Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

MambaNet: Mamba-assisted Channel Estimation Neural Network With Attention Mechanism

Dianxin Luan, Chengsi Liang, Jie Huang, Zheng Lin, Kaitao Meng, John Thompson, Cheng-Xiang Wang

TL;DR

MambaNet tackles OFDM channel estimation in scenarios with many subcarriers by fusing self-attention with a customized Mamba block. The bidirectional selective state space scan enables non-causal, long-range subcarrier dependencies to be captured efficiently, yielding superior MSE and BER performance with fewer tunable parameters than transformer-based baselines. Simulations on 3GPP TS 36.101 ETU channels demonstrate robust gains across a wide SNR range and favorable parameter efficiency, highlighting practical benefits for large-scale subcarrier configurations. The proposed architecture thus offers a low-complexity, high-performance alternative for CSI estimation in next-generation OFDM systems.

Abstract

This paper proposes a Mamba-assisted neural network framework incorporating self-attention mechanism to achieve improved channel estimation with low complexity for orthogonal frequency-division multiplexing (OFDM) waveforms, particularly for configurations with a large number of subcarriers. With the integration of customized Mamba architecture, the proposed framework handles large-scale subcarrier channel estimation efficiently while capturing long-distance dependencies among these subcarriers effectively. Unlike conventional Mamba structure, this paper implements a bidirectional selective scan to improve channel estimation performance, because channel gains at different subcarriers are non-causal. Moreover, the proposed framework exhibits relatively lower space complexity than transformer-based neural networks. Simulation results tested on the 3GPP TS 36.101 channel demonstrate that compared to other baseline neural network solutions, the proposed method achieves improved channel estimation performance with a reduced number of tunable parameters.

MambaNet: Mamba-assisted Channel Estimation Neural Network With Attention Mechanism

TL;DR

MambaNet tackles OFDM channel estimation in scenarios with many subcarriers by fusing self-attention with a customized Mamba block. The bidirectional selective state space scan enables non-causal, long-range subcarrier dependencies to be captured efficiently, yielding superior MSE and BER performance with fewer tunable parameters than transformer-based baselines. Simulations on 3GPP TS 36.101 ETU channels demonstrate robust gains across a wide SNR range and favorable parameter efficiency, highlighting practical benefits for large-scale subcarrier configurations. The proposed architecture thus offers a low-complexity, high-performance alternative for CSI estimation in next-generation OFDM systems.

Abstract

This paper proposes a Mamba-assisted neural network framework incorporating self-attention mechanism to achieve improved channel estimation with low complexity for orthogonal frequency-division multiplexing (OFDM) waveforms, particularly for configurations with a large number of subcarriers. With the integration of customized Mamba architecture, the proposed framework handles large-scale subcarrier channel estimation efficiently while capturing long-distance dependencies among these subcarriers effectively. Unlike conventional Mamba structure, this paper implements a bidirectional selective scan to improve channel estimation performance, because channel gains at different subcarriers are non-causal. Moreover, the proposed framework exhibits relatively lower space complexity than transformer-based neural networks. Simulation results tested on the 3GPP TS 36.101 channel demonstrate that compared to other baseline neural network solutions, the proposed method achieves improved channel estimation performance with a reduced number of tunable parameters.
Paper Structure (13 sections, 17 equations, 2 figures, 2 tables)

This paper contains 13 sections, 17 equations, 2 figures, 2 tables.

Table of Contents

  1. Introduction
  2. System architecture
  3. Baseband and channel settings
  4. Conventional methods and baseline neural networks
  5. MambaNet: Mamba-assisted framework
  6. Attention-assisted Mamba module
  7. Gate coefficients generation
  8. Bidirectional selective state space scan
  9. Residual architecture with up-sampling modules
  10. Simulations
  11. MSE and BER performance over an extended SNR range
  12. Complexity analysis
  13. Conclusion

Figures (2)

  • Figure 1: MambaNet: Mamba-assisted framework with self-attention mechanism. It involves an attention-assisted Mamba block and a residual convolutional neural network.
  • Figure 2: MSE and BER results for the InterpolateNet, HA02, Channelformer and MambaNet tested on the ETU channel.