[Paper Review] Low-latency Ultra Reliable 5G Communications: Finite-Blocklength Bounds and Coding Schemes

Johan Östman, Giuseppe Durisi+4 more|elib (German Aerospace Center)|Nov 26, 2016

Wireless Communication Security Techniques21 citations

TL;DR

This paper proposes finite-blocklength information theory bounds and practical coding schemes for low-latency, ultra-reliable 5G communications, optimizing spatial and frequency diversity under strict latency and reliability constraints. It demonstrates that 32 independent fading branches (e.g., 8 Tx antennas × 4 OFDM symbols) achieve near-optimal performance at 10⁻⁹ packet error probability, with a 2.68 dB SNR gap between a practical convolutional code and the theoretical limit.

ABSTRACT

Future autonomous systems require wireless connectivity able to support extremely stringent requirements on both latency and reliability. In this paper, we leverage recent developments in the field of finite-blocklength information theory to illustrate how to optimally design wireless systems in the presence of such stringent constraints. Focusing on a multi-antenna Rayleigh block-fading channel, we obtain bounds on the maximum number of bits that can be transmitted within given bandwidth, latency, and reliability constraints, using an orthogonal frequency-division multiplexing system similar to LTE. These bounds unveil the fundamental interplay between latency, bandwidth, rate, and reliability. Furthermore, they suggest how to optimally use the available spatial and frequency diversity. Finally, we use our bounds to benchmark the performance of an actual coding scheme involving the transmission of short packets.

Motivation & Objective

• To analyze the fundamental tradeoff between latency, bandwidth, rate, and reliability in 5G systems using finite-blocklength information theory.
• To determine optimal configurations of spatial and frequency diversity for ultra-reliable (10⁻⁹) and low-latency (100 μs) transmissions.
• To benchmark practical coding schemes—specifically convolutional codes with pilot-based channel estimation—against theoretical information-theoretic limits.
• To quantify the impact of pilot overhead and diversity gain on system performance under stringent latency and reliability constraints.

Proposed method

• Derives finite-blocklength converse and achievability bounds for MIMO Rayleigh block-fading channels under both uplink (average power) and downlink (spectral power density) constraints.
• Extends random-coding error exponent analysis to multiple-antenna systems, providing an upper bound on error probability for unitary space-time modulation (USTM) with no CSI at receiver.
• Uses the Polyanskiy-type bounds from [9] to compute tight bounds on maximum achievable rate under finite blocklength and target error probability (ε = 10⁻⁵, 10⁻⁹).
• Proposes a practical coding scheme using tail-biting convolutional codes (368,92) with punctured outputs and QPSK modulation, combined with pilot-aided channel estimation.
• Employs maximum ratio combining and ordered-statistics list decoding (weight ≤3) to achieve near-ML performance with reduced complexity.
• Evaluates performance across varying pilot overhead (nₚ = 1 to 8) and compares with theoretical bounds to identify optimal pilot configuration.

Experimental results

Research questions

• RQ1 What is the fundamental tradeoff between latency, bandwidth, rate, and reliability in 5G systems under finite-blocklength constraints?
• RQ2 How should spatial and frequency diversity be optimally allocated to achieve ultra-reliability (e.g., 10⁻⁹ packet error probability) in low-latency communications?
• RQ3 What is the optimal number of pilot symbols to minimize SNR gap between practical coding schemes and theoretical information-theoretic limits?
• RQ4 How does channel estimation overhead affect the performance of short-packet transmissions in MIMO OFDM systems?
• RQ5 To what extent can practical convolutional coding with pilot training approach the theoretical finite-blocklength limits?

Key findings

• The optimal configuration for 10⁻⁹ packet error probability is 32 independent fading branches (e.g., 8 Tx antennas × 4 OFDM symbols), achieving the minimum required SNR.
• For 10⁻⁹ error probability, increasing diversity beyond 32 branches (e.g., 8 Tx × 12 symbols) is ineffective due to growing channel estimation overhead.
• The performance gap between the practical coding scheme and the theoretical achievability bound is 2.68 dB at ε = 10⁻² when using 6 pilot symbols.
• The optimal number of pilot symbols for the given scheme is nₚ = 6, with performance degrading significantly for both fewer and more pilots.
• The gap between the random-coding error exponent bound and the finite-blocklength achievability bound is 0.26 dB at ε = 10⁻⁴ for 8 Tx and 4 symbols, increasing to 3.8 dB for single-antenna systems.
• The system is highly sensitive to pilot overhead: performance degrades rapidly when nₚ deviates from the optimal value due to either insufficient channel estimation or excessive pilot overhead.