Ferkans — Interactive Telecom Tutor

The URLLC Challenge

Ultra-Reliable Low-Latency Communication (URLLC) is one of the three service categories in 5G NR (alongside eMBB and mMTC). The defining requirement is the delivery of small packets (typically 32--256 bytes) within 1 ms end-to-end latency at a reliability of $1 - 10^{-5}$ or higher.

These twin constraints --- short latency and extreme reliability --- pull in opposite directions. Short latency limits the blocklength $n$ , which by the normal approximation (Section 32.1) reduces the achievable rate and makes error probability harder to control. Extreme reliability demands either high SNR or diversity, both of which consume resources.

The design of URLLC therefore revolves around three levers: (i) controlling the blocklength penalty via short-packet coding, (ii) exploiting diversity to steepen the error probability curve, and (iii) physical-layer frame design (mini-slots) that minimises alignment and scheduling delays.

Definition:
User-Plane Latency Budget

The user-plane latency for a one-way URLLC transmission is

$T_{\mathrm{total}} = T_{\mathrm{align}} + T_{\mathrm{tx}} + T_{\mathrm{proc}} + T_{\mathrm{prop}} + T_{\mathrm{retx}},$

where:

$T_{\mathrm{align}}$ : waiting time until the next transmission opportunity (depends on slot/mini-slot structure),
$T_{\mathrm{tx}} = n \cdot T_s$ : transmission time for $n$ symbols at symbol period $T_s$ ,
$T_{\mathrm{proc}}$ : processing delay at transmitter and receiver,
$T_{\mathrm{prop}}$ : propagation delay (negligible for short distances),
$T_{\mathrm{retx}}$ : additional delay if HARQ retransmission is needed.

5G NR targets $T_{\mathrm{total}} \leq 1$ ms for downlink URLLC.

With a 30 kHz subcarrier spacing ( $T_s \approx 33.3\,\mu$ s OFDM symbol including CP), a 14-symbol slot lasts $0.5$ ms, which already consumes half the budget. This motivates mini-slots of 2, 4, or 7 OFDM symbols.

,

Definition:
Mini-Slot (Non-Slot-Based Scheduling)

A mini-slot is a scheduling unit in 5G NR that is shorter than a full slot (14 OFDM symbols). Mini-slots can span 2, 4, or 7 OFDM symbols and can start at any symbol position within a slot, enabling:

Reduced alignment delay: $T_{\mathrm{align}}$ is bounded by the mini-slot duration rather than the full slot duration.
Preemption: a URLLC mini-slot can puncture (preempt) an ongoing eMBB transmission, with the eMBB user informed via a preemption indication (DCI format 2_1).
Short blocklength: a 2-symbol mini-slot at 30 kHz SCS over 20 MHz bandwidth ( $\approx 600$ subcarriers) yields $n \approx 1200$ resource elements.

,

Example: URLLC Latency Calculation with Mini-Slots

Consider a 5G NR cell with 30 kHz subcarrier spacing ( $T_{\mathrm{sym}} = 33.3\,\mu$ s including CP). A URLLC downlink transmission uses a 2-symbol mini-slot. The processing delays are $T_{\mathrm{proc,gNB}} = 100\,\mu$ s (gNB) and $T_{\mathrm{proc,UE}} = 200\,\mu$ s (UE). Propagation delay is $10\,\mu$ s. Assuming no retransmission is needed:

(a) Compute the worst-case user-plane latency. (b) Is a single HARQ retransmission feasible within the 1 ms budget?

Solution

Worst-case alignment delay

The worst case occurs when the packet arrives just after a mini-slot boundary. The alignment delay is at most one mini-slot duration:

$T_{\mathrm{align}} = 2 \times 33.3\,\mu\text{s} = 66.7\,\mu\text{s}.$

Transmission and total latency

$T_{\mathrm{tx}} = 2 \times 33.3 = 66.7\,\mu\text{s},KATEXPLACEHOLDER0ENDT_{\mathrm{total}} = 66.7 + 66.7 + 100 + 200 + 10 = 443.4\,\mu\text{s}.$ $

This is well within the 1 ms budget.

Feasibility of one HARQ retransmission

A retransmission requires: feedback (HARQ-ACK) processing at UE ( $\approx 200\,\mu$ s), uplink HARQ-ACK transmission ( $\approx 33.3\,\mu$ s), gNB processing ( $\approx 100\,\mu$ s), and a second mini-slot transmission ( $66.7\,\mu$ s). Total retransmission overhead:

$T_{\mathrm{retx}} \approx 200 + 33.3 + 100 + 66.7 = 400\,\mu\text{s}.$

Total with one retransmission: $443.4 + 400 = 843.4\,\mu$ s $< 1$ ms.

Yes, one HARQ retransmission is feasible.

Theorem: Diversity Gain under Finite Blocklength

,

URLLC: How Diversity Steepens the Reliability Curve

Error probability curves for

L = 1, 2, 4

diversity branches, showing how spatial/frequency diversity reduces the SNR required to meet the URLLC target

\varepsilon \leq 10^{-5}

.

Each additional diversity branch steepens the error probability slope by one order of magnitude per 10 dB of SNR. With

L = 4

, the

10^{-5}

target is reached at moderate SNR.

URLLC Error Probability vs SNR with Diversity

Visualise how the block error probability $\bar{\varepsilon}$ decreases with SNR for different diversity orders $L$ . The plot uses the normal approximation averaged over Rayleigh fading with MRC combining. Observe the diversity slope (steeper curves for larger $L$ ) and the finite blocklength penalty (gap to the outage probability curve at $n \to \infty$ ). Adjust the blocklength $n$ to see how shorter codes shift the curves upward.

Parameters

Diversity order

L

2

Blocklength

n

200

Rate

R

(bits/use)0.5

Resource Overhead for URLLC Reliability

Achieving $\varepsilon = 10^{-5}$ at short blocklength requires a significant resource overhead compared to eMBB. The main strategies in 5G NR are:

Low code rate: URLLC typically uses code rates $R_c \leq 1/3$ with LDPC or polar codes, sacrificing spectral efficiency for reliability.
Frequency diversity: allocating URLLC across multiple non-contiguous PRBs to exploit frequency selectivity.
Repetition / multi-slot HARQ: transmitting copies in multiple mini-slots provides time diversity.
Multi-TRP (Transmission-Reception Point): transmitting from multiple gNBs provides macro-diversity against blockage and shadow fading.

The fundamental trade-off is between URLLC reliability and the resources "stolen" from eMBB traffic, quantified by the eMBB throughput loss per URLLC user.

Definition:
Diversity-Multiplexing-Latency (DML) Trade-off

Extending the classical diversity-multiplexing trade-off (DMT) to account for finite blocklength, the DML trade-off characterises the achievable triplet $(d, r, n)$ , where:

$d$ : diversity order (error probability exponent in SNR),
$r = R / \log_2(1 + \gamma)$ : multiplexing gain (fraction of capacity used),
$n$ : blocklength (proxy for latency).

For a MIMO channel with $N_t$ transmit and $N_r$ receive antennas, the finite-blocklength DMT satisfies

$d(r, n) \leq (N_t - \lceil r \rceil)(N_r - \lceil r \rceil),$

with the bound tightened by the dispersion penalty. Shorter blocklengths reduce the effective multiplexing gain at which a given diversity order is achievable.

Quick Check

A 5G NR URLLC transmission uses a 2-symbol mini-slot at 30 kHz SCS. The OFDM symbol duration (including CP) is approximately 33.3 $\mu$ s. What is the transmission time $T_{\mathrm{tx}}$ , and how does it compare to a full 14-symbol slot?

$T_{\mathrm{tx}} \approx 66.7\,\mu$ s, which is $1/7$ of the full slot duration of $\approx 0.5$ ms

$T_{\mathrm{tx}} \approx 133\,\mu$ s, which is $2/7$ of the full slot

$T_{\mathrm{tx}} \approx 66.7\,\mu$ s, which is $1/14$ of the full slot

$T_{\mathrm{tx}} \approx 33.3\,\mu$ s, which is $1/14$ of the full slot

Correction:

T_{\mathrm{tx}} \approx 66.7\,\mu

s, which is

1/7

of the full slot duration of

\approx 0.5

ms

$T_{\mathrm{tx}} = 2 \times 33.3\,\mu$ s $= 66.7\,\mu$ s. A full slot has 14 symbols, lasting $14 \times 33.3 = 466.7\,\mu$ s $\approx 0.5$ ms. The ratio is $2/14 = 1/7$ . This dramatic reduction in $T_{\mathrm{tx}}$ and $T_{\mathrm{align}}$ is what makes mini-slots essential for meeting the 1 ms latency budget.

Outer-Loop Link Adaptation for URLLC

Input: Target BLER

\varepsilon_{\mathrm{target}}

(e.g.,

10^{-5}

);

BLER estimation window

W

; initial SNR margin

\Delta_0

(dB);

step sizes

\delta_{\mathrm{up}}

,

\delta_{\mathrm{down}}

(dB)

Output: Updated MCS for each URLLC transmission

1. Initialise: SNR margin

\Delta \leftarrow \Delta_0

2. For each URLLC transmission

k = 1, 2, \ldots

:

a. Measure CQI and compute effective SNR:

\hat{\gamma}_k

b. Adjust SNR:

\gamma_{\mathrm{eff}} = \hat{\gamma}_k - \Delta

c. Select MCS such that

R^*(n, \varepsilon_{\mathrm{target}})

at

\gamma_{\mathrm{eff}}

is achievable (using normal approximation

or pre-computed lookup table)

d. Transmit and observe HARQ outcome

a_k \in \{

ACK, NACK

\}

e. Update margin (outer loop):

- If

a_k =

NACK:

\Delta \leftarrow \Delta + \delta_{\mathrm{up}}

(e.g.,

\delta_{\mathrm{up}} = 1

dB)

- If

a_k =

ACK:

\Delta \leftarrow \Delta - \delta_{\mathrm{down}}

where

\delta_{\mathrm{down}} = \delta_{\mathrm{up}} \cdot \varepsilon_{\mathrm{target}} / (1 - \varepsilon_{\mathrm{target}})

3. The ratio

\delta_{\mathrm{down}} / \delta_{\mathrm{up}}

ensures

convergence to the target BLER:

at equilibrium,

\Pr[\text{NACK}] \cdot \delta_{\mathrm{up}} = \Pr[\text{ACK}] \cdot \delta_{\mathrm{down}}

\implies

BLER

= \varepsilon_{\mathrm{target}}

,

URLLC/eMBB Coexistence and Preemption

In 5G NR, URLLC and eMBB traffic share the same carrier. When a URLLC packet arrives during an ongoing eMBB transmission, the gNB can preempt (puncture) part of the eMBB allocation:

The URLLC mini-slot overwrites a subset of eMBB resource elements.
The eMBB UE is notified via a preemption indication (PI) in the next DCI format 2_1, enabling it to discard the corrupted coded bits before LDPC decoding.
Without PI, the eMBB decoder treats the punctured symbols as erasures, causing throughput loss.

The impact on eMBB can be quantified as follows. If a fraction $\alpha$ of eMBB resources are punctured, and the eMBB code rate is $R_c$ , then the effective code rate after puncturing is $R_c / (1 - \alpha)$ . For typical URLLC loads ( $\alpha \lesssim 5\%$ ), the eMBB throughput loss is modest.

⚠️Engineering Note

URLLC Physical Layer Design in 5G NR

5G NR Release 15/16 introduces several physical layer features specifically designed for URLLC:

Mini-slot scheduling:

Mini-slots of 2, 4, or 7 OFDM symbols (vs 14 for a full slot).
At 30 kHz SCS (FR1), a 2-symbol mini-slot has duration $2 \times 33.3\,\mu\text{s} = 66.7\,\mu\text{s}$ — enabling sub-millisecond transmission.
Mini-slot-level HARQ feedback reduces retransmission latency.

Configured grant (Type 1 and Type 2):

Type 1: RRC-configured periodic resources without dynamic grant — eliminates scheduling request + grant latency ( $\sim 0.5$ -- $1$ ms saving).
Type 2: Semi-persistent scheduling activated by DCI.
Both types pre-allocate resources for URLLC traffic, avoiding the 4-step RACH + SR + BSR + grant procedure.

PDCCH reliability:

Aggregation levels up to 16 (vs 8 for eMBB) for higher DCI detection reliability.
Compact DCI format 0_2 / 1_2 with smaller payload for URLLC scheduling.

Redundancy versions and HARQ:

Up to 4 redundancy versions (RV 0, 2, 3, 1) for incremental redundancy. URLLC typically uses RV 0 (self-decodable) to avoid waiting for retransmission.
K1 (HARQ timing) can be as short as 0 slots with mini-slot PUCCH.

Reliability target: 3GPP defines $10^{-5}$ BLER for the physical layer (transport block level) and $10^{-6}$ for the service layer (with HARQ retransmissions).

Practical Constraints

•
Mini-slot: 2/4/7 OFDM symbols (FR1: 66.7/133.3/233.3 μs at 30 kHz SCS)
•
Configured grant eliminates SR+grant latency (~0.5-1 ms saving)
•
PDCCH aggregation level up to 16 for URLLC
•
Target: 10^-5 BLER (transport block), 10^-6 (service layer)

📋 Ref: 3GPP TS 38.214 Section 5.1 (URLLC), 3GPP TR 38.824

Common Mistake: Confusing User-Plane Latency with End-to-End Delay

Mistake:

"5G NR achieves 1 ms latency for URLLC, so the end-to-end delay from application to application is 1 ms."

Correction:

The 1 ms target refers to user-plane latency — the time from when the UE's MAC layer receives a packet to when the gNB's MAC layer delivers it (or vice versa). This excludes:

Core network delay: Typically 1--5 ms for the UPF path (or $< 1$ ms with MEC edge deployment).
Application processing delay: Codec, sensor fusion, or control algorithm execution time.
Queuing delay: Contention with other traffic at the MAC scheduler.
Backhaul delay: If the gNB connects via non-ideal backhaul.

End-to-end delay for industrial automation (e.g., motion control) is typically budgeted at 5--10 ms, of which only 1 ms is the radio segment. Achieving sub-ms total latency requires Multi- access Edge Computing (MEC) to move the application close to the radio.

Why This Matters: Finite Blocklength Theory and Statistical Inference

The finite blocklength framework connects deeply to statistical hypothesis testing studied in Book FSI (Chapters 1--3). The meta-converse bound in Section 32.1 is derived from the Neyman-Pearson lemma: the converse on channel coding rate at finite blocklength reduces to a binary hypothesis test between the true channel output distribution $P_{Y^n|X^n}$ and a reference distribution $\tilde{P}_{Y^n}$ .

The channel dispersion $V$ is the variance of the log-likelihood ratio $\imath(X; Y) = \log(p_{Y|X}/p_Y)$ — exactly the test statistic studied in detection theory. The normal approximation $R^* \approx C - \sqrt{V/n}\,Q^{-1}(\varepsilon)$ is a direct consequence of the CLT applied to i.i.d. information density samples, paralleling the CLT-based approximations for detection and estimation in FSI.

URLLC (Ultra-Reliable Low-Latency Communication)

One of the three 5G NR service categories, targeting $\leq 1$ ms user-plane latency and $\geq 99.999\%$ reliability ( $\varepsilon \leq 10^{-5}$ ). Applications include industrial automation, autonomous driving, and remote surgery.

Mini-Slot

A shortened transmission time interval in 5G NR, consisting of 2, 4, or 7 OFDM symbols instead of the standard 14-symbol slot. Mini-slots reduce alignment and transmission delay for URLLC traffic.

URLLC: Ultra-Reliable Low-Latency Communication