A Survey on Modulation Techniques in Molecular Communication via Diffusion
Mehmet Sukru Kuran, H. Birkan Yilmaz, Ilker Demirkol, Nariman Farsad, Andrea Goldsmith
KKURAN et al.: SURVEY ON MODULATION TECHNIQUES IN MCVD 1
A Survey on Modulation Techniques in MolecularCommunication via Diffusion
Mehmet S¸ ¨ukr¨u Kuran,
Member, IEEE,
H. Birkan Yilmaz,
Member, IEEE,
Ilker Demirkol,
Senior Member, IEEE,
Nariman Farsad,
Member, IEEE, and Andrea Goldsmith,
Fellow, IEEE
Abstract —This survey paper focuses on modulation aspects ofmolecular communication, an emerging field focused on buildingbiologically-inspired systems that embed data within chemicalsignals. The primary challenges in designing these systems arehow to encode and modulate information onto chemical signals,and how to design a receiver that can detect and decode theinformation from the corrupted chemical signal observed atthe destination. In this paper, we focus on modulation designfor molecular communication via diffusion systems. In thesesystems, chemical signals are transported using diffusion, possiblyassisted by flow, from the transmitter to the receiver. This tutorialpresents recent advancements in modulation and demodulationschemes for molecular communication via diffusion. We comparefive different modulation types: concentration-based, type-based,timing-based, spatial, and higher-order modulation techniques.The end-to-end system designs for each modulation scheme arepresented. In addition, the key metrics used in the literature toevaluate the performance of these techniques are also presented.Finally, we provide a numerical bit error rate comparison ofprominent modulation techniques using analytical models. Weclose the tutorial with a discussion of key open issues and futureresearch directions for design of molecular communication viadiffusion systems.
Index Terms —Molecular communication, diffusion, modula-tion, nano devices, channel models, detection.
I. I
NTRODUCTION M OLECULAR communication (MC) is an emergingarea that relies on chemical signals for transferringinformation from a transmitter to a receiver [1]–[4]. Since MCsystems have different characteristics compared to traditionalcommunication systems that embed data into electromagnetic(EM) signals, MC systems can be used in areas where EMcommunication fails or is not feasible. For example, theycan be used for communication inside complex networks ofmetallic ducks and pipes [5], [6], where wireless signals fail.
M. S. Kuran is with the Department of Computer Engineering, BahcesehirUniversity, Istanbul, Turkey, (e-mail: [email protected])H. B. Yilmaz is with the Computer Networks Research Laboratory (NET-LAB), Dept. of Computer Engineering, Bogazici University, Istanbul, Turkey.(e-mail: [email protected])I. Demirkol is with the Dept. of Mining, Industrial and ICT Engi-neering, Universitat Polit`ecnica de Catalunya, Barcelona, Spain. (e-mail:[email protected])Nariman Farsad is with the Dept. of Computer Science, Ryerson University,Toronto, ON M5B 2K3, Canada. (e-mail: [email protected])Andrea Goldsmith is with the Dept. of Electrical Engineering, PrincetonUniversity, Princeton, NJ 08540 USA. (e-mail: [email protected])This research is supported in part by the Scientific and Technical ResearchCouncil of Turkey (TUBITAK) under BIDEB-2232 program with the grantnumber 118C274, NSERC Discovery under Grant RGPIN-2020-04926, theNSF Center for Science of Information (CSoI) under grant CCF-0939370,and CFI John Evans Leaders Funds.
Another benefit of MC compared to EM signalling is energyefficiency. In particular, although MC cannot achieve the datarates that are obtained by EM-based systems, they have muchlower energy spent per transmitted information bit [7], [8],making them suitable for ultra low power applications that donot require high data rates.MC can also be used to connect tiny engineered devices.Specifically, recent advances in the fields of bio-engineeringand nanotechnology have resulted in the emergence of tinydevices of sub-millimeter dimensions that can perform sensingand actuation. For example, synthetic cells are excellent bio-marker sensors and can detect bio-markers for cancer cells invivo at small concentrations [9]–[11]. As another example,micro-sized devices based on graphene could be used forremoval of nano-sized toxic contaminants [12], [13]. Althoughthese devices have been shown to work in a laboratorysetting, moving them out of the laboratory and into prac-tical settings remains an open challenge. In particular, formany applications, these tiny devices need to communicateand collaborate in swarms, or they need to transmit theirmeasurements to other devices (such as micro-sized sink nodesor health monitoring wristwatches). Since chemical signalingis already used in many biological systems to interconnectcells or regulate bodily functions, MC is a promising solutionfor interconnecting tiny devices, due to its energy efficiencyand bio-compatibility (i.e., synthetic biological devices arewell-suited to transmit and receive chemical signals withappropriate enhancement of their communication capability.)Different types of MC systems have been investigated inthe literature such as molecular communication via diffusion(MCvD) [3], [7], [14], bacteria-based communication [15],microtubule-based communication [16]–[18], calcium signal-ing [19]–[21], and pheromone signaling [5], [22]. Among theseMC systems, the MCvD has emerged as the technology ofchoice due to its versatility and applicability to many envi-ronments. The communication media and signalling schemeutilized by MCvD system are vastly different from that ofclassical communication systems. The first step towards designof these systems is modeling the MCvD system componentsincluding channel, transmitter, receiver, signal, noise, andinterference. In recent years, several key survey and tutorialson such works have been published as follows: • Farsad et al. present an overall look at the recent advancesin the greater topic of molecular communication [4]. • Jamali et al. investigate and review various works andapproaches on modeling the molecular channel [23]. • Kuscu et al. elaborate and present different approaches a r X i v : . [ c s . ET ] D ec URAN et al.: SURVEY ON MODULATION TECHNIQUES IN MCVD 2
Sec. II Sec. III Sec. IV
Sec. V
Molecular Communication via Diffusion (MCvD) Classification of Modulation Techniques
Performance Metrics for MCvD
Modulation Techniques
Diffusion EnvironmentChannel CharacteristicsMC Channel
Channel Impulse
Response Channel Capacity Error Rates (BER, SER)
Achievable Information
RateInterference to Total Received Molecule RatioMol-Eye: Eye Diagram Metrics of Molecular SignalPeak to Average Messenger Molecule Ratio
Concentration-based
Type-based Timing-based Spatial Techniques Hybrid Techniques
Molecule Emission Times
Tx Node
12 Molecule Emission Times M o l e c u l e C o n c e n t r a t i o n Modulation CharacteristicsInterference Mitigation Computational Complexity
Detection Type
Fig. 1. Organization and contents of the paper and its sections. on transmitter and receiver architectures for molecularcommunication [24].In addition to the modeling of these components, a widerange of modulation techniques have also been proposed inthe MCvD literature utilizing different aspects of the molecularsignal and the aforementioned components of the MCvD sys-tem. These techniques aim to increase the overall performanceof the communication and are another key part of the overallMCvD design. Although the previously mentioned tutorialsbriefly include some well-known modulation techniques, theydo not provide a detailed and categorized look at the plethoraof modulation techniques that have emerged in recent years. Incontrast, this paper focuses on and presents a survey of thesemodulation techniques.Since prior works have been based on different systemassumptions, the performances of the proposed techniquescannot be directly compared with one another. To overcomethis challenge, in our approach, we categorize and groupprior works according to the characteristics of the modulationscheme and the assumptions made for their performanceevaluation. We consider three key characteristics: inter-symbolinterference (ISI) mitigation, computational complexity, andmodulation type. Then, using this categorization, we offera survey of various MCvD modulation techniques. We alsoprovide a quantitative comparison between the most prominenttechniques by comparing their bit error rate performance usinganalytical models. Finally, we underline and briefly elaborateon the current open problems related to the modulation tech-nique design of MC.The main contributions of this paper are summarized asfollows: • We give a comprehensive discussion on the various per-formance evaluation metrics being used in the literatureto evaluate MCvD systems. • We present a detailed survey of the various modulationtechniques that have been proposed for the MC system. • We provide a systematic approach to compare differentmodulation techniques, including current as well as future methods. • We perform a case study on the most prevalent modula-tion techniques where we evaluate the bit error rate per-formances of each technique and provide a performancecomparison of them under the same conditions.The remainder of the paper is organized as follows, whichis also depicted in Fig. 1. In Section II, a general overviewof MC systems is presented. Section II describes the classifi-cation approach for investigating MC modulation techniques.Section IV describes a variety of performance metrics used inthe literature to evaluate the performance of MC systems. InSection V, a detailed survey of the modulation techniques thathave been proposed for MC systems is presented. Section VIelaborates on the key open problems and challenges on thedesign of modulation techniques for MC. Finally, Section VIIconcludes the paper.II. M
OLECULAR C OMMUNICATION VIA D IFFUSION
We consider an MCvD system as shown in Fig. 2 wherebya sequence of input symbols (or bits) are modulated in a timeslotted manner into 𝑀 symbols S = [ 𝑆 , 𝑆 , ..., 𝑆 𝑀 ] , (1)where 𝑆 𝑘 ∈ S refers to the 𝑘 -th symbol that is modulatedonto a chemical signal by encoding the symbol into someproperties of the chemical emission process, and S is thesymbol set. The modulated signal is transmitted through theMC channel, where the propagation environment degrades thesignal and introduces delay in a probabilistic manner. Thechannel impaired chemical signal is detected and demodulatedat the receiver. Let ˆ S = [ ˆ 𝑆 , ˆ 𝑆 , ..., ˆ 𝑆 𝑀 ] , (2)denote the set of received symbols such that ˆ 𝑆 𝑘 refers to thek-th symbol demodulated and detected at the receiver. Thedetected symbols are fed to the demodulator that can correctsome of the errors in detection to recover the information. URAN et al.: SURVEY ON MODULATION TECHNIQUES IN MCVD 3
Fig. 2. MCvD system and its three core components: the transmitter (Tx), the molecular communication channel, and the receiver (Rx).
At its core, the MCvD system utilizes small molecules orinformation particles called messenger molecules (MM) torelay information between the transmitter (Tx) and the receiver(Rx) over the MC channel. Due to the time slotted manner ofthe overall system, the Tx and Rx have to be synchronized witheach other throughout the communication. Here we followthe general MCvD literature and assume that Tx and Rx aresynchronized with each other using a synchronization methodwhich is out-of-the-scope of this paper. MMs are generallya few nanometers to a few micrometers in size and can becomposed of proteins, ions, magnetic nano particles, etc. TheTx chemically produces MMs, encodes and modulates theinformation over one or more physical properties of an MMsignal, and finally releases these MMs to the channel. Unlessstated otherwise, we assume that the MMs are released atthe start of the symbol slot. Within the MC channel, due totheir small size, the movement of MMs are chiefly governedby a type of diffusion process that depends on the particularfeatures of the channel (e.g., free diffusion or diffusion withdrift).Over time, some of these MMs reach the Rx, which uses adetection process that depends on the type of the MM and thetype of the receiver used for detection. The detection processesthat dominate the MC literature to date are passive receiversand absorbing receivers. In passive receivers, the Rx acts likea transparent entity (i.e., as if the Rx is not in the environment)and does not affect the movement of the MMs. The detectionconsists of counting the concentration of the MMs inside theRx at certain time intervals. In absorbing receivers, the Rxacts as an absorbing entity for the MMs and when the MMshit the Rx, they are removed from the environment. In thesereceivers, the detection can be modeled as counting the numberof the MMs that hit the Rx within a given time interval. A sub-category of the absorbing receivers is the partially absorbingreceivers, where only some parts of the receiver can absorbMMs while other parts simply reflect them [25].
A. Molecular Communication Channel
A key aspect of the MCvD system is the diffusion pro-cess of the MC channel, which is vastly different fromthe electromagnetic wave propagation in wired and wirelesscommunication channels. In the MCvD channel, movementof a single molecule is governed by Brownian motion, which is modelled by the Wiener process . When a group of particlescollectively exhibit Brownian motion, the resulting process iscalled a diffusion process. A detailed mathematical foundationfor the diffusion-based MC channel has been given by Hsiehet al. in [26], where it has been shown that the diffusion-basedMC channel is a stationary and ergodic channel with memory.The diffusion process can be simulated via Monte Carlosimulations. In a 3-dimensional (3D) free diffusion MC chan-nel, the displacements at each dimension, over a small timeduration Δ 𝑡 , are independent and identically distributed (iid)random variables and can be numerically simulated as: −→ 𝑟 [ 𝑘 ] = −→ 𝑟 [ 𝑘 − ] + Δ −→ 𝑟 , Δ −→ 𝑟 = ( Δ 𝑟 𝑥 , Δ 𝑟 𝑦 , Δ 𝑟 𝑧 ) , Δ 𝑟 𝑖 ∼ 𝑁 ( , 𝐷 Δ 𝑡 ) ∀( 𝑖 ) ∈ { 𝑥, 𝑦, 𝑧 } , (3)where −→ 𝑟 [ 𝑘 ] is the location of the MM at the k-th timeinstance, 𝑟 𝑖 represents 𝑖 -th component of the location vector −→ 𝑟 , Δ 𝑡 is the discrete simulation time step, 𝐷 is the diffusioncoefficient, and 𝑁 ( 𝜇 , 𝜎 ) is the Gaussian random variable withmean 𝜇 and variance 𝜎 .In the MC channel, the movement speed of MMs are chieflygoverned by the diffusion coefficient ( 𝐷 ). This coefficient isaffected by the temperature of the environment, 𝑇 , the dynamicviscosity of the fluid which the MMs diffuse in, 𝜂 , as wellas the size of the MMs via their Stoke’s radius, 𝑟 𝑠 [27]. Inparticular, 𝐷 is given by the Stokes-Einstein relation as: 𝐷 = 𝑘 𝐵 𝑇𝛼𝜋𝜂𝑟 𝑠 (4)where 𝑘 𝐵 = . · − J/K is the Boltzman constant and 𝛼 is a unitless quantity whose value is mainly governed bythe coefficient of sliding friction between the MMs and themolecules in the fluid, denoted as 𝛽 . The SE relation is derivedto describe the diffusive motion of molecules within a fluidcomposed of smaller particles [28]. Therefore, 𝛽 has twolimits: “ ∞ ” referring to a “stick” boundary and “ ” referringto a “slip” boundary [29]. The “stick” boundary refers to thecase where size of the MMs ( 𝑠 𝑚𝑚 ) are considerably biggerthan the size of the molecules in the fluid ( 𝑠 𝑓 𝑙𝑢𝑖𝑑 ) which yields 𝛼 = . On the other hand, the “slip” boundary refers to the casewhere 𝑠 𝑚𝑚 ≈ 𝑠 𝑓 𝑙𝑢𝑖𝑑 yielding 𝛼 = . Note that these are thetwo boundary conditions and depending on the relative size of Wiener process is also called Standard Brownian motion.
URAN et al.: SURVEY ON MODULATION TECHNIQUES IN MCVD 4 𝑠 𝑚𝑚 and 𝑠 𝑓 𝑙𝑢𝑖𝑑 , 𝛼 can also take other values. In this work, wefollow the general diffusion literature and select 𝛼 = since ina classical MC environment 𝑠 𝑚𝑚 is greater than 𝑠 𝑓 𝑙𝑢𝑖𝑑 such asan insulin molecule diffusing inside water (i.e., 𝑠 𝑚𝑚 = . and 𝑠 𝑓 𝑙𝑢𝑖𝑑 = .
19 nm ).As the value of 𝐷 depends on the properties of both theMMs and the environment (e.g., temperature of the fluid),assuming constant 𝐷 means that the change in the environmentis negligible. Such scenarios are called, ideal diffusion scenar-ios. On the other hand, 𝐷 can be modeled as a time-varying,space-varying, and/or molecular-density-varying value (e.g.,the diffusion coefficient can change as the concentration ofMMs in the environment increases). In such channels, theideal diffusion is not applicable and more complex diffusionprocesses are required [23]. In this survey we only considerchannels with constant 𝐷 values.Another MC channel variant is the MCvD channel withflow. In such a channel, in addition to the diffusion process,a flow also affects the movement of the MMs. Consideringan MCvD channel with flow only in the x dimension, (3)becomes: −→ 𝑟 [ 𝑘 ] = −→ 𝑟 [ 𝑘 − ] + Δ −→ 𝑟 , Δ −→ 𝑟 = ( Δ 𝑟 𝑑𝑥 + Δ 𝑟 𝑓 𝑙𝑜𝑤𝑥 , Δ 𝑟 𝑑𝑦 , Δ 𝑟 𝑑𝑧 ) , Δ 𝑟 𝑑𝑖 ∼ 𝑁 ( , 𝐷 Δ 𝑡 ) ∀( 𝑖 ) ∈ { 𝑥, 𝑦, 𝑧 } , (5)where Δ 𝑟 𝑑𝑖 represents the movement due to diffusion and Δ 𝑟 𝑓 𝑙𝑜𝑤𝑖 represents the movement due to flow.The effect of the flow depends on the type of flow consid-ered in the channel model as well as the diffusion environment.Considering a closed environment such as a pipe, flow couldeither be laminar, i.e. following constant streamlines, or tur-bulent depending on the velocity and viscosity of the fluid inthe channel. Laminar flow occurs at lower velocities, whereasturbulent flow occurs at higher velocities. This fluid velocity isdetermined by a dimensionless parameter called the Reynoldsnumber, which is the ratio of the inertial forces in a fluidand the viscous forces. Generally, studies that focus on MCchannels with flow consider laminar flow only. B. Diffusion Environment
The diffusion environment that is considered in the previoussubsection only consists of the Tx, Rx, and the MMs usedin the communication. More complex environments consideradditional physical objects or barriers that limit the overallcommunication in the channel.A constrained diffusion environment is a generalized versionof the free diffusion environment that includes environmentalobjects other than the Tx and the Rx. These objects canbe reflective, absorbing, or partially-absorbing to the MMs.Although more realistic, mathematical analysis of constraineddiffusion environments is much more challenging than that ofthe free diffusion environments due to their inherent geomet-rical asymmetry.One commonly-used special case of the constrained diffu-sion environment is a vessel-like environment that emulatesthe inside of blood vessels. In these environments, the Txand Rx pair is considered to be within a cylindrical boundary,
Fig. 3. Vessel-like environment with spherical transmitter (Tx) and sphericalreceiver (Rx). which acts as the vessel boundary (Fig. 3). In studies focusingon vessel-like environments, the vessel boundary is typicallymodeled as a reflective surface for the MMs, while the Tx andRx are located on different sides of the vessel, and a flow isexistent in the direction of the Rx. Since generally the flowin blood vessels is laminar [30], studies focusing on vessel-like environments consider laminar flow in their analysis.Due to the symmetrical nature of the vessel-like environment,analytical closed-form solutions are tractable for some of thesemodels. For instance, in [31], Dinc et al. developed a generalapproximation for flow in 3D microfluidic channels.
C. Channel Characteristics
As the MCvD channel is governed by the diffusion pro-cess, it becomes a linear system provided that the diffusioncoefficient 𝐷 does not depend on time, space, or moleculardensity, and that the detection process at the Rx does not alterthe linearity of the system [23], [32].
1) Channel Impulse Response:
The channel impulse re-sponse (CIR) function, which characterizes the channel re-sponse to an impulse input, can be used to define a com-munication channel. There are a variety of definitions in theMC literature for the CIR function (i.e., ℎ ( 𝑡 ) or ℎ ( 𝑡, 𝜏 ) ) [33]–[35]. We follow the approach of [36]–[38] and define theCIR function through another function called the CumulativeFraction of Received Response (CFRR). Please note that it isalso possible to define CFRR after defining CIR, the importantpoint is that CIR is the derivative of CFRR with respect to time(i.e., CFRR is the integration of CIR with respect to time). Definition 1.
The
Cumulative Fraction of Received Response(CFRR) is defined as the mean fraction of molecules thatare detected by the receiver on average until time 𝑡 whenmolecules are emitted at 𝑡 = . The CFRR is denoted by 𝐹 ( 𝑡 ) .After defining the CFRR, we can now define the CIR usingthe CFRR as follows: Definition 2.
Channel Impulse Response (CIR) is defined asthe derivative of CFRR with respect to 𝑡 and denoted by ℎ ( 𝑡 ) = 𝑑𝑑𝑡 𝐹 ( 𝑡 ) . (6)The CIR physically represents the rate of reception of MMsat the Rx as a function of time. The CIR and the CFRR define URAN et al.: SURVEY ON MODULATION TECHNIQUES IN MCVD 5 important aspects of the molecular channel characteristics. Forinstance, for an absorbing receiver in a 1D free diffusionenvironment, with no chemical reaction or decay, CFRR isanalytically known and shows that the emitted moleculeseventually arrive at the receiver (i.e., lim 𝑡 →∞ 𝐹 ( 𝑡 ) = ). Onthe other hand, in a 3D free diffusion environment, there isa non-zero probability that a given molecule will never reachthe Rx as time goes to infinity (i.e., lim 𝑡 →∞ 𝐹 ( 𝑡 ) < ). Asshown in [38]–[40], CIR has a heavy tailed structure in 1D-3D environments, which implies that ISI is one of the commonreasons for communication errors in an MCvD system. Inparticular, ISI is introduced by stray MMs belonging to thecurrent time slot that can interfere and impair the detection offuture symbols. This issue is exacerbated as more and morestray MMs from all previous time slots accumulate.In the MC literature, CFRR has been investigated fordifferent channel types such as free diffusion, constrained dif-fusion, and vessel-like environments. In recent years, closed-form solutions of CFRR have been derived analytically forsome key MC channels such as 1D free diffusion (withor without molecular degradation ) [41], [42], 3D free dif-fusion (with or without molecular degradation) [25], [38],[43], 3D constrained diffusion [23], and vessel-like envi-ronments [23], [44]. However, for more complex channels(e.g., constrained diffusion environment, time-variant diffusioncoefficients), finding analytical solutions for CFRR is muchmore challenging. Hence, for such environments, instead ofclosed-form solutions, infinite sum formulations have beenderived for given environmental parameters, such as the CFRRformulation given for 3D diffusion channels with a reflectivespherical source and spherical receiver in [45] and a 3D vessel-like environment with drift given in [46].
2) Channel Capacity:
Another fundamental characteristicof the MC channel is its channel capacity, which provides afundamental limit for the maximum information transmissionrate that is achievable over the MC channel. Note that this ca-pacity is independent of the modulation and coding techniquesbeing used.As explained in Section II-A, due to the diffusion dynamics,the MC channel is a channel with memory and should bemodeled as such [26], [47]. Therefore, finding analyticalexpressions for the channel capacity is challenging [8], [48].Some prior works have relied on a simplifying assumption thatthe MC channel is memoryless and evaluated channel capacityor bounds under this assumption. For example, in [40], [42],[49]–[55], upper and lower bounds on channel capacity arederived for different memoryless MCvD channels. Some otherworks have relied on achievable information rate as a measureto compare different modulation techniques. These works willbe discussed in more detail in the next sections.For the stationary channels, a third approach in channelmodeling regarding memory is to use a channel model with Molecular degradation can be defined as the alteration of the chemicalstructure of the MMs as a result of a chemical reaction. Sometimes thisreaction can be naturally occurring and sometimes it can be accelerated byadding enzymes to the environment. For MC, this effect may improve thereceived signal quality by eliminating the remaining interference moleculesor it may deteriorate the received signal by eliminating the MMs too much. a finite memory 𝑚 . These kinds of models only consider theeffects of the previous 𝑚 symbols over the current one. Inthe MC literature, studies that consider a channel with finitememory of this nature set 𝑚 = . Here it should be notedthat the selection of an adequate 𝑚 value depends on thesymbol duration. Considering the symbol duration values thatare common in the MC literature, it has been shown by Gencet al. that setting 𝑚 = is far from adequate to accuratelymodel channel memory [47]. In particular, Genc et al. showthat 𝑚 should be on the order of tens of symbols.III. C LASSIFICATION OF M ODULATION T ECHNIQUES
Similar to classical communication systems, the modulationprocess in an MCvD system varies one or more physicalproperties of the MM signal to transmit bits of information.Since the MM signal is physically different from a traditionalEM signal, the physical properties that can be used to modulatebits onto signals in an MCvD system differs from the EMsignal properties that are used to modulate bits in classicalcommunication systems, i.e., amplitude, frequency, and phase.We now describe and classify the multitude of modulationtechniques that have been proposed for MCvD systems. Thesetechniques are classified according to the physical propertybeing used to encode bits into the transmitted symbols: concentration , type , timing , and space modulation. Amongthese physical properties, concentration refers to the amountof MMs transmitted by the Tx; type refers to the type ofMMs transmitted by the Tx; timing refers to a property ofthe transmission time of MMs within the communication timeslot; and space refers to the spatially separated multiple MMrelease modules at the Tx in a molecular MIMO scenario.Also, some techniques utilize a combination of these physicalproperties as part of a single modulation technique. We classifythese techniques as hybrid techniques.We divide the characteristics and assumptions of individualmodulation techniques into two main categories: modulationcharacteristics and performance evaluation assumptions. Inmodulation characteristics, we consider ISI mitigation, compu-tational complexity (i.e., the complexity of both transmissionand reception processes), and the detection type used at theRx.We define four qualitative levels for both the ISI mit-igation and computational complexity characteristics: none,low, moderate, or high. Among these two characteristics, thecomputational complexity also depends on the implementationand the type of Tx and Rx devices. It might be the case thata given technique is much easier to implement for biologicaltype devices than described here. However, a discussion onthe implementation of these techniques is outside the scopeof this survey. As for the detection type, while most ofthe demodulation techniques utilize basic thresholding (e.g.,concentration is above a threshold or not), others use morecomplex detection types such as maximum likelihood (ML),which generally yield higher overall performance at a cost ofhigher computational complexity.Studies on these specific modulation techniques evaluatetheir performance based on differing sets of assumptions. We URAN et al.: SURVEY ON MODULATION TECHNIQUES IN MCVD 6 categorize the assumptions used in performance analysis asfollows: transmitted signal waveform (in terms of number ofMMs released over time), receiver type and the system en-vironment. Note that the proposed modulation techniques areindependent from these assumptions and that the performancesof the considered modulation techniques may differ based onthese features.IV. P
ERFORMANCE METRICS FOR M OLECULAR C OMMUNICATION VIA D IFFUSION SYSTEM
In the MC literature, authors use a variety of differentperformance metrics such as bit error rate (BER), achievableinformation rate, and interference to total received moleculeratio (ITR) to evaluate the system performance. Although thecore concepts of these metrics are general to any communica-tion system, some physical concepts and evaluation methodsare specific to the MCvD system. In this section we willenumerate such performance metrics and elaborate on theiruse in the MC literature. Note that these metrics can beinterdependent for a given channel and modulation.
A. Error Rates (BER, SER)
A common metric in evaluating the performance of a com-munication system is the BER (or symbol error rate (SER) if agiven symbol represents more than a single bit of information).In the MC literature, BER or SER is evaluated given an MCchannel and modulation technique, and is a function of signal-to-noise ratio (SNR), transmitted power, symbol duration, andenvironmental parameters (e.g., communication distance anddiffusion coefficient).In classical wireless communication, the BER is usuallyplotted against transmitter power or SNR. However, when welook at these parameters in the MC domain, the transmitterpower is usually the amount of MMs sent from the Tx forthe signal in question. In order to determine the SNR for anMC channel, noise as well as its sources must be defined, asthe noise in MC channels has different characteristics thanin wireless or wired channels. In this regard, many worksin the MC literature consider the source of noise to be thediffusion process, giving rise to the notion of diffusion noise[56]. This particular noise is based on the probabilistic natureof the arrival of molecules through the MC channel and have aconsiderable effect on the performance of the communicationsystem. However, the current state of the MC literature lacksa consistent definition of this diffusion noise, which makesperformance comparisons between different works somewhatdifficult.Alternatively, BER can be evaluated against signal-to-interference and noise amplitude ratio (SINAR). In this case,in addition to the noise sources, interference sources must alsobe defined and evaluated. There are several key interferencesources for an MC system such as inter-symbol interference(ISI), co-channel interference (CCI) [57], and inter-link inter-ference (ILI) [58], [59]. Among these sources, ISI is generallyconsidered in performance evaluations, whereas CCI and ILIare generally only considered in works that study multipletransmitters and MIMO systems, respectively.
B. Achievable Information Rate
Another commonly used performance metric for a givenmodulation technique is the achievable information rate. Sim-ilar to BER, achievable information rate is also usually eval-uated against SNR, transmitted power, symbol duration, andenvironmental parameters such as communication distance anddiffusion coefficient.In wireless communication, transmitted power is generallygiven in decibel-milliwatts (dBm) or watts. As described in [7],the transmitted power in an MCvD system mainly depends onthe energy costs of synthesizing MMs and moving these MMsto the outside of the Tx. In both of these accounts, the keymultiplier defining the overall transmitted power is the MMcount. Hence, in an MCvD systems, the transmitted power canalternatively be given in MM count or MM count per type.Consequently, the achievable information rate can be definedper energy, per MM, per MM type and per unit time.As mentioned in Section II.C.2, the MC channel typicallyexhibits memory. For a channel with memory, the mutualinformation rate can be calculated as 𝐼 ( 𝑆 ; ˆ 𝑆 ) = lim 𝑛 →∞ 𝑛 𝐼 (cid:16) 𝑆 [ ] , ...𝑆 [ 𝑛 ] ; ˆ 𝑆 [ ] , ... ˆ 𝑆 [ 𝑛 ] (cid:17) (7)for a given modulation technique, where 𝑆 [ 𝑘 ] and ˆ 𝑆 [ 𝑘 ] represent the transmitted bit value and the demodulated bitvalue during the 𝑘 𝑡ℎ symbol slot (considering the fact thatthe communication is conducted in a time-slotted manner),respectively [60]. As such, several works have evaluated theachievable information rate in such MC channels with mem-ory. These works have mainly considered the concentrationshift keying (CSK) modulation technique where informationis modulated by varying levels of emitted MM concentration.In [42], [61], the achievable information rate for a 1D MCchannel is evaluated considering a CSK modulation techniqueand varying channel memory 𝑚 . In [47], [62], [63], a 3D MCchannel is evaluated again by considering a CSK techniqueand varying channel memory. As opposed to these works, in[64] the achievable information rate for a 1D MC channel iscalculated considering a more sophisticated modulation tech-nique of the molecular concentration shift keying (MCSK),where information is modulated by varying types of emittedMM concentration. Both the CSK and the MCSK modulationtechniques are explained in detail in Section V. C. Interference to Total Received Molecule Ratio (ITR)
As previously explained, in most MCvD environments, CIRhas a heavy tail structure. The stray MMs from previoussymbol slots accumulate and impair the correct receptionability of the Rx. Two components that can be derived fromCIR, the area under the desired signal, 𝐴 SYM , and the ISI, 𝐴 ISI , both depicted in Fig. 4. For a fixed symbol duration 𝑡 𝑠 ,interference molecules reduce the communication quality byaffecting the reception process of the upcoming symbols.Hence, Interference to Total Received Molecule Ratio (ITR)for a given symbol duration 𝑡 𝑠 denotes the fraction of interfer- URAN et al.: SURVEY ON MODULATION TECHNIQUES IN MCVD 7 t s A SYM A ISI C I R -4 Fig. 4. Representation of the molecular received signal and its components:desired signal and ISI. The area A
SYM corresponds to the cumulative numberof received molecules until 𝑡 𝑠 . Molecules inside the interference area (A ISI )are received after 𝑡 𝑠 and may cause erroneous detection for upcoming symbols. ence molecules for a single emission of molecules. It is givenby ITR ( 𝑡 𝑠 ) = 𝐴 ISI 𝐴 SYM + 𝐴 ISI = ∫ ∞ 𝑡 𝑠 ℎ ( 𝑡 ) 𝑑𝑡 ∫ ∞ ℎ ( 𝑡 ) 𝑑𝑡 . (8)Therefore, ITR for a given 𝑡 𝑠 is evaluated by considering ℎ ( 𝑡 ) or an empirical approximation to it.Similar to ITR, signal-to-interference and noise amplituderatio (SINAR), signal-to-interference ratio (SIR) and signal-to-interference difference (SID) can also be used as a molecularsignal quality metric. SINAR is analogous to SINR in conven-tional communications by considering noise and interferencemolecules. SIR is the fraction of the amount of desired signalmolecules to interference molecules and SID is the differ-ence between the number of desired signal and interferencemolecules. There are various works in the literature thatconsider ITR [43], [58], [65]–[67], SINAR [68], SIR [58],[69], and SID [70], [71] for the performance evaluation.In [68], it is shown that the values of the system parametersthat give maximum SID or SINAR values also give close tooptimal BER or SER value. Therefore, SID or SINAR can beused as a representative metric for finding close to optimalsystem parameters in terms of BER or SER [68], [70], [71]for an MCvD system with similar parameters in the papers. Itis observed that in cases where 𝐴 𝑆𝑌 𝑀 is not high, but 𝐴 𝐼 𝑆𝐼 is very small (i.e., nearly zero), SIR attains extremely highvalues. Therefore, it is argued that SID is a more reliableperformance metric than SIR, for the system parameters thatare mentioned in [70], [71]. However, these important findingsneed more analysis before generalizing it.
D. Peak-to-Average Messenger Molecule Ratio (PAMR)
In typical MC systems, for a given symbol the Tx emitsall the MMs at once. However, the capabilities of the Txmight prohibit such an emission and instead it might distributethe emission times and the total amount of MM per symbolover the symbol duration. In such an emission pattern, anothermetric, called the Peak-to-Average Messenger Molecule Ratio C I R -3 bit-1bit-0bit-1 (mean)bit-0 (mean) Fig. 5. An example MOL-Eye diagram considering a binary system composedof curves of the number of received molecules representing each bit-0 andbit-1 transmissions and their mean curves (adapted from [73]). (PAMR) at the Tx, becomes relevant to the performance ofthe system. PAMR is a similar concept to Peak-to-Average-Power-Ratio (PAPR) in a traditional EM-based communicationsystem. It is defined asPAMR Tx = max 𝑁 𝑇 𝑥 ( 𝑡 𝑒 ) avg 𝑁 𝑇 𝑥 ( 𝑡 𝑒 ) , 𝑡 𝑒 ∈ 𝒯 𝑒 , (9)where 𝒯 𝑒 is the set of emission times and 𝑁 𝑇 𝑥 ( 𝑡 𝑒 ) stands forthe number of emitted molecules by the Tx at the emissiontime 𝑡 𝑒 [72]. Note that in an MC system where the Tx releasesall of the MMs at the beginning of a symbol slot, 𝑡 𝑒 is thetime at the beginning of each symbol slot. PAMR is related totransmitter capabilities and implies constraints on the numberof emitted molecules over a given time duration. If the Txhas a limited PAMR due to its transmission constraints, it willresult in a degradation of the communication performance. E. MOL-Eye
The eye diagram is another tool used to evaluate the qualityof a signal after transmission through a channel [74]. Givena communication channel and modulation technique, an eyediagram is obtained by superimposing the received signals ofconsecutive bit transmissions on top of each other. Various fea-tures of the resulting diagram can be checked to give a generalunderstanding of the underlying communication system suchas when to sample the signal, amount of jitter, and the SNRat the sample point. Turan et al. propose utilizing a similartool called the MOL-Eye for the MC domain [73] to evaluatethe performance of a modulation technique. They proposeusing three performance metrics associated with the MOL-Eyediagram: the maximum eye height, standard deviation of thereceived molecules (
𝑆𝑇 𝐷 ( Δ 𝑐 ) ), and counting SNR (CSNR).Considering a binary system where each symbol representsa single bit value, curves of the number of received moleculesrepresenting each realization of bit-1 and bit-0 transmissionsare evaluated and are called bit-0 and bit-1 curves, respectively(Fig. 5). Then, the maximum eye height is calculated asthe biggest distance between the mean of bit-0 and meanof bit-1 curves. To evaluate 𝑆𝑇 𝐷 ( Δ 𝑐 ) and CSNR, first the URAN et al.: SURVEY ON MODULATION TECHNIQUES IN MCVD 8 integral difference (that is indicating the area) between everycombination of bit-0 and bit-1 curves are calculated as Δ 𝑐 ( 𝑖, 𝑗 ) = ∫ 𝑡 𝑠 𝑐 ( 𝑖 ) − 𝑐 ( 𝑗 ) 𝑑𝑡, (10)where 𝑐 ( 𝑖 ) and 𝑐 ( 𝑗 ) are the 𝑖 -th bit-1 and 𝑗 -th bit-0 sampledcurves. Using this integral difference, 𝑆𝑇 𝐷 ( Δ 𝑐 ) is calculatedas the standard deviation of all possible Δ 𝑐 ( 𝑖, 𝑗 ) values.Similarly, CSNR is evaluated as 𝐶𝑆𝑁 𝑅 = 𝜇 Δ 𝑐 𝑆𝑇 𝐷 ( Δ 𝑐 ) , (11)where 𝜇 Δ 𝑐 represents the mean of all possible Δ 𝑐 ( 𝑖, 𝑗 ) values.Among these three metrics, it is observed in [73] that CSNRhas a one-to-one relation to BER. Assuming a channel withfinite memory 𝑚 (i.e., a model that only considers the effect ofprevious 𝑚 symbols over the current one), BER is calculatedby evaluating all the 𝑚 bit sequence combinations. On theother hand, CSNR is evaluated over a fixed number of bitcurves (e.g., in [73]). Therefore, it is computationallyeasier to calculate than BER, hence is used instead of BER tofine tune the operational parameters of devices.V. M ODULATION T ECHNIQUES
All MC modulation techniques based on releasing MMsfrom the Tx and modulating the bit-values of a given messageonto the features of the emission of MMs. On the Rx end ofthe system, these molecules interact with special biochemicalprotein structures called “receptors” and trigger an eventwithin the cell based on the receptor and the signaling pathwaywithin the Rx .There are different modulation techniques proposed in theMC literature. These techniques can be broadly classified intofive groups based on the feature of the emission of MMs: • Concentration-based Techniques : Information is repre-sented by the varying concentration level of the transmit-ted signal. • Type-based Techniques : Information is represented bythe type of the MMs of the transmitted signal. • Timing-based Techniques : Information is represented byvarious time related features of the transmitted signal. • Spatial Techniques : Information is represented by thespatial location of emission, especially for multi-antennasystems. • Hybrid Techniques : Techniques utilizing more than oneof the four features of the transmitted signal listed aboveto represent the information.
A. Concentration-Based Techniques
The main idea of concentration-based techniques is carryinginformation on the released MM concentration over discreteperiod time slots (i.e., symbol slots), where each slot is used tocarry a single symbol of the overall message. In most works, Signaling pathway is a molecular biology term describing a group ofmolecules in a cell that work together to control cell functions. After the firstmolecule in a pathway receives a signal, it activates another molecule. Thisprocess is repeated until the last molecule is activated and the cell functionis carried out hence a “pathway” [75]. each time slot has a fixed period. In its simplest form whereeach symbol represents a one-bit value (called on-off keying(OOK)), if the corresponding bit value ( 𝑆 [ 𝑘 ] ) is bit-1, the Txreleases a fixed number of MMs (i.e. 𝑛 ). On the other hand, ifit is bit-0 the Tx does not release any molecules for that symbolslot [76]. At the receiver side, the Rx counts the number ofMMs that arrive within each symbol slot (i.e., 𝑁 𝑅𝑥 [ 𝑘 ] ) andmakes a threshold based decision to decode the bit value ofthe given symbol slot ( ˆ 𝑆 [ 𝑘 ] ). If 𝑁 𝑅𝑥 [ 𝑘 ] ≥ 𝜆 , ˆ 𝑆 [ 𝑘 ] is decodedas bit-1; else it is decoded as bit-0, where 𝜆 is a thresholdvalue for signal detection.A more generalized version of OOK is called ConcentrationShift Keying (CSK) where, depending on the system de-sign, each symbol represents 𝑚 -bits of information. Followingthe classical modulation terminology, if a symbol representsone-bit of information, this technique is called Binary CSK(BCSK), and if a symbol represents two-bits of information itis called Quadrature CSK (QCSK), and so on [7], [57], [65],[77], [91]. In CSK, for the 𝑘 -th symbol in the message, theTx releases 𝑁 𝑇 𝑥 [ 𝑘 ] number of MMs depending on the currentsymbol value as 𝑁 𝑇 𝑥 [ 𝑘 ] = 𝑛 𝑆 [ 𝑘 ] , 𝑆 [ 𝑘 ] ∈ { 𝑠𝑦𝑚 , 𝑠𝑦𝑚 , ..., 𝑠𝑦𝑚 𝑚 − } , (12)where 𝑛 𝑆 [ 𝑘 ] denotes the number of molecules to be emittedfor the symbol value of 𝑆 [ 𝑘 ] and please note that 𝑆 [ 𝑘 ] can take one of the 𝑚 symbol-values (e.g., 𝑠𝑦𝑚 , 𝑠𝑦𝑚 )in the modulation alphabet. In order to demodulate ˆ 𝑆 [ 𝑘 ] from the received signal, the Rx uses 𝑚 − thresholds (i.e., 𝜆 , 𝜆 , ..., 𝜆 𝑚 − ) as ˆ 𝑆 [ 𝑘 ] = 𝑠𝑦𝑚 , 𝑁 𝑅𝑥 [ 𝑘 ] < 𝜆 𝑠𝑦𝑚 𝑖 , 𝜆 𝑖 − ≤ 𝑁 𝑅𝑥 [ 𝑘 ] < 𝜆 𝑖 , ≤ 𝑖 ≤ 𝑚 − 𝑠𝑦𝑚 ( 𝑚 − ) , 𝜆 𝑚 − ≤ 𝑁 𝑅𝑥 [ 𝑘 ] (13)where 𝑁 𝑅𝑥 [ 𝑘 ] denotes the number of received moleculesduring the 𝑘 -th symbol slot.From a communication point-of-view, CSK is akin to theAmplitude Modulation (AM) method in analog modulationand Amplitude Shift Keying (ASK) in digital modulation. Itinherits the main advantages and disadvantages of ASK asbeing a simple system to implement, but also considerablysensitive to interference and noise in the environment. Asexplained in Section II, the CIR of the MC channel has a long-tail component. This long-tail component causes considerableISI, which in turn reduces the correct decoding probability ofthe threshold based receiver.In the literature, numerous CSK-variants have been pro-posed to increase the communication performance, e.g., todecrease the error probability of the system (Table I). Most ofthe CSK-variants consider the Tx releasing all the MMs basedon the selected 𝑁 𝑇 𝑥 [ 𝑘 ] value at the start of the correspondingsymbol slot. In [62], [82] , Singhal et al. consider sending twodirac pulses during a single symbol slot, one at the beginningand another one in the middle. Correspondingly, Rx utilizestwo thresholds, one within the first half of the symbol duration URAN et al.: SURVEY ON MODULATION TECHNIQUES IN MCVD 9
TABLE IC
ONCENTRATION -B ASED M ODULATION T ECHNIQUES FOR M OLECULAR C OMMUNICATION
Technique Identification Modulation Characteristics Performance Evaluation Assumptions
Name References ISI Mitig. Comp. Complex. Detection Type Tx Waveform Rx Type EnvironmentCSK [7], [65], [77] None Low Threshold Pulse Absorbing 3D, No DriftCSK-AD [78] None Low Inst. Threshold Pulse Passive 3D, No DriftCSK-SD [79]–[81] None Low Threshold Pulse Passive 3D, No DriftCSK-SubTS [62], [82] None Low Threshold Imperfect Pulse Absorbing 1D, with DriftCSK-PA [49], [83], [84] High High Threshold Pulse Absorbing 1-2D, No DriftCSK-CPA [73] High Moderate Threshold Pulse Absorbing 1-2D, No DriftCSK w/ ATD [85] Low Low Adaptive Threshold Pulse Absorbing 3D, No DriftCSK w/ ML, MAP,MMSE [86], [87] Moderate High ML, MAP, MMSE Pulse Absorbing 3D, No DriftNC-CSK-Diff [87] None Low NC Diff. Imperfect Pulse Passive 3D, No DriftNC-CSK-Convex [88] Moderate Low NC Convex Imperfect Pulse Passive 3D, No DriftNC-CSK-Gamma [89] Moderate Low NC ML Pulse Passive 3D, with DriftCSK w/ Eq. Sig. [90] High Moderate Threshold Pulse Passive 2D, with Drift ( 𝑘 𝑓 ℎ ) and another for the second half ( 𝑘 𝑠ℎ ). It demodulatesthe received signal as ˆ 𝑆 [ 𝑘 ] = 𝑠𝑦𝑚 , 𝑁 𝑅𝑥 [ 𝑘 𝑓 ℎ ] < 𝜆 & 𝑁 𝑅𝑥 [ 𝑘 𝑠ℎ ] < 𝜆𝑠𝑦𝑚 , 𝜆 ≤ 𝑁 𝑅𝑥 [ 𝑘 𝑓 ℎ ] & 𝜆 ≤ 𝑁 𝑅𝑥 [ 𝑘 𝑠ℎ ] 𝑠𝑦𝑚 , otherwise . (14)The resulting technique, called CSK Sub-Timeslot (CSK-SubTS), carries one out of three symbol values per symbolslot.In order to minimize the error probability due to ISI, apower adjustment (CSK-PA) technique has been proposed inwhich the Tx regulates its emission amount (i.e., 𝑛 𝑺 𝑠𝑦𝑚 𝑖 ), where 𝑺 = ( 𝑆 [ 𝑘 − ] , 𝑆 [ 𝑘 − ] , ..., 𝑆 [ 𝑘 − 𝑚 ]) denotes the valuesof the past 𝑚 -symbols. By this adjustment, CSK-PA aims tominimize the difference between the 𝑁 𝑅𝑥 [ 𝑘 ] values with thesame 𝑆 [ 𝑘 ] values [49], [83], [84]. Considering a BCSK-PAsystem, the goal is to minimize the variation between the 𝑁 𝑅𝑥 [ 𝑘 ] values where 𝑆 [ 𝑘 ] = 𝑠𝑦𝑚 . Since the Tx knows thepast 𝑚 -symbols, if the channel parameters (i.e, 𝐷, 𝑟 𝑟 𝑥 , 𝑑, ) arealso known, this adjustment can be accomplished with nearcertainty. The calculation of different 𝑛 𝑺 𝑠𝑦𝑚 𝑖 values depend onthe past 𝑚 -symbol values. As 𝑚 increases, calculation of the 𝑛 𝑺 𝑠𝑦𝑚 𝑖 values becomes impractical in terms of computationalpower and memory requirements. In all three works focusingon the CSK-PA technique, 𝑚 is selected as very small to keepthe system simple and practical (In [49], 𝑚 is selected as 1(i.e., 1-bit memory) while in [83], [84] it is selected as 2).A variation of the CSK-PA technique, called the consecutivepower adjustment (CSK-CPA) technique, only focuses on theworst case scenarios [73]. In this technique, while focusingon a binary system the Tx only considers the past 𝑙 -symbolsof consecutive 𝑠𝑦𝑚 values (i.e., 𝑘 ∈ { , , .., 𝑙 } ) for poweradjustment. As an example, considering a binary system where 𝑺 = { , , , , , , } , 𝑆 [ 𝑘 ] = 𝑠𝑦𝑚 − , and the past six bit-values are used for power adjustment, CSK-PA will consider1,1,1,1,0,1 while CSK-CPA will only consider 1,1,1,1 forpower adjustment. Consequently, CSK-CPA greatly reducesthe possible Tx states and 𝑛 𝑺 𝑠𝑦𝑚 values, while only slightlyreducing the benefits of the PA mechanism.In CSK, the number of thresholds used is one less than the cardinality of the symbol set. Therefore, CSK suffers fromnot being able to differentiate between 𝑠𝑦𝑚 and no commu-nication (i.e., silence) during a symbol duration. Mahfuz et al.proposed a silence detection technique (CSK-SD) that addsone more threshold at the Rx to distinguish between these twocases [79]–[81]. Considering a BCSK modulation, BCSK-SDutilizes two thresholds 𝜆 , 𝜆 for demodulating the receivedsignal as ˆ 𝑆 [ 𝑘 ] = 𝑆𝑖𝑙𝑒𝑛𝑐𝑒, 𝑁 𝑅𝑥 [ 𝑘 ] < 𝜆 𝑠𝑦𝑚 , 𝜆 ≤ 𝑁 𝑅𝑥 [ 𝑘 ] < 𝜆 𝑠𝑦𝑚 , 𝜆 ≤ 𝑁 𝑅𝑥 [ 𝑘 ] . (15)In [78], Llatser et al. consider utilizing an instantaneousthreshold-based receiver (CSK Amplitude detection, CSK-AD) instead of a classical threshold-based receiver. The maindifference of the instantaneous threshold-based receiver is that,instead of accumulating all the MMs received within the samesymbol slot and taking a thresholding-based decision overthis accumulated value, the Rx counts the MMs receivedwithin a much shorter unit time (i.e., Δ 𝑡 ) and applies thethresholding over this much smaller number. Although thismethod can be much cheaper and simpler in design thanthe traditional threshold-based technique, it greatly increasesthe error probability of the system. In particular, Aijaz andAghvami show that this amplitude detection requires timesmore MMs to be sent than a threshold-based receiver toachieve the same error probability [92].Another variation of the basic threshold-based receiver, theadaptive threshold detector (CSK w/ATD), proposes changingthe threshold values, 𝜆 𝑖 , adaptively at every symbol slot.Considering a binary case, the Rx adaptively changes the 𝜆 value by setting it to the 𝑁 𝑅𝑥 [ 𝑘 − ] value [85]. CSK w/ATD isa fairly simple variation that does not require considerable ad-ditional complexity at the Rx. According to [85], performance-wise, CSK w/ATD decreases the error probability for short 𝑡 𝑠 and high 𝑑 values, where the ISI component is significant. Onthe other hand, it does not perform well when the informationhas long sequences of consecutive 𝑠𝑦𝑚 or 𝑠𝑦𝑚 values.As in any communication system, more complex receiverscan be used to minimize the detection errors due to ISI andnoise in the communication channel. In a sense, these solutions URAN et al.: SURVEY ON MODULATION TECHNIQUES IN MCVD 10 are the receiver-based alternatives to the CSK-PA techniquementioned above. While in CSK-PA, the Tx tries to minimizethe variation between the 𝑁 𝑅𝑥 [ 𝑘 ] values by regulating the 𝑛 𝑺 𝑠𝑦𝑚 values, in these receiver-based techniques the Rx triesto correctly demodulate ˆ 𝑆 [ 𝑘 ] by utilizing likelihood functionsand equalizers over 𝑁 𝑅𝑥 [ 𝑘 ] values and the channel stateinformation (CSI). However, since the Rx does not exactlyknow the 𝑆 [ 𝑘 ] values, these techniques are susceptible to errorpropagation.In [86], Kilinc and Akan evaluate the performance of fourreceivers designed to mitigate ISI in a time-varying diffusionenvironment: sequence based ML, linear equalization based onminimum mean-square error (MMSE), Maximum a posteriori(MAP), and decision-feedback equalization (DFE). Althoughthese receivers outperform basic threshold-based receivers interms of communication performance, they utilize complicatedmatrix or polynomial operations, which greatly increase theircomputational complexity [87]. Hence, their suitability tonanomachines, which are expected to have low computationalcapabilities, may be limited.It has been argued that in an MCvD system, the CSI maybe time-varying or very costly to be exactly acquired bythe Rx at each symbol slot. Considering such cases, severalworks propose receivers that work without having perfectCSI information, which are called non-coherent detectorsin MC domain. These detectors/receivers utilize the generalcharacteristics of the received signal regardless of the valuesof the environmental parameters (i.e, 𝐷, 𝑟 𝑟 𝑥 , 𝑑 ). In [87], areceiver that utilizes the peak time of the molecular signal anddoes the demodulation using the difference of the accumulatedconcentration in two consecutive symbol slots is proposed(i.e., NC-CSK-Diff). Here the peak time of the molecularsignal refers to the time value where the maximum numberof molecules are received. A similar work, [88], exploitsthe fact that a molecular signal, i.e., ℎ ( 𝑡 ) function, has asingular convex region, which is depicted by investigatingthe second derivative of ℎ ( 𝑡 ) . Based on this convex region,the receiver can approximate the ISI effect of a given signalover the subsequent symbol slots without knowing the exactCSI of the channel, and use this estimate to correctly decodethe signal. A similar approach is proposed in [93] and itsperformance evaluated on a testbed. All of these works arelow-complexity solutions, as is needed in practical applicationsof nanomachines.A third work builds an ML-based non-coherent receiverthat utilizes the fact that for a point Tx, spherical passiveRx and an environment with destroyer MMs, the CIR canbe approximated by a Gamma distribution [89]. By observingthe channel’s behavior for several symbol slots, the Rx canestimate the distribution parameters and use them to correctlydemodulate ˆ 𝑆 [ 𝑘 ] from the signal. The authors propose twoversions of the same receiver, one having higher performanceand also higher computational complexity, another having verylow computational complexity at a cost of worse performance.Another direction for the receiver design in concentration-based techniques is to take the geometries of the environment,the Tx, and the Rx into consideration in the reception process.In [90], such a technique, named CSK w/Eq. Sig., the authors design a reception process that combines the thresholding andlow memory symbol-by-symbol detection in the equilibriumstate of the diffusion channel. Although it is robust to varia-tions in the diffusion coefficient as well as variations in theoverall geometry of the system components, since it requiresthe system to reach the equilibrium state, the symbol durationis very high and consequently the technique is suitable forapplications with very low bit rate requirements. B. Type Based Techniques
A second group of modulation techniques, called type-based techniques, focuses on using multiple types of MMsin the communication system as the basis of the modulationtechnique. In these techniques, the transmitter has the capa-bility of releasing different types of MMs ( 𝑚𝑚 type ), which aresimilar to each other in composition (i.e., radius, diffusivity)but can only be received by a particular type of receptor atthe receiver surface. Using different types of receptors eachcorresponding to a particular type of MM, the receiver iscapable of receiving multiple molecular release signals, whichare practically orthogonal to each other, within a single symbolslot. The quantitative feature of these orthogonal molecularrelease signals used to represent bit-values depend on theparticular technique in question.In the first type-based modulation technique, MolecularShift Keying (MoSK), each different symbol-value ( 𝑆 [ 𝑘 ] ) isrepresented by a specific type of MM [53], [57], [77], [94].MoSK uses two types of MMs to modulate one-bit of infor-mation within a symbol (called Binary MoSK - BMoSK) orfour types of MMs to modulate two-bits of information withina symbol (called Quadruple MoSK - QMoSK). ConsideringBMoSK, for each symbol slot the Rx counts the number ofarriving MMs for each MM type and demodulates ˆ 𝑆 [ 𝑘 ] basedon the thresholding decisions for each MM type as: ˆ 𝑆 [ 𝑘 ] = 𝑠𝑦𝑚 , 𝑁 𝑅𝑥𝑚𝑚 𝑎 [ 𝑘 ] ≥ 𝜆 & 𝑁 𝑅𝑥𝑚𝑚 𝑏 [ 𝑘 ] < 𝜆𝑠𝑦𝑚 , 𝑁 𝑅𝑥𝑚𝑚 𝑏 [ 𝑘 ] ≥ 𝜆 & 𝑁 𝑅𝑥𝑚𝑚 𝑎 [ 𝑘 ] < 𝜆𝑒, otherwise , (16)where 𝑎 and 𝑏 represent the types of the MMs used, 𝑀 𝑀 𝑡 ∈( 𝑚𝑚 𝑎 , 𝑚𝑚 𝑏 ) , and 𝑁 𝑅𝑥𝑚𝑚 type [ 𝑘 ] represents the number of 𝑚𝑚 type received at the Rx during the 𝑘 𝑡ℎ symbol slot.An alternative approach in type-based techniques is to use amajority based detection instead of thresholding [69]. In sucha binary type-based technique, ˆ 𝑆 [ 𝑘 ] is decoded as ˆ 𝑆 [ 𝑘 ] = (cid:40) 𝑠𝑦𝑚 , 𝑁 𝑅𝑥𝑚𝑚 𝑎 [ 𝑘 ] > 𝑁 𝑅𝑥𝑚𝑚 𝑏 [ 𝑘 ] 𝑠𝑦𝑚 , 𝑁 𝑅𝑥𝑚𝑚 𝑏 [ 𝑘 ] ≥ 𝑁 𝑅𝑥𝑚𝑚 𝑎 [ 𝑘 ] . (17)A more generalized version of MoSK has been proposedby removing the constraint of a single type emission duringa symbol slot named generalized MoSK (GMoSK) [101]. Byutilizing multiple types of MMs, the cardinality of the symbolset is increased which results in higher data rates at highertransmission powers at the cost of higher receptor complexitythan the MoSK.Compared to CSK, MoSK considerably reduces the ISIelement of the signal due to the fact that each bit value is URAN et al.: SURVEY ON MODULATION TECHNIQUES IN MCVD 11
TABLE IIT
YPE -B ASED T ECHNIQUES FOR M OLECULAR C OMMUNICATION
Technique Identification Modulation Characteristics Performance Evaluation Assumptions
Name References ISI Mitig. Comp. Complex. Detection Type Tx Waveform Rx Type EnvironmentMoSK [53], [57], [77], [94] Moderate Moderate Threshold Pulse Absorbing 3D, No DriftIMoSK [95], [96] Moderate Moderate Threshold Pulse Absorbing 1D, DriftMoSK w/ ML [97] High High ML Pulse Passive 2D, No DriftRSK [98] Moderate Moderate Threshold Pulse Absorbing 1D, DriftMCSK [99] High Moderate Threshold Pulse N/A N/AMTSK [83] High High Threshold Pulse Absorbing 2D, No DriftPre-eq. CSK [69] High High Threshold (Diff) Pulse Absorbing 3D, No DriftZebra-CSK [100] High High Threshold Pulse Absorbing 3D, No Drift represented by a different type of molecule. On the otherhand, it increases the complexity of both the Tx and theRx since the Tx is required to synthesize different types ofmolecules and the Rx is required to have multiple receptorson its surface. Moreover, the average number of emittedmolecules is higher in MoSK compared to CSK since there isan emission of molecules at each symbol slot, while thereare symbol slots with no emission in CSK such as whiletransmitting sym-0 values in BCSK. Unlike CSK, MoSK doesnot have a direct counterpart among modulation techniquesutilized in classical telecommunication. Since each moleculetype is practically orthogonal to each other, one can think ofthem as signals of different frequencies with no overlappingcomponents but the properties of RF signals and molecularsignals are widely different. Beside the basic MoSK, othertype-based techniques have also been proposed in the literatureeach focusing on different methods of utilizing multiple MMtypes in representing the received molecular signal (Table II).As an alternative to the simple thresholding-based detectorproposed for MoSK receiver, ShahMohammadian et al. pro-pose an ML detector for MoSK in a linear, time-invariantdiffusion environment with both ISI and noise componentsin [97]. Similar to the work by Kilinc and Akan in [86]for concentration-based techniques, although this ML receiveroutperforms the simple thresholding-based detector, it requiresconsiderable computation power at the receiver due to theViterbi algorithm used for calculations, which reduces itspracticality inside a nanomachine.In [98], Kim and Chae propose a type-based modulationtechnique called ratio shift keying (RSK) using two types ofMMs ( 𝑚𝑚 𝑎 , 𝑚𝑚 𝑏 ) and the information is modulated over thereceived ratio of these two MM types during each symbolduration (i.e., 𝑅 𝑅𝑥𝑚𝑚 𝑎 / 𝑏 [ 𝑘 ] = 𝑁 𝑅𝑥𝑚𝑚 𝑎 [ 𝑘 ]/ 𝑁 𝑅𝑥𝑚𝑚 𝑏 [ 𝑘 ] ). In BinaryRSK (BRSK), the received signal is demodulated by using asingle threshold value 𝜆 as ˆ 𝑆 [ 𝑘 ] = (cid:40) 𝑠𝑦𝑚 , 𝑅 𝑅𝑥𝑚𝑚 𝑎 / 𝑏 [ 𝑘 ] < 𝜆𝑠𝑦𝑚 , 𝑅 𝑅𝑥𝑚𝑚 𝑎 / 𝑏 [ 𝑘 ] ≥ 𝜆 , (18)whereas in Quadruple RSK (QRSK) it is decoded by usingthree threshold values 𝜆 , 𝜆 𝜆 as ˆ 𝑆 [ 𝑘 ] = 𝑠𝑦𝑚 , 𝑅 𝑅𝑥𝑚𝑚 𝑎 / 𝑏 [ 𝑘 ] < 𝜆 𝑠𝑦𝑚 , 𝜆 ≤ 𝑅 𝑅𝑥𝑚𝑚 𝑎 / 𝑏 [ 𝑘 ] < 𝜆 𝑠𝑦𝑚 , 𝜆 ≤ 𝑅 𝑅𝑥𝑚𝑚 𝑎 / 𝑏 [ 𝑘 ] < 𝜆 𝑠𝑦𝑚 , 𝜆 ≤ 𝑅 𝑅𝑥𝑚𝑚 𝑎 / 𝑏 [ 𝑘 ] . (19) As shown in their work, RSK has a similar performanceto MoSK in the binary case but its performance degradesto the level of CSK in the quadruple case due to the factthat it utilizes multiple threshold values, which increases thereception error.Arjmandi et al. propose an alternative type-based modula-tion technique, called Molecular Concentration Shift Keying(MCSK), where, if 𝑆 [ 𝑘 ] = 𝑠𝑦𝑚 , the Tx releases ( 𝑚𝑚 𝑎 ) typeof MMs for an odd-numbered symbol and releases ( 𝑚𝑚 𝑏 ) type of MMs for an even-numbered symbol [99]. Similar toBCSK, in case 𝑆 [ 𝑘 ] = 𝑠𝑦𝑚 the Tx does not release any MMs.The Rx demodulates the signal by checking the concentrationof ( 𝑚𝑚 𝑎 ) or ( 𝑚𝑚 𝑏 ) depending on the symbol parity as ˆ 𝑆 [ 𝑘 ] = 𝑠𝑦𝑚 , (cid:40) 𝑘 𝑚𝑜𝑑 = 𝑁 𝑅𝑥𝑚𝑚 𝑎 [ 𝑘 ] < 𝜆𝑘 𝑚𝑜𝑑 = 𝑁 𝑅𝑥𝑚𝑚 𝑏 [ 𝑘 ] < 𝜆𝑠𝑦𝑚 , (cid:40) 𝑘 𝑚𝑜𝑑 = 𝑁 𝑅𝑥𝑚𝑚 𝑎 [ 𝑘 ] ≥ 𝜆𝑘 𝑚𝑜𝑑 = 𝑁 𝑅𝑥𝑚𝑚 𝑏 [ 𝑘 ] ≥ 𝜆 . (20)According to the results provided in [99], in the binary case,MCSK outperforms MoSK in terms of BER due to the fact thatthe ISI component of the molecular signal after a sequence of 𝑠𝑦𝑚 -valued symbols grows large. Hence it is highly probablethat the next symbol with 𝑠𝑦𝑚 will be decoded incorrectlyas 𝑠𝑦𝑚 . The alternating approach of MCSK remedies sucherroneous decodings. However as shown in [99], MCSK losesits edge over MoSK in terms of BER in the quadruple case. So,its advantage is more specific to the binary implementations.In a communication system, interference can be either con-structive or destructive. Considering the BMoSK technique,the ISI part of a 𝑠𝑦𝑚 transmission becomes a constructiveinterference source for the upcoming 𝑠𝑦𝑚 symbols (i.e., forall 𝑆 [ 𝑘 + 𝑥 ] = 𝑠𝑦𝑚 where 𝑥 ∈ Z + ). On the other hand, for thesymbol values 𝑠𝑦𝑚 after a 𝑠𝑦𝑚 (i.e., for all 𝑆 [ 𝑘 + 𝑥 ] = 𝑠𝑦𝑚 where 𝑥 ∈ Z + , if 𝑆 [ 𝑘 ] = 𝑠𝑦𝑚 ), the same ISI becomes de-structive. Tepekule et al. propose a technique called moleculartransition shift keying (MTSK), which focuses on constructiveand destructive ISI differentiation by using two types of MMs[83]. In MTSK, if 𝑆 [ 𝑘 ] = 𝑠𝑦𝑚 , the transmitter checks thevalue of 𝑆 [ 𝑘 + ] and chooses one of the two types of MMsaccording to this value to release in this symbol duration as 𝑇 𝑚𝑚 [ 𝑘 ] = 𝑚𝑚 𝑎 , 𝑆 [ 𝑘 ] = 𝑠𝑦𝑚 & 𝑆 [ 𝑘 + ] = 𝑠𝑦𝑚 𝑚𝑚 𝑏 , 𝑆 [ 𝑘 ] = 𝑠𝑦𝑚 & 𝑆 [ 𝑘 + ] = 𝑠𝑦𝑚 ∅ , 𝑆 [ 𝑘 ] = 𝑠𝑦𝑚 . (21) URAN et al.: SURVEY ON MODULATION TECHNIQUES IN MCVD 12
Similar to BCSK, in case 𝑆 [ 𝑘 ] = 𝑠𝑦𝑚 the Tx does notrelease any MMs. At the Rx, the signal is demodulated bychecking the concentration levels of both ( 𝑚𝑚 𝑎 ) and ( 𝑚𝑚 𝑏 ) each symbol slot as ˆ 𝑆 [ 𝑘 ] = (cid:40) 𝑠𝑦𝑚 , 𝑁 𝑅𝑥𝑚𝑚 𝑎 [ 𝑘 ] < 𝜆 & 𝑁 𝑅𝑥𝑚𝑚 𝑏 [ 𝑘 ] < 𝜆𝑠𝑦𝑚 , 𝑁 𝑅𝑥𝑚𝑚 𝑎 [ 𝑘 ] ≥ 𝜆 | 𝑁 𝑅𝑥𝑚𝑚 𝑏 [ 𝑘 ] ≥ 𝜆 . (22)This way, the Tx utilizes the advantage of the constructiveISI component while at the same time reducing the detrimentaleffect of the destructive ISI by switching to the second typeof molecules when the symbol transition occurs from 𝑠𝑦𝑚 to 𝑠𝑦𝑚 . As expected, MTSK considerably reduces the BERin the system compared to both basic CSK and MoSK tech-niques, while only increasing the system complexity slightly.Beside the communication aspects of the type-based tech-niques, MMs must be selected carefully to be chemically simi-lar to each other and to have high bio-compatibility, especiallyfor the in-vivo environments. Also, each MM type should havemore or less the same diffusion capabilities (i.e., diffusion co-efficient) to eliminate any inequality between different symbol-values. In [77], hydrofluorocarbon-based MMs are proposed,which are chemically similar to each other and can be easilysynthesized by a biological nanomachine. Alternatively, Kimand Chae propose using aldohexoses-based isomers as MMs,called Isomer Molecular Shift Keying (IMoSK), which havemuch higher bio-compatibility than the hydrofluorocarbonsand have similar chemical and synthesis features to them[95], [96]. However, these works are very preliminary studiesof MM type selection and none of them shows the bio-compatibility of these molecules in a quantitative fashion.Additionally, these MMs may not be suitable in an in-vitroenvironment with completely different environmental aspects.In [100], Pudasini et al. propose an advanced version ofthe MCSK technique called Zebra CSK, where destroyermolecules (called inhibitors) are used to reduce the ISI effectof the received signal. In Zebra CSK, when 𝑆 [ 𝑘 ] = 𝑠𝑦𝑚 , inaddition to alternatively releasing ( 𝑚𝑚 𝑎 ) or ( 𝑚𝑚 𝑏 ) dependingon the symbol number as in MCSK, the Tx also releasesdestroyer molecules of type ( 𝑑𝑚 𝑏 ) when it releases ( 𝑚𝑚 𝑎 ) ,or ( 𝑑𝑚 𝑎 ) when it releases ( 𝑚𝑚 𝑏 ) . These destroyer moleculesinteract with the MMs in the environment and practically“destroy” them in the environment. Therefore, the effect ofMMs from previous symbol slots are reduced to have a lowerBER.In a similar work, Tepekule et al. approach the problem ofISI mitigation by utilizing two types of molecules, ( 𝑚𝑚 𝑎 ) and ( 𝑚𝑚 𝑏 ) , where the Rx measures the received concentrationsof each MM type separately and consider the difference ofthese two concentrations as the actual received signal [69].This pre-equalization technique, Pre-eq CSK, works similar tothe BCSK technique using the ( 𝑚𝑚 𝑎 ) type of MMs. However,when 𝑆 [ 𝑘 ] = 𝑠𝑦𝑚 , after releasing type ( 𝑚𝑚 𝑎 ) molecules, theTx also releases type ( 𝑚𝑚 𝑏 ) MMs after a short delay. Thereceiver demodulates the signal by checking the concentrationdifference between these two types of MMs as ˆ 𝑆 [ 𝑘 ] = (cid:40) 𝑠𝑦𝑚 , 𝑁 𝑅𝑥𝑚𝑚 𝑎 [ 𝑘 ] − 𝑁 𝑅𝑥𝑚𝑚 𝑏 [ 𝑘 ] < 𝜆𝑠𝑦𝑚 , 𝑁 𝑅𝑥𝑚𝑚 𝑎 [ 𝑘 ] − 𝑁 𝑅𝑥𝑚𝑚 𝑏 [ 𝑘 ] ≥ 𝜆 . (23) This short delay choice and the ratio between 𝑁 𝑅𝑥𝑚𝑚 𝑎 and 𝑁 𝑅𝑥𝑚𝑚 𝑏 are tunable parameters whose ideal values can be foundvia heuristic methods depending on the environment. Pre-eqCSK is a less-complex technique from an implementationpoint-of-view and it achieves a much lower BER than CSK,CSK with MMSE, and MoSK. C. Timing Based Techniques
The third group of modulation techniques encodes informa-tion on the time of release of MM. Here, we will refer to thissubclass of MC channels as molecular timing channels (MTC).This modulation is different from the previous methods inthat the channel input is fundamentally continuous instead ofdiscrete. The MTC model was first proposed in [102], whichis more suitable for evaluating timing-based modulations. Inits simplest form, a MTC is based on a single MM releasedby the Tx at time 𝑡 𝑟 with information encoded on this releasetime (RT). The MM goes through some random propagationand arrive at the destination at time 𝑡 𝑦 = 𝑡 𝑟 + 𝑡 𝑛 , (24)where 𝑡 𝑛 is some random delay due to MM propagation.Note that unlike previous modulations where the symbol set isfinite, in pure timing channels, the symbol set is a continuousinterval.An important parameter of the MTC is the probabilitydistribution of the random delay 𝑡 𝑛 . Considering a 1D en-vironment with an absorbing receiver, this random delayis L´evy distributed for free diffusion and inverse Gaussiandistributed for diffusion with drift [50]. In a 3D free diffusionenvironment, as described in subsection II-C, the MM maynever arrive at the Rx, and therefore the channel model mustbe modified to reflect this effect, as we will show later. A3D vessel-like environment with drift can be approximatedas a 1D environment [103] so the same distributions canbe utilized for such environments. It can be argued that thisprobability distribution of 𝑡 𝑛 has a connection with the CIRof the corresponding MC channel. Particularly, if we assumethat the MMs move independent of each other, the expectednumber of MMs that arrive at the Rx as a function of time (i.e., 𝐸 [ 𝑁 𝑅𝑥 [ 𝑡 ]] ) is proportional to this probability distribution.The first series of works that explore timing-based tech-niques focus on understanding the fundamental limits ofMTCs. In [40], an MTC with additive inverse Gaussian noise(AIGN) was introduced where 𝑡 𝑛 in (24) is inverse Gaussiandistributed. The authors derive the upper and lower boundson capacity per channel use for this MTC with AIGN case.Then, in [50], tighter bounds on capacity of the same MTCwith AIGN case were derived. The capacity-achieving MMinput distribution was also characterized for this case.One of the drawbacks of the model in (24) is that it onlyconsiders a single MM release. In practical systems, manyMMs can be released by the Tx. Another drawback is thatthe notion of channel use (i.e., how long is the channel useduration) is not rigorously defined. It is also assumed that A channel use is a single use of the channel to transmit one symbol ofthe symbol set from the Tx to Rx.
URAN et al.: SURVEY ON MODULATION TECHNIQUES IN MCVD 13 st Symbol Slot 2 nd Symbol Slot 3 rd Symbol Slot
00 01
10 11
00 01 10 11 00 01 10 11
EmissionSub-Slot T x M o l e c u l e s Fig. 6. Example scenario of 4-PPM modulation. The information symbols are 01 10 00 and the emissions are done during the corresponding sub-slots.
MMs will eventually arrive at the Rx and they are not degradedor destroyed prior to arrival. Finally, ISI is not consideredin this model. To overcome some of these limitations, twoapproaches have been considered in the literature.First, in [39], [52], [102], [104] the single particle releaseis extended to multiple MM releases. In this scenario, infor-mation is encoded in the vector (cid:174) 𝑡 𝑠 , where each element in thevector is the time of release of one of 𝑀 MMs. It is assumedthat each particle goes through a random independent path(hence a random independent delay). The Rx observes thevector (cid:174) 𝑡 𝑦 , which contains a set of 𝑀 arrivals correspondingto each MM. Note that here again it is assumed that all theparticles eventually arrive at the Rx. If all 𝑀 MMs are ofdifferent MM types, which do not react or interact with eachother, this channel reduces to 𝑀 parallel channels given in(24). The nontrivial scenario is when the 𝑀 MMs are of thesame type and hence are indistinguishable at the Rx. In thisnontrivial scenario, the channel model becomes: ˜ (cid:174) 𝑡 𝑦 = sort (cid:0) (cid:174) 𝑡 𝑟 + (cid:174) 𝑡 𝑛 (cid:1) , (25)where (cid:174) 𝑡 𝑛 is the vector of random delays associated with eachMM, sort ((cid:174) 𝑡 ) is the sort operator that permutes the vector (cid:174) 𝑡 in ascending order of arrival times, and ˜ (cid:174) 𝑡 𝑦 is the observationat the Rx. Note that the sort operation is necessary since theMMs may arrive out of order (i.e., not according to the orderthey were released) because of independent random delayassociated with their random propagation.In [52], [104], the fundamental limits of multiple MMrelease MTC in (25) is investigated. Since this is a complicatedchannel, it is assumed that the random delays have a finitemean. Particularly, it is assumed that the random delays areexponentially distributed. Then, a guard interval is placedbetween different channel uses in order to ensure that theprobability of ISI is low. Using these assumptions, the 𝜀 -capacity of (25) is analyzed by presenting its upper andlower bounds. In 𝜀 -capacity, instead of driving the probabilityof error to zero (i.e., 𝜀 = ) as the block length increases,the probability of error is driven down to 𝜀, 𝜀 > . Notethat for 𝜀 > , despite the guard interval, there is a nonzeroprobability that there will be an ISI event, which is counted asan error in this setup. Note that calculating the capacity of thischannel such that the error probability approaches zero is verychallenging and that is why the authors consider 𝜀 -capacity. 𝜀 -capacity refers to the maximum rate of sequence of codes that can attaina decoding error probability less than 𝜀 ∈ [ , ) Although [52], [104] consider multiple MMs and ISI, aswell as rigorously defining the channel use interval, it assumesthat all the MMs arrive at the Rx. One of the challengesin MTC channels is that, without this assumption, analyzingits the fundamental limits in (25) becomes very difficult.Therefore, in [51], [55] a new MTC is presented wherethe MMs have a finite lifetime 𝜏 𝑛 . In this channel model,information is encoded in the time of release of 𝑀 MMs andall of the 𝑀 MMs are released simultaneously at the sametime within a symbol interval 𝜏 𝑠 . Consequently, the overallarrival time, (cid:174) 𝑌 [ 𝑖 ] , in this MTC channel is given as (cid:174) 𝑌 [ 𝑖 ] = (cid:40) (cid:174) 𝑡 𝑦 [ 𝑖 ] = 𝑡 𝑟 + (cid:174) 𝑡 𝑛 [ 𝑖 ] , (cid:174) 𝑡 𝑛 [ 𝑖 ] ≤ 𝜏 𝑛 𝜙, (cid:174) 𝑡 𝑛 [ 𝑖 ] > 𝜏 𝑛 , (26)where the 𝑡 𝑟 is the time that the 𝑀 MMs are released, (cid:174) 𝑡 𝑛 [ 𝑖 ] is the random propagation delay associated with the i-th MM,and (cid:174) 𝑡 𝑦 [ 𝑖 ] is the arrival time of the i-th MM. Here, the first casedepicts the case if the MM arrives at the Rx, and the secondcase corresponds to the case where the MM is degraded beforearriving at the Rx. In [55], the capacity of the MTC in (26) isanalyzed by deriving two lower bounds on capacity as well asan upper bound on the capacity. For the case where the randomdelay is L´evy distributed (i.e., for free diffusion-based MTC),it is shown that the capacity scales at least polylogarithmicallywith the number of MMs released.Besides these information theoretic works, which aim toevaluate the fundamental limits of the MTC, another seriesof works study system design (e.g., receiver design) for theMTC. Some of these works focus on pulse position modulation(PPM), where information is encoded in the time of releaseof pulses (i.e., release of a large number of MMs). In thisscheme, the symbol interval is divided into |S| sub-intervals, 𝑡 , 𝑡 , · · · , 𝑡 |S |− , where S is the symbol set. Then a pulse of 𝑀 MMs are released at the beginning of the sub-intervals cor-responding to the transmission symbol. Let 𝑦 , 𝑦 , · · · , 𝑦 |S |− be the number of MMs that arrive during each sub-interval.Then, assuming that the CIR has its maximum within eachsub-interval and that there is no ISI, the symbol can bedemodulated using ˆ 𝑆 = 𝑠𝑦𝑚 ℓ , ℓ = argmax ≤ 𝑠< |S | 𝑦 𝑠 . (27)An example scenario of 4-PPM is depicted in Fig. 6. A sum-mary of various MTC and non-MTC timing-based techniquesproposed in the literature are given in (Table III).In [106], the performance of CSK and PPM has beencompared and it is demonstrated that for binary symbols, CSK URAN et al.: SURVEY ON MODULATION TECHNIQUES IN MCVD 14
TABLE IIIT
IMING -B ASED T ECHNIQUES FOR M OLECULAR C OMMUNICATION
Technique Identification Modulation Characteristics Performance Evaluation Assumptions
Name References ISI Mitig. Comp. Complex. Detection Type Numb. MM Rx Type EnvironmentRT-Single [40], [50] None Medium ML Single Absorbing 1D, DriftRT-Multi [39], [52], [102], [104] None High N/A Multi Absorbing 1D, No DriftRT-Multi w/Viterbi [105] None High ML/Viterbi Multi Absorbing 1D, No DriftPPM [106], [107] Low Low Threshold Multi Absorbing 3D, No driftPPM w/ ML [108]–[110] None Medium ML/Max Multi Absorbing 1-3D, DriftPPM w/ decay [51], [55] None Medium First Arrvl/Avg. Multi Absorbing 1D, No DriftTime btwn pulses [106], [111]–[113] Low Medium Threshold Multi/Single Absorbing 1-3D, DriftMFSK [72], [114]–[116] None High Bandpass Multi Absorbing 1-3D, No Drift is better than PPM in an ideal diffusion scenario. Later in [107]it was demonstrated that when the symbol set contains moreelements (e.g. 16-PPM), PPM can outperform CSK in termsof BER performance again. This is because PPM is exposedto less ISI. The PPM technique is also investigated in [108]with ML and the proposed (MAX) detectors. The proposedMAX detector records the maximum values in each time binand detects the intended symbol accordingly.The optimal ML detector for one-shot PPM (i.e., wherethere is no ISI), for a detector that can detect the arrival time ofindividual MMs, is derived in [109]. Since this ML detectorcan have a large computational complexity, a new detectoris proposed that can detect the symbols using the time ofarrival of the first MM. It is shown that in some regimes,the performance of this detector approaches to that of the MLdetector which can observe the arrival time of all the MMs.These results are extended in [110], where it is shown that fora PPM scheme that is similar to the channel in (26), the firstarrival receiver, last arrival receiver or the average receivercan each achieve a performance close to the optimal detector,depending on the distribution of (cid:174) 𝑡 𝑛 . The first arrival detectordetects the symbols using the time of arrival of the first MMthat arrives at the receiver. Similarly, the last arrival uses thelast arrival time of the final MM that arrives at the receiver,while the average arrival time detector can observe the averagearrival time of MMs. The optimal detector can observe thearrival time of all the MMs.In [105], the MTC in (25) is considered, where the Tx canselect the release time of individual MMs, and the Rx candetect the arrival time of each MM. The optimal ML detectorfor such a system is derived and shown to have an exponentialcomputational complexity. Then a sequence detector based onthe Viterbi algorithm is derived that achieves a performanceclose to that of the optimal detector.One of the drawbacks of encoding information on to thetime of release of MMs or PPM is that the Tx and the Rxneed to be synchronized. One way to perform asynchronousmodulation is to encode information on the time intervalbetween two consecutive releases of molecules or pulses [106].One of the first works to consider asynchronous communi-cation over MTCs was [111], where a system that encodesinformation on the time interval between molecule releasesas well as molecule types was considered. The probability oferror and achievable information rates are then derived forthe asynchronous scheme and compared with the synchronous schemes.In [112], an MC system that relies on bacteria colonies asreceivers is considered. Such a system suffers from a largedelay since, besides the time it takes for the MMs to travelfrom the Tx to the Rx, at the Rx itself there is a long delaybefore detection. This is because the bacteria activate the path-ways that generate green fluorescent proteins (GFP), whichcan then be used to indicate the arrival of MMs. In such MCsystems with long delays, it is demonstrated that by encodinginformation on the time between pulses, information rate canbe significantly increased compared to CSK, especially byusing techniques such as differential coding.In [113] it was demonstrated that when information isencoded between the time of release of single MMs, andthere is no ISI, the channel can be represented as an additivenoise channel where the noise is distributed according to astable distribution. For diffusion based propagation, this stabledistributed noise has an infinite mean and variance, becauseof heavy tails. Hence geometric signal-to-noise ratio (GSNR)is introduced for stable distributed noises. The GSNR uses thegeometric power of the noise 𝑆 ( 𝒩 ) , which is defined as 𝑆 ( 𝒩 ) (cid:44) 𝑒 𝐸 [ log | 𝒩 |] (28)where | . | stands for the norm operator and 𝒩 is the noisesignal. It is shown that GSNR exhibits the same benefit asSNR in AWGN channel where the probability of bit error isthe same for a given GSNR value regardless of the transmitor noise power [113], [117].The final time-based technique changes the rate of release,i.e., the frequency of the release. This technique we callthe Molecular Frequency Shift Keying (MFSK), which issimilar to frequency shift keying (FSK), was first proposedand evaluated under idealistic settings in [114]. A subsequentwork [115] proposed a technique for detection of frequencymodulated molecular signals. In particular, reactions-diffusionand reaction kinetic equations are used to model and design thereception process at the receiver. In the design of the receiver,ISI is also considered, and silence periods are used to mitigateISI. The performance of the MFSK modulation is comparedto CSK in [72] using a particle based simulator and scenariosunder which MFSK can achieve good performance are dis-cussed. Finally, in [116], passband modulation is implementedby precisely controlling the longitudinal wave properties ofMMs. Longitudinal waves are generated by the Tx oscillatingtowards or away from the Rx. This creates distinguishable URAN et al.: SURVEY ON MODULATION TECHNIQUES IN MCVD 15
Tx Node Rx Node d r circ Fig. 7. Example × molecular MIMO system with uniform circular antenna(i.e., emission and absorption apparatus) placement. Indices of the antennasare shown in small circles starting from 1 to 6. In index modulation (IM),indices of the antennas represent the symbols and the Rx node detects theemission antenna index to demodulate the intended symbol. carrier waves through a simple oscillation of the Tx, andcan result in molecular frequency division multiplexing, wheremultiple transmitters can communicate with a single Rx withlimited interference. D. Spatial Techniques
Another group of techniques, called Spatial Techniques,aims to convey information by utilizing spatial diversity. Multi-antenna systems can embed the information into the emissionlocation/antenna index and the Rx demodulates the symbolby estimating the emission location/antenna index. A × example system is depicted in Fig. 7 where the Tx and theRx are 𝑑 apart and their antennas are placed in a circular waywith a radius of 𝑟 circ . Please note that the distance betweenany given antenna pair (i.e., 𝑑 𝑖, 𝑗 : distance between the 𝑖 -thantenna of the Tx and 𝑗 -th antenna of the Rx) varies.Spatial diversity is utilized in a simple way by consideringmolecular MIMO scenarios [58], [118]–[122]. In these works,the achievable throughput is improved by utilizing multipleantennas with classical modulation techniques. Index modula-tion (IM), which utilizes indices of the transmitting antennas toconvey additional information bits, was subsequently adaptedto the MCvD system by Gursoy et al. who coined the term molecular IM . In molecular IM, the message is modulated ontothe index of emission antenna(s) and detected/demodulated byestimating the emission antenna index or indices. The mostfundamental version of molecular IM, which only uses a singleantenna, has been proposed as a new modulation techniquecalled molecular space shift keying (MSSK) [59]. Authorscompared two parallel MSSK with Molecular Spatial Modu-lation (MSM) that utilizes MoSK for the symbol constellation(please note that both techniques use two types of molecules).The former technique is better for coping with ISI, on theother hand, MSM copes with Inter Link Interference (ILI)better. In addition to MSSK, Gursoy et al. incorporated theGray coding scheme for the antenna indices to reduce theBER. In [123], Gursoy et al. combined MSSK with PPMtechnique and showed that this proposed combination tech-nique achieved a significant improvement over both baselineMSSK and MSSK with CSK techniques. Spatial modulationis also analyzed in [124], [125] by considering selectioncombining and additionally equal gain combining in [124] andmaximum ratio combining in [125] at the Rx. In both studies, it is shown that the communication performance (in terms ofBER) is improved significantly even with simple versions ofindex modulation, which only activates a single antenna formodulation at each symbol slot.For 𝑚 -ary MSSK, only one of the 𝑚 antennas( 𝑎 , 𝑎 , ...𝑎 𝑚 − ) is active at each symbol slot, which reduces theILI effect significantly. At the Tx, the antenna that emits thepre-determined amount of MMs, 𝐴 Tx [ 𝑘 ] , is selected accordingto the value of the symbol as 𝐴 Tx [ 𝑘 ] = 𝑎 𝑖 , where 𝑆 [ 𝑘 ] = 𝑠𝑦𝑚 𝑖 . (29)Then, the received symbol is demodulated as ˆ 𝑆 [ 𝑘 ] = 𝑠𝑦𝑚 ℓ ℓ = argmax 𝑗 ∈{ ,...,𝑚 − } 𝑁 𝑅𝑥, 𝑗 [ 𝑘 ] , (30)where 𝑁 𝑅𝑥, 𝑗 [ 𝑘 ] denotes the total number of receivedmolecules by the 𝑗 -th antenna of the Rx (i.e., 𝑎 𝑗 ) in the 𝑘 -th symbol slot. In other words, the Rx performs maximumcount detection. The possible error sources can be caused byISI, ILI, and antenna misalignment. By checking the channelcoefficients, it can be seen that the most prominent ILI-caused errors are due to the two adjacent Rx antennas ofthe intended one. Therefore, authors incorporated Gray codingfor the antenna indexing to reduce the number of bit errorsin connection with ILI [59]. Even if the most basic versionof molecular IM (i.e., MSSK) has promising results, themore generic molecular IM versions that allow multiple activeantennas for MC have not been studied in the literature. E. Hybrid Techniques
As discussed in the previous subsections, most of the mod-ulation techniques in the molecular communication literaturefocus on a single property of the molecular signal to varyaccording to the data to be transmitted. However, a handful ofworks propose techniques that utilize more than one of theseproperties. Although these techniques yield higher achievablebit rates, they require more complex mechanisms both at theTx and at the Rx, which may limit their practicality.In [7], [126], two similar methods have been proposed thatuse multiple molecular signals, where these signals utilizeCSK but with different MM types ( 𝑚𝑚 𝑡 𝑦 𝑝𝑒 ). From a con-ceptual point of view, this method, referred to as m-ChannelCSK (m-CCSK) in this paper, where 𝑚 denotes the numberof 𝑚𝑚 𝑡 𝑦 𝑝𝑒 s being used, is similar to orthogonal channelsin wireless communication. Considering a complete indepen-dency among the different 𝑚𝑚 𝑡 𝑦 𝑝𝑒 s being used, the m-CCSKincreases the achievable information rate by 𝑚 , where each 𝑚𝑚 𝑡 𝑦 𝑝𝑒 is considered a parallel independent communicationchannel. However, such a technique requires the synthesis,storage, and release of 𝑚 MMs at the Tx, which will increasethe energy cost of the communication by the same amount.At the Rx, there must be receptors suitable for each 𝑚𝑚 𝑡 𝑦 𝑝𝑒 covering the whole surface of the Rx which will make itsdesign more complex and even impossible as 𝑚 increases.Another technique that utilizes both the concentration andthe type of MMs is the hybrid modulation scheme (HMS)that is proposed in [127]. Similar to the CSK-PA technique, URAN et al.: SURVEY ON MODULATION TECHNIQUES IN MCVD 16
HMS also leverages the fact that in some parts of the inputsignal there will be consecutive symbols having the samesymbol value. As such, the number of consecutive symbolshaving the same symbol value becomes one input signal (i.e.,repetition length signal - ˆ 𝑙 [ 𝑘 ] ) and the symbol value of eachrepetition becomes another input signal (i.e., symbol valuesignal - ˆ 𝑉 [ 𝑘 ] ). To give an example, considering a binarysystem for the following information 𝑆 = { , , , , , , , , } , (31)the two signals will be ˆ 𝑉 [ 𝑘 ] = { , , , , } , ˆ 𝑙 [ 𝑘 ] = { , , , , } . By superimposing these two signals on two orthogonalmolecular signal properties, where 𝑁 𝑇 𝑥 [ 𝑘 ] is used for therepetition length signal, and 𝑚𝑚 𝑡 𝑦 𝑝𝑒 for the symbol valuesignal, the Tx is able to send a complex signal that carriesmany bits of information. As an example, in an HMS tech-nique that uses two types of symbols and considers at mostfour consecutive symbols with the same symbol value (i.e., 𝐻 𝑀𝑆 , ), the reception ( ˆ 𝑉 [ 𝑘 ] , ˆ 𝑙 [ 𝑘 ] ) is defined as ˆ 𝑉 [ 𝑘 ] = (cid:40) 𝑠𝑦𝑚 , 𝑁 𝑅𝑥𝑚𝑚 𝑎 [ 𝑘 ] > 𝑁 𝑅𝑥𝑚𝑚 𝑏 [ 𝑘 ] 𝑠𝑦𝑚 , 𝑁 𝑅𝑥𝑚𝑚 𝑏 [ 𝑘 ] ≥ 𝑁 𝑅𝑥𝑚𝑚 𝑎 [ 𝑘 ] , (32)and ˆ 𝑙 [ 𝑘 ] = , 𝑁 𝑅𝑥𝑚𝑚 𝑚𝑎𝑥 [ 𝑘 ] < 𝜆 , 𝜆 ≤ 𝑁 𝑅𝑥𝑚𝑚 𝑚𝑎𝑥 [ 𝑘 ] < 𝜆 , 𝜆 ≤ 𝑁 𝑅𝑥𝑚𝑚 𝑚𝑎𝑥 [ 𝑘 ] < 𝜆 , 𝜆 ≤ 𝑁 𝑅𝑥𝑚𝑚 𝑚𝑎𝑥 [ 𝑘 ] , (33)where 𝑚𝑚 𝑚𝑎𝑥 = (cid:40) 𝑚𝑚 𝑎 , ˆ 𝑆 [ 𝑘 ] = 𝑠𝑦𝑚 𝑚𝑚 𝑏 , ˆ 𝑆 [ 𝑘 ] = 𝑠𝑦𝑚 . (34)As with m-CCSK, 𝐻 𝑀𝑆 𝑙, 𝑚 offers a much higher achiev-able data rate over CSK or MoSK at a cost of Tx and Rxcomplexity. The authors show that the highest achievable datarate gain can be attained where 𝑚 = (i.e., each symbolrepresents 1-bit of information) and 𝑙 ∈ [ , ] . However, suchhigh 𝑙 values will necessitate a high number of thresholds,which will lead to high BER levels. The authors also proposedan alternative technique, HMS-D2, where 𝑁 𝑇 𝑥 [ 𝑘 ] representsthe symbol value signal and 𝑚𝑚 𝑡 𝑦 𝑝𝑒 represents the repetitionlength signal. This technique reduces the amount of thresholdthat is needed at the Rx, however, in this case the Tx will haveto use − different types of MMs which will increase theenergy requirements at the Tx to impractical levels.A different approach is to combine the type-based tech-niques with timing-based techniques. In [128], authors proposemolecular type permutation shift keying (MTPSK) which isin essence a combination of the CSK-SubTS and MoSK withML detection. Here the system considers 𝑀 different typeof molecules. Then, similar to the CSK-SubTS technique ina more generalized fashion, the 𝑡 𝑠 is divided into 𝑀 sub-timeslots. At the first sub-TS, the Tx can choose to send anyone of the 𝑀 molecule types. Then, in the second sub-TSs there will be 𝑀 − options for selecting which moleculetypes to use. Therefore, for a given 𝑀 value, there will be 𝑀 ! factorial permutations/symbols, and a symbol represents log 𝑀 ! bit(s) of information. As for the reception part, theyhave utilized an ideal ML detector as well as a practicalimplementation using a Viterbi-like algorithm. Using highsignal power as well as selecting 𝑀 = , the proposedtechnique outperforms CSK, MoSK, as well as PPM. Thisimprovement comes at a cost of high Tx and Rx complexitycompared to these simpler techniques. F. Discussion on Appropriate Applications
We have presented the comparison tables for each groupof modulation techniques among the group itself, e.g., theconcentration-based techniques are listed in Table I withoutconsidering the cross comparison. Now, we will discuss thecross comparison in a qualitative and hypothetical frameworkwith considering the challenges of MCvD system such as highISI, synchronization, physical limitations, etc.Concentration-based modulation techniques are, in general,simpler and may be more appropriate for less-capable de-vices/cells. When ISI mitigation capabilities are integratedto concentration-based modulations, the complexity of thetechnique increases directly. Therefore, less-capable devicesshould use simple concentration-based modulation techniqueswith long symbol duration to cope with ISI and they are ap-propriate for low-data rate applications or applications wherea predefined message (e.g., alarm or sensed phenomenon) issent only once.Type-based techniques require different types of moleculesfor different symbols. Complexity of these techniques ishigher than the concentration-based techniques. Hence, type-based techniques may not be the best choice for less-capabledevices. However, if the device is capable of complex op-erations such as differentiating molecule types at the Rxside or synthesizing and sending different molecule types atthe Tx side, then achieving ISI mitigation becomes easiercompared to concentration-based techniques. Coping with ISIenables higher data rates. Therefore, these techniques maybe promising for applications that are requiring higher datarates. Another advantage of using type-based techniques canbe differentiating the nodes by assigning different moleculetypes to different links. However, this technique does not scaleand may not be appropriate for crowded multi node topologies.If we focus on the Timing-based techniques, they require ac-curate synchronization in time domain to decode the symbolsaccurately. These kind of modulation techniques are not appro-priate for the application scenarios where the synchronizationis hurdled. If the synchronization issue is handled adequately,then the channel capacity of the timing-based modulationtechnique scales at least polylogarithmically [55]. Therefore,timing-based modulations have a great potential with a costof synchronization requirements.Spatial techniques are also promising for MCvD systems,however compared to the previous types they are less maturetechniques and require further refinement. These techniqueshave potential to push the current data rate limits to even
URAN et al.: SURVEY ON MODULATION TECHNIQUES IN MCVD 17 P e B-CSKB-MoSKB-MTSKB-CSK-CPA (a) Short 𝑡 𝑠 ( 𝑡 peak ) -2 -1 P e B-CSKB-MoSKB-MTSKB-CSK-CPA (b) Moderate 𝑡 𝑠 ( . 𝑡 peak ) -6 -4 -2 P e B-CSKB-MoSKB-MTSKB-CSK-CPA (c) Long 𝑡 𝑠 ( 𝑡 peak )Fig. 8. BER curves of Slow and Medium modulation techniques (i.e., B-CSK, B-MoSK, B-MTSK, and B-CSK-CPA). The 𝑡 𝑠 values have been selectedaccording to these categories. higher values. Due to having multiple antennas, such MCvDsystems with spatial techniques can also be very appropriatefor localization applications. However, they require advancedTx and Rx structures with multiple emission and receptionsites. Moreover, they may not be appropriate for crowdedtopologies due to multi-user interference and the requiredsensitivity for estimating the transmitting/emitter antenna. G. Performance Comparison
As we have now described the working principles of themodulation techniques in the literature, in this section weanalytically evaluate the performances of key modulation tech-niques (i.e., CSK, MoSK, MTSK, CSK-CPA, MCSK, and Pre-Eq. CSK) in terms of BER for different average transmittedpower values. Regarding the MoSK technique, we utilize thescheme with the majority based detection variant. In order tohave a fair comparison among these techniques, we normalizedthe 𝑁 𝑇 𝑥 [ 𝑘 ] values for each symbol value so that the averagetransmitted power per symbol is the same for all the evaluatedtechniques. For example to compare the performances of B-CSK and B-MoSK techniques, where the average transmittedpower is , we set the 𝑁 𝑇 𝑥 [ 𝑘 ] values as 𝑁 𝑇 𝑥 [ 𝑘 ] B-CSK = (cid:40) 𝑠𝑦𝑚 , 𝑠𝑦𝑚 , , (35) 𝑁 𝑇 𝑥 [ 𝑘 ] B-MoSK = (cid:40) 𝑠𝑦𝑚 , 𝑠𝑦𝑚 , . (36)We use a 3D environment composed of a single point Txdevice and a single fully absorbing spherical Rx device andwithout a drift component in the environment. Both the Txand the Rx devices have a radius of µ m and the shortestdistance between these two devices is also µ m . The diffusioncoefficient, 𝐷 , has been selected as . µ m / s considering theviscosity of blood plasma at ° C [129]. As for the MM, weuse insulin molecules, whose Stokes radius is .
68 nm [129].The transmitted MC signal has a pulse waveform and weuse the CIR calculation formula for 3D environments with apoint source and fully absorbing receiver given in [38]. TheBER performance has been evaluated where the effect of past -symbols is taken into consideration (i.e., the ISI window length is ). We use a binary scenario, where each symbolrepresents a 1-bit information. As for the 𝑡 𝑠 value, we use threedifferent values as short 𝑡 𝑠 , moderate 𝑡 𝑠 , and long 𝑡 𝑠 . Each oneof these values is given as multiples of the 𝑡 peak value, whichdenotes the time of peak point of CIR function (i.e., ℎ ( 𝑡 ) ) andit represents the time when the maximum number of MMs isreceived by the Rx device. This 𝑡 peak value is also calculated bythe aforementioned CIR formulation given the environmentalparameters’ values.The BER results of the chosen modulation techniques aregiven in Fig. 8 and Fig. 9. In order to facilitate clear readabilityof the results, we categorize the modulation techniques inthree categories as slow (i.e., CSK and MoSK), medium (i.e.,MTSK, CSK-CPA) and fast (i.e., MCSK and PreEq CSK)techniques based on the 𝑡 𝑠 values under which they exhibit abetter BER performance. Fig. 8 depicts the result of the slow(i.e., CSK and MoSK) and medium (i.e., MTSK, CSK-CPA)techniques, while Fig. 9 shows the results of the medium andfast (i.e., MCSK and PreEq CSK) techniques.As seen in all three subfigures of Fig. 8, B-CSK and MoSKhave similar performances after normalizing their averagetransmitted power values regardless of the 𝑡 𝑠 value. Also,in all three cases, both MTSK and CSK-CPA outperformsthese fundamental techniques. As seen in Fig. 8(a), CSK-CPAoutperforms MTSK in the short 𝑡 𝑠 scenario. However, as seenin Fig. 8(b) and Fig. 8(c), when the 𝑡 𝑠 value increases, CSK-CPA loses this performance edge and its performance becomessimilar to MTSK.For the second group of techniques, due to their fast naturewe use shorter 𝑡 𝑠 values in all three cases. As seen in Fig. 9,both MCSK and Pre-Eq CSK outperforms the best of the slowtechniques in all 𝑡 𝑠 values. Among the two fast techniques,Pre-Eq CSK gives a lower BER value considering a short 𝑡 𝑠 value and where the average transmission power is greater than (Fig 9(a)). As 𝑡 𝑠 increases though, MCSK outperformsthe Pre-Eq CSK by a great margin both in the moderate 𝑡 𝑠 and long 𝑡 𝑠 scenarios. This is due to the fact that Pre-Eq CSKis in particular designed to operate under small 𝑡 𝑠 values andthe advantage of the equalization goes away as the 𝑡 𝑠 valueincreases. One important take away from these results is thefact that MCSK not only has a much lower BER value thanthe other techniques considered, but it also has a moderate URAN et al.: SURVEY ON MODULATION TECHNIQUES IN MCVD 18 -2 -1 P e B-CSK-CPAB-MCSKPre-Eq CSK (a) Short 𝑡 𝑠 ( . 𝑡 peak ) -6 -5 -4 -3 -2 -1 P e B-CSK-CPAB-MCSKPre-Eq CSK (b) Moderate 𝑡 𝑠 ( 𝑡 peak ) -10 -8 -6 -4 -2 P e B-CSK-CPAB-MCSKPre-Eq CSK (c) Long 𝑡 𝑠 ( . 𝑡 peak )Fig. 9. BER curves of Medium and Fast modulation techniques (i.e., B-CSK-CPA, B-MCSK, B-Pre-Eq. CSK). The 𝑡 𝑠 values have been selected accordingto these categories. computational complexity. One potential drawback of MCSKis the synchronization requirement of the Rx, which might bea hard limitation in implementation.Considering these results one can observe that the bestmodulation technique, in terms of BER, depends heavily onthe design requirements of the target MC system, in particularthe symbol duration, 𝑡 𝑠 , and the desired complexity. If asimpler system is desired clearly the CSK-CPA is the bestchoice since it is a concentration-based technique workingwith a single MM type. But it would necessitate a longer 𝑡 𝑠 value to achieve a low BER. On the other hand, if it ispossible to utilize multiple MMs, the question becomes whatis the acceptable 𝑡 𝑠 value which will have a serious impactover the system throughput. In case a short 𝑡 𝑠 is desired, Pre-Eq CSK yields the lowest BER while if a longer 𝑡 𝑠 will notcause too many problems MCSK clearly outperforms all othertechniques. Though, as stated before MCSK requires strictersynchronization which may further complicate its adaptabilityin a real-life system.VI. O PEN I SSUES & F
UTURE D IRECTIONS
This article has reviewed the modulation techniques thathave been developed for MC over the last decade. Sinceinformation in MC systems is carried by molecules, manynew challenges in modulation design arise, which have beenaddressed in the recent literature. However, the field of MC isstill in its infancy, with many open challenges remaining. Inthis section we briefly discuss some of these challenges alongwith relevant future research directions to address them.Due to its nature, the received molecular signal insidea diffusion channel is inherently a positive valued signal.Therefore, methods that utilize both positive and negativecomponents of a signal cannot be directly applied to themolecular communication. Similar to the molecular signal,the optical signal also exhibits such a behavior. Consequently,modulation techniques inspired by optical systems are envi-sioned to be more suitable for MCvD systems. Some of theexisting solutions can be applied to the problems that areencountered in MC domain but have similarity to alreadysolved problems in optical systems.
Synchronization is one of the critical hurdles for modula-tion techniques in MC systems. As previously stated in Sec-tion II, most of the works that propose modulation techniquesfor MC systems assume perfect synchronization betweencommunication nodes, the Tx and the Rx. There are severalstudies proposing different synchronization techniques forMC systems by quorum sensing [130], blind synchronizationwith channel delay [131], and signal peak observation [132]–[134]. However, these methods are either complex for thenanomachines or they heavily rely on CSI - please note thatacquiring CSI is a challenging task for time-varying MCchannels. Moreover, the integration of these synchronizationmethods with the proposed modulation techniques is not in-vestigated adequately. One important research direction is thenthe integration of synchronization methods with the existingmodulation techniques and the investigation of asynchronousmodulation techniques to increase the implementability of MCsystems.
Physical properties of the transmitter and the re-ceiver nodes in MC communications are critical for designingeffective modulation techniques. Physical properties of thetransmitter node determines the limitations on the emittedmolecules. For example, if the transmitter node does not haveenough storage, then the emission capabilities are limitedby the storage capacity. Similarly, physical properties of thereceiver nodes affect the CIR and CFRR, which also affectsthe modulation performance. In general, these physical prop-erties and limitations are not considered adequately. For morerealistic studies, these limitations should be considered in aholistic approach.
Properties of the MC channel is also critical for designingeffective modulation techniques. MC channels possess slowpropagation, high ISI, and channel memory, hence the channelsignificantly affects the modulation performance. Moreover,lack of a common SNR definition for the molecular com-munication systems is one of the critical issues for stan-dardized comparison for modulation techniques. Much ofthe past MC research focused on ISI elimination and wediscussed the modulation techniques that reduce ISI. Due toimplementation challenges at small scales, new metrics can be
URAN et al.: SURVEY ON MODULATION TECHNIQUES IN MCVD 19 proposed to compare modulation techniques while consideringthe complexity in a quantitative manner. For future directions,the tradeoff between ISI reduction and complexity can beanalyzed in a detailed and quantitative manner. In addition,new modulation techniques can be introduced that are consid-ering implementation issues (i.e., computational limitation ofdevices) more than the communication performance.
Channel coding and precoding techniques should also beconsidered to reduce the deteriorating effect of interferencemolecules. Precoding and channel coding techniques havepotential to improve the reliability of the transmitted data.Moreover, long transmission duration suggest the use of errorcorrecting codes instead of costly re-transmission strategies asin the case of satellite communications. The compatibility ofthe modulation and coding techniques is an open issue andshould be investigated thoroughly.All these MC specific challenges along with others such asmultiuser techniques, networking, and potentially the physicallayer security issues should be considered for future MCsystems. In fact, the IEEE has established a standardizationgroup (IEEE 1906.1) to address the various challenges of MCand develop standards around these solutions. Establishmentof IEEE 1906.1 has the potential to push the technologytowards commercialization if these challenges can be solvedand adopted into a widely-used standard.VII. C
ONCLUSION
A key design component of the MC system is its modulationand associated demodulation mechanism. In the last decade orso a variety of modulation techniques have been proposed forthe molecular communication in the literature. In this survey,we have summarized various modulation techniques and theirtradeoffs. In order to analyze, categorize, and elaborate uponthese techniques we provide a framework that can also beutilized for future modulation techniques. Moreover, we pro-vide a numerical performance comparison between the mostprevalent of these modulation techniques and show that allof the techniques entail tradeoffs in different environments,with no single technique emerging as the best choice in allenvironments. Finally, we provide a brief elaboration on theopen design issues about modulation techniques for MC andthe future research directions to address them.R
EFERENCES[1] I. F. Akyildiz, F. Brunetti, and C. Bl´azquez, “Nanonetworks: A newcommunication paradigm,”
Computer Networks , vol. 52, no. 12, pp.2260–2279, Aug. 2008.[2] T. Nakano, M. J. Moore, F. Wei, A. V. Vasilakos, and J. Shuai, “Molec-ular communication and networking: Opportunities and challenges,”
IEEE Transactions on NanoBioscience , vol. 11, no. 2, pp. 135–148,June 2012.[3] T. Nakano, A. W. Eckford, and T. Haraguchi,
Molecular Communica-tion . Cambridge University Press, 2013.[4] N. Farsad, H. B. Yilmaz, A. Eckford, C. Chae, and W. Guo, “A compre-hensive survey of recent advancements in molecular communication,”
IEEE Communications Surveys Tutorials , vol. 18, no. 3, pp. 1887–1919, 2016.[5] L. P. Gin´e and I. F. Akyildiz, “Molecular communication options forlong range nanonetworks,”
Computer Networks , vol. 53, no. 16, pp.2753 – 2766, 2009. [6] Song Qiu, W. Guo, S. Wang, N. Farsad, and A. Eckford, “A molecularcommunication link for monitoring in confined environments,” in
Proc.of IEEE Int. Conference on Communications Workshops (ICC) , 2014,pp. 718–723.[7] M. S. Kuran, H. B. Yilmaz, T. Tugcu, and B. Ozerman, “Energy modelfor communication via diffusion in nanonetworks,”
Nano Communica-tion Networks , vol. 1, no. 2, pp. 86 – 95, 2010.[8] C. Rose, I. S. Mian, and M. Ozmen, “Capacity bounds on point-to-point communication using molecules,”
Proceedings of the IEEE , vol.107, no. 7, pp. 1342–1355, 2019.[9] J. C. Anderson, E. J. Clarke, A. P. Arkin, and C. A. Voigt, “Environ-mentally controlled invasion of cancer cells by engineered bacteria,”
Journal of molecular biology , vol. 355, no. 4, pp. 619–627, 2006.[10] T. Danino, A. Prindle, G. A. Kwong, M. Skalak, H. Li, K. Allen,J. Hasty, and S. N. Bhatia, “Programmable probiotics for detection ofcancer in urine,”
Science translational medicine , vol. 7, no. 289, 2015.[11] S. Slomovic, K. Pardee, and J. J. Collins, “Synthetic biology devices forin vitro and in vivo diagnostics,”
Proceedings of the National Academyof Sciences , vol. 112, no. 47, pp. 14 429–14 435, 2015.[12] W. Zhu, J. Li, Y. J. Leong, I. Rozen, X. Qu, R. Dong, Z. Wu, W. Gao,P. H. Chung, J. Wang et al. , “3d-printed artificial microfish,”
Advancedmaterials , vol. 27, no. 30, pp. 4411–4417, 2015.[13] D. Vilela, J. Parmar, Y. Zeng, Y. Zhao, and S. Sanchez, “Graphene-based microbots for toxic heavy metal removal and recovery fromwater,”
Nano Letters , vol. 16, no. 4, pp. 2860–2866, 2016.[14] T. Suda, M. Moore, T. Nakano, R. Egashira, and A. Enomoto,“Exploratory research on molecular communication between nanoma-chines,” in
Proc. of Conference on Genetic and Evolutionary Compu-tation (GECCO 2005), ACM , 2005.[15] M. Gregori and I. F. Akyildiz, “A new nanonetwork architectureusing flagellated bacteria and catalytic nanomotors,”
IEEE Journal onSelected Areas in Communications , vol. 28, no. 4, pp. 612–619, May2010.[16] M. Moore, A. Enomoto, T. Nakano, R. Egashira, T. Suda, A. Kayasuga,H. Kojima, H. Sakakibara, and K. Oiwa, “A design of a molecular com-munication system for nanomachines using molecular motors,” in
Proc.of IEEE Int. Conference on Pervasive Computing and CommunicationsWorkshops (PERCOMW’06) , March 2006, pp. 6 pp.–559.[17] K. Darchini and A. S. Alfa, “Molecular communication via mi-crotubules and physical contact in nanonetworks: A survey,”
NanoCommunication Networks , vol. 4, no. 2, pp. 73 – 85, 2013.[18] Y. Chahibi, I. F. Akyildiz, and I. Balasingham, “Propagation modelingand analysis of molecular motors in molecular communication,”
IEEETransactions on NanoBioscience , vol. 15, no. 8, pp. 917–927, Dec2016.[19] T. Nakano, T. Suda, T. Koujin, T. Haraguchi, and Y. Hiraoka, “Molec-ular communication through gap junction channels: System design, ex-periments and modeling,” in
Proc. of Bio-Inspired Models of Network,Information and Computing Systems , Dec 2007, pp. 139–146.[20] ——,
Molecular Communication through Gap Junction Channels .Berlin, Heidelberg: Springer Berlin Heidelberg, 2008, pp. 81–99.[21] M. S. Kuran, T. Tugcu, and B. O. Edis, “Calcium signaling: overviewand research directions of a molecular communication paradigm,”
IEEEWireless Communications , vol. 19, no. 5, pp. 20–27, Oct 2012.[22] B. D. Unluturk and I. F. Akyildiz, “An end-to-end model of plantpheromone channel for long range molecular communication,”
IEEETransactions on NanoBioscience , vol. 16, no. 1, pp. 11–20, Jan 2017.[23] V. Jamali, A. Ahmadzadeh, W. Wicke, A. Noel, and R. Schober,“Channel modeling for diffusive molecular communication—a tutorialreview,”
Proceedings of the IEEE , vol. 107, no. 7, pp. 1256–1301,2019.[24] M. Kuscu, E. Dinc, B. A. Bilgin, H. Ramezani, and O. B. Akan,“Transmitter and receiver architectures for molecular communications:A survey on physical design with modulation, coding, and detectiontechniques,”
Proceedings of the IEEE , vol. 107, no. 7, pp. 1302–1341,2019.[25] A. Akkaya, H. B. Yilmaz, C.-B. Chae, and T. Tugcu, “Effect of receptordensity and size on signal reception in molecular communication viadiffusion with an absorbing receiver,”
IEEE Communications Letters ,vol. 19, no. 2, pp. 155–158, 2015.[26] Y.-P. Hsieh and P.-C. Yeh., “Mathematical foundations for infor-mation theory in diffusion-based molecular communications,” arXivpreprint:1311.4431 , 2013.[27] K. R. Harris,
Diffusion in liquids: A theoretical and experimental study .Butterworths, 1984.
URAN et al.: SURVEY ON MODULATION TECHNIQUES IN MCVD 20 [28] M. Cappelezzo, C. A. Capellari, S. H. Pezzin, and L. A. F. Coelho,“Stokes-einstein relation for pure simple fluids,”
The Journal of Chem-ical Physics , vol. 126, no. 22, p. 224516, 2007.[29] L. Costigliola, D. M. Heyes, T. B. Schrøder, and J. C. Dyre, “Revisitingthe stokes-einstein relation without a hydrodynamic diameter,”
TheJournal of Chemical Physics , vol. 150, no. 2, p. 021101, 2019.[30] L. Back, J. Radbill, and Y. Cho, “Measurement and prediction of flowthrough a replica segment of a mildly atherosclerotic coronary arteryof man,”
Journal of Biomechanics , vol. 19, no. 1, pp. 1–17, 1986.[31] F. Dinc¸, B. C. Akdeniz, A. E. Pusane, and T. Tugcu, “A general ana-lytical approximation to impulse response of 3-d microfluidic channelsin molecular communication,”
IEEE Transactions on Nanobioscience ,2019.[32] N. Farsad, N. Kim, A. W. Eckford, and C. Chae, “Channel and noisemodels for nonlinear molecular communication systems,”
IEEE Journalon Selected Areas in Communications , vol. 32, no. 12, pp. 2392–2401,Dec 2014.[33] A. Ahmadzadeh, V. Jamali, and R. Schober, “Statistical analysis oftime-variant channels in diffusive mobile molecular communications,”in
Proc. of IEEE Global Communications Conference (GLOBECOM) ,2017, pp. 1–7.[34] W. Wicke, T. Schwering, A. Ahmadzadeh, V. Jamali, A. Noel, andR. Schober, “Modeling duct flow for molecular communication,” in
Proc. of IEEE Global Communications Conference (GLOBECOM) ,2018, pp. 206–212.[35] V. Jamali, A. Ahmadzadeh, N. Farsad, and R. Schober, “Scw codesfor optimal csi-free detection in diffusive molecular communications,”in
Proc. of IEEE Int. Symposium on Information Theory (ISIT) , 2017,pp. 3190–3194.[36] S. Wang, W. Guo, and M. D. McDonnell, “Transmit pulse shaping formolecular communications,” in
Proc. of IEEE Conference on ComputerCommunications Workshops (INFOCOM WKSHPS) , Apr. 2014, pp.209–210.[37] C. Luo, X. Wu, L. Lin, C. Wang, and F. Liu, “Non-coherent signaldetection technique for mobile molecular communication at high datarates,” in
Proc. of IEEE Global Communications Conference (GLOBE-COM) , 2019, pp. 1–6.[38] H. B. Yilmaz, A. C. Heren, T. Tugcu, and C. B. Chae, “Three-dimensional channel characteristics for molecular communications withan absorbing receiver,”
IEEE Communications Letters , vol. 18, no. 6,pp. 929–932, June 2014.[39] A. W. Eckford, “Molecular communication: Physically realistic modelsand achievable information rates,” arXiv preprint:0812.1554 , 2008.[40] K. V. Srinivas, A. W. Eckford, and R. S. Adve, “Molecular commu-nication in fluid media: The additive inverse gaussian noise channel,”
IEEE Transactions on Information Theory , vol. 58, no. 7, pp. 4678–4692, July 2012.[41] A. Noel, D. Makrakis, and A. Hafid, “Channel impulse responsesin diffusive molecular communication with spherical transmitters,” in
Proc. of Biennial Symp. Commun. , 2016.[42] T. Nakano, Y. Okaie, and J. Liu, “Channel model and capacityanalysis of molecular communication with brownian motion,”
IEEECommunications Letters , vol. 16, no. 6, pp. 797–800, June 2012.[43] A. C. Heren, H. B. Yilmaz, C.-B. Chae, and T. Tugcu, “Effect ofdegradation in molecular communication: Impairment or enhance-ment?”
IEEE Transactions on Molecular, Biological and Multi-ScaleCommunications , vol. 1, no. 2, pp. 217–229, 2015.[44] M. Turan, B. C. Akdeniz, M. S. Kuran, H. B. Yilmaz, I. Demirkol,A. E. Pusane, and T. Tugcu, “Transmitter localization in vessel-like diffusive channels using ring-shaped molecular receivers,”
IEEECommunications Letters , vol. 22, no. 12, pp. 2511–2514, Dec 2018.[45] G. Genc, Y. E. Kara, T. Tugcu, and A. E. Pusane, “Reception modelingof sphere-to-sphere molecular communication via diffusion,”
NanoCommunication Networks , vol. 16, pp. 69 – 80, 2018.[46] W. Wicke, A. Ahmadzadeh, V. Jamali, R. Schober, H. Unterweger, andC. Alexiou, “Molecular communication using magnetic nanoparticles,”in
Proc. of IEEE Wireless Communications and Networking Conference(WCNC) , 2018, pp. 1–6.[47] G. Genc, Y. E. Kara, H. B. Yilmaz, and T. Tugcu, “ISI-aware modelingand achievable rate analysis of the diffusion channel,”
IEEE Commu-nications Letters , vol. 20, no. 9, pp. 1729–1732, Sept 2016.[48] A. Gohari, M. Mirmohseni, and M. Nasiri-Kenari, “Information theoryof molecular communication: Directions and challenges,”
IEEE Trans-actions on Molecular, Biological and Multi-Scale Communications ,vol. 2, no. 2, pp. 120–142, 2016. [49] A. Einolghozati, M. Sardari, A. Beirami, and F. Fekri, “Capacity ofdiscrete molecular diffusion channels,” in
Proc. of IEEE Int. Symposiumon Information Theory , July 2011, pp. 723–727.[50] H. Li, S. Moser, and D. Guo, “Capacity of the memoryless additiveinverse gaussian noise channel,”
IEEE Journal on Selected Areas inCommunications , vol. 32, no. 12, pp. 2315–2329, Dec 2014.[51] N. Farsad, Y. Murin, A. Eckford, and A. Goldsmith, “On the capacityof diffusion-based molecular timing channels,” in
Proc. of IEEE Int.Symposium on Information Theory (ISIT) , 2016, pp. 1023–1027.[52] C. Rose and I. S. Mian, “Inscribed matter communication: Part i,”
IEEE Transactions on Molecular, Biological and Multi-Scale Commu-nications , vol. 2, no. 2, pp. 209–227, Dec 2016.[53] G. Aminian, M. Mirmohseni, M. N. Kenari, and F. Fekri, “On thecapacity of level and type modulations in molecular communicationwith ligand receptors,” in
Proc. of IEEE Int. Symposium on InformationTheory (ISIT) , June 2015, pp. 1951–1955.[54] N. Farsad, C. Rose, M. M´edard, and A. Goldsmith, “Capacity ofmolecular channels with imperfect particle-intensity modulation anddetection,” in
Proc. of IEEE Int. Symposium on Information Theory(ISIT) , 2017, pp. 2468–2472.[55] N. Farsad, Y. Murin, A. Eckford, and A. Goldsmith, “Capacity limitsof diffusion-based molecular timing channels with finite particle life-time,”
IEEE Transactions on Molecular, Biological and Multi-ScaleCommunications , 2019.[56] W. Guo, T. Asyhari, N. Farsad, H. B. Yilmaz, B. Li, A. Eckford,and C. Chae, “Molecular communications: channel model and physicallayer techniques,”
IEEE Wireless Communications , vol. 23, no. 4, pp.120–127, 2016.[57] M. S. Kuran, H. B. Yilmaz, T. Tugcu, and I. F. Akyildiz, “Interferenceeffects on modulation techniques in diffusion based nanonetworks,”
Nano Communication Networks , vol. 3, no. 1, pp. 65 – 73, 2012.[58] B.-H. Koo, C. Lee, H. B. Yilmaz, N. Farsad, A. Eckford, and C.-B.Chae, “Molecular MIMO: From theory to prototype,”
IEEE Journal onSelected Areas in Communications , vol. 34, no. 3, pp. 600–614, 2016.[59] M. C. Gursoy, E. Basar, A. E. Pusane, and T. Tugcu, “Index modulationfor molecular communication via diffusion systems,”
IEEE Transac-tions on Communications , vol. 67, no. 5, pp. 3337–3350, 2019.[60] H. D. Pfister, J. B. Soriaga, and P. H. Siegel, “On the achievableinformation rates of finite state ISI channels,” in
Proc. of IEEE GlobalCommunications Conference (GLOBECOM) , vol. 5, 2001, pp. 2992–2996.[61] B. Atakan, “Optimal transmission probability in binary molecularcommunication,”
IEEE Communications Letters , vol. 17, no. 6, pp.1152–1155, June 2013.[62] A. Singhal, R. K. Mallik, and B. Lall, “Performance analysis ofamplitude modulation schemes for diffusion-based molecular commu-nication,”
IEEE Transactions on Wireless Communications , vol. 14,no. 10, pp. 5681–5691, Oct 2015.[63] Q. Liu and K. Yang, “Channel capacity analysis of a diffusion-basedmolecular communication system with ligand receptors,”
InternationalJournal of Communication Systems , vol. 28, no. 8, pp. 1508–1520,2015.[64] G. Aminian, H. Arjmandi, A. Gohari, M. Nasiri-Kenari, and U. Mitra,“Capacity of diffusion-based molecular communication networks overlti-poisson channels,”
IEEE Transactions on Molecular, Biological andMulti-Scale Communications , vol. 1, no. 2, pp. 188–201, June 2015.[65] M. U. Mahfuz, D. Makrakis, and H. T. Mouftah, “On the char-acteristics of concentration-encoded multi-level amplitude modulatedunicast molecular communication,” in
Proc. of Canadian Conferenceon Electrical and Computer Engineering (CCECE) , May 2011, pp.312–316.[66] H. B. Yilmaz, Y.-J. Cho, W. Guo, and C.-B. Chae, “Interferencereduction via enzyme deployment for molecular communication,”
IETElectronics Letters , vol. 52, no. 13, pp. 1094–1096, 2016.[67] Y. J. Cho, H. B. Yilmaz, W. Guo, and C.-B. Chae, “Effective inter-symbol interference mitigation with a limited amount of enzymes inmolecular communications,”
Transactions on Emerging Telecommuni-cations Technologies , vol. 28, no. 7, 2017.[68] M. C. Gursoy, H. B. Yilmaz, A. E. Pusane, and T. Tugcu, “Simulationstudy and analysis of diffusive molecular communications with anapertured plane,”
IEEE Transactions on NanoBioscience , vol. 19, no. 3,pp. 468–476, 2020.[69] B. Tepekule, A. E. Pusane, M. S. Kuran, and T. Tugcu, “A novelpre-equalization method for molecular communication via diffusionin nanonetworks,”
IEEE Communications Letters , vol. 19, no. 8, pp.1311–1314, Aug 2015.
URAN et al.: SURVEY ON MODULATION TECHNIQUES IN MCVD 21 [70] B. C. Akdeniz, N. A. Turgut, H. B. Yilmaz, C.-B. Chae, T. Tugcu,and A. E. Pusane, “Molecular signal modeling of a partially countingabsorbing spherical receiver,”
IEEE Transactions on Communications ,vol. 66, no. 12, pp. 6237–6246, 2018.[71] B. C. Akdeniz, A. E. Pusane, and T. Tugcu, “Optimal reception delayin diffusion-based molecular communication,”
IEEE CommunicationsLetters , vol. 22, no. 1, pp. 57–60, 2018.[72] H. B. Yilmaz and C.-B. Chae, “Simulation study of molecular com-munication systems with an absorbing receiver: Modulation and ISImitigation techniques,”
Elsevier Simulation Modelling Practice andTheory , vol. 49, pp. 136–150, 2014.[73] M. Turan, M. S. Kuran, H. B. Yilmaz, C. Chae, and T. Tugcu, “Mol-eye: A new metric for the performance evaluation of a molecularsignal,” in
Proc. of IEEE Wireless Communications and NetworkingConference (WCNC) , April 2018, pp. 1–6.[74] A. Grami, “Chapter 6 - baseband digital transmission,” in
Introductionto Digital Communications
Molecular biology of the cell . W. W. Norton andCompany, 2014.[76] M. U. Mahfuz, D. Makrakis, and H. T. Mouftah, “On the character-ization of binary concentration-encoded molecular communication innanonetworks,”
Nano Communication Networks , vol. 1, no. 4, pp. 289– 300, 2010.[77] M. S. Kuran, H. B. Yilmaz, T. Tugcu, and I. F. Akyildiz, “Modulationtechniques for communication via diffusion in nanonetworks,” in
Proc.of IEEE Int. Conference on Communications (ICC) , June 2011, pp. 1–5.[78] I. Llatser, A. Cabellos-Aparicio, M. Pierobon, and E. Alarcon, “Detec-tion techniques for diffusion-based molecular communication,”
IEEEJournal on Selected Areas in Communications , vol. 31, no. 12, pp.726–734, December 2013.[79] M. U. Mahfuz, D. Makrakis, and H. T. Mouftah, “A general-ized strength-based signal detection model for concentration-encodedmolecular communication,” in
Proc. of Int. Conference on Body AreaNetworks , ser. BodyNets ’13, 2013, pp. 461–467.[80] ——, “A comprehensive analysis of strength-based optimum signaldetection in concentration-encoded molecular communication withspike transmission,”
IEEE Transactions on NanoBioscience , vol. 14,no. 1, pp. 67–83, Jan 2015.[81] ——, “Concentration-encoded subdiffusive molecular communication:Theory, channel characteristics, and optimum signal detection,”
IEEETransactions on NanoBioscience , vol. 15, no. 6, pp. 533–548, Sept2016.[82] A. Singhal, R. K. Mallik, and B. Lall, “Molecular communicationwith brownian motion and a positive drift: performance analysis ofamplitude modulation schemes,”
IET Communications , vol. 8, no. 14,pp. 2413–2422, Sept 2014.[83] B. Tepekule, A. E. Pusane, H. B. Yilmaz, and T. Tugcu, “Energyefficient ISI mitigation for communication via diffusion,” in
Proc. ofIEEE Int. Black Sea Conference on Communications and Networking(BlackSeaCom) , May 2014, pp. 33–37.[84] M. Movahednasab, M. Soleimanifar, A. Gohari, M. N. Kenari, andU. Mitra, “Adaptive molecule transmission rate for diffusion basedmolecular communication,” in
Proc. of IEEE Int. Conference onCommunications (ICC) , June 2015, pp. 1066–1071.[85] M. Damrath and P. A. Hoeher, “Low-complexity adaptive thresh-old detection for molecular communication,”
IEEE Transactions onNanoBioscience , vol. 15, no. 3, pp. 200–208, April 2016.[86] D. Kilinc and O. B. Akan, “Receiver design for molecular communi-cation,”
IEEE Journal on Selected Areas in Communications , vol. 31,no. 12, pp. 705–714, December 2013.[87] B. Li, M. Sun, S. Wang, W. Guo, and C. Zhao, “Low-complexity non-coherent signal detection for nanoscale molecular communications,”
IEEE Transactions on NanoBioscience , vol. 15, no. 1, pp. 3–10, Jan2016.[88] ——, “Local convexity inspired low-complexity noncoherent signaldetector for nanoscale molecular communications,”
IEEE Transactionson Communications , vol. 64, no. 5, pp. 2079–2091, May 2016.[89] V. Jamali, N. Farsad, R. Schober, and A. Goldsmith, “Non-coherentmultiple-symbol detection for diffusive molecular communications,” in
Proc. of ACM Int. Conference on Nanoscale Computing and Commu-nication , ser. NANOCOM’16, 2016, pp. 7:1–7:7. [90] B. C. Akdeniz, M. Egan, and B. Q. Tang, “Equilibrium signaling:Molecular communication robust to geometry uncertainties,”
IEEETransactions on Communications , pp. 1–1, 2020.[91] W. A. Lin, Y. C. Lee, P. C. Yeh, and C. h. Lee, “Signal detection andISI cancellation for quantity-based amplitude modulation in diffusion-based molecular communications,” in
Proc. of IEEE Global Commu-nications Conference (GLOBECOM) , Dec 2012, pp. 4362–4367.[92] A. Aijaz and A. H. Aghvami, “Error performance of diffusion-basedmolecular communication using pulse-based modulation,”
IEEE Trans-actions on NanoBioscience , vol. 14, no. 1, pp. 146–151, Jan 2015.[93] H. Zhai, Q. Liu, A. V. Vasilakos, and K. Yang, “Anti-isi demodulationscheme and its experiment-based evaluation for diffusion-based molec-ular communication,”
IEEE Transactions on Nanobioscience , vol. 17,no. 2, pp. 126–133, 2018.[94] S. Galm´es and B. Atakan, “Performance analysis of diffusion-basedmolecular communications with memory,”
IEEE Transactions on Com-munications , vol. 64, no. 9, pp. 3786–3793, Sept 2016.[95] N. R. Kim and C. B. Chae, “Novel modulation techniques usingisomers as messenger molecules for molecular communication viadiffusion,” in
Proc. of IEEE Int. Conference on Communications (ICC) ,June 2012, pp. 6146–6150.[96] N. R. Kim, A. W. Eckford, and C. B. Chae, “Symbol interval opti-mization for molecular communication with drift,”
IEEE Transactionson NanoBioscience , vol. 13, no. 3, pp. 223–229, Sept 2014.[97] H. ShahMohammadian, G. G. Messier, and S. Magierowski, “Opti-mum receiver for molecule shift keying modulation in diffusion-basedmolecular communication channels,”
Nano Communication Networks ,vol. 3, no. 3, pp. 183 – 195, 2012.[98] N. R. Kim and C. B. Chae, “Novel modulation techniques usingisomers as messenger molecules for nano communication networksvia diffusion,”
IEEE Journal on Selected Areas in Communications ,vol. 31, no. 12, pp. 847–856, 2013.[99] H. Arjmandi, A. Gohari, M. N. Kenari, and F. Bateni, “Diffusion-based nanonetworking: A new modulation technique and performanceanalysis,”
IEEE Communications Letters , vol. 17, no. 4, pp. 645–648,April 2013.[100] S. Pudasaini, S. Shin, and K. S. Kwak, “Robust modulation techniquefor diffusion-based molecular communication in nanonetworks,”
CoRR ,vol. abs/1401.3938, 2014.[101] X. Chen, Y. Huang, L. L. Yang, and M. Wen, “Generalized molecular-shift keying (gmosk): Principles and performance analysis,”
IEEETransactions on Molecular, Biological and Multi-Scale Communica-tions , vol. 6, no. 3, pp. 168–183, 2020.[102] A. W. Eckford, “Nanoscale communication with brownian motion,”in
Proc. of Conference on Information Sciences and Systems (CISS) ,Baltimore, MD, 2007, pp. 160–165.[103] Y. Chahibi, M. Pierobon, S. O. Song, and I. F. Akyildiz, “A molecularcommunication system model for particulate drug delivery systems,”
IEEE Transactions on Biomedical Engineering , vol. 60, no. 12, pp.3468–3483, 2013.[104] C. Rose and I. S. Mian, “Inscribed matter communication: Part ii,”
IEEE Transactions on Molecular, Biological and Multi-Scale Commu-nications , vol. 2, no. 2, pp. 228–239, Dec 2016.[105] Y. Murin, N. Farsad, M. Chowdhury, and A. Goldsmith, “Time-slottedtransmission over molecular timing channels,”
Nano CommunicationNetworks , vol. 12, pp. 12–24, 2017.[106] N. Garralda, I. Llatser, A. Cabellos-Aparicio, E. Alarcon, and M. Pier-obon, “Diffusion-based physical channel identification in molecularnanonetworks,”
Nano Communication Networks , vol. 2, no. 4, pp. 196–204, 2011.[107] B. C. Akdeniz, A. E. Pusane, and T. Tugcu, “Position-based modula-tion in molecular communications,”
Nano Communication Networks ,vol. 16, pp. 60–68, 2018.[108] A. Zare and A. Jamshidi, “Receiver design and performance analysisfor pulse position modulation technique in diffusion-based molecularcommunication,”
Nano Communication Networks , vol. 21, p. 100256,2019.[109] Y. Murin, N. Farsad, M. Chowdhury, and A. Goldsmith, “Optimaldetection for one-shot transmission over diffusion-based moleculartiming channels,”
IEEE Transactions on Molecular, Biological andMulti-Scale Communications , 2019.[110] ——, “Exploiting diversity in one-shot molecular timing channelsvia order statistics,”
IEEE Transactions on Molecular, Biological andMulti-Scale Communications , 2018.[111] Y.-P. Hsieh, Y.-C. Lee, P.-J. Shih, P.-C. Yeh, and K.-C. Chen, “Onthe asynchronous information embedding for event-driven systems in
URAN et al.: SURVEY ON MODULATION TECHNIQUES IN MCVD 22 molecular communications,”
Nano Communication Networks , vol. 4,no. 1, pp. 2–13, 2013.[112] B. Krishnaswamy, C. M. Austin, J. P. Bardill, D. Russakow, G. L.Holst, B. K. Hammer, C. R. Forest, and R. Sivakumar, “Time-elapsecommunication: Bacterial communication on a microfluidic chip,”
IEEE Transactions on Communications , vol. 61, no. 12, pp. 5139–5151, 2013.[113] N. Farsad, Y. Murin, W. Guo, C. Chae, A. W. Eckford, and A. Gold-smith, “Communication system design and analysis for asynchronousmolecular timing channels,”
IEEE Transactions on Molecular, Biologi-cal and Multi-Scale Communications , vol. 3, no. 4, pp. 239–253, 2017.[114] M. U. Mahfuz, D. Makrakis, and H. Mouftah, “Spatiotemporal distri-bution and modulation schemes for concentration-encoded medium-to-long range molecular communication,” in
Proc. of Biennial Symposiumon Communications , 2010, pp. 100–105.[115] C. T. Chou, “Molecular circuits for decoding frequency coded signalsin nano-communication networks,”
Nano Communication Networks ,vol. 3, no. 1, pp. 46–56, 2012.[116] W. Guo, B. Li, S. Wang, and W. Liu, “Molecular communications withlongitudinal carrier waves: Baseband to passband modulation,”
IEEECommunications Letters , vol. 19, no. 9, pp. 1512–1515, 2015.[117] J. G. Gonzalez, J. L. Paredes, and G. R. Arce, “Zero-order statistics: Amathematical framework for the processing and characterization of veryimpulsive signals,”
IEEE Transactions on Signal Processing , vol. 54,no. 10, pp. 3839–3851, Oct 2006.[118] L.-S. Meng, P.-C. Yeh, K.-C. Chen, and I. F. Akyildiz, “Mimo com-munications based on molecular diffusion,” in
Proc. of IEEE GlobalCommunications Conference (GLOBECOM) , 2012, pp. 5380–5385.[119] B. H. Koo, H. B. Yilmaz, C.-B. Chae, and A. Eckford, “Detectionalgorithms for molecular MIMO,” in
Proc. of IEEE Int. Conference onCommunications (ICC) , 2015, pp. 1122–1127.[120] C. Lee, B. Koo, N.-R. Kim, B. Yilmaz, N. Farsad, A. Eckford, and C.-B. Chae, “Molecular mimo communication link,” in
Proc. of IEEEConference on Computer Communications Workshops (INFOCOMWKSHPS) , 2015, pp. 13–14.[121] M. Damrath, H. B. Yilmaz, C.-B. Chae, and P. A. Hoeher, “Spatialdiversity in molecular communications,” arXiv preprint:1707.07400 ,2017.[122] ——, “Array gain analysis in molecular MIMO communications,”
IEEE Access , vol. 6, pp. 61 091–61 102, 2018.[123] M. C. Gursoy, E. Basar, A. E. Pusane, and T. Tugcu, “Pulse position-based spatial modulation for molecular communications,”
IEEE Com-munications Letters , vol. 23, no. 4, pp. 596–599, 2019.[124] Y. Huang, M. Wen, L.-l. Yang, C.-B. Chae, and F. Ji, “Spatialmodulation for molecular communication,” arXiv preprint:1807.01468 ,2018.[125] Y. Huang, M. Wen, L.-L. Yang, C.-B. Chae, and F. Ji, “Spatialmodulation for molecular communication,”
IEEE Transactions onNanobioscience , 2019.[126] M. H. Kabir, S. M. R. Islam, and K. S. Kwak*, “D-mosk modulationin molecular communications,”
IEEE Transactions on NanoBioscience ,vol. 14, no. 6, pp. 680–683, Sep. 2015.[127] S. Pudasaini, S. Shin, and K. S. Kwak, “Run-length aware hybrid mod-ulation scheme for diffusion-based molecular communication,” in
Proc.of Int. Symposium on Communications and Information Technologies(ISCIT) , Sep. 2014, pp. 439–442.[128] Y. Tang, M. Wen, X. Chen, Y. Huang, and L. L. Yang, “Molecular typepermutation shift keying for molecular communication,”
IEEE Transac-tions on Molecular, Biological and Multi-Scale Communications , vol. 6,no. 2, pp. 160–164, 2020.[129] R. Freitas,
Nanomedicine Vol. I: Basic Capabilities . Landes Bio-science, 1999.[130] S. Abadal and I. F. Akyildiz, “Bio-inspired synchronization fornanocommunication networks,” in
Proc. of IEEE Global Telecommu-nications Conference (GLOBECOM) , 2011, pp. 1–5.[131] H. ShahMohammadian, G. G. Messier, and S. Magierowski, “Blindsynchronization in diffusion-based molecular communication chan-nels,”
IEEE Communications Letters , vol. 17, no. 11, pp. 2156–2159,2013.[132] M. Mukherjee, H. B. Yilmaz, and B. B. Bhowmik, “Joint synchroniza-tion and symbol detection for diffusion-based molecular communica-tion systems,” arXiv preprint:1804.06136 , 2018.[133] V. Jamali, A. Ahmadzadeh, and R. Schober, “Symbol synchronizationfor diffusive molecular communication systems,” in
Proc. of IEEE Int.Conference on Communications (ICC) , 2017, pp. 1–7. [134] ——, “Symbol synchronization for diffusion-based molecular commu-nications,”