Galvanic-coupled Trans-dural Data Transfer for High-bandwidth Intra-cortical Neural Sensing

Abstract

A digital-impulse galvanic coupling as a new high-speed trans-dural (from cortex to the skull) data transmission method has been presented in this paper. The proposed wireless telemetry replaces the tethered wires connected in between implants on the cortex and above the skull, allowing the brain implant to be “free-floating” for minimizing brain tissue damage. Such trans-dural wireless telemetry must have a wide channel bandwidth for high-speed data transfer and a small form factor for minimum invasiveness. To investigate the propagation property of the channel, a finite element model is developed and a channel characterization based on a liquid phantom and porcine tissue is performed. The results show that the trans-dural channel has a wide frequency response of up to 250 MHz. Propagation loss due to micro-motion and misalignments is also investigated in this work. The result indicates that the proposed transmission method is relatively insensitive to misalignment. It has approximately 1 dB extra loss when there is a horizontal misalignment of 1mm. A pulse-based transmitter ASIC and a miniature PCB module are designed and validated ex-vivo with a 10-mm thick porcine tissue. This work demonstrates a high-speed and miniature in-body galvanic-coupled pulse-based communication with a data rate up to 250 Mbps with an energy efficiency of 2 pJ/bit, and has a small module area of only 26 mm².

I. Introduction

With the strong demand for helping patients suffering from neural disorders (e.g., Parkinson, stroke, etc.) to regain their sensory-motor or cognitive functions, the need for neural interfaces is becoming acute in the coming decades. Intra-cortical microelectrode arrays (MEAs) continue to remain the gold standard for neural interfaces and enable electrical sensing with sub-millisecond temporal resolution and 10’s of μm spatial resolution, making this the most widely adopted clinical method at the moment. With the recent introduction of high-density MEAs (e.g., silicon probe [1],[2] or polymer probe [3] with up to 1000 electrodes), neuroscientists can increase the density of neural sensing [4]. Assuming each recording channel is sampled with a 10-bit ADC at 25 ksps, the total amount of data that needs to be transferred can reach 250 Mbps. Although data compression techniques can relax the bandwidth required for data transmission, the capability of providing raw data from an implantable device is still highly preferred, for diagnosis and optimization reasons. Currently, most of the implantable neural sensors reported in the literature, for recording intra-cortical extracellular single-unit neural activities from the brain, do not apply data compression before wireless transmission [5],[6]. High fidelity data are crucial for neuroscience research, and potential distortion due to compression is avoided.

MEAs are typically implanted via craniotomy, which requires surgery on a large skull surface area, resulting in a long period of convalescence. With the recent advancement in craniotomy, a new burr-hole craniotomy can safely open a small hole [7], with a diameter less than 10 mm, and allows the skull to also heal much more quickly than conventional large-area craniotomies. However, the existing telemetry modules are still too large to be fully implanted below the skull. Having a direct cable connection between the telemetry module and the MEA, the latter is bounded in its movements and is not completely free to follow the motion of the brain. Therefore, it will also have relative movements with respect to the brain, causing damage to the surrounding brain tissue. This leads to serious scarring and affects the quality of the recording [8]. Hence, a miniature and wideband wireless link that transmits sensor data across the dura (a brain protection tissue) and scalp is a critical part to make the MEA “free-floating,” less invasive surgically, and offer a highly preferable option in clinics [8]-[10].

A few of the existing pre-clinical intra-cortical neural interface systems with one or multiple MEAs were validated on primates [4],[11],[12], including a large head-mounted telemetry module, as shown in Fig. 1(a). Patients with such modules suffer from huge risk due to physical accidents and infections. These head-mounted telemetry modules have been miniaturized and implanted below the scalp in [13],[14], i.e., as a transcutaneous telemetry module as shown in Fig. 1(b), avoiding risks mentioned above. However, one remaining critical issue is the strain and micromotions resulting from the MEA being tethered by bundles of cumbersome wires to either the head-mounted or the transcutaneous telemetry module, as shown in Fig. 1(a) and 1(b), which leads to severe scarring [8]. Furthermore, the neural decoding of iBCIs requires frequent (daily basis) calibrations to accommodate the signal drift due to MEA moved by the tethering wire [15].

Fig. 1. State-of-the-art head-mounted data acquisition module and the proposed untethered data acquisition module with free-floating MEAs.

(a), a tethered head-mounted telemetry module. (b), a tethered transcutaneous telemetry module. (c), the proposed implantable telemetry with a tether-less trans-dural link.

As illustrated in Fig. 1(c), the neural sensing data from MEAs will be transmitted across the channel containing cerebrospinal fluid (CSF) and dura and be received by a cranial implant (or “cranial plug”) that resides in the skull-hole volume (with a diameter smaller than 10 mm), i.e., a “trans-dural” link. The cranial implant with less volume constraints than the sub-dural implant then “relays” the data transcutaneously to the external device [13].

To the best of the author’s knowledge, there is no wireless transmission method reported in the literature that can support such trans-dural link, coupling high-speed data (>250 Mbps) efficiently across biological tissues, in this case, approximately 1-4 mm dura and 3-5 mm cerebrospinal fluid CSF, and fitting to a limited module area well below 100 mm².

The solutions presented in the literature based on far-field electromagnetic (EM)-radiation [16]-[18] suffer from the bandwidth-area trade-off of antennas [19]. The dimension of the antenna must be comparable to the half wavelength of the transmitted signals to achieve optimum radiation efficiency. A high-frequency signal with a shorter wavelength can reduce the antenna dimension, but it suffers from the shadowing effect caused by the body which increases the attenuation [20]. Alternatively, an electrically small antenna (i.e., antenna dimension < λ/10) has a smaller area but has limited bandwidth and degraded radiation efficiency [21].

Inductive coupling [22]-[24] is widely adopted in implantable devices for wireless power and data transfer, but it also does not offer sufficient bandwidth due to a high quality-factor resonance required for efficient magnetic induction [25]. A large cross-section is also needed to have low path loss and to minimize the impact of misalignment. The relation between the pathloss and the quality factor is listed in Eq. (1), assuming the transmitter and the receiver have coils with the same quality factor Q_Tx = Q_Rx = Q,

$$Path\_gain\approx k{Q}^2$$ (1)

where k is the coupling coefficient between coils [26]. Wideband communication requires coils with low quality factor Q (=frequency/bandwidth), but this reduces path gain. Note the coil diameter and the coupling coefficient k are linked by the Eq. (2)

$$k=\frac{1}{{\left[1+2^{2/3}{\left(\frac{d}{\sqrt{r_{Tx}{r}_{Rx}}}\right)}^2\right]}^{3/2}}$$ (2)

where d is the distance between the two coils, r_Tx and r_Rx are the transmitter and the receiver coils radius, respectively [27]. Eq. (1) and Eq. (2) indicate that there is a fundamental trade-off between the coil size, path loss, and transmission bandwidth in inductive coupling.

Optical communication has been demonstrated to have sufficient bandwidth for data transmission up to hundreds of Mbps, while potentially having a small footprint with miniature components like VCSEL [28]. However, the mesoscopic structure of CSF and dura is too scattered for optical communication, so the propagation distance is limited. In addition, it is reported that the link quality of the optical communication has poor immunity to the horizontal misalignment (< 0.2 mm) [28].

Body channel communication (BCC) relies on the dielectric properties of the tissue to transmit electrical signals using the body as the communication medium, and galvanic coupling uses two electrodes, as the dipole coupler, to couple a differential signal to the tissue. The communication channel is typically operated in the reactive near-field, i.e., the propagation distance much shorter than the length of the dipole coupler and the signal wavelength. In such an operation region, a dipole is not radiative and its electrical field strength decays cubically with the distance (1/d³). In this case, the dipole becomes more like a coupler, and its length is no longer constrained by the wavelength of the signal. The dipole coupler of the transmitter generates a primary current flow between two electrodes of the coupler, and a secondary current flow further away from the coupler. The secondary current flow induces a voltage difference between two electrodes of the dipole coupler of the receiver.

Galvanic-coupled body channel communications (GC-BCCs) reported in the literature focus primarily on-body to on-body [29]-[32] or on-body to implant [33],[34] communication using skin-attached electrodes with relatively large dimensions, which does not well represent the channel properties of implant-to-implant trans-dural communication in this work. Although galvanic coupling has been employed for implant-to-implant intra-cardiac communication [35], it only demonstrates an operation frequency below MHz for the ECG application. It is unclear whether such an implant-to-implant channel has sufficient bandwidth to transfer data up to 100’s Mbps for the target application. In this work, we aim to explore the bandwidth boundary of galvanic-coupled body channel communications. Furthermore, we also investigated the impact of misalignments caused by the micromotion of the brain relative to the skull [2]. Finally, we designed a miniature CMOS GC-BCC transmitter IC employing a proposed “digital-impulse” modulation and validated the trans-dural data link with a biological tissue with the similar dielectric property.

The overall system concept is illustrated in Fig. 2. The proposed digital-impulse galvanic coupled BCC is designed with the goal of transferring the data collected by the MEA neural sensor across the CSF and dura tissue, with high data rate. By leveraging broadband nature of the trans-dural channel identified in this work, the neural data is encoded and modulated by an Encoding and Pulse Modulation Circuit (EPMC) to return-to-zero impulses with only nano-seconds pulse width. This not only greatly improve energy consumption, but also minimizes the charge accumulation at the electrode-tissue interface [36]. The binary data is represented by the polarity of the impulse. Both the transmitter and the receiver employ a dipole coupler to generate and sense, respectively, a confined electrical field. The diameter of each circular shape electrode of the dipole coupler is designed to be 0.55-1 mm, and the inter-electrode distance (center-to-center) is 5.5-10 mm, such that the length of the dipole coupler can be fit within the burr hole. The choice of electrode dimension and the inter-electrode distance (i.e., the length of the dipole coupler) is a trade-off between implant area, path gain, and the H-PA’s power consumption. A shorter length of the dipole coupler has more current flowing in between two electrodes of the transmitter’s dipole coupler, thus resulting in a lower path gain and higher H-PA power consumption.

Fig. 2.

Concept of the proposed trans-dural digital-impulse galvanic-coupled body-channel communication link. The transmitter is positioned under the dura and near the MEA, it receives the acquired data acquired by the probe. The data is modulated and sent through the GC-BCC to minimize power consumption. This minimizes the power consumption and increases the operation time of the deep-implanted device. The receiver, positioned under the skull, acquires the signal and transmits the data to an external unit through the UWB which can then be positioned at a greater distance.

The impulse is coupled to the trans-dural channel through the dipole coupler of the transmitter. The corresponding GC-BCC receiver is to be placed in the cranial implant, right above the sub-dural implant. Since the cranial implant has a relatively larger volume and surface area above the skull, its power consumption requirement can be relaxed, as long as it does not cause tissue heating. The received data can then be relayed and transmitted transcutaneously by e.g., an impulse-radio ultra-wideband (IR-UWB) we developed in [13].

To deliver power wirelessly from the cranial implant to the millimeter-scale sub-dural implant, the tradition inductive powering approach will be inefficient. Inductive powering can also potentially distort the signal from GC-BCC, due to the electrical coupling between them. To avoid such interference, powering solutions based on a non-electrical method can be adopted, such as optical [37] or ultrasound [38],[39]. Since optical and ultrasound waves have short wavelength, they have been demonstrated in millimeter-scale implants. This work focuses on the high-speed trans-dural data link, so the wireless powering systems is beyond the research scope.

The paper is organized as follows: Section II describes the numerical analysis of the body channel communication through the development of a finite element model. To verify the simulated model, Section III describes the setup and the result of the in-vitro channel characterization. In section IV, we analyze the effect of the misalignment on the BCC. In section V, the design of a CMOS transmitter IC is discussed in detail, and it is measured in an ex-vivo setup to validate the trans-dural data communication. Finally, the discussion and conclusions are provided in Section VI and VII, respectively.

II. Finite Element Modelling

To study the propagation properties of the trans-dural channel, a simplified human-head model composed of human-brain tissues, i.e., gray matter, cerebrospinal fluid (CSF), dura, and the skull, is developed in COMSOL™ Multiphysics simulator, as illustrated in Fig. 3(a). The model of a transmitter is positioned on top of the grey matter (i.e., the surface of the cortex), while the model of a receiver is positioned below the skull. The resulting transmission channel is therefore composed of a CSF layer and a dura layer. The frequency-dependent dielectric properties of each tissue composing the human-head model are based on the clinically validated Cole-Cole model reported in [40],[41] and the thickness of each tissue layer is listed in Fig. 3(a), according to [42]. Note that the thickness of CSF and dura has a large variation between 4 mm to 9 mm, depending on the location of the cortex, gender, the age of the subjects, etc.

Fig. 3. FEM model representing a human head and magnification on an implanted transceiver that uses GC-BCC for communication.

(a), the cross-section view of the electrical potential propagation through the trans-dural channel having a transmitter placed on the surface of the brain and a receiver placed above the dura. (b), the measured dielectric properties of the liquid phantom and dielectric properties of CSF, dura, and muscle tissue reported in the literature [40],[41]. The frequency band we are interested in is also indicated.

A. Finite element model

The simplified 3D model of the human head is developed in COMSOL™ Multiphysics software using the AC/DC electric current module. The simulation assumes the quasi-static approximation to compute Maxwell’s equation given the geometry of the model. This assumption is valid when the ratio between the channel length (L=10 mm) and wavelength (λ) is lower than 1/10 [17],[43]. For the model in analysis, a signal at the frequency of 300 MHz has a wavelength in the tissue of about 100 mm, which therefore corresponds to the maximum frequency. Two electrodes of the dipole coupler are modeled as two metal discs, which are then excited with a sinusoidal voltage signal of 1-V (peak-to-peak) amplitude.

We also assume that the data is DC-balanced with proper encoding, e.g., Manchester coding, so the low-frequency response of the medium below 50 MHz is neglected in the investigation. The space surrounding the head model is set as the ground to consider its electrical insulation [31].

The electrical potential distribution given by the simulation is shown in Fig. 3(a). The simulation results indicate that the frequency response of the transmission channel has a rather flat value of about -34 dB up to ~250 MHz frequency as shown in Fig. 6. This matches well with the calculated propagation gain based on the theoretical electric field distribution of the electric dipole in [29], which is approximately -38 dB. In addition, further simulations were carried out to verify the variation of the frequency response according to the type of tissue that makes up the transmission channel. These simulations were carried out maintaining the same geometries of the model and assigning to the channel the dielectric property of the biological tissue to be analyzed. The simulation results in Fig. 6 show that the path gain is relatively insensitive (< 1dB) to the dielectric property difference between tissues, e.g., CSF, dura, and muscle in the frequency band of interest (i.e., 50-1000 MHz). The simulated phase response versus frequency is also relatively constant with a small variation of ±0.1 rad.

Fig. 6.

Measured frequency response of galvanic coupling (GC) BCC obtained from the in-vitro measurement (yellow asterisk), and from the finite element simulation (solid lines) including phantom (red), CSF (blue), dura (green), and muscle (purple).

III. In-Vitro Channel Characterization

In order to verify the results obtained from the simulations, two custom boards were designed using discrete components. To model the transmission channel, a liquid phantom that has similar dielectric properties to a trans-dural channel was produced.

A. Electronic Module

The channel response is characterized with commercially available Voltage Controlled Oscillators (VCOs), MAX2606, 2607, 2608 from Maxim Integrated as the signal source, as shown in Fig. 4(a). To cover a wide tuning range, each VCO covers a fraction of the frequency ranges, from 70 MHz to 500 MHz, with an amplitude of 0.6 V (peak-to-peak). The input tuning voltage of each VCO, i.e., a DC control voltage (Vtune), is provided by a low-noise power supply Toellner TOE8733, which also provides the 3.3 V supply voltage. The inductance value for the VCO’s resonator is selected according to the operating frequency. Capacitors C1 and C2 are required for DC-blocking.

Fig. 4. Schematic of the electronic module.

(a), the schematic of the differential VCO. The output frequency of the VCO is controlled by the voltage at Vtune. The DC component is blocked by the series capacitors. The values for the inductance and the capacitance used are also indicated in the corresponding model of VCO. (b), the schematic of the receiver consists of two operational amplifiers (OPA855) in a non-inverting configuration. Both the VCO’s and the receiver’s electrodes have a diameter of 1 mm and an inter-electrode distance of 10 mm.

The receiver consists of two commercially available operational amplifiers (OPAMPs) in a non-inverting configuration, as shown in Fig. 4(b). The OPAMPs are OPA855 from Texas Instruments, which are chosen for their 8-GHz gain-bandwidth product. They have a high input impedance of 5 KΩ and a low input capacitance of 0.2 pF. Note that a high input impedance and low input capacitance minimize the path loss due to loading effects. In this channel characterization, the electrodes of the dipole couplers have a diameter of 1 mm and an inter-electrode distance of 10 mm, and the material of the electrodes is copper. Since the primary objective of this work is to demonstrate the feasibility of galvanic coupling as a high-speed implant-to-implant communication method, the PCB, except the dipole couplers, is insulated with epoxy to ensure it is watertight. It can be assumed that no wave propagation and reflection occur between the dipole coupler and the transmitter or the receiver, so impedance matching is not required.

To calibrate the path gain of the whole measurement setup, a reference module is implemented, which directly connects the VCO output to the receiver input without dipole couplers. This reference setup serves the purpose of de-embedding the effects of the instruments, connection cables, and the electrical property of the VCO’s and receiver’s electronics.

B. Liquid Phantom and Testing Setup

A phantom that mimics the dielectric properties of the tissue is developed for the in-vitro channel characterization. The liquid phantom is built by adjusting the salt and sugar ratio dissolved in the water according to [44]. The phantom was built to match as close as possible to the dielectric property of CSF and dura reported from the literature [40],[41] within the frequency range of interest. The measured dielectric properties (both permittivity and conductivity) of the phantom are shown in Fig. 3(b), which also includes data on CSF and dura reported in the literature.

The measurement was carried out with the spectrum analyzer P9375A in combination with the dielectric probe 85070E from Keysight Technologies. The liquid phantom also allows a flexible adjustment of the position of the VCO and receiver. The phantom is contained inside a transparent watertight container, and a 3D printed structure holds the devices for stable testing, as shown in Fig. 5. The lower frame holds the VCO PCB, while the top frame holds the receiver PCB and allows for easy rotation and horizontal movement.

Fig. 5.

In-vitro characterization setup. The VCO and the receiver are hold in place by an 3D printed structure contained inside a plastic box. The container is filled with the liquid phantom.

Ground isolation and cable radiation are carefully considered when performing the channel characterization of the galvanic coupling [29],[45]. Unwanted ground coupling can produce a ground loop between the VCO and receiver, and it can lead to the erroneous characterization of the path loss. For this reason, an isolation transformer, REOMED 1000, provides isolated supplies and grounds for the VCO and the receiver. Furthermore, long signal cables can introduce radiation, coupling from unwanted sources, as well as signal attenuation [45]. Similar to the approach presented in [32],[35],[45],[46], we use differential signals at both transmitter output and also receiver input/output to minimize the impact in case of ground coupling. The choice of implementing a VCO transmitter rather than using a signal generator is to avoid as much as possible such couplings between signal cables from the transmitter’s input and the receiver’s output. The connection wires of the transmitter input contain only DC signals for controlling the oscillation frequency of the VCO. Hence, it will not lead to wrong channel characterization even if the coupling between cables exists.

C. In-Vitro Channel Frequency Response

The frequency response of the channel composed of the liquid phantom is illustrated in Fig. 6 (yellow points). It is then compared with the result obtained from the simulation by defining the dielectric properties of the channel with those measured with the phantom (red line). Furthermore, the results of the simulations for different biological tissues are shown for comparison in the same figure. Due to the limited frequency control range of the VCOs, only discrete points within the frequency range of interest are measured.

The results of in-vitro channel characterization match well with the developed finite element model, with a difference below ±2 dB across the interested frequency range. The results also show a rather flat frequency response from 50 MHz to 250 MHz frequency range. The path gain, defined as G=V_RX / V_TX, where V_RX is the signal voltage at the receiver and V_TX is the transmitted signal, is approximately -34 dB with a slightly lower value at the higher frequency.

IV. Influence Of Misalignment

A galvanic coupled BCC with a mm-scale dipole coupler can be influenced by misalignments, especially due to brain micromotion which can be up to ±4 mm [47]. In this work, we investigated the effect of the two types of misalignments: orientational and horizontal.

A. Horizontal Misalignment

Investigations for horizontal misalignment are performed in three directions: x, y, and x+y, as illustrated in Fig. 8(a).

Fig. 8. Misalignment characterization.

(a), the normalized path gain measurement results for horizontal misalignment in x, y, and x+y directions with shift of a 5 mm and 10 mm. (b), the normalized path gain measurement results for orientational misalignment for two different frequencies with progressive rotations of 10 degrees.

Measurement of the path gain is taken for 5 mm and 10 mm horizontal shifts in three directions. The measurements were performed at the frequency of 200 MHz.

For either x or y directions, the path gain decrease is ~1 dB/mm. For the horizontal misalignment in the x+y direction (i.e., moving toward the direction of 45 degrees), the result shows a lower path gain compared to the pure x or y shift. The path gain decreases up to 20.6 dB for a 10 mm misalignment (i.e., ~2 dB/mm). The reason for this characteristic can be explained by the pattern of the equipotential line on the receiving plane, i.e., 10 mm above the plane of the transmitter’s dipole coupler, as shown in Fig 7. As gradually moving the receiver’s dipole coupler in the x direction, the potential of one electrode of the receiver dipole coupler increases initially and then decreases, while the potential of another electrode continuously decreases. For movements in the y direction, the potential of both electrodes of the pair continuously decreases. Hence, the path loss is less sensitive to the x direction (blue line in Fig. 8(a)) than to the y direction (green line in Fig. 8(a)) for the length of the shifts approximately equal to the length of the coupler (i.e., 10 mm). Extra path loss due to the movement in the x+y direction is approximately a superposition of the loss in both x and y directions.

Fig. 7.

Equipotential lines distribution at the receiver plane, i.e., 10 mm from the transmitter. The positions of the transmitter’s dipole coupler is also indicated in the figure.

B. Orientational Misalignment

An investigation for orientational misalignment is performed by setting the VCO at two frequencies, 100 and 200 MHz. A measurement of the path gain is taken every 10 degrees of rotation between 0 and 180 degrees. The results are shown in Fig. 8(b). For small rotation angle up to ±40 degrees, the path gain degradation is limited to -3 dB. For larger angular rotation, the degradation in the path gain increases rapidly. For the worst case of 90-degree misalignment, the degradation is up to approximately 20 dB.

Note that brain micromotion typically does not include significant orientational movements. The static orientational misalignment is only introduced during the implantation, so it can be removed with one-time calibration, e.g., by visually aligning or by measuring the amplitude.

V. Cmos Digital-Impulse Gc-Bcc Ic And Ex-Vivo Validation

The possibility of using the GC-BCC as a method of transmitting impulsive signals has been validated with a CMOS ASIC with the biological tissue having similar dielectric properties as dura and CSF. This chip was fabricated in 28-nm CMOS technology and can generate pulses in the order of ns. The verification was carried out using porcine muscle tissue, which, as shown in Section III, has dielectric properties very similar to trans-dural tissues in the frequency range of interest.

A. High-Speed Digital Impulse Transmitter ASIC

A transmitter ASIC is designed to provide high-speed digital-impulse modulation and occupies a core silicon area of only 190×180 μm². The transmitter module area, including the ASIC and the dipole coupler, is within 4×6.5 mm², as shown in Fig. 9(a).

Fig. 9. GC-BCC digital-impulse transmitter ASIC and module.

(a), the PCB with the transmitter ASIC, and the zoomed-in area of the transmitter. The transmitter module area is 26 mm². (b), the schematic of the encoding and pulse modulation circuits (EPMC). (c), the schematic of the H-bridge power amplifier (H-PA). (d), the reset circuit. At the reset signal, the two electrodes are connected together allowing the redistribution of the charge accumulated on the electrode-tissue interface.

This ASIC composes of a 9-bit pseudo-random bit sequence generator (9b-PRBS), an encoding and pulse modulation circuit, an H-bridge power amplifier (H-PA), and a reset circuit, as shown in Fig. 9 (b), (c), and (d). The 9b-PRBS generates a pseudo-random bit stream to simulate the recorded neural sensor data at 250 Mbps, 4 ns period. The data is encoded to a return-to-zero ternary data and modulated to biphasic impulses by a pulse modulation circuit. Positive and negative pulse width can be adjusted independently between 1-3 ns to test the frequency response and the impact of the charge imbalance [48]. The data is then driven from the dipole coupler to the tissue by the H-PA as shown in Fig. 9(c). After each impulse, the reset circuit, shown in Fig. 9(d), discharges the residual charge accumulated at the electrode-tissue interface to maintain the charge balance.

Fig. 9(b) shows the encoding and pulse modulation circuit (EPMC). The clock signal and its delayed version control the “Set” and “Reset” input of the SR flip-flop. While the data signal controls the “Enable” input of the SR flip-flop. By adjusting the delay generator, the pulse width of the output signal can be adjusted. According to the input data, either the Up and $\overline{Up}$ switches are closed or the Dn and $\overline{Dn}$ switches are closed at the same time. After each pulse, the reset circuit connects the two electrodes and discharges the residual charge at the electrode-tissue interface. The output of the self-biased inverter in the reset circuit (Fig. 9(d)) defines the reset voltage, approximately half VDD. The reset time is primarily determined by the strength of the reset circuit and the tissue impedance. The double-layer capacitor from the electrode-tissue interface affects the reset time. A smaller dipole coupler and its electrodes have lower double-layer capacitance, and thus a shorter reset time. The sizing of the reset switch also affects the achievable speed. Smaller reset switch has longer reset time, thus leading to longer symbol period. A timing control circuit is implemented in EPMC to ensure that the timing between, Up, $\overline{Up}$, Dn and $\overline{Dn}$ signals do not overlap, avoiding a low impedance path between supply and ground.

B. Ex-vivo Validation with Biological Tissue

To further validate the actual signal propagation property with biological tissue under heterogeneous conditions, the body-channel communication was performed with porcine muscle with a thickness of ~10 mm, as illustrated in Fig. 10(a). As shown in Fig. 10(b), a motherboard connected to an FPGA and a notebook is used as an interface to control the ASIC and to provide the required clock. The connection between the motherboard and the ASIC is through a flexible cable. The data is generated by the on-chip PRBS to avoid possible coupling between connection cables due to the high-speed data link, while the configuration signal for the ASIC is provided by a notebook through a serial-peripheral interface (SPI) for one-time configuration. The receiver is connected to an oscilloscope Lecroy SDA816Zi with 40 Gsps sampling speed. The transmitter ASIC is configured to send 1-ns return-to-zero pulse signal. During the idle state, i.e., when returning to zero, the reset circuit shorts two electrodes to allow charge redistribution and to remove the residual charge at the electrode-tissue interface [48]. The encapsulation was done with non-conductive epoxy glue, and the boards are fully coated except for the dipole coupler.

Fig. 10. Ex-vivo validation setup.

(a), the representation of the setup per ex-vivo validation. (b), the measurement setup with the ASIC transmitter and porcine muscle.

Fig. 11 (a) shows the partition of the average energy consumed by the ASIC. Approximately 25 % of the energy is consumed by the encoding and pulse modulation circuits, while the remaining 75 % is used by the H-PA and the reset circuit.

Fig. 11. Results of the ex-vivo validation.

(a), the transmitter ASIC power distribution chart. (b), the measured eye diagram of the received signal.

The measured eye diagram at the receiver output is shown in Fig. 11(b). The eye amplitude is approximately 6 mV, and the transmitter output voltage is ~0.9 V, so the path gain is 43.5 dB. This matches well with the simulation based on the finite element model presented in Section II-A. Compared to the channel characterization shown in Fig. 6, there is approximately 10 dB worse path loss. This is mainly because the length of dipole coupler used in the ex-vivo validation is smaller than the one used in channel characterization (5.5 mm versus 10 mm). The eye diagram has an impulse and symbol period of 1 ns and 4 ns, respectively, with 0.8 ns of impulse settling time and 1.8 ns of reset time. A signal at the receiver output with ~2 Mbit long was recorded. The bit-error rate (BER) of 2.4 × 10⁻⁶ was calculated by comparing the recorded data with the original transmitted PRBS data. To further lower the BER or to increase the communication distance (i.e., the thickness of the tissue), the pulse time can be further increased to increase the energy per bit transmitted.

VI. Discussion

The in-vitro characterization of the frequency response in liquid phantom shows a good agreement with the finite element model, as shown in Fig. 6. It is observed that the amplitude response starts to “ring” from above ~300 MHz, and the difference compared to the quasi-static finite element model is gradually increased. This indicates that the dipole coupler starts entering the transition zone between near and far fields, i.e., mid-field. The ex-vivo validation with porcine muscle confirms that the wide frequency response of the channel can be exploited to directly transmit high-speed data up to 250 Mbps. Signals with higher data rates can start becoming radiative, which has large amplitude fluctuation and is less efficient to couple.

Horizontal brain micromotion can reach up to ±4 mm when the head undergoes high accelerations, so the worst-case path gain degradation for the proposed BCC due to horizontal movement is expected to be less than 5 dB, according to the characterization in Fig. 8(a). Even with the smaller dipole coupler used in ex-vivo (with a length of 5.5 mm), the path degradation is expected to be less than 6 dB. In comparison to other in-body propagation methods, such as inductive coupling or optical, GC-BCC has better immunity to horizontal misalignment. Under similar coupler dimension and channel distance, optical transmission can only tolerate the horizontal misalignment below 1 mm [28], and the inductive coupling would also suffer from significant attenuation due to the degraded coupling coefficient.

In Table I the presented impulse-based galvanic coupled BCC is benchmarked with state-of-the-art high-speed (>200 Mbps) and miniature implantable transmitters (Tx), including inductive coupling, optical and Impulse-radio UWB (IR-UWB). The proposed GC-BCC demonstrates a data rate up to 250 Mbps with a small electronic area of 26 mm², while most state-of-the-art GC-BCC communication [34],[35],[49] has limited bandwidth up to 10’s MHz. Inductive coupled communication can achieve a higher data rate with a “de-Q” technique [23], but it suffers from high path loss and still requires a relatively large coil area. IR-UWB transmitters also require a large antenna to match the wavelength of the transmitted signals, which is difficult to reduce the area below 100 mm². Optical communication offers a high data rate but consumes high power and is sensitive to horizontal misalignment.

Table I. Comparison With State-Of-The-Art High-Speed Implantable Transmitters.

	This work	[23] TBIOCAS ’ 19	[28] TBIOCAS ’ 19	[13] ISSC’ 21
Propagation method	Digital-pulse GC-BCC	Inductive Coupling	Optical	IR-UWB
Data rate (Mbps)	>250	200	300	1660
Tx area (mm 2)	26	100 (coil area)	N.A. (discrete comp.)	105
Misalignment tolerance	> 4 mm	N.A.	0.2 mm	N.A.
Tissue thickness	10 mm (muscle)	11 mm (sku11+skin)	4,25 mm (skin)	15 mm (skin)
Tx Power cons.	0.50 mW	0.30 mW	11 mW	9,69 mW
Tx Energy Efficiency	2 pJ/bit	1,5 pJ/bit	37 pJ/bit	5,8 pJ/bit

VII. Conclusions

We present an untethered trans-dural data transfer method based on digital-impulse galvanic coupled body channel communication for high-bandwidth intra-cortical neural sensing. We have demonstrated that the bandwidth-volume-pathloss constraints in the conventional in-body far-field (e.g., EM-radiation, optical) or near-field (e.g., inductive coupling) propagations can be overcome by using impulse-based galvanic coupling.

A liquid phantom mimicking the biological tissue and galvanic coupled devices were developed to characterize the channel response and the impact of misalignments. It shows that the galvanic-coupled BCC can tolerate the horizontal misalignment up to ±4 mm, which corresponds to the reported relative motion range of the brain.

A CMOS galvanic-coupled impulse-based transmitter ASIC is designed, and the transmitter module including the dipole coupler occupies a small area of 26 mm². The transmitter device is validated ex-vivo with the 10 mm thick porcine muscle tissue. The work demonstrates a high-speed and miniature in-body galvanic-coupled communication, with a data rate up to 250 Mbps and an energy efficiency of 2 pJ/bit. In comparison to other in-body communication methods, e.g., inductive, ultrasound, EM radiation, or optical, the proposed digital-impulse GC-BCC offers several benefits simultaneously, including high data rate, low power consumption, miniature form factor, and misalignment tolerance, which serves as the perfect candidate for trans-dural data transfer of high-channel-count and free-floating intra-cortical neural sensors.