An Energy-Efficient and High-Data-Rate IR-UWB Transmitter for Intracortical Neural Sensing Interfaces

Abstract

This paper presents an implantable impulse-radio ultra-wideband (IR-UWB) wireless telemetry system for intracortical neural sensing interfaces. A 3-dimensional (3-D) hybrid impulse modulation that comprises phase shift keying (PSK), pulse position modulation (PPM) and pulse amplitude modulation (PAM) is proposed to increase modulation order without significantly increasing the demodulation requirement, thus leading to a high data rate of 1.66 Gbps and an increased air-transmission range. Operating in 6 – 9 GHz UWB band, the presented transmitter (TX) supports the proposed hybrid modulation with a high energy efficiency of 5.8 pJ/bit and modulation quality (EVM< -21 dB). A low-noise injection-locked ring oscillator supports 8-PSK with a phase error of 2.6°. A calibration free delay generator realizes a 4-PPM with only 115 μW and avoids potential cross-modulation between PPM and PSK. A switch-cap power amplifier with an asynchronous pulse-shaping performs 4-PAM with high energy efficiency and linearity. The TX is implemented in 28 nm CMOS technology, occupying 0.155mm² core area. The wireless module including a printed monopole antenna has a module area of only 1.05 cm². The transmitter consumes in total 9.7 mW when transmitting -41.3 dBm/MHz output power. The wireless telemetry module has been validated ex-vivo with a 15-mm multi-layer porcine tissue, and achieves a communication (air) distance up to 15 cm, leading to at least 16× improvement in distance-moralized energy efficiency of 45 pJ/bit/meter compared to state-of-the-art.

I. Introduction

Intra-cortical extracellular neural sensing is being widely adopted in neuroscience research as well as therapeutic brain-computer interfaces (BCIs), and the number of sensing channels continues to double every 6 years since 1960’s. Intra-cortical micro-electrode 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. A recent neuroscience research trend is to further distribute multiple high-density extracellular MEAs across the brain, as conceptually illustrated in Fig. 1(a), each with 1000’s of sensing channels, such that neuroscientists can begin to map the correlation of neuronal activity across different brain regions, with the single-neuron precision [4]–[6]. However, since each neural sensing channel typically samples at 20-50 kSps with a high-resolution (above 10-bit) analog-to-digital converter (ADC), multiple MEAs with total > 1000 recording channels demand a data transfer rate above 1 Gbps, including also communication and protocol overhead (e.g., error correction, etc.). Although many 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 perform data compression before wireless transmission [7]. High fidelity data are crucial for neuroscience research, and potential distortion due to compression is avoided.

Fig. 1. (a) Conceptual diagram of intra-cortical neural sensing and (b) the proposed prototype of the wireless IR-UWB wireless telemetry.

A few of the existing pre-clinical intra-cortical neural interface systems with one or multiple MEAs were validated on primates [8],[9], including a large head-mounted wireless telemetry module. However, patients with such modules suffer from huge risk due to physical accidents and infections. These neural sensing interfaces are severely hindered in many clinical uses due to the lack of a high-data-rate and miniature wireless telemetry solution that can be implanted below the scalp, i.e., transcutaneously, as shown in Fig. 1. To aim to a form factor similar to a cochlear implant, the area of such transcutaneous wireless telemetry module should be miniaturized to less than 5 cm² and the thickness to be less than 3 mm, due to neurosurgical implantation constraints. The transmission range up to 10 cm is highly desirable, in order to improve the reliability of the wireless link against e.g., antenna misalignment, signal attenuation from tissue, etc. Finally, the power consumption of the wireless telemetry should be limited to ~10 mW to minimize thermal flux from the module’s surface area, avoiding excessive tissue heating [10].

Most of the conventional transcutaneous wireless telemetry adopt inductive coupling [11]–[13], but the data rate is limited to a few Mbps, due to a high quality-factor resonance required for efficient magnetic induction. Inductive coupled communication can achieve a higher data rate with a “de-Q” technique [14], but this suffers from high path loss. Similarly, magnetoelectric [15] or ultrasound [16] propagation methods also do not provide sufficient bandwidth for high-bandwidth communication.

Body-channel communication (BCC) relies on the dielectric properties of the tissue to transmit electrical signals using the body as the communication medium, and prior arts in [17],[18] presented a date rate up to 16 Mbps. However, it has not been demonstrated in literature that such body channel has sufficient capacity to support data rate above 100 Mbps. Furthermore, the link quality is heavily influenced by electrode-skin contact conditions.

A near-infrared (NIR) optical transcutaneous transmitter (TX) using a vertical cavity surface emitting laser (VCSEL) [19],[20] demonstrated a data rate up to 300 Mbps but suffers from a limited transmission range (4 mm) and requires a sub-mm precise alignment between the implant TX and a wearable receiver (RX).

Impulse-radio Ultra-Wideband (IR-UWB) is promising to achieve high data rate above 100 Mbps [21]–[23]. Since IR-UWB transmitters operate at relatively high frequency, the antenna can also be implemented on chip, as an IR-UWB wireless system on chip (SoC) [24], which further miniatures the form factor to millimeter scale. However, to the best of the authors’ knowledge, none of IR-UWB TXs prior works demonstrate the scalability of the data rate beyond 1 Gbps, and Their communication range (between the implanted IR-UWB TX and the external RX) are also limited to below 1 cm.

This work presents a transcutaneous IR-UWB TX with a record high data rate, while achieving up to 10× longer transmission range and with performance well within the volumetric and heat dissipation constraints [25]. Three innovations are introduced. First, the proposed hybrid impulse modulation, combining M-array pulse position modulation (M-PPM), M-array phase shift keying (M-PSK) and M-array pulse amplitude modulation (M-PAM), to maximize the link budget. Second, a power-efficient delay generator for a calibration-free M-PPM. Third, a low-noise 8-phase ring oscillator (RO) with duty-cycle correction for an accurate M-PSK.

This paper is organized as follows. Section II introduces design consideration for the wireless telemetry in system level, including design challenges on high-data rate IR-UWB transceivers are explained in Section III. After discussing the proposed hybrid modulation, detailed circuit implementation will be described in Section III. Section IV explains the measurement results. Finally, Section V concludes this work.

II. Wireless System Considerations

The wireless communication system for the implantable neural sensing interfaces should be able to transfer data at high rates, even with high tissue loss. Narrow-band wireless solutions, e.g., Bluetooth [26] or Medical Device Radiocommunications Service (MedRadio) [27] can mitigate the tissue loss, but the data throughput is limited to few Mbps. IR-UWB solutions [21]–[24] are good candidate for such applications, as they possess a wide transmission bandwidth (> 500 MHz), e.g., 3.1 to 10.6 GHz band allocated in Federal Communication Commission (FCC) [28]. However, owing to its low effective isotropic radiated power (EIRP) limit (below - 41.3 dBm/MHz), the UWB radio has limited link margin. The low band of UWB, e. g., 3 to 5 GHz, can be chosen for lower the tissue loss [22], [24], but this band is crowded with or adjacent to many popular wireless communication protocols, e. g., Wi-Fi, 5G-New Radio (5G-NR), etc. To address aforementioned issues, this work adopts the higher UWB band, e. g., 6 – 9 GHz.

A. Implantable UWB Antenna and Path Loss

To accommodate the neurosurgical constrains, an antenna and a UWB transmitter integrated circuit (IC) should be co-designed [29], [30]. The antenna design is targeted to have a thickness below 3 mm, and an area smaller than 1 cm², while supporting the higher band of UWB. We opt for a classical printed, circular arm, monopole antenna [31], as illustrated in Fig. 2. At the bottom of the higher UWB frequency band, the antenna behaves as a resonant monopole antenna. At the higher frequencies, the circular arm and ground planes form a dual tapered slot antenna. Thus, impedance matching can be realized over a wide bandwidth. The dimensions have been tuned for operation in the human head as the envisaged scenario. For the design, a planar layered model based on [32] has been applied as well as a voxel head model, shown in Fig. 3. As shown in Fig. 4, the simulated reflection coefficient (S₁₁) is below -10 dB over the target frequency range. The simulated path loss with the envisaged scenario at 6.5-7 GHz varies between 36 to 58 dB, which is heavily influenced by the location, relative distance and the orientation of two antennas. The realized gain function patterns for 6 GHz, 7.5 GHz and 9 GHz are shown in Fig. 5. The simulated gains for these frequencies are -8.97 dBi, -6.01 dBi and -7.65 dBi, respectively.

Fig. 2.

Printed UWB monopole antenna. The antenna is printed on a 0.5 mm thick FR4 dielectric slab (ε_r = 4.28, tanδ = 0.012). The monopole arm is on the top, the ground plane is on the bottom of the printed circuit board (PCB).

Fig. 3. Antenna in its simulation environment.

Fig. 4. Simulated reflection coefficient as a function of frequency for the antenna.

Fig. 5. Realized gain patterns for: Left: 6.0 GHz (G = -8.97 dBi), Middle: 7.5 GHz (G = -6.01 dBi), Right: 9.0 GHz (G = -7.65 dBi).

The path loss is characterized ex vivo to establish a solid link budget analysis, as shown in Fig. 6. The designed antenna is inserted in a 15-mm thick, multi-layer porcine tissue, and is placed at various distances away from an UWB RX. The ex-vivo setup will be detailed in section IV.B.

Fig. 6. Ex vivo path loss measurement setup for both path loss and wireless measurement.

The characterized path loss which comprises tissue loss, free space loss, and antenna gain, at 6.7 GHz versus different transmission range is shown in Fig 7. The path loss measured at a (air-) distance of 10 cm at 6.7 GHz is 52 dB._

Fig. 7. Measured path loss result at variable distances and the tissue thickness of 15 mm at 6.7 GHz.

B. High-Data-Rate UWB Wireless Telemetry

High data rate IR-UWB above 100’s of Mbps can be achieved by either increasing the symbol rate [33] or using multi-band operation [34], [35]. However, these approaches significantly increase power consumption, especially in digital baseband (DBB). Increasing the modulation order [21], [36] is more energy-efficient way to increase the data rate without significantly increasing the power consumption of the TX, although this can increase E_S/N₀ requirement [37] for demodulation.

M-PPM [38] is promising for high data rate and energy-efficient IR-UWB systems as it simply splits more time slots within one symbol period. For instance, at the symbol rate of 250 MHz (i.e., symbol period 4 ns) and a pulse width of 2 ns, a binary PPM can be performed. To further increase the modulation order, sub-pulse width PPM solutions have been presented in [21]. Digitalized multi-pulse-position-modulation (D-MPPM) [36] can further increase the channel capacity by modulating the time difference between multiple consecutive pulses. With a fine time step of ~50 ps, it is reported that the D-MPPM can achieve 1.125 Gbps data rate. However, based on Euclidean distance, high E_S/N₀ is required to discriminate between two adjacent pulse positions in sub-pulse width PPM as the modulation order increases [37]–[39]. With a 62.5 ps time step, the required E_S/N₀ for the symbol error rate (SER) of 10^-4 is 37 dB. Adopting M-PPM with fine time step is not feasible in the targeted scenarios where the miniature antenna and tissue at UWB frequency can introduce up to 52 dB of path loss, as estimated in section II.A.

C. Proposed Hybrid Impulse Modulation

To overcome the high path loss and maximize the communication range, we propose a 3-dimensional (3-D) hybrid impulse modulation scheme (PAM-PSK-PPM) as shown in Fig. 8. This approach achieves the high data rate by increasing the modulation order, without significantly increasing power consumption of the transmitter.

Fig. 8. The proposed 3-D hybrid modulation: (a) timing diagram and (b) 3-D constellation diagram in case of 4PPM-8PSK-4PAM.

To achieve a data rate higher than 1 Gbps, this work aims to have a modulation order up to 7 (i.e., M= 128). Fig. 9(a) shows a computed E_S/N₀ requirement (according to theoretical Euclidean distance [37]) that can achieve a SER below 10^-4 versus different modulations and orders. As shown in Fig. 9(a), PAM and PPM require a higher RX demodulation requirement. In addition, generating PPM and PAM in the TX with finer resolution typically require higher hardware complexity, e.g., in delay and amplitude control, while generating more phases from the LO signal for PSK is relatively easy, e.g., with ring oscillator as proposed in this work. Hence, this work employs a combination of 4-PAM, 4-PAM and 8-PSK that optimize between the E_S/N₀ requirement and TX power consumption.

Fig. 9. (a) Computed E_S/N₀ requirements for SER<10^-4 of various modulations, according to Euclidean distance, (b) eye diagram illustration of the impact of PSK/PPM-to-PAM and PSK/PAM-to-PPM cross modulations.

This hybrid modulation also reduces the E_S/N₀ requirement by 12 – 25 dB compared to other standalone high-order modulations, e.g., M-PSK, M-PPM or M-PAM.

Due to the multi-dimensionality of the proposed modulation scheme, the modulation order has been increased while the Euclidean distance between symbols is not drastically reduced. Hence. The E_S/N₀ requirement is more relaxed, e.g., 27 dB for hybrid modulation order of 7 versus 50 dB for e.g. 128-PPM. M-QAM is an alternative modulation that has similar E_S/N₀ demodulation requirement as the proposed hybrid impulse modulation (Fig. 9(a)), but its transmitter implementation is very power consuming when combining with impulse modulation since IQ-based up-conversion architecture is needed (to be discussed in section III).

Since highest achievable E_S/N₀ on the RX will be limited by the error vector magnitude (EVM) from the TX, i.e., E_S/N₀ ≤ BW·T_SYMBOL/EVM² where BW is the signal bandwidth, all possible source of EVM degradation from the TX should be avoided. To ensure a sufficient E_S/N₀ to achieve an SER below 10^-4, the EVM of TX should be better than the minimum E_S/N₀ requirement. Assuming Bandwidth is 1 GHz and T_SYMBOL is 4.2 ns, the TX EVM requirement for the TX to support the target hybrid modulation (4-PAM, 8-PSK and 4-PAM) is calculated to be -21 dB.

There are several possible sources of the EVM degradation in the proposed IR-UWB TX with the hybrid modulation: cross-modulation, phase error and inter-symbol interference (ISI). One common cross-modulation is between phase and amplitude of the carrier, i.e., PSK and PAM, as shown in Fig. 9(b). It is also possible to have the cross-modulation between PSK and PPM, since both are modulated in the time domain. In this work, the potential cross-modulation between PPM and PSK is avoided by implementing the step of PPM to be multiple integers of the carrier period.

Phase error from the multi-phase LO is another source of EVM degradation, and this can come from either LO phase noise or (multi-)phase accuracy. The EVM of the transmit symbol with the root mean square (RMS) phase error (ψ_RMS,DCO) can be derived by the following equation [37], [40]

$${\phi }_{RMS}<EV{M}_{TX}\underset{˜}{<}{\left(\sqrt{\frac{E_{S,req}}{N_0}\ast \frac{1}{T_{SYMBOL}\cdot BW}}\right)}^{-1}$$ (3)

The RMS phase error should be less than 5° to achieve the target EVM of -21 dB.

Last but not least, to minimize the potential EVM degradation due to the ISI, the PPM range and T_PULSE is designed to be only half of the symbol period (T_SYMBOL).

D. Link Budget Analysis

The link budget can be calculated (in dB) as

$$\begin{array}{c}\frac{E_S}{N_0}+NF+N_0=\\ P_{Pulse}+10\cdot \log \left(T_{PULSE}\right)+10\cdot \log \left(1/3\right)+G_R+PL\end{array}$$ (2)

where E_S/N₀, NF, N₀ and PL are the required E_S/N₀ for achieving a SER below 10^-4, the noise figure, the noise spectral density and the path loss including TX antenna gain, respectively. This work aims to achieve a modulation order of 7, so the E_S/N₀ requirement is calculated to be 27 dB for the hybrid modulation of 4-PPM, 8-PSK and 4-PAM (Sec. II. C). If the pulse shape is triangular, as designed in this work, the average power of each pulse (P_Pulse) is reduced to one third, compared to a square pulse. Assuming the NF of the RX is 2.5 dB and Gr is 7 dBi, the required P_Pulse should be higher than -8.1 dBm, to achieve an air distance longer than 10 cm. Assuming T_Symbol is 4.2 ns, i.e., a duty cycle of 47.6%, the average output power will be – 11.3 dBm.

Since the 4-PAM has a PAPR of 3.3 dB, the required peak pulse power should be at least -4.8 dBm, Then, the required PA peak power is designed to be ~0 dBm, which can still be implemented with a supply voltage of 0.9 V in this work.

III. Transceiver Architecture and Circuit Implementations

One key challenge to realize the proposed 3-D hybrid impulse modulation is to ensure a high level of independence among different sub-modulations with low power overhead when performed simultaneously. Prior IR-UWB TXs perform the impulse modulation with either carrier-less [21], [22], [36] or up-conversion approaches [23], [34], [35]. The carrier-less TXs use edge-combining which leverage the fast-switching digital logic in nanoscale CMOS, thus power-efficient and requiring no dedicated LO. However, simultaneously performing M-PPM and M-PSK in carrier-less TXs is challenging, since both phase and time need to be modulated via delay edges. The classical IQ-based up-conversion method can perform the high order modulation, but limit to achieving high energy efficiency. The polar-based up-conversion TX with asynchronous pulse only BPSK shaping [23] consumes low power, but it is limited support.

Fig. 10 shows the proposed low power polar-based IR-UWB TX capable of performing the proposed 3-D hybrid impulse modulation. The amplitude and the phase modulation can be performed independently by the polar architecture. Unlike the carrier-less topology, the proposed TX modulates the pulse delay independently apart from the carrier phase, as conceptually illustrated.

Fig. 10. Block diagram of the proposed UWB transmitter.

The impulse waveform is shaped by the digitally controlled power amplifier (DPA) and a pulse shaper (PS) [41] which uses eight delay cells to perform FIR filtering in RF domain. The output of each delay cell enables eight PA cells. The shape and the width of the impulse can be adjusted by the delay of the PS output. An injection-locked ring based digitally controlled oscillator (DCO) is adopted to provide low jitter, 8 phases for 8-PSK in a wide frequency range. 7-b 238 Msym/s digital data is distributed to three different modulation paths, PAM, PSK, and PPM, after being synchronized with a 476 MHz system clock (SYS_CLK). The digital PA with 32-unit cells supports up to 4 -PAM. The 8-PSK is performed by selecting one of 8 phases from the ILRO using a phase selector (PHMUX).

By accurately delaying the impulse with a PPM control (PPM_CTRL), precise 4-PPM can be supported. Fig. 11 illustrates the timing diagram of the proposed 3-D hybrid impulse modulation with the proposed digital polar-based IR-UWB TX. To trigger an impulse, a 25% duty-cycled unit pulse (UNIT_PUL) at 238 MHz is generated from the SYS_CLK. The PPM controller (PPM_CTRL) modulates the delay of the UNIT_PUL according to the PPM data, and its output (PPM_OUT) is then fed to the PS to synthesize a triangularshaped impulse.

Fig. 11. Timing diagram of the 3-D hybrid impulse modulation with the proposed IR-UWB TX.

T_PULSE is set to 2 ns by the PS, and T_SYMBOL is set by two cycles of the SYS_CLK to be 4.2 ns. This ensures a sufficient PPM range (T_PPM) to perform 4-PPM and avoids potential impact from ISI or multi-path, as discussed in Sec. II. D. To avoid the cross-modulation between PPM and PSK, the carrier phase is set at the beginning of each impulse symbol.

A. The proposed negative-skewed DCO

To perform the 8-PSK modulation, the DCO needs to provide the low-phase error, 8 phases output at 6-9 GHz frequency. The RMS phase error in this work is aimed to be less than 5°. The LC oscillator is a good candidate for the low phase error, but hard to generate multi-phase outputs with a wide frequency range. Ring oscillators (ROs) operate in a wide frequency range and can provide the multi-phase inherently from the output of delay cells. However, since the multi-stages of delay cell for multi-phase generation limits the operating frequency, achieving 8 phases at 6 - 9 GHz is still challenge in 28-nm CMOS technology. In addition, conventional ROs typically have poor phase noise, which is challenging to meet the target phase error requirement for 8-PSK modulation. As shown in Fig. 21(b), the measured phase error of a conventional free-running RO is 24°.

Fig. 21. Measured result of phase noise.

To overcome these limitations, an injection-locked negative skewing ring-based DCO is proposed in Fig. 12. The DCO frequency is digitally controlled by the current bank. A negative skewing technique [42], [43] is adopted for reducing the delay of each stage. In each delay cell, its PMOS (IPP/IPN) input is forwarded to the NMOS (INP/INN) one stage ahead, resulting in a faster transition and lower phase noise. The negative skewed delay cell can increase the DCO operating frequency by 30% compared to the conventional delay cell as shown in Fig. 13. To achieve the wide tuning range as well as the fine resolution, the frequency of DCO is tuned by three steps with different resolutions: coarse, medium, and fine. To avoid the quasi-lock and minimize the spur from the output spectrum, during injection-locking [44], a small DCO frequency step of 6 kHz/LSB is chosen. However, this approach increases the duty cycle error due to the unequal behavior at rise/fall transition.

Fig. 12. (a) Block diagram of the proposed DCO with the negative-skewed delay cell and the proposed DCO buffer and (b) block diagram of injection pulse generator.

Fig. 13. Simulation result of the operating frequency enhancement by using the negative skewed delay cell.

Injection-locking [44] is adopted to achieve the low phase error, after its frequency is corrected by the frequency detector and an external frequency locked loop (FLL), as illustrated in Fig. 10. The short pulse generated based on the reference clock is differentially injected into one of the output stages of the delay cell. Dummy switches are also added to other stages for output load balancing. The injection strength is controlled by the injection pulse width. The DCO is “re-synchronized” at every injection cycle, limiting the phase noise accumulation. A stronger injection sanitizes the DCO phase error further but causes more phase offset at the injection point, affecting the EVM. Fig. 14 shows the simulated phase offset with different injection pulse widths.

Fig. 14. Simulation results of phase offset and phase error with different injection pulse widths.

B. The proposed DCO Buffer with duty-cycle correction

Since the negative-skewed DCO output has intrinsic duty-cycle error and the amplitude is not rail-to-rail, the DCO buffer with a feedback duty cycle correction is proposed to boost the DCO output level to full swing with the duty-cycle corrected. It is embedded in each delay cell with low capacitive loading and recovers the amplitude disturbance due to injection-locking, critical for minimizing PSK-to-PAM cross-modulation.

As shown in Fig. 12, a low pass filter that comprises a transmission gate, a capacitor, and an inverter-based error amplifier extracts a common-mode voltage of DCO_OUT and controls the bias of the first stage of the buffer (V_DCC) to adjust the duty cycle to 50%. It can correct the duty cycle error within 5 ns, as shown in Fig. 15. Each buffer consumes only 120 μW.

Fig. 15. Simulation result of duty cycle correction of the DCO buffer.

C. Phase Selector

The phase selector is a differential 8-to-1 multiplexer (MUX) to select the phase from the DCO outputs to the differential DPA, for the 8-PSK modulation, as illustrated in Fig. 16. To reduce the output load at each input clock path, critical at high frequency, the MUX is two-staged, 4-to-1 and 2-to-1. As discussed in section II.C, the accuracy between phases also contribute to the EVM degradation. Hence, all the control inputs are retimed and all delays in each path are carefully matched to minimize the timing skew.

Fig. 16. Block diagram of the phase selector.

D. PPM Controller

A digital-to-time converter (DTC)-based PPM [36] is simple and low-power solution for high-speed communication. However, it is sensitive to the PVT variation and thus requires the complicated calibration. The delay line with the same delay tuning range (2 ns) consumes 250 μW in 28nm technology [41] and requires extra hardware complexity for the calibration. In addition, if the step of PPM is not an integer multiple of DCO period, potential cross-modulation between PPM and PSK can be introduced. Hence, a fully synchronized delay controller for PPM is proposed as shown in Fig. 17. The 25% duty-cycled unit pulse (UNIT_PUL) is sampled by the DCO output (DCO_OUT) divided by 4 and thus the delay step is the quadruple of DCO_OUT period (4×T_DCO). As long as the DCO is injection locked by the system clock, no extra delay calibration is required. The proposed PPM controller only dissipates 115 μW.

Fig. 17. Block diagram of the PPM controller.

E. Power Amplifier and Impedance Matching Network

A switched-capacitor PA (SCPA) [41] is exploited to achieve a high linearity and efficiency, as shown in Fig. 18. The SCPA comprises 32-unit PA cells to be controlled by the 8-level pulse shaping and the 4-PAM. The unit PA cell is dual capacitively coupled [45], to reduce the common-mode voltage fluctuation that induces the spectrum leakage at low frequency. A wideband tunable matching network is implemented on-chip. Such low output power and highly sliced SCPA prefers small unit capacitors (C_u), due to a limited driving capability at low output amplitude. However, a large inductance in the matching network is needed if C_u is small, so the signal bandwidth is also reduced. To balance between these two considerations, a C_u of 11 fF is chosen in this design.

Fig. 18. Block diagram of the power amplifier.

IV. Measurement Results

The proposed UWB TX IC was fabricated in 28-nm CMOS technology, occupying only 0.155 mm², including an on-chip antenna matching network as shown in Fig. 19. The wireless module implemented on an FR-4 printed circuit board (PCB) has a core electronic area of only 1.05 cm², including a printed circular monopole antenna sized 9.4 × 7 mm² and the TX IC as shown in Fig. 20. One of the main challenges of the application-specific integrated circuits (ASIC)-antenna integration is due to the stringent thickness and area requirements, i.e., less than 3 mm and 5 cm², for the implantation. On one hand, the printed antenna prefers a thick dielectric layer below to minimize its area. On the other hand, the 50-ohm transmission line between the ASIC and the antenna will become too wide if using the same thickness of the dielectric layer, making the whole module less compact. To provide a sufficient dielectric thickness for the antenna, multiple dielectric (FR-4 prepreg) layers are stacked and compressed without copper layers at the antenna part, while using only one prepreg layer under the transmission line, as shown in Fig. 20.

Fig. 19. Die micrograph.

Fig. 20. (a) Wireless module and (b) its vertical view.

A. Electrical Performance

Fig. 21 shows the measured phase noise. Thanks to the injection-locking, the measured phase error of the proposed DCO is reduced significantly from 24° to 2.6° at 6.7GHz, meeting the target for 8-PSK as discussed in Sec. III. A. Fig. 22 shows the measured peak power of the SCPA, which indicates the linearity requirement for 4-PAM modulation can be achieved. As shown in Fig. 23, the optimum impedance matching is at around 6.5 GHz and the 10-dB S11 bandwidth is more than 1 GHz, sufficient for a 2 ns impulse modulation. A wired measurement of impulse modulation in time domain, an “eye-diagram” of the impulse envelope and a constellation in complex plane of the hybrid modulation are shown in Fig. 24 and demonstrate a high modulation quality with an EVM of -21.8 dB. No ISI due to the limited bandwidth is observed from the time-domain modulation waveform as shown in Fig. 24.

Fig. 22. Measured results of PA peak power.

Fig. 23. Measured result of S₁₁.

Fig. 24. Wired measurement of (a) time-domain impulse modulation waveform, (b) constellation in complex plane, and (c) envelope eye diagram of the proposed hybrid impulse modulation.

The measured 3-D constellation is shown in Fig. 25, and the EVM is -21 dB. It can be observed that the main contribution of EVM comes from phase noise. Up to 0-dBm TX peak power, the measured output spectrum is -41.3 dBm/MHz, or -11.3 dBm, well matched with the PAPR described in Sec. II. D. Therefore, the measured output spectrum can comply with the FCC mask, even without the attenuation of the tissue. The power breakdown of the UWB TX is shown in Fig. 27. The measured power consumption is below 10 mW. The power consumption of the hybrid modulation part consumes only 15% of the total power.

Fig. 25. Measured 3-D constellation.

Fig. 27. Power breakdown.

B. Ex-vivo Wireless Link Measurements

To test the transmission range, the wireless measurement is setup like the ex-vivo path loss measurement setup as introduced in Sec. II.B. Fig. 28 shows the cross-section of the procaine tissue and the placement of the wireless TX module. The wireless module including the antenna is completely covered by a 15-mm thick, multi-layer porcine tissue. The thickness of each tissue layer is approximately 3-mm fat, 10-mm muscle and 2-mm skin. The frequency-dependent dielectric properties of each tissue are reported in [32] and [46], which is based on the clinically validated Cole-Cole model. The encapsulation was done with non-conductive epoxy glue, and the board is fully coated. Note typical human scalp thickness is approximately 5-8 mm [47], and it can significantly vary depending on the location of the scalp, gender, the age of the subjects, etc.

Fig. 28. Ex-vivo measurement setup and tissue cross-section.

The UWB RX is composed of a sinuous UWB antenna, a low noise amplifier (LNA) (Narda LNA-20-04001200-20-15P) and a high-speed oscilloscope (LeCroy SDA816Zi) as an analog-to-digital converter (ADC). The sampled data is demodulated numerically. The LNA has a noise figure and power gain of 1.8 dB and 24 dB, respectively, and the oscilloscope has a measured noise floor below -156 dBm/Hz. This corresponds to an equivalent total RX noise figure of approximately 2.5 dB.

To perform the bit error rate (BER) test, first a pseudo-random bit stream (PRBS) is provided as the TX input. The RX signal sampled by the oscilloscope at 40 GSps is numerically frequency downconverted to the baseband. The demodulator then synchronizes on the incoming cyclo-stationary signal based on a priori information on the known modulation properties, e.g., symbol period, elementary PPM-shift, phase and amplitude constellation. The phase-offset between TX and RX will be estimated based on the PRBS sequence. Finally, the signal is demodulated based on the Euclidean distance, to obtain the most-likely transmitted symbols.

The UWB RX is placed at various distances away from the surface of the porcine skin tissue. As shown in Fig. 29(a), the measured transmission ranges are 2 and 15 cm at 1.66 and 1.43 Gbps data rates, respectively, for the BER below 10^-4. Fig. 28 also shows the measured eye diagram and constellation of the received 1.43 Gbps signal at an 18-cm air distance, with a measured BER of 1.5×10^-4. The measured total EVM, including the RX noise, is -18 dB, which corresponds to E_S/N₀ of ~24 dB. This is higher than the minimum required E_S/N₀ estimated in Fig. 4, to achieve an SER of 10^-4. Note that the SER is approximately 2 times of the BER in the worst case, assuming an arbitrary bit-to-symbol mapping.

Fig. 29. Measured (a) BER of 1.43 Gb/s and 1.66 Gb/s data-rates versus distance, (b) Eye diagram and constellation of the received 1.43 Gbps signal at 18-cm air distance.

The transmission range of 15 cm at 1.43 Gbps matches well with the link budget estimation in section II, but the 2-cm range at 1.66 Gbps is shorter than the expectation. A possible reason could be that the TX EVM of -21 dB is becoming a significant contributor of symbol errors at 1.66 Gbps (as discussed in section II.C).

Such RX hardware and demodulation algorithms do not always need to be implemented with the benchtop equipment, but possible with low-power ASICs. The most hardware demanding part of the demodulator is for the PPM, which requires a large RX bandwidth to avoid inter-symbol interference (ISI) which can degrade the time resolution in PPM demodulation. The proposed hybrid modulation has relaxed the PPM demodulation requirement by having a larger position step of ~600 ps. According to the simulation result shown in Fig. 30, an RX low-pass filter (LPF) bandwidth larger than 500 MHz RX is preferred. Note the ADC sampling should be designed to at least 2× of the RX LPF bandwidth, i.e., ~1 GSps, to fulfill the Nyquist requirement. The recent IEEE 802.15.4z compatible IR-UWB radio front ends reported in literature typically have an RX baseband bandwidth wider than 250 MHz and an ADC sampling of 2 GSps, e.g., [48], which can potentially meet the RX front-end requirements of the proposed hybrid modulation with low power consumption.

Fig. 30. Simulated eye diagram of 4-PAM and 4-PPM with different RX low-pass filter bandwidth.

C. Performance Summary and Benchmarks

Table I summarizes the performance and benchmarks with state-ot-the-art transcutaneous TXs with data rate > 10 Mbps. An EM propagation with IR-UWB can achieve up to 500 Mbps data rate, but no prior works demonstrate the scalability up to Gbps. An EM propagation at millimeter-wave frequency, e.g., 60 GHz, can achieve data rate up to 6 Gbps [52] with an excellent energy efficiency, but a high tissue attenuation at this frequency will lead to a limited communication distance. [49], [52] also integrate the antenna on-chip, leading to an extremely small module form factor below 10 mm², but the robustness of the wireless link (i.e., SER) versus communication distance has not been demonstrated. The EM channel modeling in Fig. 4 suggests that the channel property is heavily influenced by the location, relative distance and orientation between TX and RX antennas, and the path loss can vary up to 25 dB. Hence, a transmission range longer than a few cm is a crucial criterion to make the wireless link reliable in such an implant scenario.

Table I. Performance Summary and Comparison Table with Prior Arts.

	This work	[22] TBioCAS’16	[49] TBioCAS’16	[24] JSSC’21	[50] JSSC’17	[51] CICC’20	[52] JBHI’15	[19] TBioCAS’20	[14] TBioCAS’19	[53] SSCL’20	[17] VLSI’21
Device technology	28nm CMOS	180nm CMOS	GaAs HBT	180nm CMOS	40nm CMOS	18nm CMOS	65nm CMOS	VCSEL	65nm CMOS	110nm CMOS	65nm CMOS
Wireless method	IR-UWB	IR-UWB	IR-UWB	IR-UWB	ISM RF	Sub-GHz RF	Millimeter-wave	Optical	Inductive	BCC	BCC
Frequency	6-9GHz	3-7GHz	8GHz	4.1GHz	1.75GHz	0.7GHz	60 GHz	>800nm NIR	0.2GHz	DC-20MHz	MHz-GHz
Signal bandwidth	1GHz	4GHz	1GHz	1GHz	0.2GHz	N.A.	2GHz	N.A.	0.2GHz	20MHz	N.A.
Modulation	4PPM+8PSK+ 4PAM impulse	BPSK impulse	OOK impulse	OOK impulse	OOK	LSK	OOK	OOK impulse	Return to zero	Return to zero	OOK
TX architecture	Digital polar (DPA+ PHMUX)	Edge combine	Edge combine	Power oscillator	Power amplifier	Power amplifier	Power amplifier	-	Pulse-based	Pulse-based	Power amplifier
Max. data rate	1.66Gb/s	500Mb/s	128Mb/s	150Mb/s	58Mb/s	27Mb/s	6Gbps	300Mb/s	200Mb/s	16.7Mb/s	10Mb/s
TX power cons.	9.69mW	5.4mW	561mW	0.69mW	0.09mW	N.A.	12.5mW	11mW	0.3mW	0.56mW	0.001mW
TX energy efficiency	5.8pJ/b	10.8pJ/b	438pJ/b	4.7pJ/b	1.6pJ/b	N.A.	2pJ/b	37pJ/b	1.5pJ/b	33.9pJ/b	52pJ/b
TX peak POUT	0dBm	N.A.	-12.3dBm	N.A.	-18.5dBm	-	-32dBm	-	-	-	-
TX antenna area	66mm2	100mm2	N.A.	4.5mm2	N.A.	-	3mm2	N.A.	100mm2	0.24mm2	N.A.
TX antenna gain	-8.5dBi	N.A.	N.A.	N.A.	N.A.	-	7.4dBi	-	-	-	-
Tissue thickness	15mm skin/fat	2mm skin/fat & 4mm bone	15-20mm phantom	10mm muscle	No tissue included	1-2.5mm tissue	No tissue included	3.5mm skin	11.8mm skull/skin	2mm brain	10mm brain
Air transmission range (in-vitro)	2cm@1.66Gb/s 15cm@1.43Gb/s	1.5cm	1cm	N.A.**	N.A.**	N.A.**	N.A.**	0.4cm	N.A.**	Contact based	Contact based
Normalized energy effi.*	45pJ/bit/m (@1.43Gb/s)	720pJ/bit/m	43.8nJ/bit/m	-	-	-	-	9.25nJ/bit/m	-	-	-

^*TX energy efficiency normalized to 1-m air-transmission range; tested with biological tissue of phantom.

^**No air transmission range reported with tissue or phantom.

The presented IR-UWB TX achieves Gbps of the data rate with the longest air transmission range under the implant condition, leading to a range normalized energy-efficiency of 45 pJ/bit/meter, which is at least 16× improvement compared to the state-of-the-art.

V. Conclusion

This paper presented an energy-efficient and high data-rate IR-UWB telemetry system for intracortical neural sensing interfaces. The implantable IR-UWB solution enables a transcutaneous wireless communication at an at least 3 × higher data rate and a 10× longer communication range compared to the state of the art. A printed circular monopole antenna was designed for the small form factor and supporting the high band of UWB. The proposed 3-D hybrid impulse modulation that modulates the phase, the delay, and the amplitude simultaneously provides high data rate while optimizing the link margin. The link budget was analyzed with the measured path loss. The proposed digital polar IR-UWB TX realizes the 3-D hybrid impulse modulation with high independence among each modulation. The proposed low-noise injection locked DCO enables the 8-PSK impulse modulation at the high UWB band by adopting the negative skewed delay cell and a buffer with the duty cycle correction. The proposed synchronized delay controller performed the 4-PPM without the need of complex delay calibration. The dual capacitively coupled SCPA with the on-chip matching network performs the 4-PAM with excellent linearity and efficiency. Fabricated in 28 nm CMOS technology, this work achieved 0.155 mm² and 1.05 cm² of the core die area and the system module area including the antenna printed at the same PCB, respectively. Consuming 9.7 mW, the proposed wireless telemetry system supports up to 1.66 Gb/s data rate, and up to 15 cm transmission range from 15 mm below skin at 1.43 Gb/s, which achieves at least 16 times better range normalized efficiency than prior arts.

Fig. 26. Measured output spectrum result with the proposed 3-D hybrid modulation.