5G NR numerology: what SCS tells you and what it cannot prove

A carrier uses 30 kHz subcarrier spacing. Does that mean it is faster than a 15 kHz carrier?

No. It means the time-frequency grid is organised differently.

Subcarrier spacing affects symbol and slot structure. It does not tell you how many PRBs a UE received, which MCS was selected, how many layers were realised, how much retransmission occurred, whether the TDD pattern offered a downlink opportunity, or whether the test path was limited beyond the RAN.

The field sequence is:

1. establish the configured grid;
2. verify that the band, channel bandwidth and SCS combination is supported;
3. identify the carrier and BWP context actually used;
4. separate the design objective from the observed symptom;
5. test other explanations before recommending a numerology change.

Original Takwa Sebai decision map. The two deployment examples are synthetic and are not an A/B performance comparison.

The basic NR numerology relationship

For NR:

Δf = 2^μ × 15 kHz

where μ is the numerology index.

For the normal cyclic-prefix examples used here:

The reference slot duration is the 1 ms subframe divided by the number of slots in that subframe. It is useful for reading the grid. It is not an end-to-end latency promise.

As μ increases:

• the SCS doubles;
• the useful OFDM symbol period becomes shorter;
• the number of slots in the same 1 ms interval doubles;
• the grid can represent scheduling opportunities at a finer time granularity.

The word can is essential. A shorter slot does not prove the scheduler served a UE sooner or the packet crossed the entire service path faster.

Original Takwa Sebai time-frequency comparison. More slot boundaries do not mean more user throughput by themselves.

Read the time and frequency axes together

Time view

Inside the same 1 ms subframe:

• 15 kHz SCS has one reference slot;
• 30 kHz has two;
• 60 kHz has four;
• 120 kHz has eight.

This does not show which symbols are downlink, uplink or flexible in a TDD configuration. It also does not show whether the scheduler assigned resources to the UE.

Frequency view

A larger SCS places adjacent subcarriers farther apart. The resource grid therefore changes in frequency as well as time.

The deployed design also has a band, channel bandwidth, BWP context, SSB configuration, duplex mode, antenna system and real propagation environment. “Shorter slot” describes only one part of that design.

What the grid does not contain

A numerology diagram does not tell you:

• allocated PRBs;
• modulation order or target code rate;
• realised MIMO layers;
• BLER, HARQ or other retransmission pressure;
• cell load and scheduler decisions;
• TDD direction opportunities;
• protocol overhead;
• transport and core limitations;
• subscription, server or device limits.

Treat the grid as a coordinate system, not a performance result.

Check constraints before preferences

The first engineering question is not “Which SCS is best?” It is “Which combinations are defined and supported for this deployment?”

At minimum, establish:

1. operating band and frequency range;
2. duplex mode;
3. configured channel bandwidth;
4. carrier SCS and the SCS of the relevant BWP;
5. SSB and common-configuration context;
6. UE and network capability;
7. software and feature baseline.

The applicable 3GPP release and product documentation still matter. A training article is not a substitute for the deployed equipment baseline.

This order also prevents false comparisons. A low-band FDD coverage layer and a mid-band TDD capacity layer differ in far more than numerology.

Why SCS is not a throughput button

Downlink transport-block determination depends on inputs including:

• modulation order;
• target code rate;
• number of layers;
• number of allocated PRBs.

This supports a narrow conclusion:

SCS alone is insufficient to determine user throughput.

It does not make SCS irrelevant. It means a throughput diagnosis must include the variables that determine how much data was scheduled and delivered successfully.

For a low-throughput investigation, align:

• serving carrier and active BWP;
• UE capability and serving mode;
• allocation across the test interval;
• modulation, coding, rank/layers and retransmission evidence;
• RF condition and mobility state;
• cell load and a comparison interval;
• TDD pattern where applicable;
• transport, server and test method;
• policy or subscription limits.

Only after this should numerology remain on the list of plausible limiting design factors.

Break the false causal chain

A common shortcut is:

30 kHz SCS → shorter slot → lower latency → higher throughput

Each arrow requires evidence.

• Shorter reference slot: a property of the grid.
• Lower radio delay: depends on the configured opportunity and actual scheduling.
• Lower end-to-end latency: depends on RAN, transport, core and application.
• Higher throughput: depends on resources delivered successfully over time.

The chain may guide a design discussion. It is not a field diagnosis.

Synthetic comparison: two deployment roles

These two scenarios are synthetic. They are not measured networks, customer cases or proof of the author's deployment practice.

The cited Release 18 version of TS 38.104 supports these band, channel-bandwidth and SCS combinations.

What the comparison teaches:

• layers may have different network objectives;
• bandwidth and duplex opportunities must be considered;
• UE population and services matter;
• the evidence should be tied to the objective.

What it does not prove:

• that 30 kHz always means speed;
• that 15 kHz always means coverage;
• that SCS caused any performance difference;
• that one scenario has better latency or spectral efficiency;
• that the result would persist with different load, site grid, antenna or TDD pattern.

Those claims require controlled evidence.

Verify the deployed reality in three layers

Layer 1: planned and configured design

Preserve an authorised export or screenshot containing:

• band and duplex mode;
• channel bandwidth;
• carrier and BWP SCS;
• SSB/common configuration;
• TDD pattern where applicable;
• release, software and feature baseline.

Keep the object identity and timestamp. Do not rewrite the configuration from memory.

Layer 2: UE and signalling context

Use an authorised trace or device log to determine:

• the serving cell and carrier;
• NSA or SA context;
• the relevant BWP;
• reconfiguration during the test;
• mobility, fallback or another state transition.

Do not infer the active context from one isolated decoded field.

Layer 3: observed behaviour

Align:

• exact test start and stop time;
• location and radio conditions;
• UE model, capability and thermal state;
• allocation and load;
• retransmission and MIMO evidence;
• transport and server measurements;
• a known-good comparison.

Configuration describes intent. Traces and measurements describe what happened to this UE in this interval.

Decision workflows by symptom

Throughput below expectation

1. Prove the test is repeatable.
2. Confirm cell, carrier, mode and BWP.
3. Verify band, bandwidth and SCS against the design.
4. Inspect allocation, MCS, layers and retransmissions.
5. Check load and TDD opportunities.
6. Exclude transport, server, policy and UE limits.
7. Compare with a known-good case using the same carrier design.
8. Consider a controlled numerology trial only if the evidence still points there.

Latency above expectation

Separate:

• scheduling and retransmission delay;
• TDD waiting time;
• RAN processing and queues;
• transport and core;
• server and application response;
• measurement resolution and clock alignment.

A 0.5 ms reference slot cannot explain a 30 ms end-to-end result on its own.

Coverage or mobility unstable

Verify:

• band and antenna/beam context;
• serving and neighbour measurements;
• SSB and measurement configuration;
• handover or beam-management phase;
• speed and route repeatability;
• timing, interference and retransmission evidence.

Numerology is not a substitute label for every RF trade-off.

Before changing SCS

Reject a proposal such as “speed is low, use a wider SCS” until it defines:

• why the current combination is wrong or limiting;
• whether allocation, MIMO and delivery are healthy;
• whether transport and policy are excluded;
• whether the alternative is supported;
• what observation would confirm success;
• what would falsify the hypothesis;
• the service risk and rollback;
• the comparison window and acceptance criterion.

A change should answer a measured question. It should not be an experiment performed because the grid diagram looks faster.

Minimum evidence package

Continue the evidence chain

• Prove which BWP was active.
• Read RSRP, RSRQ and SINR together.
• Use the complete low-throughput workflow.
• Read the complete 5G NR field guide.

Download the free illustrated 111-page ebook.

Technical sources

• 3GPP TS 38.211 V18.9.0, Release 18, clauses 4.2 and 4.3.2 for SCS and normal-CP slot structure. ETSI publication.
• 3GPP TS 38.104 V18.13.0, Release 18, clause 5.3.5 for channel-bandwidth/SCS combinations. ETSI publication.
• 3GPP TS 38.214 V18.8.0, Release 18, clause 5.1.3 for transport-block determination inputs. ETSI publication.