FlowSim | Nanodevice mobility simulator

Nanoscale devices with Terahertz (THz) communication capabilities are envisioned to revolutionize precision medicine by being deployed within human bloodstreams. These microscopic devices, smaller than red blood cells, could enable fine-grained sensing applications for detecting biomarkers of various health conditions and targeted drug delivery for cancer treatment.

However, associating the locations of detected events with the events themselves presents unprecedented technical challenges. Flow-guided nanoscale localization—a new class of in-body localization—faces constraints including sub-centimeter THz communication range, energy-harvesting limitations, and high mobility of nanodevices in bloodstreams with speeds reaching 20 cm/s.

From decades of research on traditional indoor localization, we know that early-stage performance evaluation often suffers from incomplete metrics, inconsistent performance indicators, and different evaluation environments. To avoid such a 'lock-in' in flow-guided localization, there was an urgent need for standardized performance evaluation methodologies before the field matured.

We developed a comprehensive workflow for standardized performance evaluation of flow-guided nanoscale localization approaches, implemented as an open-source simulation framework that jointly accounts for multiple critical factors previously ignored in existing evaluations.

The framework addresses three key technical domains:

The approach builds on established ns-3 network simulation architecture, creating a modular framework that can generate raw data for streamlined evaluation of different flow-guided localization solutions. This enables objective back-to-back comparison of approaches using consistent environments, scenarios, and performance metrics.

Drawing from lessons learned in indoor localization standardization efforts like EVARILOS and NIST PerfLoc, the framework implements comprehensive performance metrics including point accuracy, region accuracy, localization reliability, and energy consumption analysis.

The simulation framework architecture follows a well-established ns-3 layered model with specialized modules for nanoscale medical applications:

The framework models energy-harvesting nanodevices with intermittent behavior, featuring turn-on/turn-off thresholds based on energy availability. Communication follows a passive beacon reception followed by active backscattering model, where nanodevices report circulation times and event detection bits.

Critical implementation innovations include modeling blood flow complexity through Gaussian noise components representing vortex and laminar flow characteristics, collision detection for communication interference, and configurable event detection thresholds for medical applications. A key contribution was the optimization and parallelization of the simulator for high-performance computing (HPC) environments using SLURM job scheduling, enabling large-scale evaluations that would be computationally prohibitive on single machines.

The framework implements sophisticated mathematical models for energy harvesting and nanodevice operation:

The simulator supports multiple performance evaluation strategies including Simple Random Sampling (SRS), Stratified Simple Random Sampling (SSRS), Cluster Random Sampling (CRS), Regular Grid Sampling (RGS), and Spatial Coverage Sampling (SCS) for target event location selection.

Key configurable parameters include beaconing intervals (default 100ms), energy storage capacity (800 pJ), turn-on/off thresholds (10/0 pJ), harvesting cycles (20ms), and operational frequency (1 THz). The framework generates comprehensive raw data including circulation times, event detection bits, and energy consumption profiles. Critical optimizations for HPC environments include efficient memory management, parallel job distribution across SLURM clusters, and scalable data processing pipelines that enable evaluation of thousands of scenarios simultaneously.

The framework successfully demonstrated its capability to provide realistic performance assessment of flow-guided localization approaches, revealing significant insights about the challenges facing this emerging technology when practical constraints are considered.

Key performance findings from benchmark evaluations:

The framework's sampling strategy analysis showed that Regular Grid Sampling (RGS) converges fastest toward overall performance with only 0.2% difference in region accuracy compared to dense sampling, requiring approximately 50% of evaluation points for reliable results.

The HPC optimization and parallelization implementation achieved over 6x speedup in execution times with optimal CPU usage at 12 units, making large-scale evaluations computationally feasible on SLURM-managed cluster systems. The majority of execution time is spent on raw data generation rather than performance benchmarking, enabling rapid comparative analysis across multiple scenarios and parameter configurations.

The standardized evaluation framework enables objective assessment across multiple critical domains:

For specific medical applications, the framework enables researchers to model diverse scenarios including cancer biomarker detection, hypoxia sensing for early diagnosis, and targeted drug delivery for oncology treatments. The realistic modeling of energy constraints and communication limitations provides crucial insights for practical deployment feasibility.

The open-source nature of the framework facilitates community-driven development and ensures reproducible research results across different institutions and research groups working on nanoscale medical technologies. Building on this foundation, we developed Graph Neural Network models that significantly improve event location prediction accuracy by effectively handling the highly erroneous raw data generated by energy-constrained nanodevices.

Several promising directions could enhance the framework's capabilities and biological fidelity:

The most impactful enhancement would be developing patient-specific simulation capabilities, similar to personalized anesthesia protocols. This would involve adapting bloodstream models based on individual physiological indicators and real-time health monitoring data.

Additionally, recent work by our team has already demonstrated the potential for Graph Neural Network-based approaches that achieve enhanced accuracy, reliability, and coverage compared to the benchmarks established in this framework, showing the continued evolution enabled by standardized evaluation methodologies.