Experimental Channel Characterization of Human Body Communication Based on Measured Impulse Response

Intra-body communication (IBC) will foster personalized medicine by enabling interconnection of implanted devices. Communication takes place through energy-efficient technologies such as capacitive coupling (CC) and galvanic coupling (GC); however, their modeling is still incomplete. This paper tackles characterization of the human body channel using impulse response, including a first-ever comparison of CC and GC in both wearable and implantable configurations. Experimental data are leveraged to evaluate the measured impulse response in ex-vivo chicken tissue and in-vivo human tissue in a frequency range up to 100 kHz. Pseudorandom noise (PN) sequences are transmitted in baseband and a correlative channel sounding system is implemented. Experimental results demonstrate that the channel is relatively flat in the frequency range of interest, thus offering the opportunity to simplify the design of an IBC transceiver. The relationship between the channel responses and the transmitter-to-receiver distance is also examined using linear correlation, and two regression models are developed. The results show that CC channels are not affected by distance within the range of investigation, while a negative relationship is found for GC channels. Finally, experiments reveal that implantable CC with isolated ground -not deeply investigated yet- is a very promising solution for IBC.


Experimental Channel Characterization of Human Body Communication Based on Measured Impulse Response
Anna Vizziello, Senior Member, IEEE, Pietro Savazzi, Senior Member, IEEE and Fabio Dell'Acqua, Senior Member, IEEE Abstract-The emerging research area of intra-body communication (IBC) will foster personalized medicine by enabling interconnection of implanted devices that ensure pervasive patient monitoring and characterization.Communication takes place through energy-efficient technologies such as the capacitive coupling (CC) and galvanic coupling (GC) techniques, which enable bodyarea communication without wiring; however, the relative novelty of such communication methods means their modeling is still incomplete.This paper focuses on channel characterization of the human body, directly comparing for the first time CC and GC techniques in both wearable and implantable configuration.Experimental data are considered to evaluate the measured impulse response in both ex-vivo chicken tissue and in-vivo human tissue in a frequency range up to 100 kHz.Pseudorandom noise (PN) sequences are transmitted in baseband and a correlative channel sounding system is implemented to evaluate the channel impulse and frequency response through real measurements.Experimental results demonstrate that the channel is relatively flat in the frequency range of interest, which simplifies the design of a transceiver suitable for intra-body networks (IBNs).
Index Terms-Wireless Sensor Networks, Intra-Body Communication, Intra-Body Networks, Human-Body Communication, Short Range Communications, Internet of Nano-Things, Internet of Medical Things, Body Area Networks, Coupling Technologies, Galvanic coupling, Capacitive Coupling, Experimental Testbed.

I. INTRODUCTION
Implantable medical devices will promote next generation healthcare including personalized medicine through real-time physiological monitoring and proactive drug delivery, Anna Vizziello, Pietro Savazzi, and Fabio Dell'Acqua are with the Department of Electrical, Computer and Biomedical Engineering, University of Pavia, Pavia, 27100 Italy (e-mail: name.surname@unipv.it).All authors are also with the Consorzio Nazionale Interuniversitario per le Telecomunicazioni -CNIT Manuscript received XXXX; revised XXXX.
Challenging applications are conceived, such as recovery from paralysis, which would need the communication among implants for the exchange of information inside/outside body.For these scenarios it is necessary to develop body-centric architectures based on both intranet and Internet of Medical Things (IoMT) [2], [3].
The intra-body network (IBN) paradigm enables interconnection of devices inside the human body by enabling transmission of the acquired measurements among implanted sensor devices as well as to an external monitoring center.To this purpose, proper energy-efficient communication technologies are needed.
The most common intra-body communication (IBC) links use classical radio frequency (RF) waves at frequencies below 1 GHz or in one of the standard Industrial, Scientific and Medical (ISM) bands, in the form of narrowband (NB) or ultra-wideband (UWB) signal [4].However, several studies have demonstrated that RF signals, although profitably used for the communication of on body wearable devices, experience high losses within the tissue [5].Consequently, the coverage is limited to short distances, with possible heating that may cause damage to the tissues that vehiculate waves.Hence, other technologies have been explored as profitable alternatives for sub-cutaneous communication among implants.These include ultrasounds, capacitive coupling (CC) and galvanic coupling (GC) techniques, which show lower attenuation within the human tissues compared to RF methods.Ultrasounds are acoustic waves showing good propagation properties in environments with high water content such as the human body [6].However, they suffer severe multi-path fading and long delays caused by slow propagation; these factors can be counteracted by suitable design of transceivers, which however generally results into large sizes and high power consumption, unacceptable in implants.Considering the above context and related constraints, in this paper we investigate the so called coupling technologies for human body communication (HBC) that are capable of mitigating the above-mentioned issues.They consists in EM-based methods operating at low frequencies, up to 100 MHz.
Coupling technologies include capacitive and galvanic coupling methods [5], [7], and are already included in the Standard for Wireless Body Area Networks [8].Capacitive and galvanic coupling rely on a couple of electrodes at the transmitter end and another one at the receiver end, although in a different configuration.CC is usually employed for wearable-type scenarios: only one transmitter electrode is attached to the body because the other (ground electrode) floats.The same setup is used for the receiver [7].Recently, preliminary explorations were made for implanted CC with an isolated ground electrode [9], [10].In GC both the pairs of transmitter and receiver electrodes are attached to or implanted in the body [7].These coupling technologies involve lower power levels compared to RF methods, enabling longer-distance transmission within the body, while at the same time avoiding tissue heating [7].

A. Research Motivation
In order to design proper transmission methods leveraging on these coupling technologies, an accurate channel model is essential for characterizing the electrical behavior of tissues.The main approaches consist of quasi-static approximations [11], full wave numerical techniques, such as Finite Difference Time Domain (FDTD) Method, Finite Element Method (FEM) [5], [7] and Equivalent Circuit Analysis (ECA)-based models [5], [7], [12], [13].
The quasi-static field distribution analyses are computationally efficient, but they only model low frequency approximations to Maxwell's equations and can not be used for high frequency applications [13].Field analysis using FDTD and FEM are more accurate and flexible and recently, indeed, 2-D and 3-D models based on FDTD and FEM [14] were proposed to address realistic geometrical properties of the human body.These models, however, are computationally very demanding, making them unsuitable for rapid design and deployment of an IBN.
The ECA model offers a simple transfer function valid for a wide range of frequencies, with accurate and instantaneous gain computation; this makes it useful for IBN deployment in time-sensitive healthcare applications [13].The developed methods consider a single layer of tissue or multiple heterogenous layers composed of skin, fat, muscle, and bone tissues with experimental analysis [13].
These channel models are effective in representing the dielectric properties of human tissue that may affect signal propagation.On the other hand, they do not consider essential properties of wireless channels, such as multi-path delay spread and amplitude fading statistics, that need to be taken into account when designing a communication system [15].
Impulse-response-based methods have a potential to fill this gap, but limited effort has been spent so far on characterizing IBC channels.Studies were conducted on CC in wearable configuration [16]- [19] and others on GC [15], [20].However, a comparative investigation of both CC and GC technologies based on impulse response channel method, under both wearable and implantable configurations, has not been conducted yet, to the best of the authors' knowledge.The present work intends to address such research gap.

B. Main Contributions
The main contributions of this paper are the following: • a channel model based on impulse response is derived for a comparison of CC and GC technologies.We explore and compare for the first time all the four possible configurations that include CC and GC in wearable and implantable scenarios.• a first attempt is made to characterize the human-body channel using a correlative channel sounding method based on experimental measurements, conducted on both ex-vivo and in-vivo tissues.• for the first time the frequencies up to 100 KHz are investigated, which were only preliminarly evaluated in our previous work [20].The analysis of this frequency range is essential to develop baseband UWB transceiver, whose simplicity is suitable for IBNs.In future works we aim to extend our evaluation up to 100 MHz.• safety considerations are incorporated in the employed experimental testbed by transmitting ultra-low power, in the order of tens of W, which prevents tissue heating.Extensive experiments were carried out and different parameter settings were tested, including electrode size, inter-electrode distance, and distance between transmitter and receiver.Experimental findings indicate that the channel response is relatively flat for the frequency ranges of interest and the noise can be approximated as additive white Gaussian in all the considered four configurations of the electrodes.These results allow to design simple transceivers, without complex receivers to counteract multi-path effects of the channel as required, for example, in ultrasound technology when employed for intra-body networks.Among the possible configurations, implantable CC appears to be the preferable one allowing lower attenuation while assuring less interference from the environment, as it accurs in CC wearable scenario.It is only recently that some attention was devoted to CC in implantable configuration [9], [10], since the CC technology was usually employed for onskin settings [7].However, implantable CC requires more complex hardware implementation due to the need of isolated ground.Both GC wearable and implantable configurations are less sensitive to external interference but achieve higher attenuation than CC.
The rest of the paper is organized as follows.In Sec.II the considered low-power coupling technologies and their underlying physical principles are presented, together with configuration settings of the electrodes.Sec.III presents the channel model based on impulse response and Sec.IV describes the system overview and the experimental setup.Sec.V shows the experimental results, while Sec.VI presents future directions and possible applications, and Sec.VII closes the paper with some summarizing conclusions.

II. LOW-POWER COUPLING TECHNOLOGIES
Low-frequency EM technologies are customarily classified based on their coupling principles, that use different physical methods to generate an electrical signal to propagate through the human body.The electrical signal lies below 100 MHz and has power levels in the order of W, lower than the one used in traditional RF signals, which extend up to several GHz [7].For this reason, body coupling methods have gained great attention in IBC research aimed at ensuring safety and decreasing energy consumption.Body coupling methods are classified into capacitive and galvanic coupling techniques [21].

A. Coupling Technologies
Capacitive and galvanic couplings share some features as both employ electrodes to transmit and receive, albeit in different configurations.In CC, only one of the two transmitter electrodes is attached to the body while the other (ground) electrode floats.The same configuration is used for the receiver.The physical principle is based on near-field electrostatic coupling of the human body with its surroundings (Fig. 1(a)).The signal electrode of the transmitter induces the electric field in to the human body [7].The induced electrical signal is controlled by an electrical potential and the body acts as a conductor with the ground as the return path [7].The usual carrier frequency ranges from 100 kHz up to 100 MHz [7], [22] and this approach is usually employed in wearable scenarios covering long distances, up to 170 cm.However, its operation may be affected by environmental conditions.
In GC both pairs of electrodes, transmitting and receiving are attached to or implanted in the human body.In GC an AC current flows inside the body and the body acts as a waveguide transmission line.Specifically, an electrical signal is applied differentially between the two electrodes of the transmitter.While the primary current carrying the data flows between the two transmitting electrodes, highly attenuated secondary currents can still be detected at the receiver electrodes (Fig. 1(b)).This technology is suitable for implanted scenarios and consumes two orders of magnitude less energy than RF transceivers [23].Its usual operating frequency range is 1 kHz-100 MHz with a coverage range up to 20 − 30 cm [21].
In terms of applications, GC is usually employed for communication among devices in implanted scenarios, while CC for wearable settings to establish communication between on-body devices or devices close to the body [9].Anyhow, as it will be detailed in the following, a modified CC configuration has been recently proposed for implantable scenarios [24]- [27].

B. Coupling Technology Configurations
Both coupling types require transceivers with two electrode pairs.Fig. 2 (a), (b) illustrates the different electrode configurations of CC and GC coupling in a wearable scenario.In CC, only one of the electrodes (signal electrode) of the transmitter side and receiver side is attached to the body, while the other electrode (ground electrode) floats (Fig. 2 (a)).In GC, both electrodes at transmitter and receiver side are attached to the human body [7] (Fig. 2 (b)).The different physical principle, explained above, calls for lower GC transmission rates and distances than those of CC.At the same time, there is no need for a floating ground reference nor for propagation outside the human body, hence GC does not suffer interference from external environment [5].Anyway, given the aforementioned features, CC is the usual choice in wearable scenarios.
The GC in implantable configuration consists of both electrode pairs of transmitter and receiver embedded inside the body as in Fig. 2 (d).So far, GC has been the preferred choice in implantable scenarios [9].Recently, however, it has been demonstrated that a stable capacitive return path can be achieved not only by exposing the capacitive ground electrode directly to the air, as in wearable configurations, but also in implantable settings, provided that the ground electrode is isolated from human tissue [9] (Fig. 2 (c)).In this way, the path between transmitter electrodes has higher impedance than the path to the receiver, resulting in reduced signal attenuation with respect to the implantable GC (Fig. 2 (d)).Therefore, the intra-body capacitive method emerges as a viable alternative for communication among implanted devices that can increase the transmission range currently achievable with GC technology.The results of implantable CC are very promising but this area is still in a nascent stage.Implantable CC is not widely accepted, also because a thorough investigation of its characteristics has not yet been carried out [10].A few studies have been conducted, such as [9], [10], but a proper channel modeling for this configuration is still lacking, and further investigation is required to evaluate the features of implantable CC [26].One of the objectives of the paper is hence to confirm the feasibility of this configuration and, at the same time, compare it with the implantable GC, due to the similarity of electrode configurations.The final goal is to build a first, comprehensive comparison among all the possible coupling wearable and implantable configurations, assessing benefits and drawbacks of each configuration.The investigation focuses on the impulse response of the communication channel.

III. CHANNEL MODELING BASED ON IMPULSE RESPONSE
When designing wireless communication systems, accurate channel models are required to assess how the signals of interest propagate through the medium.Different approaches to channel modeling exist, each with its own pros and cons.This work uses a stored channel impulse response approach [28], [29], which employs a correlative channel sounder.The method has been chosen because of these two advantages: (i) the measured and stored channel impulse responses are based on experimental measurements, (ii) the stored responses are reproducible and reusable, which is useful when simulating and optimizing communication systems.Before detailing the developed channel model, the theoretical foundations of CC and GC are recalled in the following sub-section.

A. Dielectric Properties of Human Tissues
Gauss's law and charge-continuity equations are given in ( 1) and ( 2): Fig. 2: Electrodes configuration placement for CC and GC technologies.
in which  is the electric displacement,  is the electric charge density,  is the current density,  is the electrical conductivity,  is the electric field intensity and   is the current density of the source.
When the product of body size and the wave number in biological tissues is much larger than 1, wave propagation and inductive effect in biological tissues may be neglected [26], [30].Therefore, Maxwell's equations can be decoupled as quasi-static electric field governing equation: in which  is the electric potential,  is the angular frequency, and  is the permittivity.The permittivity  is governed by the Cole-Cole equation, which shows how the dielectric properties of a tissue change over a large frequency range [7]: where ε is the complex relative permittivity, and Δ is the magnitude of the dispersion calculated as Δ =   −  ∞ , in which  ∞ is the permittivity at field frequencies where  ≫ 1 and   the permittivity at  ≪ 1.  is the relaxation time constant that depends on physical processes, such as ion effects, and  is a distribution parameter that lies between 0 and 1 [7] and is a measure of the broadening of the dispersion.The properties of a tissue are therefore more appropriately described by means of multiple Cole-Cole dispersion: where   is the static ionic conductivity and  0 is the permittivity of free space.Equation (5) may be used to predict the dielectric behaviour in the considered frequency range, with a proper choice of parameters for each tissue.
The complex conductivity and the complex specific impedance of a tissue may be then calculated as [7]: Electrical properties of human body tissues may be modeled by equivalent electrical components such as resistors and capacitors, and are the building bricks to develop the transfer function of the body channel based on circuit model, FEM, or circuitbased FEM model [7].As detailed later, given the dielettric properties of the human tissues, we exploit a different approach based on channel impulse response.This method is more suitable to analyze the properties of the body channel from a communication perspective.In the rest of this paper, the term channel refers to human body channel.

B. Correlative Channel Sounders
A channel sounding signal is composed of a pulse transmission that occurs with predetermined repetition intervals.When signals are received, a sounder device filters and records them for off-site storage and processing [15].The type of sounding signal sent depends on the method used [28] and in the following we consider pseudorandom noise (PN) sequences.
As is known, the received signal can be described as () = () * ℎ() + (), where () and () are the transmitted and received signal, respectively, ℎ() is the channel impulse response, n(t) is the additive white noise, and * is the convolution operator.Correlating each side of the previous equation with () yields: where   is the cross-correlation function between () and (),   () is the auto-correlation function of (),  is the delay time.() and () are assumed uncorrelated.
If the channel impulse response ℎ() changes slowly within the time interval required to measure the correlation function, then (7) can be employed to measure ℎ().This can be achieved if   approaches a delta function, as in this case   becomes a good approximation of ℎ() as shown in (7).
To this aim, PN sequences are used as the transmitted signal () since they yield an autocorrelation function with a high correlation peak and much lower components off-peak.By correlating the received signal with the transmitted PN sequence in (7), it is possible to calculate the crosscorrelation   [29], which corresponds to the CIR.
The capability to change the maximum delay and bandwidth are essential characteristics of PN sequence sounders.Furthermore, the sounder presents a processing gain, which reduces transmitter power requirements and can be increased by using longer code words.Using maximal-length PN sequences as the transmitted signal leads to an auto-correlation function characterized by a high correlation peak and extremely low side lobes (high peak-to-off-peak ratio).This feature allows any multi-path component to be detected at the receiver when correlating the channel output with the originally transmitted PN sequence by means of a convolution matched filter [20].Fig. 3 illustrates the blocks of the considered channel sounding architecture.

C. Channel Model Development
The channel response can be examined using a power delay profile, which can be calculated using the impulse response of the body channel.In a general wireless channel, the power delay profile is defined as a time-averaged function with respect to squared impulse responses.Indeed, the impulse response may change over time due to the multipath effect, where each path shows random signal loss and delay.Thus, the impulse response ℎ(, ) is expressed as a function of time  and delay .However, the body channel can be assumed to be constant over time if the person being examined does not move [17], as in the measurements conducted in this paper.Hence, under this assumption, we obtain a time-invariant impulse response ℎ() and the power delay profile  ℎ () can be calculated by squaring the impulse response without time averaging: The signal loss and power delay profiles of the conducted measurements may be used to obtain the following channel parameters: mean signal loss, mean delay, root-mean-square (rms) delay, received power, coherence bandwidth.
The mean signal loss is calculated by averaging the magnitudes of the signal loss at all frequencies, while the mean delay τ and rms delay   are obtained using the power delay profile: where   =   *   with   being the number of PN bits and   number of samples per bit.
The coherence bandwidth   may be calculated using a frequency correlation function [31] expressed as a function of the power delay profile [17], [31].Given the inverse relation between   and   , the coherence bandwidth   can be calculated from   [31].

IV. SYSTEM OVERVIEW AND EXPERIMENTAL METHOD
In order to experimentally measure the channel impulse responses, PN sequences were transmitted in baseband.A linear polynomial PN sequence of degree  = 14 was implemented using a linearfeedback shift register with a chip duration of 5.2  (corresponding to 96 kHz bandwidth) as in [20].These parameters were set according to the frequency range employed in the test system, up to around 100 KHz.
The testbed [32] was modified to implement a correlative channel sounding, only explored to a limited extent in our previous work [20] (Fig. 3).The source code of the transceiver [32] is available online on Code Ocean for sake of replicability [33].Hardware requirements are moderate, limited to two PCs with sound cards used to generate/transmit and receive the signal in a subset of the GC frequency range [32].As shown in Fig. 4(a), we use a batterypowered laptop and a PC as transmitter and receiver, respectively.The laptop was unplugged from the grid to avoid common ground return paths to the desktop, as required by the coupling technologies [32].A Matlab session must be kept open on each machine, to run the transmitter and receiver software, respectively.
The generated PN sequences are transmitted using the tx Matlab program, and then converted from digital to analog (D/A) domain to be sent over the sound card of the transmitter (see Fig. 3).The transmitted signal is injected into the biological tissue through a cable connected to the LINE OUT jack on one side and to the two transmitter electrodes on the opposite side.
After transmission, the two electrodes of the receiver detect the received signal, which is sent to the other PC via a cable connected to the LINE IN jack.The Matlab rx program includes a 50 Hz filter and a convolution matched filter to correlate the channel output with the transmitted PN sequence known at the receiver, to build the CIR estimation (Fig. 3).The audio frequency sampling   is set to 192 KHz, with 16 bits per sample.Table I shows the values of main system parameters.The audio parameter setting can be set through the control panel of each computer, while other parameters, such as the degree of the PN sequence, may be set in the tx/rx Matlab program code.

A. Experimental setup
We consider two configurations: one with electrodes implanted in ex-vivo chicken breast tissue (Fig. 4 (b)) and another one with wearable electrodes placed on in-vivo human skin (Fig. 4 (c)).Electrodes with different sizes (0.5 mm and 1 cm diameter) were tested.
For the case of ex-vivo tissue we employed low-cost regular leads, covered with an aluminum foil to avoid oxidation due to the water content of the chicken breast.We used small-sized circle electrodes (in the order of 0.5 mm diameter) to test a real configuration scenario for future miniaturized medical devices.For the case of in-vivo tests we employed commercial electrodes with 1 cm size placed on human leg skin.The ex-vivo tissue, a sample of chicken breast sized roughly 21 cm × 16 cm × 6 cm, consisted of a single-layer tissue, i. e., the muscle.The in-vivo tissue (human leg) involved heterogeneous multi-layers tissues, i. e., skin, fat, muscle and bone tissues.Indeed, although the electrodes are placed on the skin, the signal is expected to flow not only in the outmost layer of skin but rather in all of the aforementioned tissues [21].Performances were computed over 100 runs.Unless otherwise stated, the transmission power   was in the order of 10  and the inter-electrodes distance at both transmitter and receiver side was set to 1.5 cm for ex-vivo tests and 4 cm for invivo experiments, while the transmitter-to-receiver distance was varied during the experiments.

V. EXPERIMENTAL CHANNEL CHARACTERIAZION
The received signals were post-processed in MATLAB to obtain the channel impulse response (CIR) and channel frequency response (CFR) for both GC and CC technologies in wearable and implantable configurations.
Preliminary evaluation demonstrated that the GCbased communication channel is non-frequencyselective [15], [34], although this was proved only in wearable setting.In other studies [17], [35]- [37], some aspects of the CC channel were presented.In [17] a channel delay spread was evaluated, which however was not due to the multipath effects as in radio channels, but rather to capacitive coupling effects in the body.The spreading time resulted to be constant as in a resistor-capacitor circuit [17].However, in these studies the transceiver setup considered a transmitter and/or receiver sharing an Earth ground connection through the power grid.This means characterizing a channel which is in fact different from the real case of wireless body area network (WBAN) or intra-body network, lacking any common ground.Furthermore, these papers did not consider nor compare the four possible Fig. 5: Channel Impulse Response (CIR) for a wearable GC configuration in heterogeneous in-vivo biological tissues with 1 cm electrodes,   = 4 cm inter-electrodes distance, and  − = 10 cm distance between transmitter and receiver.configurations discussed in this paper and shown in Fig. 2, since the evaluation of impulse response was conducted only in wearable settings and separately for GC and CC [15], [17], [34]- [37].Moreover, the frequency range was different from the considered one, up to 100 KHz.The evaluation of this frequency range will allow to design baseband UWB transceiver, whose simplicity meets well the need of long-lasting implants in IBNs.
Fig. 5 illustrates the measured CIR for the communication scenario of heterogeneous tissues, i.e. the in-vivo tissues of a human leg.The figure shows the high peak-to-off-peak ratio discussed in Sec.III, which provides good correlation results from the experiments.All the CIRs obtained in different scenarios show a similar behaviour, suggesting no multi-path effect takes place in the body channel.

A. Experimental Results
Figs. 6 and 7 represent the CIR and its corresponding frequency domain representation, i. e., the CFR, for implanted and wearable scenarios with txto-rx distances of 6 cm and 10 cm, respectively.Similar trends with different magnitudes for channel gain are obtained in the CFRs when changing distances between the transmitter and receiver.The CFRs indicate that the channel is relatively flat within the frequency range of interest, although with lower channel gain for the GC than for the CC technology.In particular, focusing on the implantable setting shown in Fig. 6, the CC method shows a higher peak-to-off-peak ratio for CIR compared to GC technology, as well as larger channel gain of the CFR (around 76 dB vs. 36 dB).Indeed, as it will be detailed in Sec.VI, implantable CC is very promising thanks to its lower signal attenuation inside the body with respect to GC. Confirmation was then found in very recent literature [24]- [27] since up to now CC technology was only employed in wearable configuration with floating ground electrode.Figs. 8 and 9 show the mean amplitude of the received signal in time, normalized to the number of bits of the ADC converter, the maximum peakto-peak amplitude of the cross-correlation in time, and the mean amplitude of the cross-correlation in frequency, for implantable and wearable scenarios, respectively.Values were averaged over 100 runs.Comparing the two figures, it is possible to note that the amplitude of the received signal is lower for GC in implantable configuration than in wearable scenario (Fig. 8(a) vs Fig. 9(a)).However, cross correlations in both time and frequency domains are better for GC in implantable than in wearable scenario (Fig. 8(b) vs Fig. 9(b), and Fig. 8(c) vs Fig. 9(c)).This is due to a larger electrode size in wearable configuration -1 cm vs 0.5 mm of implantable setting-that allows higher signal amplitude at the receiver with, however, stronger noise also (see Fig. 7(a) vs Fig. 6 (a)).During the experiments, it was not possible to compare different electrode sizes in implantable and wearable configurations for feasibility reasons.Indeed, implantable and wearable configurations in the chicken breast showed similar results due to the single layer tissue, while human implantation was not explored since it would have required appropriate medical lab facilities.Overall Fig. 8 and 9 show better performance of CC over GC.It is however worth noting that, unlike for GC whose currents are confined inside the body, CC suffers from external interference in wearable configuration due to its physical principle based on the formation of an electrical field around the body.

B. Analysis based on noise thresholding and CIR calibration
1) Noise thresholding: Experiments were conducted also considering only the contribution of the main values of the channel impulse response discarding other noise contributions with a thresholding procedure.The method estimates the SNR and deletes the contributions of the CIR with values lower than a threshold set equal to √︁  2  , where   2  is the estimated noise variance.Fig. 10 shows results obtained in implantable configurations and compares them with a setting where a cable was used to connect the transmitter and the receiver.In particular, Fig. 10 (a), (b), (c) report the full CIR, while 10 (d), (e), (f) only the CIR contributions with values higher than the threshold, showing that the CIR in the setting with the cable presents less samples than the one with wireless transmission via CC and GC.In particular, the impulse response presents three main samples in the first case, while become around ten in the other ones.Similarly, 10 (g), (h), (i) and 10 (j), (k), (l) illustrate the full CFR and the one obtained from the CIR with thresholding, respectively, showing more clearly the flat nature of the channel with higher channel gain for CC than GC technology.In more details, Fig. 10 (j) illustrates the flat channel with cable, while showing its low pass behaviour with around 10 dB major attenuation at higher frequencies.Fig. 10 (k) depicts the high-pass behaviour of the CC channel, with higher attenuation at low frequencies due to the capacitive effect of the body.CF magnitude rises indeed from 36 dB to 80 dB at medium frequencies, to then settling on 65 dB at higher frequencies.A similar, slightly high-pass behaviour is shown for the GC channel in Fig. 10 (l) although with CFR values lower than CC.The low-pass filter effect observed at normalized frequencies above 0.9 is due to anti-aliasing filters in the sound cards.
2) CIR calibration: In order to better characterize the measured intra-body CIR, it is possible to equalize both the cable and the complete transceiver device effects using the CIR estimated in direct txto-rx cable connection conditions, as in Fig. 10.
After removing noise by thresholding, the estimated discrete-time CIR of the cable channel can be written as where    () represents the estimated CIR when the received signal   () is collected after a cable that directly connects transmitter and receiver, and   is the sampling interval.
Once estimated ℎ  (), the intra-body discretetime CIR ℎ() ≡   (  ) can be equalized by using the following inverse filter, i.e. the zero-forcing (ZF) criterion, where W  is the convolution matrix corresponding to the inverse ZF filter, while H c is the convolution matrix of the discrete-time cable CIR of length . is determined after thresholding, by considering only non-zero elements of ℎ  () In order to better mitigate noise effects, the minimum mean square error (MMSE) criterion can be also used: where  2  is the noise variance defined above.Our experiments, however, verified that the equalized estimated CIR is practically equivalent to the ones depicted in Fig. 10.This outcome confirms that the cable CIR can be effectively represented by an ideal flat channel that does not affect the measured intra-body CIR when employing CC and GC technologies.

VI. FUTURE DIRECTIONS AND APPLICATIONS
The relatively flat nature of a CC and GC human body communication channel allows to design simple communication systems, which may provide benefits in the medical and commercial application area.However, several open issues should be first investigated for the development of wireless implantable solutions.

A. Channel Modeling and Transceiver Design
Plans for future work include the investigation of the channel impulse response on a wider frequency range, up to 100 MHz, which would require a different hardware to replace the currently employed testbed.Also, the evaluation of the body movements will be included in the channel characterization, although from our preliminary analysis we expect the impact of bodyly motion on channel attenuation to be much smaller than for RF WBAN channels.
Furthermore, proper transmission techniques suitable for intra-body may be designed, leveraging on the channel model findings, including UWB, compressive sensing trasmission methods [38], and simple digital modulation schemes [39], appropriate for implanted transceivers.Also, multiple implants scenario are conceived, in which it is fundamental to develop opportunistic wake-up methods with location awareness of devices [40]- [42] or to exploit resource allocation methods [43] to minimize energy consumption as required in IBNs.

B. Technology Readiness Level (TRL) and miniaturization
Future directions of research and development include TRL increase and miniaturization of the devices.CC and GC technologies have reached a degree of maturity to move from the proof-ofconcept stage to technology development, although implantable CC is only at an early stage of investigation.The development of these miniaturized transceivers is currently ranked at TRL 4, since they have been tested in the laboratory.Further studies will encourage the development of manufacturing processes to develop powerful miniaturized wireless communication devices.
Furthermore, coupling technologies are still at the millimeter scale, and further miniaturization is required for long-term implantation in humans.Among the possible configurations, implantable CC is the most promising one in terms of miniaturization.Indeed, implantable GC requires a minimum inter-electrode distance to ensure proper operation.A tiny device would require really close electrodes at the transmitter, however, all the current would flow between those electrodes and secondary currents would become too weak to be detected at the receiver side.On the contrary, if the interelectrode distance at the transmitter is not very short, the secondary currents show higher intensity and are easy to be detected at the receiver, but this makes it difficult to miniaturize the GC transceiver.Fortunately, the isolated ground electrode of the implantable CC may be placed extremely close to the signal electrode, allowing high miniaturization of the devices.The downside is the need to use an additional layer composed of a very effective insulation material for the ground to assure very high impedance between the transmitter electrodes, as well as between receiver electrodes.Also, as mentioned in the experimental results, implantable CC performs better than GC due to the lower attenuation of the signal inside the body, making longer transmission distances possible.

C. Energy Harvesting
Regarding energy transfer to implants, designing wireless sub-mm scale implantable medical devices with centimeters-deep operation range presents power delivery challenges.Inductive coupling [44] is the conventional method adopted for power transfer in biomedical implants in the size of cm and placed under the skin, because of high power transmission efficiency and minimal dissipation of the magnetic field inside the tissue.However, miniaturization of implants from cm down to mm, and even sub-mm size, makes the inductive energy transfer difficult.The process becomes even harder when moving from under-skin devices to deeptissue implantation, since the distance from the external powering device and the implant increases.Hence, alternative technologies need to be explored for powering, such as ultrasounds, which calls for the development of solutions with a combination of intra-body technologies.

D. Medical Applications
Several applications are envisioned for IBC coupling technologies.Currently, health care systems often utilize wired solutions, for example to combat epilepsy and chronic pain, or for drug delivery, where an actuator controls the release of a medicine, based on a certain sensed biological marker [15].The specific nature of coupling technologies would allow to replace parts of these systems, to make the solution less intrusive and allow for a more scalable, implantable network of devices.
In a long-term perspective, IBC technologies may be employed for closed-loop artificial neural systems that are completely implantable.These systems can be composed of sensing implantable devices, transceivers and actuators capable to support artificial prosthesis for the motor recovery of paralyzed limbs, or to bypass spinal injuries, thus overcoming the current need for external processing.Another exciting envisioned application is a network of injectable, wireless neural sub-mm implants.Their small size and wireless features allows a high freedom in selecting their placement for neural recording.Since most neurological diseases involve several brain regions, being able to monitor neural activity in multiple sites and observe intra-region communication is crucial for a better understanding of dysfunctions.Furthermore, placing multi-ple neural recording nano-implants in and around the area of seizure activity could be helpful for surgical planning or to monitor epileptic patients.The coupling methods enables communication to, from and between implants, allowing the creation of an intranet of devices inside the body that may also communicate with an IoT architecture external to the body to accomplish more complex tasks, thus leading to the emergence of novel ICT-based solutions.

VII. CONCLUSIONS
In this paper we discussed the characterization of communication channels inside the human body through measured impulse response, with the aim of comparing CC and GC technologies in both wearable and implantable configurations.Experimental results demonstrate that the channel is relatively flat in the frequency range of interest up to 100 KHz, which makes simple baseband transceiver design suitable in principle for IBNs.Moreover, experiments revealed that implantable CC with isolated ground is a very promising solution, given its ability to cover long distances with very low transmission power.Future avenues of development have been illustrated, from investigation of the channel to include the impact of body movement, to proper transmission techniques suitable for IBNs, to the miniaturization of implanted devices, along with the current other open issues in this exciting IBC research area.

Fig. 6 :
Fig.6: The measured channel impulse response (CIR) and channel frequency response (CFR) in implanted configuration with 0.5 mm electrodes size and  − = 6 cm with ex-vivo biological tissue.The inter-electrodes distance is set   = 1.5 cm for both GC and CC configurations.Note that the ground is isolated at both transmitter and receiver in implanted CC configuration.

Fig. 7 :
Fig.7:The measured channel impulse response (CIR) and channel frequency response (CFR) in wearable configuration with 1 cm electrodes size and  − = 10 cm with in-vivo heterogeneous biological tissue.The inter-electrodes distance is set   = 4 cm for both GC and CC configurations.Note that the ground is floating at both transmitter and receiver in wearable CC.

Fig. 8 :
Fig. 8: Comparison between GC and CC in implanted configuration for different tx-to-rx distances ( − ).0.5 mm electrodes are employed with exvivo tissue.

Fig. 10 :
Fig. 10: The measured channel impulse response (CIR) and channel frequency response (CFR) when the received signal is filtered with a tresholding procedure based on SNR and noise variance to remove noise, under three configurations: (i) a cable connecting transmitter and receiver; (ii) CC implanted scenario in ex-vivo tissue with 0.5 mm electrodes size and  − = 6 cm, (iii) GC implanted scenario in ex-vivo tissue with 0.5 mm electrodes size and  − = 6 cm.

TABLE I :
Parameters setting