Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

RadMamba: Efficient Human Activity Recognition through Radar-based Micro-Doppler-Oriented Mamba State-Space Model

Yizhuo Wu, Francesco Fioranelli, Chang Gao

TL;DR

RadMamba tackles the challenge of real-time, privacy-preserving HAR with radar by introducing a radar-centric state-space approach that preserves micro-Doppler structure. It combines channel fusion with downsampling, Doppler-aligned segmentation, and convolutional projections inside a selective state-space backbone to dramatically reduce parameters and FLOPs while maintaining high accuracy. Across CW and FMCW datasets, RadMamba achieves near-state-of-the-art performance with orders of magnitude fewer parameters and substantially lower inference cost, enabling on-sensor deployment and energy-efficient edge AI. The work is complemented by ablations, open-source tooling, and a hardware-aware perspective that highlights RadMamba’s potential for hardware-software co-design in next-generation radar-based HAR systems.

Abstract

Radar-based Human Activity Recognition (HAR) is an attractive alternative to wearables and cameras because it preserves privacy, and is contactless and robust to occlusions. However, dominant Convolutional Neural Network (CNN)- and Recurrent Neural Network (RNN)-based solutions are computationally intensive at deployment, and recent lightweight Vision Transformer (ViT) and State Space Model (SSM) variants still exhibit substantial complexity. In this paper, we present RadMamba, a parameter-efficient, micro-Doppler-oriented Mamba SSM tailored to radar HAR under on-sensor compute, latency, and energy constraints typical of distributed radar systems. RadMamba combines (i) channel fusion with downsampling, (ii) Doppler-aligned segmentation that preserves the physical continuity of Doppler over time, and (iii) convolutional token projections that better capture Doppler-span variations, thereby retaining temporal-Doppler structure while reducing the number of Floating-point Operations per Inference (#FLOP/Inf.). Evaluated across three datasets with different radars and types of activities, RadMamba matches the prior best 99.8% accuracy of a recent SSM-based model on the Continuous Wave (CW) radar dataset, while requiring only 1/400 of its parameters. On a dataset of non-continuous activities with Frequency Modulated Continuous Wave (FMCW) radar, RadMamba remains competitive with leading 92.0% results using about 1/10 of the parameters, and on a continuous FMCW radar dataset it surpasses methods with far more parameters by at least 3%, using only 6.7k parameters. Code: https://github.com/lab-emi/AIRHAR.

RadMamba: Efficient Human Activity Recognition through Radar-based Micro-Doppler-Oriented Mamba State-Space Model

TL;DR

RadMamba tackles the challenge of real-time, privacy-preserving HAR with radar by introducing a radar-centric state-space approach that preserves micro-Doppler structure. It combines channel fusion with downsampling, Doppler-aligned segmentation, and convolutional projections inside a selective state-space backbone to dramatically reduce parameters and FLOPs while maintaining high accuracy. Across CW and FMCW datasets, RadMamba achieves near-state-of-the-art performance with orders of magnitude fewer parameters and substantially lower inference cost, enabling on-sensor deployment and energy-efficient edge AI. The work is complemented by ablations, open-source tooling, and a hardware-aware perspective that highlights RadMamba’s potential for hardware-software co-design in next-generation radar-based HAR systems.

Abstract

Radar-based Human Activity Recognition (HAR) is an attractive alternative to wearables and cameras because it preserves privacy, and is contactless and robust to occlusions. However, dominant Convolutional Neural Network (CNN)- and Recurrent Neural Network (RNN)-based solutions are computationally intensive at deployment, and recent lightweight Vision Transformer (ViT) and State Space Model (SSM) variants still exhibit substantial complexity. In this paper, we present RadMamba, a parameter-efficient, micro-Doppler-oriented Mamba SSM tailored to radar HAR under on-sensor compute, latency, and energy constraints typical of distributed radar systems. RadMamba combines (i) channel fusion with downsampling, (ii) Doppler-aligned segmentation that preserves the physical continuity of Doppler over time, and (iii) convolutional token projections that better capture Doppler-span variations, thereby retaining temporal-Doppler structure while reducing the number of Floating-point Operations per Inference (#FLOP/Inf.). Evaluated across three datasets with different radars and types of activities, RadMamba matches the prior best 99.8% accuracy of a recent SSM-based model on the Continuous Wave (CW) radar dataset, while requiring only 1/400 of its parameters. On a dataset of non-continuous activities with Frequency Modulated Continuous Wave (FMCW) radar, RadMamba remains competitive with leading 92.0% results using about 1/10 of the parameters, and on a continuous FMCW radar dataset it surpasses methods with far more parameters by at least 3%, using only 6.7k parameters. Code: https://github.com/lab-emi/AIRHAR.

Paper Structure

This paper contains 29 sections, 18 equations, 6 figures, 6 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Works
  3. Conventional machine learning method
  4. CNN-based method
  5. RNN-based method
  6. Hybrid CNN-RNN-based method
  7. Transformer and SSM-based method
  8. Proposed RadMamba
  9. Solution 1: Channel Fusion and Downsampling
  10. Solution 2: Doppler-Aligned Segmentation and Patch Embedding
  11. Solution 3: Convolution Projections in CP-Mamba Blocks
  12. Selective State-Space Modeling for Micro-Doppler Sequences
  13. Overall Architecture
  14. Open-source Datasets and Preprocessing for Radar-based HAR
  15. Dataset DIAT: X-Band CW Radar with Discrete Activity Instances DatasetA
  16. ...and 14 more sections

Figures (6)

  • Figure 1: The schematic of the proposed RadMamba architecture and its novel Solution 1 for radar micro-Doppler-based classification.
  • Figure 2: VisionMamba Issue 2 and RadMamba Solution 2: Doppler-Aligned Segmentation
  • Figure 3: The architecture of the State Space Model in RadMamba.
  • Figure 4: Micro-Doppler Signature Example of (a) Dataset DIAT (b) Dataset CI4R (c) Dataset UoG20
  • Figure 5: Example of comparisons between ground truth and best prediction of RadMamba model with the configuration in Table I on (a) Dataset DIAT; (b) Dataset CI4R; (c) Dataset UoG20.
  • ...and 1 more figures