QI characteristics: priority level, packet delay budget, packet error rate) — the QoS inputs of §3.

Companion documents in this set

• 02 Radio Frame Structure — slots, symbols, bandwidth parts and the TDD patterns that every K0/K1/K2 offset is counted in, plus the mini-slot picture behind §12.1.
• 03 Random Access — the RAR uplink grant, which is a stripped-down grant carried in a MAC RAR rather than a DCI.
• 04 Timing Advance — timeAlignmentTimer expiry clears every configured grant and SPS assignment.
• 05 HARQ — the NDI, RV, DAI and CBG fields of the DCIs in §6, the K1 timing of §8, and the transport block size arithmetic of §9.2.
• 06 Channel Mapping — how DL-SCH and UL-SCH relate to PDSCH and PUSCH, and where PDCCH sits.
• 07 MAC PDU and Control Elements — the BSR, PHR and Configured Grant Confirmation control elements of §3 and §10.
• 09 LCP — how the UE decides what goes into an uplink transport block once the scheduler has sized it.
• 10 BSR, PHR, SR — the three uplink reports that are the scheduler's only view of the uplink.
• 11 DRX — PDCCH monitoring is suspended outside Active Time, which interacts directly with monitoringSlotPeriodicityAndOffset.
• 17 System Information / 18 MIB and SIB1 IEs — where CORESET 0 and search space 0 come from, before any dedicated configuration exists.
• 26 UE Capability — processing capability 2, supported DCI formats, and the carrier-aggregation limits that scale the budgets in §5.3.