Published: Aug 1, 2024
Authors: Xiaochan (Luna) Xue , Saurabh Parkar , Shucheng Yu , Yao Zheng

ISAC Series

This post is part of a series all about ISAC.

Fundamental

New Trend Waveform Design in ISAC

Advanced

mmWave Breathing Pattern Detection
NextG UAV Detection Using an O-RAN-Controlled Reconfigurable Intelligent Surface (RIS)

A lightweight Integrated Sensing and Communication (ISAC) framework is presented for contactless respiration pattern recognition using a composite OFDM–FMCW waveform at 28 GHz mmWave. A narrowband FMCW radar signal is embedded into the OFDM guard band, enabling simultaneous high-resolution sensing and data communication without modifying the OFDM structure or requiring additional hardware.

Highlights

Guard-band FMCW reuse preserves 5G NR spectral integrity
Robust respiration sensing under realistic body motion
Hardware validation on a mmWave USRP testbed
End-to-end AI pipeline achieving >98% classification accuracy

Methodology

Composite Waveform design

Narrowband FMCW chirps embedded into unused OFDM guard bands
No changes to OFDM modulation, framing, or scheduling
FMCW sweep bandwidth evaluated from 0.25–2 MHz
FMCW-to-OFDM power ratio systematically analyzed to balance sensing and communication

Fig. 1. Waveform Spectrogram

Sensing Pipeline

OFDM Mode

Compensation for SFO, STO, and CFO
CSI phase extraction per active subcarrier
Linear detrending for phase sanitization

Fig. 2.1. Raw phase with SFO/STO trend.

Fig. 2.2. Zoomed view showing respiratory cycles amid noise.

FMCW Mode

Dechirping and beat-frequency extraction
Range-bin selection for slow-time respiration signal
Drift suppression using detrend filtering

Respiration Patern Extraction

Hampel filtering, Moving-average and median filtering for smoothing
Empirical Wavelet Transform (EWT) for adaptive respiration-band isolation
Hilbert transform used to extract amplitude envelopes and normalize signals

Raw and Preprocessed beat signal

EWT based decomposition for pattern isolation

Hilbert transform and pattern normalization

Fig. 3. Pattern Extraction for FMCW mode

Phase denoising and smoothing

EWT based decomposition for pattern isolation

Hilbert transform and pattern normalization

Fig. 4. Pattern Extraction for OFDM mode

Experimental Setup

28 GHz mmWave testbed based on NI-USRP-2974
16-channel transmit and 4-channel receive phased arrays

Fig. 5. Experimental Setup

Deep Learning Model for Pattern Classification

Input: normalized respiration waveforms
Model: lightweight 1D convolutional neural network (1D-CNN)
Two convolution layers followed by global max pooling
Output: multi-class respiration pattern prediction

Fig. 6. CNN model Structure

Results

Overall classification accuracy: 98–98.5%; Eupnea and Kussmaul patterns achieve 100% accuracy.
FMCW sensing maintains stable respiration extraction under body and hand movement with similarity score of 89.2%
OFDM CSI-based sensing degrades under motion due to multipath sensitivity with similarity score of 83.5%
Communication EVM : 20.36%

Fig. 7. Confusion matrix of the classification model

Testbed Devices

USRP 2974 Software-Defined Radio

The USRP 2974 is a high-performance software-defined radio platform designed for advanced wireless research and development. It supports multiple frequency bands and provides flexible signal processing capabilities.

TMYTEK BBox Lite

The X310 SDR platform offers robust performance for wireless communication research. It features wide frequency coverage and excellent signal quality for various experimental applications.

TMYTEK BBox One

Our O-RAN testbed cluster provides a comprehensive environment for testing and validating Open Radio Access Network architectures and AI-driven network optimization.

TMYTEK UD Box