Learning-Based Near-Orthogonal Superposition Code for MIMO Short Message Transmission

Chenghong Bian, Chin-Wei Hsu, Changwoo Lee, Hun-Seok Kim

Abstract

Massive machine type communication (mMTC) has attracted new coding schemes optimized for reliable short message transmission. In this paper, a novel deep learning-based near-orthogonal superposition (NOS) coding scheme is proposed to transmit short messages in multiple-input multiple-output (MIMO) channels for mMTC applications. In the proposed MIMO-NOS scheme, a neural network-based encoder is optimized via end-to-end learning with a corresponding neural network-based detector/decoder in a superposition-based auto-encoder framework including a MIMO channel. The proposed MIMO-NOS encoder spreads the information bits to multiple near-orthogonal high dimensional vectors to be combined (superimposed) into a single vector and reshaped for the space-time transmission. For the receiver, we propose a novel looped K-best tree-search algorithm with cyclic redundancy check (CRC) assistance to enhance the error correcting ability in the block-fading MIMO channel. Simulation results show the proposed MIMO-NOS scheme outperforms maximum likelihood (ML) MIMO detection combined with a polar code with CRC-assisted list decoding by 1-2 dB in various MIMO systems for short (32-64 bit) message transmission.

1. Introduction

In the era of Massive Machine Type Communication (mMTC), devices such as IoT sensors and industrial automation systems frequently transmit small bursts of data. Traditional error-correction codes, like LDPC or Polar codes, are mathematically optimized for long data blocks and suffer from significant "coding gain losses" when applied to short messages (32–64 bits).

Furthermore, transmitting these short messages over Multiple-Input Multiple-Output (MIMO) channels introduces spatial interference. Existing deep-learning approaches often lack flexibility, requiring retraining for different antenna configurations. This paper introduces a flexible, learned Near-Orthogonal Superposition (NOS) coding scheme designed to overcome these limitations.

2. Methodology

The MIMO-NOS framework utilizes an end-to-end auto-encoder structure, as illustrated in Figure 1. It consists of three primary components: a neural encoder, a learnable detection architecture for training, and a robust tree-search decoder for evaluation.

Block diagram of the proposed MIMO-NOS scheme highlighting the distinct training and evaluation phases.

2.1 Neural Encoder and Superposition

The encoder takes a bit sequence, partitions it into smaller chunks, and maps them to one-hot vectors. A Deep Neural Network (DNN) then learns to map these to high-dimensional vectors that are "near-orthogonal" to one another. These vectors are superimposed (added) and reshaped for space-time transmission across multiple antennas.

2.2 Learnable Architecture

During the training phase, the system is optimized as an auto-encoder. The receiver side includes an MMSE detector and a residual network to equalize the signal before a neural decoder predicts the bit probabilities. This allows the encoder to learn a codebook robust to channel noise and interference.

Detailed architecture of the learnable components, including the encoders, MMSE detector, and residual network.

2.3 Looped K-best Decoder

While a neural decoder is used for training, the authors propose a CRC-assisted looped K-best tree-search algorithm for high-performance inference. This decoder explores the most likely bit combinations in a tree structure and uses a "looped" mechanism to revisit earlier layers, correcting errors that might have propagated. A Cyclic Redundancy Check (CRC) is used to verify the final candidate.

3. Key Insights and Results

The performance of the MIMO-NOS scheme was evaluated using Monte-Carlo simulations on block-fading MIMO channels with message sizes of 32, 48, and 64 bits.

  • Reliability Gains: The proposed scheme outperforms the baseline (Polar codes + ML detection) by 1–2 dB in Packet Error Rate (PER) across various Signal-to-Noise Ratio (SNR) levels.
  • Robustness and Flexibility: The learned codebook is robust enough to be reshaped for different antenna counts (e.g., 2x2 to 4x4) without retraining. Interestingly, training on a smaller MIMO setup can sometimes improve performance on larger ones by forcing the encoder to learn a more conservative, robust codebook.
  • Orthogonality: The encoder naturally learns to create codewords that are nearly orthogonal, which helps the receiver "untangle" superimposed components despite the channel's interference.

4. Conclusion

MIMO-NOS offers a significant performance advantage for short-packet mMTC applications. By combining deep learning-based encoding with a robust tree-search decoder, the system achieves superior reliability compared to traditional coding schemes. Future research will focus on scaling this approach to longer bit lengths and developing differentiable versions of the K-best decoder to further optimize the training process.

Summary updated

Abstract

Massive machine type communication (mMTC) has attracted new coding schemes optimized for reliable short message transmission. In this paper, a novel deep learning-based near-orthogonal superposition (NOS) coding scheme is proposed to transmit short messages in multiple-input multiple-output (MIMO) channels for mMTC applications. In the proposed MIMO-NOS scheme, a neural network-based encoder is optimized via end-to-end learning with a corresponding neural network-based detector/decoder in a superposition-based auto-encoder framework including a MIMO channel. The proposed MIMO-NOS encoder spreads the information bits to multiple near-orthogonal high dimensional vectors to be combined (superimposed) into a single vector and reshaped for the space-time transmission. For the receiver, we propose a novel looped K-best tree-search algorithm with cyclic redundancy check (CRC) assistance to enhance the error correcting ability in the block-fading MIMO channel. Simulation results show the proposed MIMO-NOS scheme outperforms maximum likelihood (ML) MIMO detection combined with a polar code with CRC-assisted list decoding by 1-2 dB in various MIMO systems for short (32-64 bit) message transmission.
BibTeX
@article{bian2023learning,
  author = {Chenghong Bian and Chin-Wei Hsu and Changwoo Lee and Hun-Seok Kim},
  title = {{Learning-Based Near-Orthogonal Superposition Code for MIMO Short Message Transmission}},
  journal = {IEEE Trans. Communications},
  year = {2023},
  url = {https://openreview.net/forum?id=b12fTD5z82},
}
Figure