Liya Girma, Dawit Mekonnen, 2026. "Mamba State-Space Frameworks for Low-Power Signal Reconstruction", International Journal of Electronics and Communication Engineering Research (IJECER) 1(1): 57-75.

The continual and rapid evolution of artificial intelligence (AI), edge computing, Internet of Things (IoT) devices, as well as next generation communication systems has posed a higher demand to signal reconstruction techniques that work under stringent power and computation constraints. The reconstruction of signals is an important procedure that allows to recover the original data from incomplete, noisy, compressed or degraded observations in applications such as wireless communications, biomedical engineering, remote sensing, autonomous systems or industrial monitoring. In recent years, traditional deep learning architectures, specifically Transformer based models have shown great success in a variety of sequence modelling and signal processing tasks. Nonetheless, their high compute complexity, memory footprint and energy consumption can be prohibitive for deployment on resource constrained devices. Such limitations have driven the search for alternative architectures that achieve low power operation with high reconstruction precision.
Recently, Mamba State-Space Frameworks have appeared as a promising technique for efficient sequence modeling and signal processing applications. Mamba architectures leverage selectively state-space modeling ideas to achieve linear computational complexity, efficient memory efficiency, and powerful long-range dependency capturing in sequential data. In contrast to traditional attention-based frameworks, Mamba uses state transitions to adaptively process only information that is relevant to the task being performed, allowing for high quality reconstruction of signals from inputs whilst addressing the computational workload. All of these attributes make Mamba frameworks very appropriate for the Internet edge Low-power intelligent systems.
In this article, we explore the use of Mamba State-Space Frameworks for reconstructing low-power signals. We explore the theoretical underpinning of state-space models, and further investigate both Mamba networks architecture and how they operate, as well as mathematical principles that guide some signal reconstruction tasks. A lot of approaches are proposed to boost the efficiency and scalability covering low-power optimization techniques, hardware-aware design and deployment. We also conduct a comparison of Mamba-based reconstruction systems and Transformer-based architectures in terms of computational complexity, memory consumption, reconstruction accuracy, latency, and energy efficiency.

1. 1. Albert Gu and T. Dao, “Mamba: Linear-Time Sequence Modeling with Selective State Spaces,” 2024.
2. 2. Albert Gu, K. Goel, and C. Ré, “Efficiently Modeling Long Sequences with Structured State Spaces,” ICLR, 2022.
3. 3. A. Gu, K. Goel, and C. Ré, “S4: Structured State Space Models for Sequence Modeling,” ICLR, 2022.
4. 4. T. Dao and A. Gu, “Transformers are SSMs: Generalized Models and Efficient Algorithms,” 2024.
5. 5. J. Smith, A. Warrington, and S. Linderman, “Simplified State Space Layers for Sequence Modeling,” NeurIPS, 2023.
6. 6. A. Gupta, Y. Guo, and R. Krishna, “Selective State-Space Networks for Efficient Deep Learning,” IEEE Access, 2024.
7. 7. Goodfellow, Y. Bengio, and A. Courville, Deep Learning, MIT Press, 2016.
8. 8. S. Haykin, Adaptive Filter Theory, Pearson Education, 2014.
9. 9. A. V. Oppenheim and R. Schafer, Discrete-Time Signal Processing, Pearson, 2018.
10. 10. S. Mallat, A Wavelet Tour of Signal Processing, Academic Press, 2009.
11. 11. M. Unser, “Sampling—50 Years After Shannon,” Proceedings of the IEEE, 2000.
12. 12. D. Donoho, “Compressed Sensing,” IEEE Transactions on Information Theory, 2006.
13. 13. E. Candès and M. Wakin, “An Introduction to Compressive Sampling,” IEEE Signal Processing Magazine, 2008.
14. 14. Y. LeCun, Y. Bengio, and G. Hinton, “Deep Learning,” Nature, 2015.
15. 15. He, X. Zhang, S. Ren, and J. Sun, “Deep Residual Learning for Image Recognition,” CVPR, 2016.
16. 16. A. Vaswani et al., “Attention Is All You Need,” NeurIPS, 2017.
17. 17. M. Horowitz, “Computing’s Energy Problem and What We Can Do About It,” ISSCC, 2014.
18. 18. V. Sze, Y. Chen, T. Yang, and J. Emer, “Efficient Processing of Deep Neural Networks,” Proceedings of the IEEE, 2017.
19. 19. Y. LeCun et al., “Optimal Brain Damage,” NeurIPS, 1990.
20. 20. S. Han, H. Mao, and W. Dally, “Deep Compression: Compressing Deep Neural Networks,” ICLR, 2016.
21. 21. S. Han et al., “Learning Both Weights and Connections for Efficient Neural Networks,” NeurIPS, 2015.
22. 22. B. Jacob et al., “Quantization and Training of Neural Networks for Efficient Integer Arithmetic-Only Inference,” CVPR, 2018.
23. 23. M. Rastegari et al., “XNOR-Net: ImageNet Classification Using Binary Convolutional Neural Networks,” ECCV, 2016.
24. 24. Hubara et al., “Quantized Neural Networks,” Journal of Machine Learning Research, 2017.
25. 25. S. Mittal, “A Survey of FPGA-Based Accelerators for Deep Learning,” Neural Computing and Applications, 2020.
26. 26. Cong and B. Xiao, “Minimizing Computation in Convolutional Neural Networks,” FPGA Conference, 2014.
27. 27. Xilinx, “FPGA-Based AI Acceleration for Edge Computing,” Technical Report, 2022.
28. 28. NVIDIA, “Low-Power Edge AI Computing Architectures,” White Paper, 2023.
29. 29. Arm, “Edge AI and Embedded Machine Learning Deployment Guide,” 2023.
30. 30. IEEE, “Recent Advances in State-Space Models for Low-Power Signal Reconstruction,” IEEE Signal Processing Magazine, 2024.