S. Dörner, J. Clausius, S. Cammerer, and S. ten Brink, “Learning Joint Detection, Equalization and Decoding for Short-Packet Communications,” IEEE Transactions on Communications, vol. 71, no. 2, pp. 837–850, Feb. 2023, doi: 10.1109/TCOMM.2022.3228648.
A machine learning-based approach for joint detection, synchronization, equalization, and decoding in short-packet wireless communications, where messages must first be detected before decoding.
The evolution of wireless systems toward Shannon’s capacity limits but highlights gaps in practical short-packet performance, especially for Internet of Things (IoT) and massive machine-type communications (mMTC).
System Model
The system is single-input single-output (SISO) with single-carrier modulation, transmitting k bits in n symbols (including overhead).
Channel Model
A stochastic multipath model based on Proakis C (5 taps) is used: taps h_i are normally distributed with Proakis weights. The process:
- Add random delay τ_off ~ U(0, n – n_M), where n_M = 5 (memory length), yielding a 2n – n_M sequence.
- Convolve with random h (length 2n-1).
- Convolve with STO-filtered raised-cosine (span 16), adding one tap (total memory n_M=5, length 2n).
- Add AWGN n ~ N(0, σ²).
This simulates multipath, random access, and STO.
Conventional State-of-the-Art Baseline
For benchmark, a system with Zadoff-Chu preamble (n_bl,ZF=20 symbols for k=64, n=64) followed by QPSK-modulated payload (44 symbols, 88 coded bits via 5G LDPC code rate 8/11).
Receiver:
- Detection via preamble correlation + energy threshold (optimized for 0.1% false alarms per 5G PRACH specs).
- If detected, estimate channel from autocorrelation, pass snippet to BCJR equalizer.
- Iterative equalization, demapping, decoding (IEDD): BCJR (max-log MAP) + damped min-sum BP decoder (λ=0.7, 0.2 extrinsic weight), up to ℓ_IEDD iterations.
This is a competitive baseline for short packets.
