PaperlayerView exact revision on arXiv

Source-linked AI summary

Channel Modeling for Diffusive Molecular Communication - A Tutorial Review

Vahid Jamali, Arman Ahmadzadeh, Wayan Wicke, Adam Noel, Robert Schober

arXiv:1812.05492v1cs.ET

TL;DR

The paper addresses the need for simple yet sufficiently accurate channel models for diffusive molecular communication. It reviews end-to-end models under a common framework, develops deterministic and statistical formulations, and discusses models for time-varying and complex systems, along with challenges and future directions.

  • Problem

    Diffusive molecular communication requires simple yet sufficiently accurate channel models to support system design, analysis, and operation.

  • Method

    The paper conducts a tutorial review of physical-principle-based, simulation-driven, and experimentally-driven channel models spanning release mechanisms, environments, and reception mechanisms.

  • Results

    The review presents unified end-to-end models, deterministic and statistical received-signal models, and channel models for time-varying systems with moving transmitters and receivers.

  • Takeaways & Limitations

    Diffusive MC channel modeling must address receiver behavior, statistical molecule observations, complex environments, and differing microscale and macroscale application requirements.

  • Takeaways & Limitations

    The reviewed analytical models commonly assume simple boundary and initial conditions, limiting direct applicability to complex environments such as cardiovascular systems.

Abstract

from arXiv · show

Molecular communication (MC) is a new communication engineering paradigm where molecules are employed as information carriers. MC systems are expected to enable new revolutionary applications such as sensing of target substances in biotechnology, smart drug delivery in medicine, and monitoring of oil pipelines or chemical reactors in industrial settings. As for any other kind of communication, simple yet sufficiently accurate channel models are needed for the design, analysis, and efficient operation of MC systems. In this paper, we provide a tutorial review on mathematical channel modeling for diffusive MC systems. The considered end-to-end MC channel models incorporate the effects of the release mechanism, the MC environment, and the reception mechanism on the observed information molecules. Thereby, the various existing models for the different components of an MC system are presented under a common framework and the underlying biological, chemical, and physical phenomena are discussed. Deterministic models characterizing the expected number of molecules observed at the receiver and statistical models characterizing the actual number of observed molecules are developed. In addition, we provide channel models for time-varying MC systems with moving transmitters and receivers, which are relevant for advanced applications such as smart drug delivery with mobile nanomachines. For complex scenarios, where simple MC channel models cannot be obtained from first principles, we investigate simulation-driven and experimentally-driven channel models. Finally, we provide a detailed discussion of potential challenges, open research problems, and future directions in channel modeling for diffusive MC systems.

I. INTRODUCTION

The introduction motivates diffusion-based molecular communication as an alternative for applications where conventional radio-frequency systems are unsafe or impractical. It presents the tutorial as a rigorous, comprehensive framework for modeling release, propagation, reception, observed signals, and data-driven alternatives.

  • Motivation: Molecular communication uses molecules as information carriers for biological, industrial, and environmental applications.Examples include target-substance sensing, smart drug delivery, chemical-reactor monitoring, pollutant monitoring, and oil-transport monitoring.
  • Motivation: Diffusion-based MC is prevalent because it needs no special infrastructure or external propagation energy and suits resource-limited ad hoc devices.The tutorial also considers advection and chemical reaction networks alongside diffusion.
  • Scope and approach: The tutorial reviews end-to-end channel models covering molecule release, the physical environment, and reception mechanisms.Its framework includes deterministic expected-signal models, statistical observation models, and a unified treatment of timing and counting receivers.
  • Scope and approach: The tutorial incorporates simulations and experiments for analytical support and data-driven modeling when closed-form channel solutions are unavailable.The paper also reviews recent advances involving advection, reaction kinetics, moving nodes, and complex reception mechanisms.
  • Scope and approach: The review develops models from biological, chemical, and physical principles, including Fick’s laws and the general advection-reaction-diffusion equation.It also summarizes assumptions and special cases that permit closed-form channel impulse responses.
  • Physical foundations: Free-diffusion concentration rises until molecules arrive, then falls as they spread; greater distance lowers the peak and delays its occurrence.The concentration model is independent of radial and angular variables under the stated symmetry assumptions and can characterize propagation in blood vessels.

B. Advection

Advection models solute transport through a velocity field that may vary with position and time. The review distinguishes force-induced drift from bulk fluid flow and describes their physical origins and modeling conditions.

  • Advection model: Advection describes solute transport through a velocity vector v(d, t) that can depend on position and time.For particle motion, the velocity is treated as constant over a sufficiently small time step.
  • Advection mechanisms: The review categorizes advection as force-induced drift or bulk flow.This distinction separates particle motion caused by external forces from motion induced by movement of the surrounding fluid.
  • Advection mechanisms: Force-induced drift arises from external electrical, magnetic, gravitational, or combined forces acting on particles rather than the fluid.The force may vary across space and time, while the friction coefficient is related to diffusion through ζD = k_BT.
  • Advection mechanisms: Bulk flow occurs when particle movement is induced by fluid motion, as in blood vessels and microfluidic channels.For dilute suspensions, the flow velocity is typically independent of particle concentration but can vary with boundaries or obstacles.

1) Velocity Vector Field:

The paper models molecular transport using velocity fields for flow or drift, then combines advection with diffusion to characterize concentration changes. Péclet and dispersion measures identify whether diffusion, advection, or both dominate transport.

  • Velocity Vector Field:: Bulk flow transports molecules through fluid motion, with velocity depending on space when boundaries or obstacles shape the environment.In dilute suspensions, the flow velocity is independent of particle concentration.
  • Velocity Vector Field:: For straight ducts, Poiseuille flow is slowest at the wall and fastest at the center, producing a space-dependent velocity profile.The center velocity is v(0) = [0, 0, v0], while the wall velocity is v(ac) = [0, 0, 0].
  • Velocity Vector Field:: The advection equation describes concentration changes from velocity-driven transport, but general velocity fields and boundary conditions often require numerical solutions.Closed-form solutions are available for constant uniform and selected duct-flow cases.
  • Velocity Vector Field:: Uniform advection translates an initial concentration without changing its shape, whereas Poiseuille flow generally spreads concentration spatially over time.The distinction follows from whether the velocity field is space-independent or space-dependent.
  • Advection-Diffusion Equation:: As parallel flow increases, the concentration peak increases, peak time decreases, and the concentration tail shortens.These effects can enhance coverage and reduce inter-symbol interference in diffusion-based MC systems.
  • Advection-Diffusion Equation:: Péclet number determines the appropriate transport model: diffusion dominates for Pe ≪ 1, advection for Pe ≫ 1, and both matter near Pe ≈ 1.The characteristic comparison uses advection and diffusion times over relevant system lengths.

C. Chemical Reactions

Chemical reactions alter molecular propagation and can either degrade signaling molecules naturally or be deliberately used for communication objectives such as inter-symbol interference reduction. The review organizes reaction models by reaction order and mechanism, including degradation, bimolecular, and enzymatic reactions.

  • C. Chemical Reactions: Chemical reactions affect diffusive MC propagation and can be exploited to reduce inter-symbol interference or model ligand-receptor interactions.The paper studies reactions together with diffusion and advection through corresponding reaction-diffusion models.
  • C. Chemical Reactions: Reaction rate functions depend on reaction rate constants and reactant concentrations, with overall reaction order equal to the sum of individual reactant orders.Reaction equations describe local concentration changes over time.
  • C. Chemical Reactions: Unimolecular degradation converts signaling molecules into products irrelevant to communication and is commonly modeled as a first-order reaction.Zero-order and second-order degradation models may also be relevant depending on reaction speed.
  • C. Chemical Reactions: Second-order degradation decays hyperbolically, while first-order and zero-order degradation decay exponentially and linearly, respectively.For sufficiently large time, the second-order concentration can exceed the first-order concentration under the stated model comparison.
  • C. Chemical Reactions: Bimolecular reactions model interactions between two reactant species, but their nonlinear coupled PDEs are challenging to solve without approximations.The review derives MC channel impulse responses under assumptions such as negligible backward reaction and nearly constant concentration of one reactant.
  • C. Chemical Reactions: Michaelis-Menten enzymatic kinetics are simplified under fast degradation, slow backward reaction, and enzyme excess so enzyme concentration remains approximately constant.These assumptions make the intermediate complex short-lived and enable tractable modeling.

2) Advection-Reaction-Diffusion Equation:

The advection-reaction-diffusion model combines transport by flow, diffusion, and chemical reaction. Under simplified unbounded-environment assumptions, the paper obtains a closed-form example and shows how degradation changes the concentration profile.

  • Advection-Reaction-Diffusion Equation:: General advection-reaction-diffusion models combine diffusion, drift, and reaction effects for a single molecule type.The corresponding PDE describes their joint impact on molecular propagation.
  • Advection-Reaction-Diffusion Equation:: General initial and boundary conditions make the coupled model difficult to solve, motivating simplifying assumptions for tractable examples.The stated example relies on an unbounded environment, impulsive point release, uniform flow, and first-order degradation.
  • Advection-Reaction-Diffusion Equation:: A closed-form solution is obtained for impulsive release in an unbounded 3D environment with uniform flow and first-order degradation.The model assumes a point source and reaction rate f(κ, c(d, t)) = κc(d, t).
  • Advection-Reaction-Diffusion Equation:: As degradation rate increases, the concentration peak decreases, while the long-time concentration tail fades faster.The faster tail decay was exploited for inter-symbol interference reduction, although the lower peak is generally undesirable.

III. COMPONENT MODELING

The section models diffusive MC as an end-to-end chain comprising transmitter, physical channel, and receiver, and defines a unified channel impulse response (CIR) for that chain. The models account for component properties, propagation phenomena, and receiver-specific observation mechanisms.

  • System components: The transmitter generates, stores, encodes, and releases signaling molecules, whereas the physical channel governs their propagation and the receiver observes and processes them.Propagation can involve diffusion, advection, degradation, geometry, and obstacles.
  • End-to-end channel: The end-to-end channel includes the transmitter, physical channel, and receiver, while excluding coding, modulation, detection, and decoding operations.Its input is a stimulation signal representing the modulated information symbol.
  • Channel impulse response: The end-to-end CIR h(t) is the probability of observing one output molecule at time t after impulsive stimulation at t0 = 0.Defining the CIR probabilistically supports received-signal models based on molecule arrival times and observed counts.
  • Receiver observation: Observation is receiver-dependent: passive receivers count signaling molecules inside the receiver, while reactive receivers count activated receptor molecules.The CIR therefore uses a receiver-specific operational definition of the observed signal.
  • Channel formation: Obtaining h(t) for a specific system requires solving the advection-reaction-diffusion equation or a suitable simplification with appropriate initial and boundary conditions.The modeled phenomena include generation, release, diffusion, degradation or production, receptor kinetics, and signaling pathways.
  • Receiver models: Receiver models include passive and active reception; fully absorbing reception treats molecules reaching the receiver as absorbed at its surface.Reactive reception models ligand-receptor binding and unbinding on receptor-covered receiver surfaces.

C. Transmitter Models

The section reviews transmitter models that range from simple point sources to geometry-aware volume and ion-channel-based models. It also identifies particle generation through chemical reaction networks as a largely unresolved modeling challenge.

  • Transmitter properties: The main transmitter properties affecting the end-to-end CIR include geometry, particle generation through chemical reactions, and the release mechanism.These features motivate progressively more detailed transmitter models.
  • Point Transmitter: Point transmitters are widely used because of their simplicity, but they omit the geometry and other physical features of the transmitter.They are modeled as zero-dimensional points that commonly release molecules instantaneously.
  • Volume Transmitter: Volume transmitters account for transmitter geometry by initially distributing molecules throughout the transmitter volume, yielding more realistic models than point transmitters.They still assume instantaneous molecule generation.
  • Ion-Channel Based Transmitter: Ion-channel-based transmitters model spherical transmitters with membrane ion channels whose opening and closing are controlled by a gating parameter.These models incorporate both transmitter geometry and the release mechanism.
  • Ion-Channel Based Transmitter: With many open ion channels, the ion-channel-based CIR becomes the CIR of a volume transmitter because the entire transmitter surface acts as a transparent membrane.The equivalence follows from treating many open channels as an effectively transparent surface.
  • Particle generation: Most reviewed transmitter models omit particle generation via chemical reaction networks because incorporating it requires solving a challenging coupled reaction-diffusion equation.Mesoscopic models and numerical solution are used in studies that do include particle generation.

D. Physical Channel Models

Physical-channel models capture how diffusion, geometry, advection, flow, and degradation shape the channel impulse response. Bounded geometries can substantially alter observations, while some bounded cases approach unbounded-channel behavior under suitable conditions.

  • Channel effects: Physical-channel CIRs account for diffusion alongside effects such as advection, geometry, and chemical reactions.Advection may be constructive or destructive depending on velocity direction and strength.
  • Bounded diffusion channels: Bounded-channel CIRs are obtained by solving the diffusion equation with boundary conditions representing the channel geometry and physical or chemical properties.Rectangular and circular ducts use reflective-wall conditions and geometry-specific solutions.
  • Circular duct channels: Small duct radii produce larger CIRs than unbounded channels because repeated wall reflections increase the chance of molecule observation.The comparison uses ac ∈ {5, 6, 9, 12} × arx with arx = 0.15 µm.
  • Circular duct channels: For large duct radii, the bounded circular-duct CIR approaches the unbounded-channel CIR, making the latter a good approximation.The approximation applies to sufficiently large bounded circular ducts in the reported comparison.
  • Advection channels: Uniform advection permits closed-form transformations in some unbounded or aligned cases, whereas general bounded cases may require direct evaluation of the advection-diffusion equation.When velocity has both parallel and orthogonal components in bounded dimensions, the moving-reference-frame technique cannot be applied there.
  • Laminar-flow channels: The accuracy of the reported dispersion- and flow-dominant CIR expressions depends on αd; increasing D and dz improves the dispersion-regime approximation.The passage does not establish accuracy outside the corresponding regime conditions.
  • Reaction effects: First-order degradation multiplies the surviving probability by exp(−κt), so the CIR is smaller than without degradation at any instant.The survival probability decreases monotonically as time increases.

E. Summary of End-to-End CIR Models

The review organizes end-to-end CIR models by transmitter, physical channel, and receiver properties, while recording dimensionality, numerical solution, and reaction-diffusion treatment. Its summary table provides a compact vocabulary for comparing these model components.

  • Transmitter models: The CIR-model summary distinguishes point and volume transmitters, including volume or surface release and transparent surfaces.The transmitter notation records the assumed release geometry.
  • Physical-channel models: Physical-channel entries record diffusion, advection type, boundedness, and degradation or production reactions.Advection is classified as uniform or laminar, while geometry is marked bounded or unbounded.
  • Receiver and solution descriptors: Receiver entries specify whether sensing uses a volume, surface, or partial surface, and the table also records channel dimension and numerical evaluation.The summary identifies the corresponding CIR equation when available.
  • Reaction-diffusion treatment: When reaction-diffusion equations are used, the summary indicates whether reaction and diffusion are solved separately or jointly.The notation also identifies whether reaction processes may represent active reception mechanisms.

IV. RECEIVED SIGNAL MODELING

The received-signal framework unifies timing and counting observations across recurrent and non-recurrent receivers. It then supports signal models that depend on the CIR and can incorporate noise molecules, inter-symbol interference, and temporal correlation.

  • Signal-model framework: The section develops signal models for parameter estimation and data detection, beginning with a unified representation and multiple receiver-observation time scales.The framework is subsequently generalized to noise molecules and time-slotted communication with ISI.
  • Received-signal choices: Different MC signal models describe instantaneous transparent-receiver counts, receptor-bound molecules, or accumulated arrivals at absorbing receivers.These quantities correspond to distinct reception mechanisms and observation choices.
  • Unified signal definition: The unified representation is based on molecule arrival times and, for recurrent receivers, departure times.Arrival and departure vectors together form a complete representation of the received signal.
  • Receiver classes: A non-recurrent receiver observes each molecule at most once, whereas a recurrent receiver can observe the same molecule multiple times.Transparent and unbinding reactive receivers are recurrent; absorbing receivers and reactive receivers with κb = 0 are non-recurrent.
  • Receiver classes: The framework treats non-recurrent receivers as a special case of recurrent receivers with no departure events.This corresponds to ndpr(t) = 0 for all t.
  • Timing channels: Timing-channel models relate transmitter release times to corresponding receiver arrival times for non-recurrent receivers.The arrival-time vector is paired with the release-time vector of the transmitted molecules.

2) Timing-based Receivers:

Timing-based models describe receiver observations through molecule release times and random propagation delays, but their general formulation becomes difficult to simplify and extend. The review therefore distinguishes timing and counting receivers, emphasizing more tractable counting models.

  • Timing-based Receivers: Timing-channel models treat observation delays as nonnegative random variables, typically under independent molecule behavior.For an unbounded one-dimensional environment without flow, the delay follows a Levy distribution.
  • Timing-based Receivers: The receiver cannot directly access the full arrival-time vector because molecules of the same type are indistinguishable and only observed molecules are known.Among Ntx released molecules, the receiver observes only narv(t) molecules by time t.
  • Timing-based Receivers: The general arrival-time formulation applies to non-recurrent receivers but is difficult to simplify or extend to recurrent receivers, interfering molecules, and ISI.The review notes that the formulation is especially cumbersome when noise molecules or inter-symbol interference are present.
  • Counting Receivers: Counting receivers use molecule counts as the received signal and are classified by recurrent versus non-recurrent behavior and accumulative versus instantaneous counting.AMC counts arrivals within an observation window, whereas IMC counts molecules observed at a specified time.
  • Counting Receivers: The four counting categories include nR-AMC, R-AMC, R-IMC, and nR-IMC receivers, with recurrence allowing molecules to be counted multiple times.For nR-IMC receivers, the received signal is non-decreasing because received molecules do not leave the receiver.
  • Counting Receivers: Models for R-IMC receivers are extended to other receiver types by substituting the appropriate probability of observation over time intervals.R-AMC modeling remains cumbersome because a molecule may be counted repeatedly within the observation window.

C. Signal Models

The review develops deterministic and statistical signal models from the channel impulse response, then compares Binomial, Poisson, and Gaussian descriptions of observed molecule counts. Approximation accuracy depends on h(t, τ) and the number of released molecules.

  • General Signal Models: The received-signal analysis first derives expected molecule counts, then statistical fluctuations, and finally stochastic models for time-variant channels.The channel impulse response is interpreted as the probability that a released molecule is observed at a later time.
  • Deterministic Models: Under independent molecule behavior, the expected count from Ntx molecules released at time zero is determined by the CIR h(t).For time-invariant channels, h(t, τ) depends on t−τ.
  • Deterministic Models: Finite receptor occupancy violates the independent-molecule linear model because the expected received signal becomes nonlinear in Ntx.The effect arises when a reactive receiver has a finite number of receptors.
  • Statistical Models: The Binomial model represents each released molecule as either observed or unobserved, with Ntx trials and success probability h(t, τ).Its support is the integer range from zero through Ntx, but the distribution complicates analysis.
  • Approximation Comparison: Gaussian modeling is analytically convenient but can be inappropriate when the expected count is small because it permits continuous, negative, and non-integer values.Large Ntx does not guarantee a large expected observed count.
  • Statistical Models: The Poisson model approximates Binomial counts when trials are numerous and the mean Ntxh(t, τ) is small, while the Gaussian model uses the Binomial mean and variance when the expected count is sufficiently large.Both approximations improve mathematical tractability, but they rely on different operating conditions.
  • Approximation Comparison: For typical MC systems with h(t, τ)<0.1, the Poisson model is generally more accurate, whereas Gaussian accuracy improves with Ntx and can suit macroscale systems.The comparison uses RMSE between approximate Gaussian or Poisson CDFs and the Binomial CDF.

3) Time-Variant Models:

Time-variant models treat the CIR and mean received signal as functions of both observation time and release time. For mobile transmitters or receivers, diffusion makes the channel non-stationary and causes the mean signal to decay over long release times.

  • Time-Variant Models: Changes in diffusion-related channel parameters make h(t, τ) and the mean received signal depend on both t and τ.The time-variant analysis focuses on variations in system parameters such as D and flow-related quantities.
  • Model Assumptions: The mobile-transceiver model assumes an unbounded diffusive channel, a point transmitter, a passive receiver, and mobility represented by three-dimensional diffusion.The transmitter and receiver have diffusion coefficients Dtx and Drx.
  • Time-Variant Models: Relative transmitter-receiver motion is captured through effective diffusion coefficients D2=Dtx+Drx and D1=D+Drx.D2 describes relative transmitter-receiver motion, while D1 describes relative signaling-molecule and receiver motion.
  • Time-Variant Models: With at least one mobile transceiver, the mean received signal is non-stationary and approaches zero as τ approaches infinity in an unbounded environment.The average transmitter-receiver separation increases because the transceivers diffuse away from each other.
  • Time-Variant Models: The normalized variance increases with τ, indicating growing uncertainty about the mean received signal over longer release times.This uncertainty is attributed to the random walk.
  • Stochastic Approximation: When D2τ≤d0^2/200, the CIR probability density can be accurately approximated by a log-normal distribution.The resulting stochastic model can support design and performance analysis of time-variant MC systems.

D. Interfering Noise Molecules

The review models environmental interfering molecules and repeated transmissions within a unified statistical framework. Under stated assumptions, interference is Poisson, while continuous transmission introduces ISI and signal-dependent diffusion noise.

  • Interfering Noise Molecules: Environmental noise molecules arise from natural or synthetic sources rather than the transmitter’s intended signaling release.The review develops statistical models for their observed count at the receiver.
  • Noise Statistics: Under assumptions of a constant observation-time mean and sufficiently many receiver subvolumes, the observed noise count follows a Poisson distribution.The Binomial model over subvolumes approaches Poisson as the number of subvolumes tends to infinity.
  • Noise Statistics: The Poisson noise model is also justified as the maximum-entropy, worst-case noise distribution under the stated assumptions.This interpretation treats the spatial coordinates of observed noise molecules jointly with their count.
  • Continuous Transmission: Continuous transmission is modeled with time slots and an ISI memory of L symbol intervals, while interfering noise is included in the received count.ISI in interval k originates from the L−1 preceding symbol intervals.
  • Continuous Transmission: The total received count can be modeled as a Poisson random variable by superposing independent signal and interference contributions.The signal contribution is determined by the CIR and transmitted symbols, while the interference contribution has mean r̄int.
  • Gaussian Approximation: When expected information and interference counts are large, a Gaussian model is available, but MC diffusion noise remains signal dependent rather than AWGN-like.The diffusion-noise variance follows the received signal level.
  • ISI Mitigation: Reactive molecules and enzymes can shorten CIR decay and thereby mitigate ISI without requiring a symbol interval long enough for complete decay.Using a sufficiently large symbol interval can reduce transmission rate.

2) Simplifications for Extreme SNR Regimes:

Extreme SNR assumptions reduce the received-signal model to diffusion-noise-limited or interference-limited forms. The section also relates signal correlation to sampling intervals and channel coherence under mobility.

  • 2) Simplifications for Extreme SNR Regimes:: When signaling molecules dominate interference, Poisson and Gaussian diffusion-noise models apply, with SNR = ¯rsig.This is the diffusion-noise-limited regime, defined by ¯rsig ≫ ¯rint.
  • 2) Simplifications for Extreme SNR Regimes:: When interference molecules dominate signaling molecules, Poisson and Gaussian interference-noise models apply, with SNR = ¯r²sig/¯rint.The Gaussian form becomes signal-independent, as in conventional wireless communication models.
  • 2) Simplifications for Extreme SNR Regimes:: For OOK, interference noise dominates when s[k] = 0, whereas diffusion noise dominates when s[k] = 1.The resulting noise power is larger for symbol s[k] = 1 in the illustrated diffusion-noise-limited regime.
  • 2) Simplifications for Extreme SNR Regimes:: The minimum sampling spacing needed to reduce absolute sample correlation below ζt = 0.2 decreases as the diffusion coefficient increases.Correlation decreases as the time separation ∆t increases.
  • 2) Simplifications for Extreme SNR Regimes:: For mobile transmitters and receivers, the channel mean decorrelates as ∆τ increases, and coherence time decreases as their diffusion coefficients increase.The stated example uses fixed threshold ζτ = 0.5 and Dtx = Drx = {0.01, 0.05, 0.1} × D.

V. SIMULATION- AND EXPERIMENT-DRIVEN MODELS

Simulation- and experiment-driven models extend channel modeling to complex systems where analytical solutions are unavailable or rely on restrictive assumptions. Simulation classes trade physical resolution against computational efficiency, while mesoscopic methods represent stochastic molecular events on grids.

  • V. SIMULATION- AND EXPERIMENT-DRIVEN MODELS: Analytical channel models require assumptions tied to specific parameters and conditions, such as sufficiently distant boundaries or observation points.The validity of an assumption depends on whether its physical effects can be neglected in the region of interest.
  • V. SIMULATION- AND EXPERIMENT-DRIVEN MODELS: Simulations efficiently model complex, coupled phenomena, whereas reliable experiments may be more representative but are time-consuming and expensive.The review therefore considers both simulation-driven and experimental approaches.
  • V. SIMULATION- AND EXPERIMENT-DRIVEN MODELS: Continuum simulations solve spatio-temporal PDEs on grids and update molecule concentrations over time steps.They are appropriate when the physical scale and molecule number are sufficiently large for PDE descriptions.
  • V. SIMULATION- AND EXPERIMENT-DRIVEN MODELS: Mesoscopic accuracy and validity depend on grid resolution, time-step choices, and well-stirred-subvolume assumptions.Increasing continuum grid resolution and decreasing time steps improve accuracy but increase computational demands; well-stirred conditions require scale constraints.
  • V. SIMULATION- AND EXPERIMENT-DRIVEN MODELS: Mesoscopic simulations discretize space into subvolumes while stochastically generating reaction and molecule-transition events.They assume molecules are uniformly distributed within each subvolume and the solvent behaves as a homogeneous continuum in thermal equilibrium.

3) Microscopic Simulations:

Microscopic simulations track individual molecules through reactions, diffusion, flow, and boundary interactions, making them suitable for nanometer-to-micrometer systems. The review also situates them among hybrid and data-driven approaches for complex environments.

  • 3) Microscopic Simulations:: Microscopic simulations track every molecule individually, using discrete time and continuous spatial positions rather than well-stirred subvolumes.The solvent is still modeled as a continuum in the common implementation.
  • 3) Microscopic Simulations:: Each microscopic time step tests molecules for reactions and then applies stochastic diffusion displacements, with flow adding deterministic v∆t motion.Reaction conversion uses a probability based on the reaction rate and time step.
  • 3) Microscopic Simulations:: Microscopic tools are common for cellular and molecular-communication systems because they support simulations from nanometers to micrometers.Examples include Smoldyn, BiNS2, N3Sim, MUCIN, and AcCoRD.
  • 3) Microscopic Simulations:: Molecular-dynamics simulations model solvent molecules and intermolecular interactions at a more detailed physical scale.They may include charge-induced forces and collision dynamics, but the high molecular density makes them computationally demanding.
  • 3) Microscopic Simulations:: Hybrid simulations combine simulation classes when a system contains multiple physical scales of interest.This avoids constraining the entire simulation to the most granular approach required by one region or phenomenon.
  • 3) Microscopic Simulations:: Simulation-driven models address environments whose analytical channel response is difficult or impossible to derive, including connected microfluidic chambers.The example releases 500 molecules uniformly in one chamber for diffusion through a connecting pipe.

2) Data-Driven Model:

Data-driven channel models address unpredictable randomness and nonlinearities that analytical and simulation models cannot fully capture. The review combines physics-motivated fitting, neural-network models, experimental case studies, and future modeling challenges.

  • 2) Data-Driven Model:: Physically motivated parametric models use first-principles equations while fitting selected parameters to experimental measurements.The reviewed examples modify analytical-model parameters to match observed testbed data.
  • 2) Data-Driven Model:: Neural-network models can jointly learn a channel model and its parameters when no suitable physically motivated model exists.GANs are identified as one architecture for generating artificial data resembling measurements.
  • 2) Data-Driven Model:: A biological testbed models bacteria as uniformly distributed transmitters receiving a common light stimulus and instantaneously pumping protons.The simple model assumes rectangular illumination and represents measurement noise as additive Gaussian noise.
  • 2) Data-Driven Model:: The simple saturation model fails to follow measurements because it omits a persistent decreasing bias in the measured proton concentration.The discrepancy appears during the constant-illumination and darkness experiment.
  • 2) Data-Driven Model:: Adding a linear bias offset produces an enhanced parametric model that fits the biological measurements well.The example shows that physical-principles models may require modification for experimental systems with living-cell dynamics.
  • 2) Data-Driven Model:: Future work requires experimentally verified general channel models, including models for complex networks, turbulent flow, and macroscale environments.The review notes that current models often focus on microscale systems and that reactive receivers can invalidate independent-link decompositions.
Loading 1812.05492v1…