Ferkans — Interactive Telecom Tutor

The URLLC Challenge

Ultra-Reliable Low-Latency Communication (URLLC) is one of the three pillars of 5G, alongside enhanced mobile broadband (eMBB) and massive machine-type communication (mMTC). URLLC targets a packet error rate (PER) of $10^{-5}$ or lower at a user-plane latency of 1 ms.

These requirements dramatically change the information-theoretic operating point. At 1 ms latency with a 30 kHz subcarrier spacing in 5G NR, the available time-frequency resources correspond to roughly $n \sim 100$ - $500$ channel uses (depending on the bandwidth and number of OFDM symbols). At these blocklengths, the rate penalty from the normal approximation is substantial, and the classical capacity formula is a poor design guide.

The finite-blocklength analysis of Sections 26.1-26.2 provides the correct framework for URLLC system design.

URLLC Rate Penalty at Short Blocklengths

The achievable rate as a function of blocklength for progressively stricter error probability targets (

10^{-1}

to

10^{-7}

). At

n = 200

and

\varepsilon = 10^{-5}

, the rate is only about 74% of the Shannon capacity.

Definition:
The Rate-Reliability-Blocklength Tradeoff

For a given channel, the rate-reliability-blocklength tradeoff is the three-dimensional surface:

$\{(R, \epsilon, n) : R = R^*(n, \epsilon)\}$

where $R^*(n, \epsilon) = C - \sqrt{V/n}\,Q^{-1}(\epsilon) + O(\log n/n)$ .

This surface reveals three fundamental tensions:

Rate vs reliability (fix $n$ ): Higher rate $\Leftrightarrow$ higher error probability
Rate vs latency (fix $\epsilon$ ): Higher rate $\Leftrightarrow$ longer blocklength
Reliability vs latency (fix $R$ ): Lower error probability $\Leftrightarrow$ longer blocklength

URLLC requires operating in the corner where all three are demanding: high rate, low $\epsilon$ , short $n$ .

Rate-Reliability-Blocklength Tradeoff for URLLC

Visualize the achievable rate as a function of blocklength for different reliability targets. The gap between the capacity line and the finite-blocklength curve grows dramatically as $\epsilon$ decreases.

Parameters

SNR (dB)10

\log_{10}(\epsilon)

-5

Number of antennas (diversity order)1

Theorem: SNR Penalty for Finite-Blocklength AWGN

For the real AWGN channel, the SNR penalty (in dB) for operating at finite blocklength $n$ and error probability $\epsilon$ instead of at capacity is:

$\Delta_{\text{SNR}}(n, \epsilon) \approx \frac{Q^{-1}(\epsilon)}{\sqrt{n}} \cdot \frac{\sqrt{V} \cdot \text{SNR} \cdot (1 + \text{SNR}) \ln 2}{\text{SNR}/2} \text{ dB}$

For $\text{SNR} = 0$ dB, $n = 200$ , and $\epsilon = 10^{-5}$ : $\Delta_{\text{SNR}} \approx 3.5$ dB.

This means URLLC requires 3.5 dB more SNR than what the capacity formula suggests.

The SNR penalty translates the rate gap from the normal approximation into the additional power needed to compensate. For URLLC, this penalty is typically 2-5 dB, which is a significant cost. It explains why 5G NR allocates dedicated resources (mini-slots, reserved scheduling) for URLLC rather than simply coding at the Shannon limit.

Proof

Rate gap to SNR gap

The rate gap is $\Delta R = \sqrt{V/n}\, Q^{-1}(\epsilon)$ . The capacity is $C(\text{SNR}) = \frac{1}{2}\log(1 + \text{SNR})$ . The SNR penalty satisfies $C(\text{SNR} + \Delta_{\text{SNR}}) = C(\text{SNR}) - \Delta R$ .

Linearization

For small $\Delta R$ : $\Delta_{\text{SNR}} \approx \Delta R / C'(\text{SNR}) = \Delta R \cdot 2(1 + \text{SNR})\ln 2 / 1 = \Delta R \cdot 2(1+\text{SNR})\ln 2$ . Converting to dB: $\Delta_{\text{SNR}}[\text{dB}] = 10\log_{10}(1 + \Delta_{\text{SNR},\text{linear}}/\text{SNR})$ .

Example: URLLC Capacity in 5G NR

A 5G NR URLLC system operates at 3.5 GHz with:

Bandwidth: 10 MHz ( $\approx 52$ resource blocks)
Subcarrier spacing: 30 kHz
OFDM symbols per mini-slot: 2
Available channel uses: $n = 52 \times 12 \times 2 = 1248$
Target PER: $\epsilon = 10^{-5}$
SNR at cell edge: 0 dB

(a) Compute the achievable spectral efficiency using the normal approximation.

(b) Compare with the Shannon capacity.

(c) If 4-antenna diversity is available (diversity order 4), how does this improve?

Solution

Shannon capacity

$C = \log_2(1 + 1) = 1$ bit/use. Spectral efficiency: 1 bit/s/Hz.

Normal approximation

AWGN dispersion at $\text{SNR} = 1$ : $V = 1 \times 3 / (2 \times 4) = 0.375$ nats $^2 = 0.780$ bits $^2$ .

$R^* = 1 - \sqrt{0.780/1248} \times Q^{-1}(10^{-5}) = 1 - \sqrt{6.25 \times 10^{-4}} \times 4.26 = 1 - 0.025 \times 4.26 = 1 - 0.107 = 0.893$ bits/use.

Spectral efficiency: 0.893 bits/s/Hz, a 10.7% loss vs capacity.

With 4-antenna diversity

With receive diversity order $d = 4$ , the effective channel is a $4 \times 1$ SIMO with SNR scaling. The effective SNR concentrates around $4 \times \text{SNR} = 4$ (6 dB), and the dispersion decreases as $V_{\text{eff}} \approx V/d$ . At SNR = 6 dB: $C = \log_2(5) = 2.32$ bits/use, $V_{\text{eff}} \approx 0.22$ bits $^2$ . $R^* = 2.32 - \sqrt{0.22/1248} \times 4.26 = 2.32 - 0.057 = 2.26$ bits/use. Diversity dramatically reduces the finite-blocklength penalty.

Capacity vs Finite-Blocklength Rate at Common Operating Points

Operating point	Capacity (bits/use)	$R^*(n, 10^{-5})$ (bits/use)	Efficiency
AWGN, 0 dB, $n=128$	1.00	0.63	63%
AWGN, 0 dB, $n=512$	1.00	0.81	81%
AWGN, 0 dB, $n=2048$	1.00	0.91	91%
AWGN, 10 dB, $n=128$	3.46	2.50	72%
AWGN, 10 dB, $n=512$	3.46	2.98	86%
BSC(0.1), $n=128$	0.531	0.22	41%
BSC(0.1), $n=512$	0.531	0.38	72%
BEC(0.5), $n=128$	0.500	0.26	52%
BEC(0.5), $n=512$	0.500	0.39	78%

Why This Matters: 5G NR URLLC Design and Finite-Blocklength Theory

The finite-blocklength framework directly influenced 5G NR URLLC design. Key features include: (1) Mini-slots of 2 or 4 OFDM symbols (vs 14 for eMBB), enabling blocklengths of 100-500 symbols; (2) Configured grant for uplink (no scheduling request delay); (3) HARQ-less operation mode where the first transmission must meet the reliability target without retransmission; (4) Short LDPC codes with adjusted lifting sizes for blocklengths down to 256; (5) Polar codes for control information at blocklengths 32-1024.

The normal approximation is used in 3GPP link-level evaluations to benchmark code performance and determine the minimum SNR needed for a given (rate, blocklength, reliability) triple.

See Book telecom, Ch. 32 for the full treatment of URLLC system design.

Historical Note: Strassen's Pioneering Work on Second-Order Rates

1962, 2009-2010

The idea that the maximum coding rate has a $\sqrt{V/n}$ correction to capacity was first established by Volker Strassen in 1962, long before the PPV framework. Strassen proved the result for the BSC using combinatorial arguments. However, his work did not provide computable bounds for finite $n$ — only the asymptotic expansion. The result was largely forgotten until Hayashi (2009) and then Polyanskiy, Poor, and Verdu (2010) independently rediscovered and dramatically generalized it. The PPV contribution was to provide computable, tight bounds (not just asymptotic expansions) and to develop the meta-converse as a universal tool. It is a beautiful example of how a fundamental result can lie dormant for decades until the right application (URLLC) and the right tools (hypothesis testing) come together.

🚨Critical Engineering Note

Link Adaptation with Finite-Blocklength Constraints

Classical link adaptation (AMC) uses the Shannon capacity or its AWGN approximation to select the modulation and coding scheme (MCS) for a given channel quality indicator (CQI). For URLLC, this is insufficient:

Rate selection must use $R^*(n, \epsilon)$ instead of $C$ , which requires knowing the channel dispersion (not just the SNR).
BLER target shifts from $10^{-1}$ (eMBB, with HARQ) to $10^{-5}$ (URLLC), amplifying the finite-blocklength penalty by a factor of $Q^{-1}(10^{-5})/Q^{-1}(10^{-1}) \approx 3.3$ .
Channel uncertainty becomes critical: with few symbols, the channel estimate is noisy, and outage from CSI errors can dominate the coding error.

Modern URLLC systems use lookup tables mapping (SNR, $n$ , $\epsilon$ ) to MCS, precomputed from the PPV bounds rather than from Shannon capacity.

Practical Constraints

•
3GPP BLER target for URLLC: 10^-5 (one-shot) or 10^-6 (with diversity)
•
Mini-slot duration: 2 OFDM symbols = 71 us at 30 kHz SCS
•
Max user-plane latency: 1 ms (for most URLLC use cases)

📋 Ref: 3GPP TS 38.214, TS 38.331

Common Mistake: Dispersion Changes Under Fading

Mistake:

Using the AWGN dispersion formula for a fading channel by simply replacing $\text{SNR}$ with the average SNR $\bar{\text{SNR}}$ .

Correction:

In fading channels, the information density has two sources of randomness: the additive noise and the fading coefficient. The effective dispersion is:

$V_{\text{fading}} = \text{Var}[\iota(X; Y, H)] = V_{\text{AWGN}}(\text{SNR}) + \text{Var}_H[C(\text{SNR} \cdot |H|^2)]$

The second term can dominate, especially in Rayleigh fading where $|H|^2$ varies widely. This dramatically increases the finite-blocklength penalty. The fix is to use diversity (spatial, temporal, frequency) to reduce the fading dispersion, which is precisely why URLLC benefits enormously from multi-antenna reception.

Quick Check

A URLLC system requires PER $= 10^{-5}$ with blocklength $n = 200$ at SNR $= 5$ dB. Using the AWGN normal approximation ( $C = 1.16$ bits/use, $V = 0.97$ bits $^2$ ), what is the achievable rate?

$R^* \approx 0.87$ bits/use

$R^* \approx 1.10$ bits/use

$R^* \approx 0.58$ bits/use