How is the Transport Block Size (TBS) calculated in 5G NR?

Per TS 38.214: first compute the resource elements per PRB available for data, N'RE = 12 × Nsymb − NDMRS − Noh, then cap it, NRE = min(156, N'RE) × nPRB. Next the information bits, Ninfo = NRE × R × Qm × v, where R is the target code rate from the MCS table, Qm the modulation order and v the number of layers. Finally Ninfo is quantised to a valid TBS using the 38.214 rules (a table for Ninfo ≤ 3824, a formula above it). Dividing the TBS by the slot duration gives the instantaneous bit rate; multiplying by (1 − BLER) gives the delivered throughput.

How does the TDD DL/UL slot ratio affect 5G throughput?

In TDD the same carrier is time-shared between downlink and uplink, so only a fraction of slots carry your direction. With a DDDSU pattern (3 full downlink, 1 special, 1 uplink per 5 slots), roughly 70–75% of the airtime is downlink, so the delivered downlink throughput is about 0.7–0.75 of the all-downlink peak, and uplink is far lower. The special slot contributes partial symbols. Choosing the TDD pattern (e.g. DDDSU vs DDDDDDDSUU) directly trades downlink against uplink capacity, which is why it is one of the biggest levers on real-world rates.

How many MIMO layers does 5G NR support and how do they scale throughput?

5G NR supports up to 8 layers in downlink and up to 4 in uplink per UE (more across multi-user MIMO at the cell level). Throughput scales almost linearly with the number of spatial layers because each layer carries an independent stream over the same time-frequency resource — so 4 layers roughly quadruples the single-layer rate, provided the radio channel actually has enough spatial richness (rank) and SINR to support them. In poor or line-of-sight channels the usable rank collapses and extra antenna ports do not help, which is why measured layer counts, not configured ones, drive real throughput.

What is the maximum code rate Rmax in 5G NR?

The maximum code rate used in the peak data-rate formula is Rmax = 948/1024 ≈ 0.926. The actual code rate for a given transmission comes from the MCS table in TS 38.214 (target code rate ÷ 1024), which together with the modulation order defines the spectral efficiency. Higher MCS indices use higher code rates and higher-order modulation, giving more bits per resource element but needing more SINR; the scheduler picks the highest MCS the channel can carry at the target BLER (typically 10%).

Why is real 5G throughput lower than the peak number?

The peak figure assumes the best modulation (256QAM), maximum layers, full bandwidth, all slots in your direction and zero errors. Reality subtracts on every axis: the TDD pattern gives you only part of the slots; the scheduler shares PRBs with other users; the channel rarely supports max rank or 256QAM everywhere, so the average MCS and layer count are lower; DMRS, control and other overhead consume resource elements; and BLER plus HARQ retransmissions reduce delivered bits. A realistic figure is typically 30–60% of the marketing peak, and the calculation in this article lets you predict it for your own configuration.

5G NR Throughput Calculation — PRB, MCS, MIMO Layers, TDD & Overhead

Q: What is the 5G NR peak data-rate formula?

3GPP TS 38.306 gives the approximate peak data rate as a sum over component carriers: data rate (Mbps) = 10^-6 × Σ ( v × Qm × f × Rmax × (NPRB × 12 / Ts) × (1 − OH) ), where v is the number of MIMO layers, Qm is the modulation order (2/4/6/8 for QPSK/16QAM/64QAM/256QAM), f is the UE scaling factor (1, 0.8, 0.75 or 0.4), Rmax = 948/1024 is the maximum code rate, NPRB is the number of resource blocks for the bandwidth and numerology, 12 is the subcarriers per RB, Ts is the average OFDM symbol duration (10^-3 / (14 × 2^μ)), and OH is the overhead (0.14 DL FR1, 0.18 DL FR2, 0.08 UL FR1, 0.10 UL FR2). It is the theoretical ceiling; real throughput is lower because of the TDD slot ratio, scheduling, channel conditions and BLER.

CHAPTER ONE

Throughput is a calculation, not a number

“How fast is 5G?” is a trick question, because the honest answer is a formula, not a figure. A 100 MHz cell can hand a single UE 2.3 Gbps in a lab and 200 Mbps on a busy street, and both numbers are correct — they are the same formula evaluated with different inputs. Every engineer who plans capacity, validates a site, sizes transport or explains a slow speed test is really doing one thing: walking the throughput equation and seeing which term collapsed. Learn the equation once and you stop guessing.

There are two layers to know. The first is the peak data-rate formula in TS 38.306 — a clean closed form the industry uses for the headline “up to” numbers and UE category limits. The second is the Transport Block Size (TBS) calculation in TS 38.214 — the exact per-slot bit count the scheduler actually uses. The peak formula tells you the ceiling; the TBS calculation tells you what a given slot really carries. This article walks both, then strips the peak down to a realistic delivered rate, factor by factor.

Where this lives in 3GPP. The peak data-rate formula and UE capability scaling are in TS 38.306 §4.1.2. The TBS determination, MCS tables and code rates are in TS 38.214 §5.1.3. The bandwidth-to-PRB mappings are in TS 38.101-1 (FR1) and 38.101-2 (FR2). The frame/slot structure that sets the TDD ratio is in TS 38.211 and configured per TS 38.213.

CHAPTER TWO

The peak data-rate formula, term by term

Here is the whole thing, straight from TS 38.306. It looks dense; it is actually just “bits per resource element × resource elements per second, summed over carriers.”

Approximate peak data rate · TS 38.306 §4.1.2 DataRate (Mbps) = 10⁻⁶ · Σ_j=1..J [ v_Layers^(j) · Q_m^(j) · f^(j) · R_max · (N_PRB^BW,μ · 12 / T_s^μ) · (1 − OH^(j)) ]

T_s^μ = 10⁻³ / (14 · 2^μ) is the average OFDM symbol duration. R_max = 948/1024. The 12 is subcarriers per resource block. J = number of aggregated carriers.

Term	Meaning	Typical value
v_Layers	Number of MIMO layers (spatial streams)	1–8 DL, 1–4 UL
Q_m	Modulation order (bits per symbol)	2 / 4 / 6 / 8 (QPSK…256QAM)
f	UE scaling factor (capability)	1, 0.8, 0.75, 0.4
R_max	Maximum code rate	948/1024 ≈ 0.926
N_PRB	Resource blocks for the bandwidth & numerology	e.g. 273 @ 100 MHz / 30 kHz
12 / T_s	Resource elements per second per PRB	derived from μ
OH	Overhead fraction	0.14 DL FR1 · 0.18 DL FR2 · 0.08 UL FR1 · 0.10 UL FR2

The structure is the key insight: throughput is a product. Halve any factor and you halve the result; collapse any factor to a fraction and the whole rate follows. That is why a single weak link — rank-1 channel, a downlink-poor TDD pattern, a UE that only does 64QAM — caps everything regardless of how generous the others are. The rest of this article is just the seven terms, one chapter each.

CALCULATOR

Interactive throughput calculator

Try the formula yourself. Set the carrier, layers, modulation, code rate, TDD pattern and BLER, and the peak and delivered rates update live — the same maths as TS 38.306, running in your browser.

5G NR Throughput Calculator

Peak per TS 38.306 · delivered = peak × TDD DL fraction × (1 − BLER)

Carrier (bandwidth · SCS)

MIMO layers

Modulation

Code rate R — 0.93

TDD pattern (DL fraction)

Overhead (OH)

Carriers (CA)

First-tx BLER — 0%

UE scaling f

Peak data rate (all-DL ceiling)

—

Delivered (× TDD × (1−BLER))

—

Peak follows TS 38.306 with R_max replaced by your chosen R. “Delivered” applies the TDD DL fraction and BLER only; real rates also depend on measured rank, average MCS and PRB scheduling (see the chapters below).

CHAPTER THREE

Bandwidth → PRBs — numerology & the RB table

Bandwidth does not enter the formula directly — the number of resource blocks does. A resource block is always 12 subcarriers, but how many fit in a channel depends on the subcarrier spacing (SCS), set by the numerology μ (SCS = 15 × 2^μ kHz). Wider SCS means fewer, fatter subcarriers, so the PRB count is not simply proportional to MHz. 3GPP tabulates the exact maximums; here are the ones you will actually use.

Channel BW	15 kHz (μ=0)	30 kHz (μ=1)	60 kHz (μ=2)	120 kHz (FR2)
20 MHz	106	51	24	—
50 MHz	270	133	65	—
100 MHz	—	273	135	66
200 MHz (FR2)	—	—	—	132
400 MHz (FR2)	—	—	—	264

The other half of the bandwidth term is resource elements per second. There are 12 subcarriers per PRB and, per the formula, 14 × 2^μ OFDM symbols per millisecond. So a higher numerology packs more symbols per second — which exactly cancels the lower PRB count, keeping the RE-per-second-per-MHz roughly constant. Numerology is about latency and frequency range, not raw capacity.

Worked example — resource elements per second at 100 MHz

μ = 1 (30 kHz) → T_s = 10⁻³ / (14 × 2) = 35.71 µs

N_PRB = 273 → RE/s = 273 × 12 / 35.71µs = 3276 / 35.71e−6

≈ 91.7 million resource elements per second — the raw “canvas” every other factor paints on.

CHAPTER FOUR

Modulation & MCS — Q_m, code rate, spectral efficiency

Each resource element carries Q_m raw bits, but not all of them are your bits — the channel code adds redundancy. The product of modulation order and code rate is the spectral efficiency: the information bits per resource element. The scheduler picks both at once through the Modulation and Coding Scheme (MCS) index, which maps to a (Q_m, target code rate) pair in TS 38.214.

Modulation	Q_m (bits/RE)	Max code rate	Max spectral efficiency
QPSK	2	~0.12–0.93	up to ~1.9 b/RE
16QAM	4	~0.37–0.74	up to ~2.9 b/RE
64QAM	6	~0.45–0.93	up to ~5.6 b/RE
256QAM	8	~0.67–0.93	up to 7.41 b/RE
1024QAM (Rel-17, FR1 DL)	10	up to ~0.93	up to ~9.3 b/RE

The catch is that higher MCS needs more SINR. The scheduler runs link adaptation: it picks the highest MCS the channel can carry at the target block-error rate (typically 10% first-transmission BLER), guided by the UE's CQI reports. Near the tower you get 256QAM at a high code rate (7+ b/RE); at the edge you fall to QPSK at a low code rate (well under 1 b/RE). That single factor can swing throughput by 8× across a cell.

The scheduler walks this curve in real time — the measured MCS, not the configured maximum, sets your bits per resource element.

Premium interactive lab

Build your own throughput calculator

Set bandwidth, numerology, MCS, layers and TDD pattern and watch the peak and delivered rate update live — the same maths as this article, made interactive. Lifetime access — ₹999 / $9.99.

Open the 5G PHY Lab

CHAPTER FIVE

MIMO layers — the linear multiplier

Layers are the cleanest term in the formula: each spatial layer carries an independent data stream over the same time-frequency resource, so throughput scales almost linearly with layer count. Two layers double the rate; four quadruple it. 5G NR supports up to 8 layers in downlink and 4 in uplink per UE.

Linear — if the channel has the spatial rank and SINR to carry the streams. In line-of-sight or low-SINR conditions the usable rank collapses.

The asterisk on “linear” matters in the field. Layers only help if the radio channel has the rank to support them — rich multipath scattering that keeps the streams separable. A clean line-of-sight link or a low-SINR cell edge may support only rank 1–2 no matter how many antenna ports the gNB has. This is why measured average layer count, reported via the rank indicator (RI), is one of the most revealing throughput KPIs: configured 4×4 with a measured rank of 1.3 explains a lot of disappointing speed tests.

CHAPTER SIX

TBS — how a slot's bits are really counted

The peak formula is an average; the scheduler needs an exact integer bit count for each transmission. That is the Transport Block Size, computed per TS 38.214 in four steps. This is the calculation that actually decides what a slot carries.

The cap at 156 RE/PRB (12 subcarriers × 13 usable symbols) is the spec's ceiling on usable resource elements per PRB.

Worked example — TBS for one slot at 100 MHz

273 PRB, 14 symbols, 1 DMRS symbol (12 RE/PRB), N_oh=0 → N'RE = 12×14 − 12 = 156

N_RE = min(156,156) × 273 = 42,588 RE

256QAM (Q_m=8), R=948/1024, v=4 → N_info = 42,588 × 0.926 × 8 × 4 ≈ 1.26 Mbit

≈ 1.26 Mbit per slot ÷ 0.5 ms slot ≈ 2.52 Gbps instantaneous — the all-downlink ceiling for this config.

CHAPTER SEVEN

The TDD pattern — the slot-ratio reality

Here is the factor that most often surprises people. Almost all mid-band 5G is TDD: downlink and uplink share the same carrier in time, so only a fraction of slots carry your direction. The TDD pattern (configured via tdd-UL-DL-ConfigurationCommon) is a repeating sequence of Downlink, Uplink and Special slots — and it directly multiplies your throughput by the fraction of slots you actually get.

Pick the pattern and you pick the DL/UL split. DDDSU favours downlink; DDDDDDDSUU even more so; symmetric patterns trade DL for UL.

So the 2.52 Gbps all-downlink ceiling from the TBS example becomes roughly 0.7 × that — about 1.75 Gbps — the instant you account for a realistic DDDSU pattern, before any other loss. Uplink, with one full U slot in five plus part of the special slot, lands far lower. The special slot's split into downlink, guard and uplink symbols (nrofDownlinkSymbols / nrofUplinkSymbols) adds a little of each direction back. The TDD pattern is, after MCS and layers, the single biggest lever on real-world rates.

CHAPTER EIGHT

Overhead, BLER & HARQ — peak to delivered

Three more subtractions stand between the slot rate and what the user sees.

Overhead

Not every resource element carries your data. DMRS (demodulation reference signals), control channels (PDCCH), reference signals (CSI-RS, SSB) and the like consume REs. The peak formula bundles this into a single OH term (0.14 DL FR1, 0.18 DL FR2, 0.08 UL FR1, 0.10 UL FR2); the TBS calculation models it explicitly via the DMRS RE count and the configured xOverhead (0, 6, 12 or 18 RE per PRB). Either way, expect to lose roughly 14–18% off the top in downlink.

BLER and HARQ

The scheduler targets ~10% block-error rate on the first transmission, so about one in ten transport blocks needs a HARQ retransmission. Those retransmissions cost airtime that could have carried new data, so delivered throughput is the slot rate times roughly (1 − residual BLER). With HARQ combining the residual error rate after retransmissions is tiny, but the retransmitted slots are still “spent,” which is why you model the first-transmission BLER as an efficiency loss, not just an error rate.

Peak vs delivered, in one sentence. Take the peak, then multiply by your TDD DL fraction, by (measured layers / max layers), by (measured spectral efficiency / max spectral efficiency), by the scheduling/PRB-sharing fraction, and by (1 − first-tx BLER). The product is what the speed test shows — usually 30–60% of the headline number.

CHAPTER NINE

Two worked examples — peak & realistic

Example A — downlink peak (lab conditions)

100 MHz, μ=1, N_PRB=273, v=4, 256QAM (Q_m=8), f=1, R_max=0.926, OH=0.14

RE/s = 273 × 12 / 35.71µs = 91.7e6

rate = 4 × 8 × 1 × 0.926 × 91.7e6 × (1−0.14) × 10⁻⁶

≈ 2,337 Mbps — the headline “up to 2.3 Gbps” per 100 MHz carrier.

Example B — realistic downlink (busy cell, mid-cell UE)

Start from the 2,337 Mbps peak, then apply real-world factors:

× TDD DDDSU (DL airtime ~0.70) → 1,636 Mbps

× layers 2 instead of 4 (rank-2 channel, ×0.5) → 818 Mbps

× 64QAM @ R~0.75 instead of 256QAM @ 0.926 (SE 4.5 vs 7.41, ×0.61) → 499 Mbps

× PRB sharing with other users (~0.6) × (1−BLER ~0.9) → ~270 Mbps

≈ 250–300 Mbps delivered — exactly the range a real speed test shows on a live mid-band cell.

Both numbers came from the same formula. The peak is not a lie and the delivered figure is not a failure — they are the two ends of the same product, and now you can compute either for any configuration you are handed.

CHAPTER TEN

From numbers to the field — what limits throughput

When a cell underdelivers, the calculation tells you exactly which term to interrogate — and each maps to a measurable KPI.

Symptom	Collapsed term	KPI to check / action
Low rate everywhere	Bandwidth / PRBs	Carrier aggregation; check allocated BWP & PRB grant
Low rate at the edge only	MCS (low SINR)	Avg MCS / CQI distribution; coverage, interference, power
Configured 4×4 but low rate	Layers (low rank)	Avg RI; antenna config, channel scattering, SINR for rank
Good DL, poor UL	TDD slot ratio	Review UL/DL pattern; consider supplementary uplink
Peak fine, busy-hour poor	PRB sharing	PRB utilisation; add capacity / carriers / load balance
Throughput sawtooths	BLER / HARQ	BLER & retransmission rate; link-adaptation tuning

This is the real value of knowing the formula: a slow cell is never a mystery, it is a term that went to a fraction. Find the term, check its KPI, fix the cause. Throughput is not magic — it is arithmetic with radio inputs.

CHAPTER ELEVEN

Frequently asked questions

What is the 5G NR peak data-rate formula?

DataRate = 10⁻⁶ · Σ ( v · Qm · f · Rmax · (NPRB · 12 / Ts) · (1 − OH) ) over carriers, where v is layers, Qm modulation order, f the UE scaling factor, Rmax = 948/1024, NPRB the resource blocks, Ts the OFDM symbol duration, and OH the overhead. It is the ceiling; delivered rate is lower.

How is TBS calculated?

N'RE = 12·Nsymb − NDMRS − Noh per PRB; NRE = min(156, N'RE) × nPRB; Ninfo = NRE × R × Qm × v; then Ninfo is quantised to a valid TBS. Throughput = TBS / slot-duration × (1 − BLER).

How does the TDD pattern affect throughput?

TDD time-shares the carrier, so you only get the slots in your direction. DDDSU gives ~70% downlink airtime, so DL throughput is ~0.7× the all-downlink peak and UL is much lower. The pattern is a direct DL-vs-UL trade.

How do MIMO layers scale throughput?

Almost linearly — each layer is an independent stream on the same resource, so 4 layers ≈ 4× one layer — if the channel has the rank and SINR. Measured rank (RI), not configured ports, drives real rates.

Why is real throughput far below the peak?

The peak assumes 256QAM, max layers, full bandwidth, all slots downlink and no errors. Reality subtracts the TDD ratio, PRB sharing, lower average MCS and rank, overhead and BLER — typically leaving 30–60% of the headline figure.

• • •

Throughput is the most-searched 5G number and the most misunderstood — because it is a product of seven factors masquerading as a single figure. Walk the equation, find the collapsed term, and you can predict, explain or fix any rate on any cell. For the coding that sets R_max and the BLER behaviour, see the companion deep dive on LDPC & Polar coding.