Pulse: Multi-Domain Signal Processing & Control

• Pulse is a temporally localized variation in physical signals, fundamental for measurement, control, and signaling across diverse domains.
• Recent advances fuse multi-modal sensor data and deep learning architectures to enhance pulse detection, artifact rejection, and real-time processing.
• Applications range from wearable biomedical monitors and quantum pulse programming to astrophysical measurements and distributed machine learning systems.

A pulse, across scientific and engineering domains, refers to a temporally localized variation in a physical quantity—such as electromagnetic intensity, optical power, electrical current or voltage, mechanical motion, biomedical signals, or radiation dose—characterized by discrete onset and cessation, often used for signaling, measurement, control, or estimation of periodicity. Pulses are fundamental constructs in classical and quantum control, communication, sensing, medical instrumentation, astrophysics, and high-performance computation. The following sections provide a comprehensive technical survey of representative pulse-related paradigms, methodologies, astrophysical phenomena, and biomedical or quantum applications, with explicit emphasis on architectures, performance, and illustrative case studies derived from recent literature.

1. Pulse Signal Processing in Biomedical Sensing

Pulses in biomedicine frequently denote the cyclical fluctuations of blood volume or flow associated with cardiac activity, transduced via diverse sensing modalities:

• Photoplethysmography (PPG): PPG sensors exploit light absorption changes in vascularized tissue to estimate pulse waveform morphology. However, motion artifacts (MA) impair robustness. The PULSE model, a lightweight deep network, fuses single-channel PPG and three-axis accelerometry via temporal convolutions and a multi-head cross-attention (MHCA) mechanism, enforcing temporal alignment and motion artifact suppression. Temporal convolutions with dilations (kernel K=3, dilation d=2, channels {32,48,64}) generate deep time embeddings for both modalities. The MHCA layer, with PPG as Query and acceleration as Key/Value, supports explicit calculation and visualization of per-axis attention weights, exposing semantic relevance of motion epochs (e.g., y-axis for eating, x/z-axes for walking) and increasing explainability. On the PPG-DaLiA dataset, PULSE* achieves a mean absolute error (MAE) of 4.03 BPM, outperforming prior models with comparable (<132k) parameter counts and enabling on-device implementation (inference latency <100 ms on Cortex-M4) (Kasnesis et al., 2022).
• Ultrasound-based Pulse Detection: The PuLsE system demonstrates a wrist-worn, low-power ultrasound transceiver for continuous heart-rate monitoring. The signal chain utilizes a 10 MHz pulser, analog envelope detection (bandwidth compression >5×), and fixed-point DSP (Q1.15) for on-device spectral analysis. A two-dimensional (pulse-by-depth) M-mode matrix is built, differentiated, transformed (RFFT along depth, CFFT along pulses), and frequency-accumulated to robustly extract heart rate. Correlation with ECG gold standard achieves r=0.99 and a mean error of 0.69±1.99 BPM at the optimal lateral wrist location. Power consumption is 5.8 mW, enabling >7 days continuous operation on smartwatch batteries, with RAM use of 68 kB (Giordano et al., 2024).
• Non-contact Pulse Wave Reconstruction: Fusion-E2Pulse leverages asynchronous event cameras and RGB video for high-fidelity pulse waveform estimation, surpassing frame-based methods in capturing fine morphological features (e.g., the dicrotic notch). An attentional fusion architecture couples band-pass filtered RGB priors with event-derived micro-dynamics in a dual-stream encoder, SE bottleneck, and temporal self-attention, jointly optimized with adversarial, spectral, and morphological loss. Performance metrics include heart-rate MAE (0.78 BPM), Pearson waveform correlation (0.89), and systolic phase duration error (16.74 ms) (Feng et al., 14 Jun 2026).

2. Pulses in Astrophysics and Radiation Science

In astrophysics and high-energy instrumentation, pulse phenomena and pulse-based control underpin both observation and dosimetric precision:

• X-ray Pulsars: In accreting millisecond X-ray pulsars (AMXPs) such as Aql X-1, the term pulse refers to the quasi-coherent periodic modulation of X-ray flux at the neutron star spin frequency. Temporal analysis using the $Z^2_n$ statistic demonstrates pulse detectability concentrated in the soft (3–13 keV) band (e.g., $Z^2_1\approx 100$ for 3–13 keV) and is spectrally modeled using absorbed blackbody and disk blackbody plus iron line. During pulse-on intervals, additional blackbody residuals ($kT\approx1.7$ keV, $R\approx1$–2 km) consistent with hotspot emission from the neutron star surface are required for statistical consistency. Phase-resolved spectroscopy isolates the hotspot component to the pulse-high spin phase, affirming the surface footprint interpretation (Kocabıyık et al., 8 Jan 2025).
• Radiation Dosimetry and Control: In pre-clinical FLASH radiotherapy (FLASH-RT), individual pulse monitoring is achieved using plastic scintillators and FPGA-based real-time controllers. Each LINAC radiation pulse is integrated to yield the per-pulse dose using $$D_{\text{pulse}} = k\int_{t_0}^{t_0+\tau_p} I(t) dt$$; cumulative dose and pulse count trigger beam gating for precise dose delivery. Empirical studies show per-pulse linearity (≤3%) and characterize ramp-up transients and steady-state dose delivery, essential to mapping the FLASH beam parameter space $(D_{\text{pulse}}, \bar{D}, f)$ for biological effect studies (Ashraf et al., 2021).

3. Pulses in Quantum and Classical Control

Pulse-level programming and hardware abstraction form the foundation of advanced quantum operations and hybrid quantum–classical computing:

• Quantum Pulse Programming: Gate-level abstractions obscure the analog dynamics of quantum hardware. Qiskit Pulse introduces explicit pulse-control, exposing waveforms $d=[d_0,\dots,d_{N-1}]$ over time-parameterized channels (Drive, Control, Measure, Acquire). Gate calibrations (such as cross-resonance CZ or CNOT) entail direct Hamiltonian tomography, waveform synthesis (e.g., GaussianSquare, Drag pulses), and pulse-level randomized benchmarking, yielding gate fidelities $F=0.979$–0.981 that match native device gates. Pulse-level access supports Hamiltonian engineering, dynamical decoupling, error mitigation, and custom readout discrimination across the entire readout chain (Alexander et al., 2020).
• Pulse-level HPC-Quantum Integration: MQSS-Pulse formalizes three primitives—ports (I/O physical channels), frames (carrier context: $(t_0,\omega,\phi)$), and waveforms (sampled envelopes)—at every layer (API, compiler IR, backend interface). C-based APIs eliminate interpreter overhead, MLIR dialects and QIR profiles encode pulse operations, and portable exchange formats coordinate device capabilities and constraints. The end-to-end workflow supports in-line HPC quantum variational algorithms (pulse-VQE) with low-latency, portable execution on arbitrary QPU hardware (Echavarria et al., 30 Oct 2025).

4. Pulses in Parallel and Distributed Machine Learning

Pulses, in the context of large-scale machine learning, refer to architectural units or dataflow partitions designed to optimize resource utilization and communication:

• Pipeline Training Acceleration: Training large UNet-style diffusion models with skip connections using standard pipeline parallelism incurs prohibitive activation communication. PULSE introduces skip locality as an explicit optimization, enforcing device-level collocation for skip-connected encoder–decoder pairs and caching skips. The system includes (1) a skip-aware dynamic-programming partitioner enforcing symmetric collocation constraints, (2) an ILP-based schedule synthesizer for bubble-efficient pipeline execution, and (3) a hybrid tuner to find optimal pipeline/data parallelism and batch size under memory constraints. This design yields 89% reduction in inter-node communication and up to 2.3× throughput gains over baselines on communication-bound clusters (Sun et al., 17 Jun 2026).

5. Pulses in Sensing and Remote Interaction

Diverse pulse fusion and prompting methods exploit the discrete, informative nature of pulses across sensor modalities:

• Sensor Fusion for Pulse and Pose Reconstruction: PULSE architectures exploit deep sensor fusion for robust estimation tasks. In wearables, fusing PPG with inertial signals via cross-attention enables reliable heart rate tracking in dynamic conditions (Kasnesis et al., 2022). In wireless pose estimation, pulses in the Doppler domain are screened as confidence-aware “motion prompts” and injected via cross-attention into spatial features, yielding more stable frame-wise human pose estimation from mmWave radar data and suppressing spurious spectral motion cues (Zheng et al., 13 May 2026).
• Remote Physiological Measurement and 3D Mapping: The Pulse3DFace dataset provides multi-view RGB and PPG signals mapped into 3D facial UV space, generating signal-to-noise, amplitude, and phase maps across the face and neck. This enables physiologically meaningful synthetic video generation for remote pulse estimation (rPPG), illuminates the effect of lighting and geometry, and establishes benchmarks for spatially resolved pulse measurement (Rohr et al., 11 Dec 2025).

6. Pulses in Observational Astronomy and Optical Communication

Pulses are central to high-contrast imaging and space-based downlinks by providing precise time- or polarization-based distinctions:

• Adaptive Optics with Pulse Lasers: The PULSE system for the 5.1-m Hale telescope employs a pulsed 355 nm UV Rayleigh laser to generate a high-fidelity artificial guide star. The high-order wavefront sensor, operating in open loop, measures low-altitude turbulence and drives a 3,388-actuator deformable mirror. Low-order natural guide star pathways provide tip–tilt and truth sensing. Simulations show that PULSE can maintain K-band Strehl ≥20% at $m_V=16$–17, enabling direct imaging of exoplanets and debris disks around faint, cool stars—capabilities out of reach for prior AO systems (Baranec et al., 2013).
• Space-Based Optical Downlink: The PULSE-A CubeSat mission demonstrates a <1.5 U circular-polarization shift-keying (CPolSK) laser payload for a 10 Mbps downlink from LEO to an amateur telescope-based ground station. The system leverages MEMS-class optics, EDFA amplification, and open-source bus/payload integration. Link budgets predict ≈5–10 dB margin at −50 dBm received, with lab BER < 10⁻⁷ at 10 Mbps. Student-led systems engineering and open-source release are emphasized for knowledge transfer and accessibility (Hanssler et al., 8 Jul 2025).

7. Terminological Scope and Domain-Specific Implications

The term pulse thereby encompasses: (1) waveform features in biological and astrophysical systems; (2) discrete events or instructions in quantum and classical control; (3) batch/activation units in distributed computation; and (4) modulation primitives in communication and sensing. The multi-modal pulse fusion paradigm is especially pronounced in recent sensing, wearable, and radar literature, where temporal, spatial, and spectral pulses are synthesized for artifact rejection, precision, or explainability. A plausible implication is that advances in cross-modal pulse abstraction and fusion architectures will progressively blur the distinction between analog biological phenomena, computational primitives, and quantum-control protocols as data-driven systems unify their treatment of pulses as both signal and operator (Kasnesis et al., 2022, Echavarria et al., 30 Oct 2025, Zheng et al., 13 May 2026, Sun et al., 17 Jun 2026).