A Multi-Channel, Impedance-Matching, Wireless, Passive Recorder for Medical Applications (2019 Version)
Dissertation
Presented in Partial Fulfillment of the Requirements for the Degree Doctor of Philosophy in the Graduate School of The Ohio State University
By
Wei-Chuan Chen, B.Eng.
Graduate Program in Electrical and Computer Engineering
The Ohio State University
2019
Dissertation Committee:
Asimina Kiourti, Co-Advisor John L. Volakis, Co-Advisor Liang Guo Daniel Rivers © Copyright by
Wei-Chuan Chen
2019 Abstract
This dissertation presents a new technology for batteryless and wireless neu- rorecording system which can be applied clinically. Two clinical issues of this type of neural implant are the 1) multichannel operation and 2) high impedance and DC voltage offset from the brain electrode impedance. To resolve these two problems, one wireless multichannel system and one brain electrode interface impedance-matching system are proposed respectively. To achieve multichannel operation, one photo- activated multiplexer is employed in the implant circuit. The interrogator additionally sends an infrared control signal for channel selection. Experimental results show that the proposed neuropotential recorder exhibits 20 µVpp sensitivity at all eight channels.
The system is also in compliance with the strictest Federal Communications Com- mission standards for patient safety. Notably, the proposed approach is scalable to a much higher number of channels. On the other hand, to mitigate the high impedance and DC voltage offset of the brain-electrode interface, one self-biasing PNP Bipolar
Junction Transistor (BJT) is adopted in the brain circuits. This self-biasing PNP
BJT increases the overall system’s impedance and maintains the system sensitivity while the high impedance is present. Measurement results demonstrate that emulated neuropotentials as low as 200 µVpp can be detected at a 33 kΩ electrode impedance.
Together, these proposed techniques would lead the wireless neuro recorders to be applicable in real, in-vivo clinical applications.
ii This is dedicated to my parents, my family, my friends and everyone ever helps me.
iii Acknowledgments
I would like to thank my advisors, Dr. Volakis and Dr. Kiourti, who have encour- aged and given their best support to me during the journey of my Ph.D. Its never easy to give guidance to the latest, state of art research topics. I am really indebted to them for their valuable advice on both academics and my personal development.
I would like to thank my parents, who are always there for me and encourage me to pursue my dream, even though deep in their heart they know their son will not be able to be around them for a long time.
I would like to thank my colleagues Cedric Lee, Brocks Delong, Md Asiful Islam
(Asif), Jingni Zhong, Satheesh Bojja Venkatakrishnan, and Shubhendu Bhardwaj. It is an incredible experience to work alongside these talented people. I would like to thank my junior colleagues Jack Blauert, Vigyanshu Mishra, Saad Alharbhi, Katrina
Guido, Keren Zhu, and Ramandeep Vilhkhu. I really enjoyed every event we shared and celebrated together in the ESL. Hope you guys have a good time in the lab and earn the degree smoothly.
I would like to thank my friends who have ever helped me and accompanied me through the toughest time. I will not be able to accomplish my degree without your support.
Finally, I would like to thank my significant other, I-Shan Tsai, for her love and patience.
iv Vita
May 12, 1992 ...... Born - Taipei, Taiwan
2014 ...... B.Eng.
2015-present ...... Graduate Research Associate, Electrical & Computer Engineering, The Ohio State University.
Publications
Research Publications
W. Chen, K. Guido and A. Kiourti, ”Passive Impedance Matching For Implanted Brain-Electrode Interfaces,” in IEEE Journal of Electromagnetics, RF and Microwaves in Medicine and Biology (published).
W. Chen, C. W. L. Lee, A. Kiourti and J. L. Volakis, ”A Multi-Channel Passive Brain Implant for Wireless Neuropotential Monitoring,” in IEEE Journal of Electro- magnetics, RF and Microwaves in Medicine and Biology, vol. 2, no. 4, pp. 262-269, Dec. 2018.
W. Chen, B. Delong, R. Vilkhu, and A. Kiourti, ”Enabling Batteryless Wearables and Implants,” in 2018 Applied Computational Electromagnetics Society Journal, vol. 33, no. 10, pp. 11061108, Oct. 2018.
W. Chen, B. DeLong, R. Vilkhu and A. Kiourti, ”Towards batteryless wearables and implants,” in 2018 International Applied Computational Electromagnetics Society Symposium (ACES), Denver, CO, 2018, pp. 1-2.
W. Chen, A. Kiourti and J. L. Volakis, ”A Passive Multi-Channel Brain Implant for Wireless Neuropotential Monitoring,” 2018 IEEE International Symposium on
v Antennas and Propagation & USNC/URSI National Radio Science Meeting, Boston, MA, 2018, pp. 1133-1134.
Fields of Study
Major Field: Electrical and Computer Engineering
Studies in: Electromagnetics RF and Microwave Circuits Antennas
vi Table of Contents
Page
Abstract ...... ii
Dedication ...... iii
Acknowledgments ...... iv
Vita...... v
List of Tables ...... x
List of Figures ...... xi
1. Introduction ...... 1
1.1 Motivation and Challenges of Wireless and Fully-Passive Neurosens- ing Systems ...... 1 1.2 Neural Signal Types and Deep Brain Sensing Applications . . . . .2 1.2.1 Action Potential ...... 2 1.2.2 Field Potential ...... 3 1.3 State-of-the-Art Neural Implants ...... 5 1.3.1 Wired Neurosensors ...... 5 1.3.2 Wireless Neurosensor ...... 6 1.4 Former Single-Channel Wireless Fully-Passive Recorder ...... 20 1.5 Organization of this Thesis ...... 21
2. Wireless and Fully-Passive Neurosensing System Overview ...... 26
2.1 Block Diagram and Operation Principle ...... 26 2.2 Mixer Circuit ...... 27 2.3 Antenna Interface ...... 29 2.4 Interrogator System ...... 32
vii 3. Improved Interrogator System: Analysis and Design ...... 36
3.1 Phase Noise Analysis ...... 36 3.2 Self-Mixing and DC Offset ...... 38 3.3 Phase Noise Reduced Interrogator System ...... 39
4. Wireless and Fully-Passive Multi-Channel Neurorecorder ...... 41
4.1 Block Diagram and Operation Principle ...... 42 4.2 Antenna Interface ...... 45 4.3 Implanted Mixer ...... 48 4.4 Infrared Transceiver and Receiver Design ...... 49 4.5 System Performance ...... 51 4.5.1 Fabricated Prototype ...... 51 4.5.2 Measurement Setup ...... 51 4.5.3 Minimum detectable signal ...... 53 4.5.4 Specific Absorption Rate (SAR) ...... 55
5. Wireless and Fully-Passive Implant Match to High-Impedance Electrodes 57
5.1 Introduction to Electrodes and Detection Mechanism ...... 57 5.2 Modeling and Measurement of Electrode Impedance and DC offset voltage ...... 60 5.3 Block Diagram and Operation Principle ...... 63 5.4 System Performance ...... 66 5.4.1 Measurement Setup ...... 66 5.4.2 Stand-Alone Circuit Performance ...... 67 5.4.3 Integrated System Performance ...... 68 5.4.4 DC Offset Tolerance ...... 71
6. Conclusion ...... 74
6.1 Summary ...... 74 6.2 Key Contributions ...... 75 6.3 Future Work ...... 76 6.3.1 In-V ivo Measurement and Result ...... 76 6.3.2 In − vivo Measurements in Rats ...... 78 6.3.3 In − vivo Measurements in Human Subjects ...... 79 6.3.4 Recorded Signal Processing ...... 80 6.3.5 Channel Scalability ...... 81 6.3.6 System Miniaturization ...... 82 6.3.7 Antenna Misalignment and Choice ...... 83
viii Appendices 86
A. MATLAB Code for Genetic Algorithm ...... 86
A.1 Main Code ...... 86 A.2 Script Writing ...... 96 A.3 Restraint Check ...... 97 A.4 Simulation ...... 100
Bibliography ...... 103
ix List of Tables
Table Page
1.1 Voltage And Frequency Range Of Signals Generated By The Human Brain ...... 5
2.1 Gain and Noise Figure of Components Applied in the Interrogator . . 34
4.1 Comparison Between Proposed vs. Previously Wireless Implanted De- vices Without Integrated Circuits ...... 46
5.1 Node Voltage and BJT Operation Regions ...... 65
x List of Figures
Figure Page
1.1 Typical action potential of a neuron [1]...... 2
1.2 Integrated Neural Interface system block diagram [2]...... 6
1.3 The neural recording microsystem: (a) illustration of the overall sys- tem; (b) simplified block diagram of the implanted microsystem dia- gram [3]...... 8
1.4 Functional block diagram of the NPU-32 [3]...... 9
1.5 Schematics of the dual panel brain implantable microsystem featuring an active brain sensor (microelectrode array integrated to amplifier IC) in the cortical unit, and hybrid A/D, control, and RF-IR(Infrared) telemetry in the cranial unit [4]...... 10
1.6 (a) Photographic images showing an implantable 16-channel microsys- tem with a dual-panel liquid crystal polymer substrate. A spiral pat- tern of RF power receiving coil is clearly visible in the backside image. (b) Block diagram of the dual-panel microsystem showing neural sig- nal and power/clock flows among various micro- and optoelectronic components [4]...... 11
1.7 Design overview of HermesC [5]...... 12
1.8 HermesC System (A) Setup with animal in a metal home cage and receiving antenna on plastic cage window (B) Aluminum enclosure with stub antenna in lid (C) PCB with INI chip (D) Example video data [5]. 13
xi 1.9 Block diagram of Bluetooth-based single-unit recording system. The wireless data acquisition module is carried by the rat in a backpack that transmits data to a PC-based control subsystem. Microdrive wire electrodes were implanted in the CA1 of the hippocampus for sensing single-unit activities which were first amplified by the analog-front-end amplifier and then digitized by the ADC that connects to the MCU for Bluetooth wireless transmission to the host PC [6]...... 14
1.10 Block diagram of the neural data acquisition system [7]...... 15
1.11 Functional block diagram of INI3 microchip [8]...... 16
1.12 Block diagram of NeuralWISP [9]...... 17
1.13 Exploded model view of the hermetically packaged 100-ch fully im- plantable wireless neurosensing device [10]...... 18
1.14 Circuit block diagram [10]...... 19
1.15 Schematic of a general BMI system architecture. Such systems require multichannel sensors for neural recording, specialized IC for signal pro- cessing, programs for brain simulation, a wireless link to communicate with external unit, and finally a real-time brain mapping to effectively track and decode the neural activities. In this paper, we focus on the wireless link design and characterization between reader and implanted loop antennas. [11]...... 20
1.16 (A) An external transducer powers and communicates with a neural dust mote placed remotely in the body. Driven by a custom transceiver board, the transducer alternates between transmitting a series of pulses that power the device and listening for reflected pulses that are modu- lated by electrophysiological signals. (B) A neural dust mote anchored to the sciatic nerve in an anesthetized rat. Inset shows neural dust mote with optional testing leads. (C) Components of a neural dust mote. The devices were assembled on a flexible PCB and consist of a piezoelectric crystal, a single custom transistor, and a pair of recording electrodes. (D) The transceiver board consisted of Opal Kelly FPGA board, application-specific integrated circuit (ASIC) board (Seo et al., 2015; Tang et al., 2015), and the transducer connector board. [12]. . . 24
1.17 Neurosensing system block diagram [13]...... 25
xii 2.1 Block diagram of the proposed neurosensing system...... 26
2.2 Block diagram of the proposed subharmonic mixer...... 28
2.3 (a) Measured permittivity, and (b) measured conductivity of pig-skin versus the reference skin properties reported in [14]...... 30
2.4 Block diagram of the interrogator system...... 32
3.1 The phase noise behaves as a skirt centered at at the carrier signal in frequency domain...... 37
3.2 The phase noise behaves as fluctuation in time domain after demodu- lation...... 38
3.3 Proposed phase noised reduced interrogator setup...... 39
3.4 Frequency domain of the proposed phase noise reduced interrogator. . 40
3.5 Demodulated version of -115 dBm backscattered neural signal at 100 Hz. The plot compares the currently reported system vs. the previous one...... 40
4.1 Block diagram of the proposed mulitchannel neurosensing system. . . 42
4.2 Proposed implanted infrared receiver and exterior transceiver used to toggle different neuro-channels in a wireless and passive manner. The transceiver employs 3 photodiodes to form a 3 digit code for selecting each of the 8 channels. In this condition, the three digit code is 0,1,0 and channel 3 is selected...... 43
4.3 Proposed system layout: (a) brain implant, and (b) exterior interroga- tor. The corresponding areas both are 40 mm × 40 mm ...... 47
4.4 Simulated transmission coefficient, |S21|, between the transmitting and receiving antennas for different pig skin depth...... 49
4.5 Radiation and reception pattern for the (a) infrared emitter, (b) pho- todiodes and (c) the geometry of the photodiodes and photo emitters 50
xiii 4.6 Fabricated prototypes: (a) brain implant, (b) exterior interrogator . . 52
4.7 The neurosensing measurement setup with the layered head phantom. 53
4.8 (a) Measured permittivity, and (b) measured conductivity of pig-skin versus the reference skin properties reported in [14]...... 54
4.9 Time-domain performance while recovering the minimum detectable
neural signal (Vin = MDSneuro = 20µVpp)...... 55
4.10 SAR performance averaged over 1 g with input carrier power at 2.4 GHz and 6 dBm...... 56
5.1 Patch clamp configurations: A diagram showing five commonly used patch clamp configurations. [15] ...... 59
5.2 Models of the electrical double layer at a positively charged surface: (a) the Helmholtz model, (b) the GouyChapman model, and (c) the Stern model, showing the inner Helmholtz plane (IHP) and outer Helmholtz plane (OHP). The IHP refers to the distance of closest approach of specifically adsorbed ions (generally anions) and OHP refers to that of the non-specifically adsorbed ions. The OHP is also the plane where the diffuse layer begins. d is the double layer distance described by the
Helmholtz model. φ0 and φ are the potentials at the electrode surface and the electrode/electrolyte interface, respectively [16]...... 60
5.3 Equivalent circuit model of the electrode...... 61
5.4 Clinical brain electrode impedance measurement...... 62
5.5 Block diagram of the proposed impedance-matching neurosensing system. 63
5.6 Proposed neural implant design: (a) DC mode, and (b) RF mode. . . 64
5.7 Measurement set-up used to assess the neurosensing system performance. 66
5.8 Fabricated BJT Circuit (a) directly connected to circulator, (b) con- nected to antenna in air, and (c) connected to antenna in pig skin . . 67
xiv 5.9 The demodulated time domain 100 Hz signal of (a) circuit alone Vin = 100 µVpp, (b) with the antenna in air Vin = 100 µVpp, and (c) with the antenna in pig skin Vin = 200 µVpp in series with a 33 kΩ resistor . . 69
5.10 a) Measured permittivity, and (b) measured conductivity of pig skin versus the reference skin properties reported in [14]...... 70
5.11 Measured transmission coefficient (S21) of the implanted and interroga- tor antenna system (a) through air, and (b) through pig skin. . . . . 71
5.12 SAR performance averaged over 1 g with input carrier at 2.4 GHz. . . 72
5.13 Demodulated waveform of a 100 µVpp and 100 Hz signal subject to: (a) 0 V offset, (b) +50 mV offset, and (c) -50 mV offset in series with a 33 kΩ resistor ...... 73
6.1 Wired ECG recording v.s. wireless ECG signal recorded by the pro- posed wireless implant ...... 77
6.2 Neural activity recording. (a) Schematic of the area of interest (pri- mary somatosensory cortex, hind limb region contralateral to the stim- ulated hindpaw) and probe placement during recording. (b) Represen- tation of signal averaging procedure used during the signal processing stage of neural recording. (c) Overlay of WiNS and wired extracted somatosensory evoked potential (SSEP) after processing the signal. . 78
6.3 The experiement setup for the proposed implant system during Parkin- son’s disease surgery ...... 79
6.4 Wired recording neural signal in total 10 seconds length ...... 81
6.5 Transmission loss of the interrogator-received signal at a) 2.4 GHz and b) 4.8 GHz. (Credit to Katrina Guido) ...... 83
6.6 (a)side view of the published antenna, and (b) fabricated prototype. . 84
6.7 Transmission loss at 2.4 GHz between the Bio-Matched Horn and an implantable patch antenna immersed at: (a) 4 mm, and (b) 20 mm, as a function of positional misalignment...... 85
xv Chapter 1: Introduction
1.1 Motivation and Challenges of Wireless and Fully-Passive Neurosensing Systems
Countless people are suffering from various neurological disorders today. The reasons and outcomes of such disorders vary widely. Some of the disorders may be attributed to chronic neurological diseases like epilepsy, Parkinson’s and Alzheimer’s.
Others might stem from external forces such as car accident, sports injury, etc. In turn, these disorders may result in limb tremor, paralysis of part or whole body and loss of consciousness, depending on the affected area. As a result, the quality of the patient’s daily life is greatly degraded. To resolve these conditions and rehabilitate the dysfunctional capabilities, brain-computer interfaces (BCI) have long been proposed.
BCI is the communication gateway between the brain and the external devices. With the help of BCI, today’s patients can stand, sit and move with their prosthetics being controlled by their own brain. As such, the quality of the patient’s life is greatly improved.
The goal of this dissertation is to propose neural implants to fulfill reliable BCI with the following features: 1) safe and unobtrusive neural potential sensing, 2)
1 patient safe, 3) reliable detection of neural signals, 4) wireless transmission and 5) low power operation.
1.2 Neural Signal Types and Deep Brain Sensing Applica- tions
In the human brain, information can be created and conveyed by both chemical and electrical signals [17]. Typically, there are two types of neural signals that can be detected depending on the quantity of firing neurons: a) action potentials (neural spikes) b) field potentials [18].
1.2.1 Action Potential
Figure 1.1: Typical action potential of a neuron [1].
2 Fig. 1.1 presents a typical action potential of neurons. When the neuron is in the rest state, the membrane’s potential across the internal to extracellular fluids is around - 70 mV. After the neuron is excited over the threshold potential (-55 mV), the voltage-gated ion channels embedded in the cell membrane open/close to create voltage action potentials or neural spikes [19]. This action potential can be as high as +40 mV and propagates through the axon to the synapse.
However, it is difficult to directly record the individual neural spikes clinically.
This is because spikes may be produced by an unknown quantity of neurons. As a result, it’s still unclear how to associate each neural spike to specific neurons [20].
Therefore, while a multielectrode array is applied to the brain for recording, multiple spikes from nearby neurons can be detected and at times only single spike is observed.
From Fig. 1.1, the neural spikes are biphasic and typically 0.3 - 1.0 ms in length. In terms of frequency, the spike’s energy is concentrated in the range from 250 Hz to 5
KHz [21].
1.2.2 Field Potential
When a group of neurons produce action potentials that propagate congruently to the recording electrode, the recording electrode experiences a linear combination of the synchronized neurons action potentials. These are called field potentials [21–
23]. Generally, there are three types of field potentials that can be detected: 1) electroencephalographic (EEG) signals, 2) electrocorticographic (ECoG) signals, and
3) local field potential (LFP) signals.
EEG is the current most widely used method to monitor brain activity. By placing the electrodes on the surface of the scalp, EEG signals can be collected and recorded in
3 a non-invasive way. However, EEG signals may lose spatial and temporal information due to the attenuation through the scalp. This feature makes EEG recordings highly unpredictable and unreliable [23,24].
Instead of placing the electrodes on the scalp, the ECoG signal is recorded by plac- ing the electrodes under the skull, either below (epidural) or above the dura matter
(subdural) [24]. The unattenuated ECoG signal provides higher spatial resolution, wider bandwidth (0 - 500 Hz vs. 0.5 - 40 Hz for EEG signal) and higher amplitude
(peak 100 - 200 µVpp vs. peak 20 - 40 µVpp for EEG signal). Since the ECoG signal has a higher signal to noise ratio (SNR) and higher tolerance to the electromyography
(EMG) or electrooculography (EOG) signals, it requires less user training to control non-muscular 1D or 2D movement than EEG signal [23, 24]. Compared to neural spikes or local field potentials, the ECoG technique can provide a more robust inter- face over a long term period. As a result, it is preferred in terms of clinical feasibility and signal accuracy. However, a major drawback of ECoG recording is the invasive intracranial surgery requirement.
If the electrodes are placed closer to the neurons, the local field potentials (LFPs) can be recorded. LFPs are sometimes considered as the internal version of EEG, but with less attenuation and spatial averaging [21]. Therefore, LFPs provide the best spatial resolution and the largest signal amplitude from 60 µVpp to 2-4 mVpp within
10 -200 Hz frequency range. LFPs also have more stable performance than neural spikes due to being less vulnerable to the scar tissues formed around the electrode tips. Still, the LFPs brain-implanted electrodes are not stable and reliable enough in the long-term period and lead to the frequent replacement of the implanted electrodes through surgery.
4 Table 1.1: Voltage And Frequency Range Of Signals Generated By The Human Brain Neural Signals Voltage Range Frequency Range
Action Potential (Neural Spikes) 100 ∼ 2000 µVpp 300 Hz ∼ 5000 Hz
Electroencephalogram (EEG) 20 ∼ 40 µVpp 0.5 Hz ∼ 40 Hz
Electrocorticographic (ECoG) 100 ∼ 200 µVpp < 500 Hz
Local Field Potential (LFPs) 60 ∼ 4000 µVpp 10 ∼ 200 Hz
Table. 1.1 summarizes the voltage and frequency range of signals generated by the human brain.
1.3 State-of-the-Art Neural Implants
Since there are many clinical conditions to consider, it is difficult to compare different types of neural recorders. Below, we summarize past works [2–11, 25, 26] , analyze their pros and cons, and discuss measurement setups with their limitations.
1.3.1 Wired Neurosensors
Perlin et al. [26] introduced a novel overlay cable to address integration and pack- aging issues for neural sensors. The parylene overlay cable is employed to interconnect the microelectrode array and the chips in 3-D to reduce the platform area greatly.
The experiment was performed using guinea pigs but only measured with the end of the polymer cable coming out of the head. Therefore, although the experiment is conducted in-vivo and the system is implantable. However, the wired connection is still unavoidable for the setup.
5 Another classic wired neurosensor is BrainGate [25]. By placing the 96-microelectrode array into the primary motor cortex and routing the brain signal around the damaged part to the external devices, the paralyzed patients can restore lost motor functions.
After receiving the raw signal from the brain implants through the wire, the exter- nal devices perform the signal processing and send control signals to the robotic or prosthetic limb. Although still in tethered connection, this approach is still viable for tetraplegic patients.
1.3.2 Wireless Neurosensor
Figure 1.2: Integrated Neural Interface system block diagram [2].
6 Harrison et al. [2] proposed a fully-implantable 100-channel neural recording sys- tem (see Fig. 1.2) with a 10 by 10 Utah electrode array (UEA). Each of the elec- trodes/channels has its own low-noise amplifier to increase the signal to noise ratio and match the high impedance of the brain microelectrode array. As a result, the power consumption of the chip is relatively high and may lead to excessive heat on the neighboring tissues. To transmit the signals from multiple electrodes, an analog multiplexer is employed for further digitization. Another feature of this work relates to the spike detection mechanism. Notably, the spike detector in each low-power amplifier detects the existence of the neural spikes by setting a threshold. Next, the analog waveform and the selected spike are transmitted to the external circuit wire- lessly using an inductive coil. This coil also serves for power transmission to avoid the use of batteries in the implants.
A challenge of this system is the crosstalk and digital interference to the analog circuits and neural amplifiers. As a result, this implanted system could only transmit the single channel digitized analog waveform by powering with the battery instead of the coil and could not detect spiking signals.
To detect the spike signal, Sodagar et al. proposed a 64-channel wireless im- plantable microsystem (see Fig. 1.3) [3]. It consisted of two 32-channel passive silicon recording probes, four 16-channel amplifier chips with signal conditioning front-end and a bidirectional telemetry module (see Fig. 1.4). There are also two neural pro- cessor units in the system, each having four 8-channel modules (SD-8) and eight pro- grammable analog spike detectors for a single module. Each SD-8 module also has an analog channel selector (multiplexer) to select a single channel for further digitization.
The bidirectional telemetry module serves as the data and power transmission route.
7 Figure 1.3: The neural recording microsystem: (a) illustration of the overall system; (b) simplified block diagram of the implanted microsystem diagram [3].
In [3], there are two operational modes: a) scan mode and b) monitor mode.
When operating in the scan mode, all neural channels are searched to detect spike
occurrence. Once the recorded signal is above the threshold, the address of the active
channel is further processed and transmitted externally. When in the monitor mode,
only one or two channels are selected and digitized by the ADC. During the in-vitro setup, the three different programmable spike detection capabilities (positive, negative and bilateral) are verified in the scan mode. Only one channel with the preconditioned neural signal is tested during the scan mode due to bandwidth limitations. While in the monitor mode, the recorder is able to switch among different channels, digitize the analog signal and transmit to the outside. Although, the system structures are for multichannel recording ( [2,3]), real issues, like limited bandwidth and cross talk, still restrict the system to the single channel operation. Also, the telemetry module in the current prototype can’t support such a complex system. Thus a battery is
8 Figure 1.4: Functional block diagram of the NPU-32 [3].
required for in-vitro measurements. Heat is another crucial problem when it comes
to concurrent multichannel recording.
Another method for brain implants is to transmit the signal wirelessly via infrared
telemetry. Song et al. [4] proposed a 16-channel of intracortical signal recording
(neural spikes and LFPs) implants and transmitted the signal through the skin with
infrared telemetry (see Fig. 1.5). A major drawback of this system (see Fig. 1.6) is the
relatively high power consumption (12 mW). That is, above the tolerable range of ∼
9 Figure 1.5: Schematics of the dual panel brain implantable microsystem featuring an active brain sensor (microelectrode array integrated to amplifier IC) in the cortical unit, and hybrid A/D, control, and RF-IR(Infrared) telemetry in the cranial unit [4].
10 mW noted in Harrison et al. [8] proposed. In addition, although infrared telemetry is novel, an external battery is still required to power the implanted electronics. This makes this system not fully implantable.
Chestek et al. [5] proposed a novel system called HermesC for the fully wireless recording of neural spikes and LFPs (see Fig. 1.7). This recorder (see Fig. 1.8) consists of one integrated neural interface version 3 (INI3) microchip that was previously used in [27]. HermesC samples one channel out of the 96-channel array with the
10-bit ADC and transmits the signal externally using wireless telemetry. In-vivo measurement of this system were conducted with rhesus monkeys. The INI3 chip and Lithium battery are enclosed within an aluminum lid and connected to the head- protruding wires from the microelectrode array. This setup enhances data and power
10 Figure 1.6: (a) Photographic images showing an implantable 16-channel microsystem with a dual-panel liquid crystal polymer substrate. A spiral pattern of RF power receiving coil is clearly visible in the backside image. (b) Block diagram of the dual- panel microsystem showing neural signal and power/clock flows among various micro- and optoelectronic components [4].
transmission by avoiding tissue loss in the monkey’s head. Although this system is fully implantable, the protrusion brings inconvenience to patients and increases the risk of infection. Also, there is still only one channel that can be selected and transmitted at a time [2,3].
Chen et al. [6] proposed an 8-channel neural spike recording system which consists of 8 amplifiers, a 12-bit ADC and a Bluetooth wireless telemetry module. The record- ing system (see Fig. 1.9) could support a lower sampling rate of 1 kHz in previewing all channels and a higher rate of 10 kHz for single channel monitoring via remote
11 Figure 1.7: Design overview of HermesC [5].
control. Measurements were carried out by placing the implant in a rat navigated in a vertical maze. The challenge with this implant was its very high power consumption
(185 mW) and the requirement for a rechargeable battery. Again, this large power consumption was much larger than the safe limit (10 mW) [8] noted. In addition, rechargeable batteries also require additional surgery for replacement, making this approach impractical in terms of clinical application.
Rizk et al. [7] proposed a wireless 96-channel, fully implantable, neural data ac- quisition system. This recorder consisted of 3 digitizing headstage modules, one implantable central communication module, and one inductively-coupled coil (see
Fig. 1.10). Each headstage module recorded the neural waveforms from the two sets of 16 electrodes. There were three operating modes in this system. In the first mode, the spike waveform is extracted with the timestamp and sent out wirelessly. In the
12 Figure 1.8: HermesC System (A) Setup with animal in a metal home cage and re- ceiving antenna on plastic cage window (B) Aluminum enclosure with stub antenna in lid (C) PCB with INI chip (D) Example video data [5].
second mode, the system provides one 1 milliseconds window for all 96-channels wave- forms to be sent out. In the third mode, the full streaming data for a specific single channel is transmitted. Notably, each headstage consumes high power of 300 mW, viz much higher than the safe limit of 10 mW [8]. The in-vivo experiment of this work is conducted using a sheep. The system is implanted acutely and not chroni- cally due to packaging issues. So far, this is the only system that is fully-implantable, wireless and doesn’t need a battery. The drawback of this system is also very clear and relates to the very high power consumption. Such power would eventually lead to high-temperature in the neighboring tissue.
Harrison et al. [8] proposed a 100-channel wireless neural recording system (see
Fig. 1.11). This recorder is similar to that in the previous work [2] but with improved
13 Figure 1.9: Block diagram of Bluetooth-based single-unit recording system. The wire- less data acquisition module is carried by the rat in a backpack that transmits data to a PC-based control subsystem. Microdrive wire electrodes were implanted in the CA1 of the hippocampus for sensing single-unit activities which were first amplified by the analog-front-end amplifier and then digitized by the ADC that connects to the MCU for Bluetooth wireless transmission to the host PC [6].
performance. As in the in-vivo experiment, only one channel signal can be recorded at a time, digitized and transmitted wirelessly. However, although this system is able to detect neural spikes wirelessly and is batteryless, it’s still not fully implantable and only allows for single channel operation. However, this proposed neural recorder circuit only consumes 8 mW which falls into the tolerable range of 10 mW.
Holleman et al. [9] also proposed a neural interface, referred NeuralWISP (see
Fig. 1.12), this interface employs a commercial radio frequency identification (RFID) reader. The implanted recorder consists of one custom low-noise, low-power amplifier
14 Figure 1.10: Block diagram of the neural data acquisition system [7].
and an analog spike detector for reducing bandwidth and power requirement. The employed RFID reader transmitted 30 dBm (1 W) of RF power to the NeuroWISP.
The microcontroller of the implants received the power and woke to record the neural spikes, digitize the analog spike waveform and transmit it to the RFID reader. The in-vivo experiment was carried out using a Manduca Sexta moth and a macaque monkey. Results show that the system could achieve wireless and batteryless neural recording operation. However, it’s still difficult to scale up to multichannel operation even with high RF input power levels.
15 Figure 1.11: Functional block diagram of INI3 microchip [8].
Yin et al. [10] proposed a parallel 100-channel recording system which is fully implantable and wireless. The system consisted of a 100-electrode intracortical neu- ral sensing array, a rechargeable Li-ion battery, a pre-amplifier board (PCB-A) and an RF power/data board (PCB-B) as shown in Fig. 1.13. These components were packaged in a hermetically sealed module and placed below the skin. Fig. 1.14 shows the circuit block diagram of the pre-amplifier board (PCB-A) and the RF power/data board (PCB-B). As in the in-vivo experiment, the results verify the 100-channel par- allel recording capabilities using two adult awake Yorkshire swines and two rhesus macaque monkeys. The neural spikes and the LFPs were observed in the Yorkshire
16 Figure 1.12: Block diagram of NeuralWISP [9].
swine or macaque monkey experiments by changing the bandpass filter cutoff fre- quency. One advantage of this system is the simultaneous 100-channel recorder and digitizer without any neural spike detector. Although the measurements are promis- ing, the issues of high power consumption (50 mW) and rechargeable batteries remain unsolved problems. Our ultimate goal is to achieve the same recording capabilities as in [10], but with reduced power consumption and without battery use.
Notably, the aforementioned publications focus on the integrated circuit design and, typically, ignore the importance of the wireless telemetry interface. Usually, inductive couplers are employed for wireless telemetry. However, the RF signal could be strongly attenuated due to losses in the head tissue. Song et al. [11] proposed the implanted and reader antenna design based on the attenuation of the head tissue and
17 Figure 1.13: Exploded model view of the hermetically packaged 100-ch fully im- plantable wireless neurosensing device [10].
the impedance matching of the neural chips (see Fig. 1.15). Results showed that up to -25 dB link efficiency could be achieved with this approach.
Dongjin et al. [12] proposed a neuro sensor of the peripheral nervous system, re- ferred to as neural dust. Instead of using the electromagnetic waves for the wireless telemetry, this system emphasis on ultrasonic backscatter system to power and trans- mit the signal from the implant as shown in Fig. 1.16. The main advantage of using
18 Figure 1.14: Circuit block diagram [10].
ultrasound is the extremely small size of the sensors (0.8mm × 3mm × 1mm). The in-vivo experiment verified that the evoked EMG responses can be recorded using a rat model. Although the experiment showed a very promising result, the difficulty to scale up for the multichannel recording operation and the associated muscle pain from the ultrasound remains unsolved.
19 Figure 1.15: Schematic of a general BMI system architecture. Such systems require multichannel sensors for neural recording, specialized IC for signal processing, pro- grams for brain simulation, a wireless link to communicate with external unit, and finally a real-time brain mapping to effectively track and decode the neural activi- ties. In this paper, we focus on the wireless link design and characterization between reader and implanted loop antennas. [11].
1.4 Former Single-Channel Wireless Fully-Passive Recorder
To overcome the aformentioned challenges of power, wireless, and batteryless con- nections, a new class of fully-passive and wireless implants were proposed [13,28–32].
To be specific, the term ’fully-passive’ implies that no battery is used, and no energy- harvester, rectifier or regulators are employed.
These type of neural recorders operate in a manner similar to RFIDs. Specifically, an external interrogator is used to send the RF signal to the implant, that includes a
20 mixer. The latter mixes the recorded brain neural signal with the carrier to form an
amplitude-modulated signal that is further transmitted back to the interrogator.
In our team’s latest work [13] (see Fig. 1.17), one wireless and a fully-passive brain
implant was fabricated and validated for a single channel operation. This implanted
recorder exhibited a very small footprint of 8.7 mm × 10 mm and was shown to have
sensitivity as high as 20 µVpp. As such, all signals generated by the human brain,
see Table. 1.1, can be detected. However, the single channel operation of our earlier
device [13] is not suitable for realistic clinical applications. For the latter, we require
100 or even 1000s of channels of concurrent neuropotential recording.
This dissertation builds upon our team’s brain implant by first optimizing its
impedance matching to low level signal detection. This is done by developing and
demonstrating new active matching within the implant while still retaining its wireless
and batteryless characteristics. Specifically, the new implant provides for much lower
signal detection of 200 µVpp vs. 5 mVpp without the active matching circuits. Also, the DC offset is suppressed to further increase sensitivity.
In addition to the above key improvement, this dissertation presents the first multichannel recorders that are also batteryless and wireless. That is the newly proposed implant enables: 1) multichannel operation, 2) high impedance of brain electrodes interface, and 3) DC voltage offset.
1.5 Organization of this Thesis
This dissertation is organized as follows: Chapter 2 provides an overview of the developed wireless fully-passive neuro recorders. The proposed recorders contain a mixer circuit, antenna interface, and interrogator system.
21 Chapter 3 describes the effect of phase noise on the interrogator and a novel tech-
nique to reduce it. The phase noise created by the signal generator leaks to the
amplifying and filtering blocks due to lack of port isolation in the circulator. Af-
ter downconverting with the interrogator mixer, the phase noise appears as random
fluctuations and degrades the desired neuropotential signal integrity. A phase noise
reduction interrogator is therefore proposed to improve the desired signal integrity.
By adding two bandpass filters between the circulator and the splitter, the afore-
mentioned phase noise is reduced greatly before reaching the amplifier and filtering
blocks. Hence, the demodulated signal integrity is improved accordingly.
Chapter 4 discusses a scalable multichannel neural recorder to address the multi-
probe requirements of clinical setups. Infrared transceiver/receiver and a multiplexer
are employed to increase the channel number of the neuro recorders. Using the pro-
posed solution, N infrared photoemitters/diodes are used to accomplish 2N channels
while keeping the system sensitivity at 20 µVpp.
Chapter 5 illustrates the high impedance and DC offset of the clinical brain elec-
trodes and the effect to the wireless neuro recorders. The impedance and DC offset
issues are addressed by implementing a self-biasing PNP bipolar junction transistor
(BJT) technique. The self-biasing BJT increases the overall system impedance and
maintains the system sensitivity while connecting with the clinical brain electrode.
The unique operating regions of the PNP BJT guarantee the normal operation while
addressing the high DC offset voltage. With the proof-of-concept fabrication cir-
cuits, the implant can recover 100 µVpp signal level in series of 33 kΩ emulated brain electrode impedance and ± 50 mV DC offset.
22 Chapter 6 summarizes the dissertation, the contributions of this work, and provide suggestions for future work.
Together, these proposed techniques would lead the wireless neuro recorders to be applicable in real, in-vivo clinical applications.
23 Figure 1.16: (A) An external transducer powers and communicates with a neural dust mote placed remotely in the body. Driven by a custom transceiver board, the trans- ducer alternates between transmitting a series of pulses that power the device and listening for reflected pulses that are modulated by electrophysiological signals. (B) A neural dust mote anchored to the sciatic nerve in an anesthetized rat. Inset shows neural dust mote with optional testing leads. (C) Components of a neural dust mote. The devices were assembled on a flexible PCB and consist of a piezoelectric crystal, a single custom transistor, and a pair of recording electrodes. (D) The transceiver board consisted of Opal Kelly FPGA board, application-specific integrated circuit (ASIC) board (Seo et al., 2015; Tang et al., 2015), and the transducer connector board. [12].
24 Figure 1.17: Neurosensing system block diagram [13].
25 Chapter 2: Wireless and Fully-Passive Neurosensing System Overview
2.1 Block Diagram and Operation Principle
Figure 2.1: Block diagram of the proposed neurosensing system.
The block diagram of the proposed neurosensing system is shown in Fig 2.1. The system consists of two parts: 1) a brain implant placed under the scalp and attached to a recording electrode that penetrates through the bone to the cortical cortex sur- face, and 2) an external interrogator placed outside the scalp to communicate with
26 the implanted sensor. First, the external interrogator transmits a 2.4 GHz carrier
signal via the interrogator antenna to activate the brain implant. The implanted
diode acts as a mixer that uses the 2.4 GHz carrier to upconvert the brain signal
(at frequency fneuro) to 4.8 GHz ± fneuro. This upconverted third-order product is
then backscattered by the implant’s antenna and eventually received by the inter-
rogator. This signal can then be directly observed in the frequency domain using
a spectrum analyzer and/or can be demodulated and observed in the time domain
using an oscilloscope.
In this chapter, the three major parts of the system which entail the mixer circuit,
antenna interface and interrogator system will be discussed. These three parts are
interconnected with each other and have almost equal contributions to the most crit-
ical parameter, system sensitivity. The system sensitvity is defined as the minimum
neural potential that can be recovered and observed either in the frequency domain or
time domain. By analyzing, designing and finally optimizing these three factors, the
system sensitivity can be increased, hence, detecting a wider range of brain signals.
2.2 Mixer Circuit
The typical RF mixer is tured on by the local oscillator (LO) which is subse-
quently mixed with the intermediate frequnecy (IF) to generate a radio output (RF).
The latter frequency is typically at fLO ± fIF , which is known as second-order inter- modulation. However, in the proposed system, using second-order intermodulation as backscattered signal will interfer with both the 2.4 GHz carrier signal and its as- sociated phase noise. To avoid these undesired frequency components, 2fLO ± fIF , which is the third-order intermodulation, is chosen for the major backscattered signal.
27 To further increase this third-order intermodulation product, the subharmonic mixer circuit is adopted due to its lower conversion loss of higher order intermodulation. In the proposed system, fLO is 2.4 GHz, fIF is the frequency of neual signal (fneuro) and the desired radio ouput is third-order intermodulation (fRF = 4.8 GHz ± fneuro).
Figure 2.2: Block diagram of the proposed subharmonic mixer.
The proposed subharmonic mixer consists of a diode pair, an inductor, a capac- itor, and a matching circuit as shown in Fig 2.2. In the pratical implementation of the RF mixer, Schottky diode is widely used because of its consistent RF properties.
Single Schottky diode generally produces second-/third- order intermodulation simul- taneously. To lower the conversion loss of the Schottky diode, an anti-parallel diode pair (APDP) circuit strucure is adopted. The mechanism of the APDP is summa- rized as follows: The positive waveform of the RF carrier would turn on one branch of the diode and turn off the other one. On the other hand, the negative waveform
28 would just behave in the opposite way. Hence, the LO frequnecy is doubled and the
conversion loss of the third-order intermodulation is lowered.
Sitting on the two ends of the ADPD is the inductor and capacitor pair. The
inductor’s impedance could be modeled as jwL and acts like short circuit to the
lower frequency (fneuro) and like open circuit to the higher frequency (fLO). Also,
the capacitor’s impedance behaves as 1/jwC and acts like short circuit to the higher frequency (fLO) and like open cirucit to the lower frequency (fneuro). The special
frequency response of the inductor/capacitor pair provides the proper ground to the
fneuro and fLO, hence, improving the mixer efficiency.
Finally, one matching circuit is required between the implanted antenna and the
subharmonic mixer. This matching cirucit is intended to match the design at the
dual band (2.4 GHz and 4.8 GHz). Within this dual band operation, it would ensure
the 2.4 GHz carrier reaches the mixer and the generated 4.8 GHz ± fneuro product passes to the implant antenna efficiently. The mixer cirucit analysis and optimization is carried in Keysight’s Advanced Design System (ADS).
2.3 Antenna Interface
In order to design the implanted and interrogator antenna pair, there are several things needed to be considered in advance: the operating frequency, the operating region and the antenna type. Generally, the footprint of the antenna is inversely propotional to the wavelength λ. In this sense, higher operating frequency is better since the size of the antenna is smaller. However, the loss tangent of the head tissue layer would also increase along with the frequency as shown in Fig 2.3 which forms one classical trade-off between the size and the loss. Considering these two factors
29 and the generic behavior of the subharmonic mixer, 2.4 GHz in ISM band and 4.8
GHz for its second harmonics are adopted for the antenna operaitng frequency.
Figure 2.3: (a) Measured permittivity, and (b) measured conductivity of pig-skin versus the reference skin properties reported in [14].
The operating region of the antenna could be divided in three areas: reactive near-field, radiative near-field and far-field. Typically, the antenna is working in the far-field region because it’s easier to calculate the analytical solution of the field patern. The boundary of the near-field and far-field could be described as in [33]:
2 Rnearfield = 2D /λ (2.1) where D is the longest dimension of the antenna aperture and λ is the wavelength of the operating frequency. Assuming the longest dimension D of the antenna is roughly
λ/2, the near- and far- field boundaty could be obtained as λ/2 by 2.1 which is 6.2 cm at 2.4 GHz. However, this boundary distance would introduce higher propogation loss, which is inversely propotional to the distance square, into our system. As such, the
30 system will be unable to detect the backscattered signal. To reduce the propagation loss of the implant and the interrogator, it is unavoidable to operate the antenna coupler in the near-field. Since the antenna coupler is working in the near-field, the radiation pattern of the antenna is not so crucial any more. Instead, the transmission coefficient (S21) of the antenna is used to characterize the transmission properties.
Finally, to determine the antenna type, the implanted mixer should be also in- cluded for consideration. To reduce the complexity of the integration between the antenna and the implanted circuits, the printed circuit board (PCB) antenna is nat- urally the first choice among others. Also, to fulfill the dual band capability, the
E-shaped patch antenna is employed. However, since the antenna is operatng in the reactive near-filed region and has too many parameters for tuning, an in-house genetic algorithm is adopted for further optimization. The genetic algorithm is realized in
Matlab. Given the simulation results from Ansys HFSS, the genetic algorithm calcu- lates the cost of each antenna geometry. Based on the cost of each generation, the genetic algorithm creates a new set of the antenna geometry as the next generation.
Once the goal of the cost is reached, the genetic algorithm is stops. The readers are referred to Appendix 1 for further genetic algorithm details.
In the proposed implants, the antenna coupler is firstly optimized. Then, the simulated transmission coefficient is imported to Keysight ADS for further integra- tion. After the ADS built-in optimization and method of moment simulation, the implant neuro recorders are either fabricated within in-house milling machine or by manufacturer.
31 Figure 2.4: Block diagram of the interrogator system.
2.4 Interrogator System
The interrogator is defined as the system which outputs the 2.4 GHz carrier signal and demodulats the backscattered signal of 4.8 GHz ± fneuro. It conisits of the signal generator, circulator, filtering and amplifying block, multiplying chain, mixer and equipment for display like spectrum analyzer and oscilloscope as shown in Fig 2.4.
The signal generator outputs 2.4 GHz signal and splits it into two routes: the first one flows into the circulator and is then directed to the interrogator antenna and the second one serves as the input of the multiplying block. The first one further reaches to the implant via the interrogator and implant antenna coupler. After the subharmonic mixer creates the third-order intermodulation and transmits it back to the interrogator, the circulator directs this backscattered signal (4.8 GHz ± fneuro) to the filtering and amplifying block. This block is composed of several bandpass filters
32 and low-noise amplifiers (LNA) which are centered at 4.8 GHz to filter unnecessary
2.4 GHz components. With sufficient filtering and amplifying, the backscattered
signal reaches to the RF port of the interrogator mixer. Meanwhile, the second route
of the signal generator output produces 4.8 GHz harmonics after the multiplying
blocks which reach to the LO port of the interrogator mixer. The interrogator mixer
downconverts the 4.8 GHz ± fneuro backscattered signal from the RF port using the
4.8 GHz signal from the LO port to recover neural signals at fneuro. After appropriate
filtering and amplifying at the pre-amplifier, the downconverted neural signal fneuro
is displayed on osiclloscope.
The main challenge of the interrogator is the minimum backscattered signal level
that can be recovered. To determine this crucial property, the noise figure of the
interrogator should be calculated first. The noise figure is the ratio of the input
signal noise ratio to the output signal noise ratio of the system. The cascaded noise
figure (NF) could be calculated from the following formula:
F2 − 1 F3 − 1 F4 − 1 Fcascaded = F1 + + + + ... (2.2) G1 G1G2 G1G2G3
Table 2.1 shows the noise and figure of the components used in the filtering and amplifying chain. Using 2.2, the noise figure of the filtering and amplifying block which is also the noise figure of overall interrogator is 3.8 dB.
The noise of the interrogator system could be derived as follows:
Noiseinterrogator[dBm] = kT + 10 log10(BIF ) + NFRX
= −174 [dBm/Hz] + 10 log10(5 kHz) + 4 [dB]
= −133 [dBm] where -174 dBm/Hz is the thermal noise power in dBm per 1 Hz of bandwidth at room temperature, NFRX is the receiver noise figure calculated as above. The -133
33 Table 2.1: Gain and Noise Figure of Components Applied in the Interrogator Model Gain (dB) NF(dB) Gain (linear) NF(linear) BPF -2 (1 dB loss 2 0.63 1.58 (VBF-4440+) from cable) LNA (ZX60- 24.7 1.83 295.12 1.52 542LN+) BPF -1 2 0.63 1.58 (VBF-4440+) LNA (ZX60- 24.7 1.83 295.12 1.52 542LN+) BPF -1 2 0.63 1.58 (VBF-4440+) Mixer -6.6 6.6 0.22 4.57 (ZX05-73L+) LN Preamp 60 15 1,000,000 31.62 (SR 560)
dBm shows the actual noise level after the interrogator system. According to our
experiment, to further recovered the backscattered signal and visubly observed, 13
dB signal to noise ratio (SNR) is required.
The minimum detectable signal of the recevier could be determined through the following formula:
MDSRX [dBm] = Noiseinterrogator[dBm] + SNR[dB]
= −133 [dBm] + 13 [dB]
= −120[dBm]
Furthermore, the minimum detectable neuropotential signal level can be expressed as:
MDSneuro [dBm] = Receiver Sensitivity [dBm] + Lsys [dB]. (2.3)
34 where Lsys = overall system loss, and in the above Receiver Sensitivity = minimum
detectable signal level by the receiver. From 2.4, the receiver sensitivity is -120 dBm.
Thus, to guarantees a minimum detectable signal of 20 µVpp (or -90 dBm), Lsys < 30
dB. The receiver system loss, Lsys, could be divided into three major components as
expressed in 2.4:
Lsys[dB] = Lprop[dB] + Lconv[dB] + Lmatch[dB] (2.4)
In the above, Lprop = propagation loss between the implanted and the interrogator
antenna at 4.8 GHz ± fneuro, Lconv = conversion loss at the implanted mixer, and
Lmatch = impedance mismatch losses between the antenna and the mixer in the im-
planted device. To ensure Lsys < 30 dB, Lprop, Lconv and Lmatch should be minimized to the extent possible.
35 Chapter 3: Improved Interrogator System: Analysis and Design
Although the upconverted backscattered RF signal with -120 dBm could be de- modulated and observed on the oscilloscope, the fluctuation noise in the time domainis still too high for further signal processing. In this chapter, the major reason of the bad signal integrity appeared on the interrogator, phase noise and the self-mixing effect, would be discussed and analyzed. Also, the proposed phase noise reduced in- terrogator system would be present as well. According to the experiment result, the signal integrity is improved greatly compared to the previous design.
3.1 Phase Noise Analysis
The signal generator would also produce the phase noise when generating specific frequency carrier because of its own instability. In the frequency domain, the phase noise would decrease away from the center frequency and appears as a skirt centered at the carrier signal as shown in Fig 3.1. On the other hand, in the time domain, this noise appears as fluctuations after demodulated as shown in Fig 3.2. Generally, the phase noise would not interfere the minimum detectable signal level since it has no contribution to the noise floor. However, this instability fluctuation in the time
36 Figure 3.1: The phase noise behaves as a skirt centered at at the carrier signal in frequency domain.
domain degrades the demodulated signal integrity and is required to be eliminated for further signal processing.
In previous interrogator design, the signal generator would also produce 4.8 GHz harmonic and its associated phase noise while outputting the 2.4 GHz carrier. The phase noise centered at 4.8 GHz harmonic would leak to the demodulation chain due to the imperfection of the circulator. In our interrgoator setup, the phase noise would have -71.9 dBm at an offset 10 Hz, -108.4 dBm at an offset 30 Hz and -109.9 dBm at an offset 100 Hz as shown in Fig 3.1. This leaked 4.8 GHz phase noise would cover the desired backscattered signal which is generally -115 dBm in the frequency domain. In the time domain, this phenomenon represents lots of random frequency components appeared at the demodulated signal instead of the pure sinusoidal waveform.
37 Figure 3.2: The phase noise behaves as fluctuation in time domain after demodulation.
3.2 Self-Mixing and DC Offset
The signal generator not only produce the phase noise centered at 4.8 GHz but the high harmonics at 4.8 GHz. This harmonic would also leak throguh the circulator and reach at the mixer RF port with amplifying and filtering. Eventually, it would mixes with another 4.8 GHz LO signal and generate certain level of DC offset which is genrally called self-mixing . Once the DC offset is higher than the following pre- amplifier’s tolerance, it would saturate the pre-amplifier and kill the overall system.
38 Figure 3.3: Proposed phase noised reduced interrogator setup.
3.3 Phase Noise Reduced Interrogator System
To improve the interrogator’s phase noise performance and suppress the harmon- ics, the design in Fig 3.3 is proposed. Compared to the previously employed inter- rogator system, two extra bandpass filters are added between the circulator and the splitter. Both filters are centered at 2.4 GHz and are used to suppress the 4.8 GHz phase noise before it enters into the circulator. The superior performance of the fre- quency domain is displayed in Fig 3.4. The desired backscattered signal this time is uncovered by the extra two bandpass filters. Fig 3.5 shows both the demodulation output of the proposed and previous interrogator with a 100 Hz neural signal of -115 dBm backscattered level. Clearly, the addition of the two bandpass filters significantly improves the retrieved signal integrity.
39 Figure 3.4: Frequency domain of the proposed phase noise reduced interrogator.
Figure 3.5: Demodulated version of -115 dBm backscattered neural signal at 100 Hz. The plot compares the currently reported system vs. the previous one.
40 Chapter 4: Wireless and Fully-Passive Multi-Channel Neurorecorder
The wireless and fully-passive brain implant demonstrated in [13] can be used only for single-channel recording. This implanted recorder exhibited a very small footprint of 8.7 mm × 10 mm and was shown to exhibit sensitivity as high as 20
µV pp. As such, all signals generated by the human brain can be detected. However, the single channel operation of our earlier device [13] is not suited for realistic clinical applications. For the latter, we require 100s or even 1000s of channels of concurrent neuropotential recording.
In this Chapter, we propose multi-channel neuropotential recording that is wire- less and batteryless. Specifically, we design and fabricate a proof-of-concept wireless multi-channel neuropotential monitoring system with the following unique features:
1) wireless, passive and biocompatible operation,
2) concurrent recording from 8 channels by using an infrared transceiver/receiver, scalable to 100s and up to 1000s of channels,
3) neuropotential detection as low as 20 µVpp per channel,
4) validation using fresh pig skin to emulate the human skin’s properties at RF
and infrared bands,
5) extremely small power consumption (98 µW ),
41 6) compliance with the strictest Federal Communications Commission (FCC) stan- dards for patient safety [34].
As compared to previous batteryless multi-channel neuropotential recordings [35], the proposed implant exhibits: a) 28 times higher sensitivity, b) ∼2 times smaller footprint (1600 mm2 vs 3000 mm2 in [35]), c) higher scalability (3 photodiodes for
8 channels vs. 8 photodiodes for 8 channels in [35]), and d) alternative and, likely, more practical implementation (using infrared instead of visible light as in [35] for improved skin penertration).
4.1 Block Diagram and Operation Principle
Figure 4.1: Block diagram of the proposed mulitchannel neurosensing system.
The block diagram of the proposed neuropotential transceiver and receiver is de- picted in Fig 4.1. The set-up consists of 1) an implanted recorder placed just under the scalp with the recording electrodes protruding through the bone into the brain,
42 Figure 4.2: Proposed implanted infrared receiver and exterior transceiver used to toggle different neuro-channels in a wireless and passive manner. The transceiver employs 3 photodiodes to form a 3 digit code for selecting each of the 8 channels. In this condition, the three digit code is 0,1,0 and channel 3 is selected.
and 2) an external interrogator placed outside the scalp. This operation of the pro- posed neurosensor involves two processes: 1) wireless monitoring of the neural signal using a process similar to RFIDs, 2) selection/toggling of the different channels via an infrared-enabled implanted multiplexer.
The process for wireless and batteryless neural signal collection/monitoring can be summarized as follows. First, the interrogator sends a 2.4 GHz carrier signal to “turn
43 on” the implanted recorder. The mixer in the implanted device uses the 2.4 GHz
signal to generate a 4.8 GHz ± fneuro modulated signal that is eventually transmitted back to the interrogator. To obtain low conversion loss and suppress the DC term in the mixing process, an anti-parallel diode pair (ADPD) configuration is employed, as described in [13].
For toggling different channels, an extremely low-power commercial multiplexer
(Analog Devices, ADG708) is adopted. Further, to provide power and control to the multiplexer, we employed an implanted infrared receiver and an exterior infrared transceiver as shown in Fig 4.2. For this particular multiplexer, each of the 8 channels is selected via an “on-off” scheme that employs 3 implanted photodiodes (Vishay,
VEMD2000). These three photodiodes can be individually turned on and off to generate 8 different optical combinations. The optical signal is denoted as P0, P1 and
P2 as depicted in Fig 4.2. Each of the 8 combinations is then used via the multiplexer to select one of the 8 probes for recording. Notably, generation of this P0P1P2 code is done via the external infrared emitters as in Fig 4.1 and Fig 4.2. These emitters also pass their signal through the skin to the photovoltaic cell to introduce a stable DC voltage on the multiplexer of the implant. It is noted that power consumption of the implant is due to the multiplexer operation and the resistance of the photodiodes.
But the power consumption of the multiplexer operation can be reduced to trivial level by choosing high-value resistors and high-sensitivity photodiodes. Typically, when the series resistance of the photodiode is R = 1.5M Ω, see Fig 4.2, the power consumption is only 16 µW (-18 dBm).
Table 4.1 compares the proposed multichannel neuropotential monitoring sys- tem with previously reported wireless and batteryless neural recorders. The neural
44 recorders in [13, 29, 31, 32] have only one channel and are therefore not sufficient for realistic clinical applications. A batteryless multichannel system was reported in [35] but this implant has ∼ 2 times larger footprint and achieves 28 times lower sensitivity.
Also, in [35], visible light was employed to excite the photodiodes. But visible light cannot penetrate the biological tissues. Further, the design in [35] exhibited limited scalability, requiring 8 photodiodes for 8 channels (rather than 3 photodiodes as done here).
The developed brain implant and interrogator PCB layouts are shown in Fig 4.3.
For a proof-of-concept demonstration, the implant and interrogator each employed a three-layer metallization structure, occupying a footprint of 40 mm × 40 mm. Specif- ically, for the implanted neural recorder, the photovoltaic cells and the implanted antenna were placed on the top layer to receive the RF and infrared signals from the interrogator as shown in Fig 4.3 (a). The multiplexer and mixing circuits were located on the bottom layer. The middle layer contains the common ground shared by the antenna, multiplexer, and the mixing circuits. A similar structure was employed for the interrogator circuit, as shown in Fig 4.3 (b). Co-simulations were performed in
ANSYS HFSS and Keysight ADS. An in-house genetic algorithm was also employed for design optimization [36]. Below, we discuss the various components that comprise the transceiver.
4.2 Antenna Interface
The implants in [29] and [32] were designed to operate under the skull, while the implants in [13] and this work are intended for operation under the skin. Our target distance of 2.5 mm was selected according to the thickness of the pig-skin used in this
45 Table 4.1: Comparison Between Proposed vs. Previously Wireless Implanted Devices Without Integrated Circuits
Ref Type Footprint Power Channels Operation Operation Min. detectable consumption distance frequency signal 2 [29] Implanted 12 × 4 mm N/A 1 < 1.5 cm 2.4 GHz and 4.8 3.4m Vpp (in-vitro) GHz 2 [31] Implanted 39 × 15 mm 0.008 mW 1 8 mm 2.4 GHz and 4.8 200 µVpp (in-vitro)
46 GHz 2 [32] Implanted 16 × 15 mm 0.035 mW 1 ∼1.5 cm 2.4 GHz and 4.8 63 µVpp (in-vitro) GHz 2 [13] Implanted 8.7 × 10 mm 0.305 mW 1 2 mm 2.4 GHz and 4.8 20 µVpp (in-vitro) GHz 2 [35] Implanted ∼ 50 × 60 mm N/A 3 3 mm 2.45 GHz and 4. 700 µVpp (in-vitro) 9GHz 2 Proposed Implanted 40 × 40 mm 0.598 mW 8 2.5 mm 2.4 GHz and 4.8 20 µVpp (in-vitro) GHz Figure 4.3: Proposed system layout: (a) brain implant, and (b) exterior interrogator. The corresponding areas both are 40 mm × 40 mm
experiment. This aligns well with real-world applications. Specifically the thickness of human head skin ranges from 2 mm to 4 mm, according to age [37]. In this Chapter, the implanted and interrogator antennas have been optimized for the aforementioned
2.5mm distance.
The implanted and interrogator antennas were designed to overcome propagation loss at 2.4 GHz and 4.8 GHz ± fneuro. The lower propagation loss at 2.4 GHz would
47 reduce the require LO power sent by the interrogator, hence ensure to comply with the FCC regulations [34]. On the other hand, the lower propagation at 4.8 GHz would increase the backscattered modulation signal and imporve the overall system sensitivity. To achieve dual band radiation, an E-shaped patch antenna geometry was adopted for the implanted and interrogator antennas, shown in Fig 4.3. The simulated transmission coefficient |S21| between the implanted and interrogator antennas as a function of the depth into the skin tissue is given in Fig 4.4. As seen, |S21| = −6dB|−
14dB at 2.4 GHz|4.8 GHz when the implant is 1 mm below the skin. Notably, our interrogator/implanted antenna pair is designed and optimized in the reactive near-
field mode. Because of the poor antenna gain in the far-field mode, any unintended
2.4 GHz signal is precluded from reaching the implant. Even if a strong interference
2.4 GHz signal was present, the power would not be sufficient to activate the brain implant given the poor antenna gain and tissue loss.
4.3 Implanted Mixer
The goal of the mixer design is to decrease the conversion and mismatch losses
(Lconv and Lmatch). The employed mixer consists of the following:
1) Antiparallel Diode Pair (APDP),
2) Inductor that provides the circuit’s route to ground for the low-frequency neu- ropotential signal (fneuro),
3) Capacitor that acts as short for the 2.4 GHz to isolate the DC from the neu- ropotential signal (fneuro),
4) Matching circuit to minimize mismatch losses between the antenna (50 Ω) and mixer. To further reduce the number of lumped elements, the inductor and
48 Figure 4.4: Simulated transmission coefficient, |S21|, between the transmitting and receiving antennas for different pig skin depth.
capacitor were replaced by short- and open-circuited transmission lines, respectively.
The final mixer and matching circuit PCB layout is shown in Fig 4.3 (a). It is noted that commercial APDP diodes (Avago Technologies, HSMS-286C) with good circuit balance were employed for subharmonic mixing.
4.4 Infrared Transceiver and Receiver Design
The infrared exterior transceiver and implanted receiver are intended to provide
1) stable DC voltage of 3 V, and 2) a control signal to the implanted multiplexer for toggling among the different channels. To ensure batteryless operation, the DC
49 voltage is provided via an infrared-illuminated photovoltaic cell. The control signal
is provided by the 3 digit code generated by the photodiodes. It is important that
the infrared receiver have sufficient isolation to ensure detection of the 3 digit code.
Otherwise, the infrared emitter can lead to erroneous channel turn on. To avoid this,
highly directive photodiodes and photoemitters were selected. The patterns of the
photodiodes are given in Fig 4.5 (a) and (b).
Figure 4.5: Radiation and reception pattern for the (a) infrared emitter, (b) photo- diodes and (c) the geometry of the photodiodes and photo emitters
As seen in Fig 4.5, the radiation receiving sensitivity at 20◦ from the normal inten- sity is rather low. This ensures excellent isolation between the adjacent photodiodes.
The suggested separation between two photodiodes should be:
D = H · tanθ (4.1)
where H = thickness of the skin, and θ = infrared beam angle between the photodiode
and the neighboring emitter. Typically, 5 mm < H < 8 mm and θ < 20◦. Based on these averages, the distance between the photodiodes should be at least D ≈ 2
50 mm. That is, the goal isolation between different channels can be achieved by the designated physical placement of the photodiodes.
4.5 System Performance
4.5.1 Fabricated Prototype
The fabricated brain implant and interrogator propotypes are shown in Fig 4.6.
As mentioned earlier, for this proof-of-concept demonstration, both devices occupy a footprint of 40 mm × 40 mm. They were fabricated on Rogers RO4003C substrate
(r = 3.38, tanδ = 0.0021) of thickness 40 mils (1.016 mm). To ensure biocompati- bility, the implanted recorder was coated with a 0.7 mm-thick layer of Polydimethyl- siloxane (r = 2.8, tanδ = 0.001 [32]). This relatively lossless layer decreases the
power absorbed by the human tissue, thereby reducing losses between the implant
and interrogator antennas [38].
Selection of a higher permittivity substrate will further miniaturize the implant
by shrinking the antenna and microstrip feed lines. We have already identified high
permittivity materials (TMM 13i, r = 12.2, tanδ = 0.0019) that exhibit similar loss
tangent to the low permittivity dielectric employed in this work (RO4003C, r = 3.38,
tanδ = 0.0021). We have also identified flexible polymer-ceramic composites having
tanδ ≈ 0.0025 for r ≈ 10 [39]. That is, we are confident that we can switch to
higher-permittivity dielectrics without increasing dielectric losses.
4.5.2 Measurement Setup
The measurement setup used to validate the 20 µVpp sensitivity of our neuropoten-
tial detector for all channels is shown in Fig 4.7. As depicted, the 2.4 GHz carrier was
a pure 10 dBm sinusoid that was supplied to the interrogator using a signal generator
51 Figure 4.6: Fabricated prototypes: (a) brain implant, (b) exterior interrogator
(Agilent SG386). Also, the emulated neuropotentials at fneuro = 10 Hz to 5 kHz were generated using an arbitrary function generator (Leader, LFG-1300). For the infrared control circuit, shown in Fig 4.7, manual turn on/off switches were used. To emulate the human head tissues, the neural recorder was immersed inside a four-layer head phantom shown in Fig 4.7. For this phantom, the bone, dura/gray matter, and white matter tissues were formulated using the recipes in [40]. For the skin, we used fresh
52 Figure 4.7: The neurosensing measurement setup with the layered head phantom.
pig skin layer [41]. Conventional phantoms accurately emulate the dielectric proper- ties of human tissues, but not their light penetration properties. With this in mind, pig skin (viz. actual biological tissue) is selected as to better emulate both RF and infrared signal propagaion. The permittivity and loss tangent of the pig-skin were measured using the Agilent 85070E Dielectric Probe Kit and further compared vs. the theoretical skin properties [14] shown in Fig 4.8. The reader is referred to [42–44] for infrared losses through the scalp.
4.5.3 Minimum detectable signal
As already noted, our goal is to detect neural signals down to 20 µVpp. To achieve this, we must minimize the overall losses between the interrogator and implant. This
53 Figure 4.8: (a) Measured permittivity, and (b) measured conductivity of pig-skin versus the reference skin properties reported in [14].
overall system loss, Lsys, is defined as the difference between the power of the neu- ropotential signals detected at the input of the implanted device (fneuro) and the
backscattered power received at the interrogator (4.8 GHz ± fneuro). Based on sim-
ulations, we found that Lsys = -18.2 dB when 10 Hz < fneuro < 5 kHz for a distance
of 2.5 mm between the implant and interrogator. Our measurements using a pig skin
thickness of 2.0 mm indicated that Lsys = -25 dB. Although Lsys is 7 dB larger than simulation, it still meets the system loss criteria of Lsys < 30 dB. Fig 4.9 shows exam- ple demodulated backscattered signals received at the interrogator. For these example waveforms, the emulated neuropotentials were as low as 20 µVpp. The observed higher
noise can be attributed to the preamplifier filter used at the interrogator. This noise
can be suppressed by using filters.
54 Figure 4.9: Time-domain performance while recovering the minimum detectable neu- ral signal (Vin = MDSneuro = 20µVpp)
4.5.4 Specific Absorption Rate (SAR)
To confirm that the proposed neurosensing system meets the FCC safety guide- lines, a 10-cm-radius spherical head model was employed. This model was composed of skin (9.2-cm-radius), bone (8.5-cm-radius), gray matter (7.8-cm-radius), and white matter (7.5-cm-radius) tissues [45]. The implant was placed in the middle of the skin layer.
For SAR simulation, the brain implant was placed below the skin layer and the interrogator was placed right above it. It was found that the average SAR for 1 g of tissue with 6 dBm carrier power at 2.4 GHz had a max value of SAR1g = 0.368
W/kg. This value conforms to the strictest FCC requirements of SAR1g < 1.6 W/kg for uncontrolled environment exposure [34].
55 Figure 4.10: SAR performance averaged over 1 g with input carrier power at 2.4 GHz and 6 dBm.
56 Chapter 5: Wireless and Fully-Passive Implant Match to High-Impedance Electrodes
When it comes to real clinical in-vivo experiments, two major problems are need
to be solved. They are the high impedance and the DC offset of the brain electrodes.
Implant designs in the previous chapters did not consider these two factors. This
neglected high impedance and DC offset are anticipated to degrade system sensitivity
and lead to unsatisfying experiment results during clinical testing. In this chapter,
we will provide an introduction to basic electrodes, their modeling and how they may
affect system performance. Next, a novel technique with self-biasing BJT is proposed
to solve both the high impedance and DC offset concerns. A fabricated protoype and
its in-vivo validation are also presented.
5.1 Introduction to Electrodes and Detection Mechanism
Currently there are several methods to record human brain activity like electroen- cephalography (EEG), functional magnetic resonance imaging (fMRI) and magne- toencephalography (MEG). However, none of these cab detect signals from single neurons and their precision is much better than single-unit recording. A single unit is defined as a single, firing neuron whose spike potentials are distinctly isolated by the recording microelectrodes [46]. These microelectrodes are typically fine-tipped,
57 high impedance conductors. They are primarily glass micro-pipettes filled with elec- trolyte similar to the intracellular fluid or metal microelectrodes made of platinum or tungsten based on the different types of the recording [46]. The types of the single-unit recording could be classified as intracellular, extracellular and combined recordings [47].
Intracellular recording is achieved by inserting the electrodes through the cell membrane to record neural activity within the cell. It can provide information of the steady and resting membrane voltage, postsynaptic potentials. Glass micropipettes are usually chosen for intracellular recording because they have higher chance to penetrate and retain the neural cells as compared to the metal electrodes. Fig 5.1 shows the patch clamp technique, which allows the recording from a single ion channel on the cell membrane [48].
Extracellular single unit recording is achieved by placing the electrodes close to the cell surface. It can provide spike information by measuring the voltage potential change outside the cell. Metal electrodes are usually adopted because of their high low-frequency impedance and low high-frequency impedance characteristic [49]. Gen- erally, spike signals range in the higher frequency, while local field potentialx range in the lower frequency. With this special impedance frequency characteristic, the noise, local field potential, is suppressed. This leads to high signal-noise ratio. Also, the metal electrodes have sufficient mechanical rigidity for puncturing the biological tis- sues with a minimum tip area. The majority of metal microelectrodes are composed of stainless steel, tungsten, platinum/iridium, or pure iridium according to different applications [50].
58 Figure 5.1: Patch clamp configurations: A diagram showing five commonly used patch clamp configurations. [15]
When the electrode is placed into the liquid ionic conductor (electrolyte), the cations and anions react with the electrode and create an electrolyte-electrode double layer as show in Fig 5.2 These layers were first discovered by Hermann von Helmholtz and were called Helmholtz double layers after his name [51, 52]. The first layer is composed of the ions adsorbed onto the surface of the charged electrodes. On the other hand, the second layer is composed of the opposite charge ions attracted by the
Coulomb force. With these two opposite charge layers, the metal electrodes act like a capacitor in parallel with a resistor. Then the neural potential generated by the neurons can be measured and recorded by the electrodes within this interface.
59 Figure 5.2: Models of the electrical double layer at a positively charged surface: (a) the Helmholtz model, (b) the GouyChapman model, and (c) the Stern model, showing the inner Helmholtz plane (IHP) and outer Helmholtz plane (OHP). The IHP refers to the distance of closest approach of specifically adsorbed ions (generally anions) and OHP refers to that of the non-specifically adsorbed ions. The OHP is also the plane where the diffuse layer begins. d is the double layer distance described by the Helmholtz model. φ0 and φ are the potentials at the electrode surface and the electrode/electrolyte interface, respectively [16].
5.2 Modeling and Measurement of Electrode Impedance and DC offset voltage
The equivalent model of a sub-cranial electrode is well analyzed in the litera-
ture [53] and is illustrated in Fig 5.3. As seen, the circuit model is composed of the
electrolyte solution resistance (Rs), the double layer interface resistance and capaci-
tance (Re and Ce), and the metal electrode resistance (Rm). Generally, the solution
resistance (Rs) and the metal electrode resistance (Rm) are negligible as compared to the double layer interface resistance and capacitance (Re and Ce). Because of
60 Figure 5.3: Equivalent circuit model of the electrode.
the double layer capacitance, the impedance of the electrode behaves like a complex number and changes according to the frequency. Referring to Fig 5.3, this electrode impedance degrades the signal amplitude at the input of the neuropotential monitor- ing system (Vin) by means of a voltage divider:
Za Vin(ω) = Vsig(ω) × (5.1) Za + Ze
where Vsig is the neuropotential amplitude generated within the brain, Ze is the complex electrode impedance, and Za is the complex neuro-sensor impedance. The voltage divider equation shows that the magnitude of Za would decrease Vin and cause phase distortion when Ze >> Za [53].
To better understand the effect of electrodes on the neurosensing system perfor- mance, we proceed to characterize via electrode impedance spectroscopy the impedance of clinical macro-electrodes currently used for Deep Brain Stimulation (DBS) surgery at Ohio State’s Wexner Medical Center (FHC microTargeting mTD differential elec- trode) [54]. A potentiostat with a three-electrode setup is employed to measure the electrode impedance. Measurement results of electrode impedance magnitude and
61 Figure 5.4: Clinical brain electrode impedance measurement.
phase as a function of frequency are shown in Fig 5.4. As seen, the impedance magnitude reduces while increasing frequency. At the smallest frequency where neu- ropotentials may be identified, viz. at 0.5 Hz, the electrode impedance is as high as
33 kΩ.
Added to the above, the electrochemical reaction that takes place at the electrode interface will give rise to different DC voltage levels across different recording elec- trodes [55–57]. This voltage difference, which may be as high as 50 mV , known as
the DC offset voltage, can have detrimental consequences. In conventional battery-
enabled ICs, this offset is known to saturate the first-stage neural amplifier, while in
our previous fully-passive neurosensing system [13], the offset is anticipated to change
the bias point of the employed antiparallel diode pair (APDP) mixer and deteriorate
the sensitivity.
62 Figure 5.5: Block diagram of the proposed impedance-matching neurosensing system.
These real-world concerns for clinical electrodes are hereafter taken into account.
Given that neural signals may be as low as 0.5 Hz in frequency, a capability to match
to at least 33 kΩ of electrode impedance is necessary for the neurosensing system.
Concurrently, the ability to overcome the DC offset voltage is a key requirement for
the design.
5.3 Block Diagram and Operation Principle
The block diagram of the proposed neurosensing system with impedance-matching
capabilities is shown in Fig 5.5. To resolve the aformentioned impedance issues, the
operating principle of the improved technique used to passively match the electrode-
implant interface and eliminate the DC offset is summarized in Fig 5.6. As seen, the
implant consists of: a) an implantable antenna used for wireless backscattering, b) a
Schottky diode that acts as a rectifier in DC mode and as a mixer in RF mode, c)
a matching network used to mitigate Lcircuit between the antenna and the Schottky diode (composed of two microstrip lines with open- and short-ended microstrip lines),
63 Figure 5.6: Proposed neural implant design: (a) DC mode, and (b) RF mode.
d) the high-impedance clinical electrodes, and e) a PNP Bipolar Junction Transistor
(BJT) added between the Schottky diode and the electrodes to serve as an impedance buffer.
Circuit operation is composed of two modes, viz. the DC mode (Fig 5.6 (a)) and the RF mode (Fig 5.6 (b)), as analyzed below.
1) DC Mode. To activate the neural sensor, the interrogator transmits a 2.4 GHz carrier signal. Once received by the implant, the Schottky diode acts as a rectifier that serves to create DC current and self-bias the BJT per Fig 5.6 (a). Following biasing of the BJT, brain signals coming from its base should pass through the BJT
64 Table 5.1: Node Voltage and BJT Operation Regions Voltage B-E Junction B-E Junction Mode
VE < VB < VC Reverse Forward Reverse-active
VE < VB > VC Reverse Reverse Cut-off
VE > VB < VC Forward Forward Saturation
VE > VB > VC Forward Reverse Forward-active
and, eventually, get upconverted by the Schottky diode. To do so, the BJT emitter is connected right after the Schottky diode, while its collector is connected to ground.
Comparing the voltages at the base (VB), collector (VC ), and emitter (VE) terminals, the BJT may either operate in the forward-active region (VE > VB > VC ) or the saturation region (VE > VB < VC ), Table 5.1. In both cases, signal may flow from the base to the emitter, while the DC voltage at the base may be neglected. This unique feature implies tolerance to DC offset. Simulations indicate that the input impedance of the self-biasing BJT circuit is 219 kΩ and remains almost constant across the entire neural frequency range (0.5 Hz to 1 kHz). Due to the high input impedance of the
BJT, the circuit of Fig 5.6 can readily match to the high-impedance electrodes.
2) RF Mode. The Schottky diode now serves as a mixer, as shown in Fig 5.6
(b). That is, the diode utilizes the 2.4 GHz carrier signal to upconvert the brain signals (at frequency fneuro) and give rise to the third-order harmonic component
(4.8 GHz ± fneuro). This upconverted signal is backscattered toward the interrogator and is, eventually, demodulated to recover the neuropotentials in the time domain.
65 Figure 5.7: Measurement set-up used to assess the neurosensing system performance.
5.4 System Performance
5.4.1 Measurement Setup
The in-vitro measurement setup used to validate the neurosensing system of
Fig 5.6 is shown in Fig 5.7. As depicted, a signal generator (Agilent SG386) feeds a
2.4 GHz carrier with 10 dBm signal level to the interrogator. An arbitrary function generator (Keysight 33500B) emulates neuropotentials as sinusoidal waveforms (at frequency fneuro). To consider a worst-case scenario for electrode impedance in this study, a 33 kΩ resistor is used to represent this impedance per Fig 5.4. The inter- rogator then demodulates the neuropotentials in the time domain. The demodulated neuropotentials are then visualized via an oscilloscope.
66 Figure 5.8: Fabricated BJT Circuit (a) directly connected to circulator, (b) connected to antenna in air, and (c) connected to antenna in pig skin
5.4.2 Stand-Alone Circuit Performance
As a first step, performance of the implanted circuit is tested in a stand-alone
wired configuration. That is, the implanted antenna is not considered in the design,
but rather the implanted circuit is directly connected to a circulator, as shown in
Fig 5.8. Here, a proof-of-concept circuit, 56.28 mm × 28.49 mm in size, is considered.
The circuit is fabricated on Rogers RO4003C substrate (r = 3.38, tanδ = 0.0021) of
67 thickness 32 mils (0.813 mm). Miniaturization is outside the scope of this particular work, yet can be readily performed via techniques explored in the past [13]. Referring to Fig 5.8, the 2.4 GHz carrier is set as the input to port 1 of the circulator and is routed directly to the implant through port 2. The implant mixes the carrier with the emulated neural signals and outputs the 4.8 GHz ± fneuro product to port 2 of the circulator. The latter signal is routed to port 3 of the circulator where it is, eventually, demodulated and plotted in the time domain.
Results show that neural signals as low as 100 µVpp can be retrieved for an electrode impedance of 33 kΩ. An example demodulated waveform at 100 Hz is shown in Fig 5.9
(dashed/red), at the minimum detectable level of 100 µVpp. Expectedly, higher signal levels result in less noisy waveforms, while smaller electrode impedances result in improved sensitivity. For comparison, and assuming the same electrode impedance of 33 kΩ, the sensitivity of the neurosensing system in [13] is 50 times lower.
5.4.3 Integrated System Performance
Performance of the complete wireless system is then validated in free space (Fig 5.8
(b)) and via a tissue-emulating model (Fig 5.8 (c)). In both cases, the implant circuit of Fig 5.8 (a) is attached to an antenna that serves as the wireless interface between the neuro-sensor and the interrogator. Here, the patch antenna design reported in [13] is considered, which exhibits dual-band resonances at 2.4/4.8 GHz and a footprint of
40 mm x 40 mm. Again, sensor miniaturization falls outside the scope of this work, yet can be readily performed using already available techniques [13]. The interrogator antenna follows the design in Chapter 4, while the overall system layout follows the design in the Chapter 2.
68 Figure 5.9: The demodulated time domain 100 Hz signal of (a) circuit alone Vin = 100 µVpp, (b) with the antenna in air Vin = 100 µVpp, and (c) with the antenna in pig skin Vin = 200 µVpp in series with a 33 kΩ resistor
Referring to Fig 5.8 (b), the implanted and interrogator antennas are placed in free-space with a distance of ∼ 0.1 mm between the two. Results show that neural signals as low as 100 µVpp can be retrieved for an electrode impedance of 33 kΩ.
An example demodulated waveform at 100 Hz is shown in Fig 5.9 (dotted/black), at the minimum detectable level of 100 µVpp. Referring to ig 5.8 (c), the implanted antenna is placed under a 2 mm-thick layer of pig skin. The permittivity and loss tangent of the pig skin are measured using the Agilent 85070E Dielectric Probe Kit and further compared vs. the theoretical skin properties [14] shown in Fig 5.10.
Results in this case show that neural signals as low as 200 µVpp can be retrieved for an electrode impedance of 33 kΩ. An example demodulated waveform at 100
Hz is shown in Fig 5.9 (solid/blue) at the minimum detectable level of 200 µVpp.
69 Figure 5.10: a) Measured permittivity, and (b) measured conductivity of pig skin versus the reference skin properties reported in [14].
This slight degradation in performance is expected given the losses associated with biological tissues.
Indeed, Fig 5.11 compares the transmission coefficient between the two antennas in free space and with pig skin used as a separation medium. As seen, the transmission coefficient degrades by ∼ 3 dB at 2.4 GHz and by ∼ 8 dB at 4.8 GHz. As mentioned in Section 5.4.3, higher signal levels result in less noisy waveforms, while smaller electrode impedances result in improved sensitivity. For comparison, and assuming the same electrode impedance of 33 kΩ, the sensitivity of the neurosensing system in
[13] is 25 times lower. Specific Absorption Rate (SAR) simulations are also performed for the 10-cm-radius spherical head model of [13]. The simulation is shown in Fig 5.12 and indicate that SAR averaged over 1g of tissue equals 0.862 W/kg (at 6 dBm power).
This value conforms to the strictest FCC requirements of SAR1g < 1.6 W/kg for uncontrolled environment exposure [34].
70 Figure 5.11: Measured transmission coefficient (S21) of the implanted and interrogator antenna system (a) through air, and (b) through pig skin.
5.4.4 DC Offset Tolerance
To verify the DC offset tolerance of the implant, the measurement setup of Fig 5.8
(a) is adopted. In this case, the function generator this time provides the emulated neuropotentials as well as an unwanted DC offset voltage. Referring to Fig 5.13, demodulated waveforms are presented for a 100 µVpp and 100 Hz neural signal subject to : 0 V (solid/red), +50 mV (dashed/black) and -50 mV (dotted/blue) DC offset.
71 Figure 5.12: SAR performance averaged over 1 g with input carrier at 2.4 GHz.
As seen, the proposed implant can tolerate even the most extreme ± 50 mV DC offset at the minimum detectable level of 100 µVpp.
72 Figure 5.13: Demodulated waveform of a 100 µVpp and 100 Hz signal subject to: (a) 0 V offset, (b) +50 mV offset, and (c) -50 mV offset in series with a 33 kΩ resistor
73 Chapter 6: Conclusion
6.1 Summary
This dissertation introduced the clinical consideration of wireless fully-passive neuro recorders and discussed related solutions.
Chapter 2 provided an overview of the wireless fully-passive neuro recorders.
Three major parts of the system which entailed the mixer circuit, antenna interface and interrogator system were described. The critical parameter, system sensitivity, was defined and derived from the equations 2.4 and 2.3. In order to increase the system sensitivity, the overall system loss (Lsys) was required to be reduced.
Chapter 3 described the effect of phase noise on the interrogator and proposed a technique to reduce it. The phase noise created by the signal generator was shown to leak into the amplifying and filtering blocks due to non-idealities of the circula- tor. Specifically, after downconverting with the interrogator mixer, the phase noise appeared as random fluctuations and degraded the desired neuropotential signal in- tegrity. By adding two bandpass filters between the circulator and the splitter, the aforementioned phase noise was reduced greatly. Hence, the demodulated signal in- tegrity was improved accordingly.
74 Chapter 4 discussed the scalable multichannel neuro recorder to address multi-
probe demands for clinical applications. The infrared transceiver/receiver and a mul-
tiplexer were employed to increase the number of channels in the recorders. Under
the proposed schematic, N infrared photoemitters/diodes were used to achieve 2N
channels recordings while keeping the system sensitivity at 20 µVpp.
Chapter 5 illustrated the high impedance and DC offset of the clinical brain elec- trodes and the effect to the wireless neuro recorders. Impedance matching and DC offset issues were addressed by implementing a self-biasing PNP BJT technique. The self-biasing BJT increased the overall system impedance and maintained system sen- sitivity while connecting with the clinical brain electrode. The unique operating regions of the PNP BJT guaranteed the normal operation while facing the high DC offset voltage. With the proof-of-concept fabrication circuits, the implant was shown to recover 200 µVpp signal level in series of 33 kΩ emulated brain electrode impedance and ± 50 mV DC offset.
Together, these proposed techniques would lead the wireless neuro recorders to be applicable in real, in-vivo clinical applications.
6.2 Key Contributions
The contributions of this dissertation are two-fold: a) multichannel system and b) brain electrode interface impedance-matching system.
With regards to the multichannel system, a proof-of-concept wireless neuropo- tential monitoring system was designed and demonstrated with the following unique features:
1) wireless, passive and biocompatible,
75 2) concurrent recording from 8 channels using an infrared transceiver/receiver, scalable to 100s and up to 1000s of channels,
3) neuropotential detection of 20 µVpp per channel,
4) validation using fresh pig skin to emulate human properties at RF and infrared bands,
5) extremely small power consumption (98 µW ),
6) compliance with the strictest Federal Communications Commission (FCC) stan- dards for patient safety.
With regards to the brain electrode interface impedance-matching system, the following is a list of accomplishments:
1) identified the electrode impedance and DC offset effects,
2) proposed and fabricated a proof-of-concept impedance-matching implant using a self-biased BJT circuit,
3) achieved 200 µVpp signal sensitivity even when a 33 kΩ brain electrode impedance is used to connect to the 50 Ω source impedance.
4) identified phase noise effect and proposed a reduced phase noise interrogator to ensure proper recovery of signal integrity.
6.3 Future Work
6.3.1 In-V ivo Measurement and Result
The ultimate goal of the wireless passive implanted system is to record neuropo- tentials in-vivo. In the past, we successfully recovered ECG (Electrocardiography) signals from the human body using commercial electrodes attached to our device as
76 depicted in Fig. 6.1. This experiment demonstrated that the proposed wireless im- plant can record and recover the human neural signals. To further verify our circuits in clinical settings, experiments of neural recording must be conducted as follows:
Figure 6.1: Wired ECG recording v.s. wireless ECG signal recorded by the proposed wireless implant
77 6.3.2 In − vivo Measurements in Rats
One approach is to verify the proposed wireless implant using in−vivo experiments
with rats. In [58], the paws of the rat were connected to an isolated pulse simulator.
By stimulating the limbs of the rat, somatosensory evoked potentials were produced
in the rat brain accordingly. Our wireless implant circuit was shown to recover these
evoked neuropotentials. The wired and wireless time domain signals were shown to
be similar to each other as shown in Fig. 6.2
Figure 6.2: Neural activity recording. (a) Schematic of the area of interest (primary somatosensory cortex, hind limb region contralateral to the stimulated hindpaw) and probe placement during recording. (b) Representation of signal averaging procedure used during the signal processing stage of neural recording. (c) Overlay of WiNS and wired extracted somatosensory evoked potential (SSEP) after processing the signal.
This experiment verified that the proposed circuits are capable of recording in- vivo rat brain signals. The experiments also provided confidence to pursue further complex human brain recording.
78 6.3.3 In − vivo Measurements in Human Subjects
Our team also worked with the Wexner Medical Center at the Ohio State Univer- sity to conduct experiments with human subjects. While the neurosurgeon performed
Deep Brain Stimulation surgery for patients with Parkinson’s disease, our team using the recorder to obtain human brain signal. The setup is shown in Fig. 6.3
Figure 6.3: The experiement setup for the proposed implant system during Parkin- son’s disease surgery
79 So far, we have not been able to recover human brain signals using this setup. A major reason is the noisy environment in the operating room. To address the latter, more sophisticated signal processing procedures are required. This implies further research.
6.3.4 Recorded Signal Processing
As noted, the operating room in the hospital is a noisy environment to conduct the experiment due to many equipments in the room. These noise sources contribute either constant frequency noise like 50/60 Hz from the power cable or random fluc- tuations in the time domain. Fig. 6.4 displays the typical wired recording signal in
10 seconds length.
Since the proposed neurosensing process is based on analog amplitude modula- tion, it’s relatively vulnerable to noise from surrounding hospital equipments. There- fore, additional signal processing procedures are required to obtain the desired neural signal. Typically, notch filtering, bandpass filtering, moving average and window averaging are employed for signal processing.
The notch filter is employed when the noise frequency is known (like 50/60 Hz from the power cable and 120 Hz from the line-powered electric light source). The bandpass filter is mainly used to extract and isolate specific neural signals like the beta wave from 12.5 Hz to 30 Hz. Also, the moving average and window average can be used to “smooth” the recorded signal and further reduce the random fluctuations.
However, there are still challenges on how to distinguish the neural signal from the given recording like Fig. 6.4 and the window size of the moving/window averaging.
80 Figure 6.4: Wired recording neural signal in total 10 seconds length
More techniques and methods are further required to obtain the less noisy and robust neural signal.
6.3.5 Channel Scalability
In our proposed multi-channel system, channel scalability is achieved using N photodiodes for 2N channels. For the 8-channel system reported in this work, 3 photodiodes are employed. Specifically, using 10 photodiodes, we can reach 210 = 1024 channels. To maintain a small implant size, customized photodiodes and multiplexers would also be required. The former implies frequency-dependent photodiodes that
81 use both frequency and physical separation to reduce total footprint. We would also
require miniature in-house multiplexer designs. In terms of power consumption, if the
photodiode serial resistance is kept at the MΩ level, the estimated power consumption by a 10-photodiode layout will be similar to that of the 3-photodiode layout reported in this work.
6.3.6 System Miniaturization
For our proof-of-concept circuit board, the footprint of the circuit was relatively large (either 40 mm × 40 mm in Chapter 4 or 56.28 mm × 28.49 mm in Chapter
5). This is not suitable due to its large size. The main reason of the relatively large size is because of the lower dielectric constant substrates (RO4003C, r = 3.38, tanδ = 0.0021). As is well known, higher substrate dielectric constants reduce the size of the RF circuit and antenna size. Therefore, the selection of a higher permittivity substrate will further miniaturize the implant by shrinking the antenna and microstrip feedline length. We have already identified high permittivity materials (TMM 13i,
r = 12.2, tanδ = 0.0019) that exhibits similar loss tangent to the low permittivity dielectric employed in this work. We have also identified flexible polymer-ceramic composites having tanδ ≈ 0.0025 for r ≈ 10 [39]. And these substrates should be
used in the future.
Another approach to further reduce the RF circuit size is to design different im-
plantable antenna. Misalignment of the implant and interrogator should be eliminated
for greater efficiency.
82 6.3.7 Antenna Misalignment and Choice
As described in Chapter 2, the implant and interrogator antenna coupler must operate in close proximity viz in the reactive region. Therefore coupling is vulnerable to misalignment.
Fig. 6.5 shows that even 5 mm of misalignment has a significant connection im- pact. Therefore, it may be appropriate to consider other methods of connections. As
Figure 6.5: Transmission loss of the interrogator-received signal at a) 2.4 GHz and b) 4.8 GHz. (Credit to Katrina Guido)
an alternative, a bio-matched horn antenna for on-body transmission was considered in [59]. The horn uses distilled water for better matching to the human tissue as depicted in Fig. 6.6. Also, it could provide a stable transmission coefficient to the implant’s antenna as shown in Fig. 6.7. It is clear that this approach has better mis- alignment tolerance vs the results in Fig. 6.5. This positional misalignment tolerance
83 Figure 6.6: (a)side view of the published antenna, and (b) fabricated prototype.
of the antenna makes it a potential candidate to replace the interrogator and implant antenna pair.
84 Figure 6.7: Transmission loss at 2.4 GHz between the Bio-Matched Horn and an implantable patch antenna immersed at: (a) 4 mm, and (b) 20 mm, as a function of positional misalignment.
85 Appendix A: MATLAB Code for Genetic Algorithm
A.1 Main Code
% Path and File Setup clc;
close all;
clear all;
% change this;
hfssExePath =’C:\Program Files\AnsysEM\AnsysEM17.2\Win64\ansysedt.exe’;
tmpPrjFile = [pwd,’\BJT V12 32mil.aedt’];
tmpScriptFile = [pwd, ’\miniMacro.vbs’];
tmpDataFile = [pwd, ’\tmpData.m’];
tmpDesignName = ’BJT SMA T1’;
tmpSetupName = ’Setup1’;
tmpSweepName = ’Sweep1’;
addpath(’library/’);
hfssIncludePaths(’library/’);
%% Interrogator one
geometry var = cell(20, 5);
geometry var(1, :) = {’lb’, ’mil’, 249.4, 100, 800};
86 geometry var(2, :) = {’lc’, ’mil’, 101.1, 100, 400}; geometry var(3, :) = {’ld’, ’mil’, 170, 50, 300}; geometry var(4, :) = {’le’, ’mil’, 77.4, 50, 700}; geometry var(5, :) = {’lf’, ’mil’, 198.9, 50, 600}; geometry var(6, :) = {’wb’, ’mil’, 125, 20, 200}; geometry var(7, :) = {’wc’, ’mil’, 231.2, 150, 700}; geometry var(8, :) = {’wd’, ’mil’, 7.92, 20, 200}; geometry var(9, :) = {’we’, ’mil’, 28.59, 20, 200}; geometry var(10, :) = {’wf’, ’mil’, 21.43, 20, 200}; lengthLimit ext = 1100; widthLimit ext = 1100;
%% Implanted one geometry var(11, :) = {’lb int’, ’mil’, 249.4, 100, 800}; geometry var(12, :) = {’lc int’, ’mil’, 101.1, 100, 400}; geometry var(13, :) = {’ld int’, ’mil’, 170, 50, 300}; geometry var(14, :) = {’le int’, ’mil’, 77.4, 50, 700}; geometry var(15, :) = {’lf int’, ’mil’, 198.9, 50, 600}; geometry var(16, :) = {’wb int’, ’mil’, 125, 20, 200}; geometry var(17, :) = {’wc int’, ’mil’, 231.2, 150, 700}; geometry var(18, :) = {’wd int’, ’mil’, 7.92, 20, 200}; geometry var(19, :) = {’we int’, ’mil’, 28.59, 20, 200}; geometry var(20, :) = {’wf int’, ’mil’, 21.43, 20, 200}; lengthLimit int = 1100; widthLimit int = 1100; minTraceWidth = 8;
87 timeStart = tic;
%% Modify the Vairables Here !!! npop = 16; % Number of Populations maxIters = 15; % Maximum Number of Iterations (Generation)
S target = [-10, -5, -5, -10];
%% Optimization Setup nvar = 20; % Number of Variables nbits = 8; % Number of Bits nconstr = 12; % Number of Constraints natsel = npop/2; % must be even number for easy execution rmut = 0.5; % start mutation rate cost = zeros(npop,1);
S parameter = zeros(npop, 4);
% Accuracy = 0.01; % accuracy required (1%).
% hasConverged = false;
%% Create an initial population that satisfies constraints pop = zeros(npop, nvar*nbits); chrom = zeros(npop,nvar); test number Local = zeros(1,npop); test number Global = 0; for i = 1:npop pass = 0; while pass = 1 test number Local(i) = test number Local(i) + 1; test number Global = test number Global + 1; pop(i,:, :) = round(rand(1, nvar*nbits));
88 geometry temp = cell2mat(geometry var(:, 4:5));
for j = 1:nvar
chrom(i, j) = geometry temp(j, 1) + ( geometry temp(j, 2) - geometry temp(j, 1)
) .* ([2.ˆ(-[1:nbits])] * pop(i,( (j-1) *nbits+1):( (j) *nbits))’);
end
pass = Restrant Check no via(chrom(i, :), lengthLimit ext, lengthLimit int,
widthLimit ext, widthLimit int, minTraceWidth, nconstr);
end
end
%% Start to simulate the parent generation
for i = 1:npop
Script writing(tmpPrjFile, tmpScriptFile, tmpDataFile, tmpDesignName, tmpSe- tupName, tmpSweepName, chrom(i, :), geometry var);
runScript = [’cscript’ ’ ’ tmpScriptFile]; system(runScript);
delScriptFile = [’del’ ’ ’ tmpScriptFile]; system(delScriptFile);
run(tmpDataFile);
disp([num2str(i), ’ solutions completed in the parents generation.’]);
tmpCost S11 2p4 = S target(1) - 20*log10(abs(S(1,1,1)));
tmpCost S21 2p4 = S target(2) - 20*log10(abs(S(1,2,1)));
tmpCost S21 4p8 = S target(3) - 20*log10(abs(S(2,2,1)));
tmpCost S22 4p8 = S target(4) - 20*log10(abs(S(2,2,2)));
cost S11 2p4 = tmpCost S11 2p4 * (tmpCost S11 2p4 < 0);
cost S21 2p4 = tmpCost S21 2p4 * (tmpCost S21 2p4 > 0);
cost S21 4p8 = tmpCost S21 4p8 * (tmpCost S21 4p8 > 0);
89 cost S22 4p8 = tmpCost S22 4p8 * (tmpCost S22 4p8 < 0); cost(i) = (cost S11 2p4)2ˆ + 2*(cost S21 2p4)2ˆ
+ 2*(cost S21 4p8)2+ˆ (cost S22 4p8)2;ˆ
S parameter(i, :) = [20*log10(abs(S(1,2,1))), (20*log10(abs(S(2,2,1)))),
(20*log10(abs(S(1,1,1)))), (20*log10(abs(S(2,2,2))))]; disp([’The S21 of 2.4GHz is ’, num2str(S parameter(i, 1)), ’ 4.8GHz is ’, num2str(S parameter(i, 2)), ’ S11 of 2.4GHz is ’, num2str(S parameter(i, 3)),
’ S22 of 4.8GHz is ’, num2str(S parameter(i, 4)),
’. The cost is ’, num2str(cost(i)), ’.’]); disp(’ ’); t = toc(timeStart); time = datevec(t./(60*60*24)); disp(time(3:6)); end
[cost, ind] = sort(cost); % arrange from best to worst
% Store parameters costbest = zeros(maxIters+1,1); costbest(1) = cost(1); var n cost = zeros(npop,nvar + 5,maxIters+1); var n cost(:,:,1) = [chrom(ind(1:npop),:) S parameter(ind(1:npop), :), cost(:)]; disp(’Parents generation are all simulated.’); disp(’ ’); disp(var n cost(:, :, 1)); toc(timeStart);
90 temp = var n cost(:, : , 1);
save(’Result of parents generation.txt’, ’temp’, ’-ASCII’, ’-append’);
rmut = rmut*2; % to start the mutation with the initial mutation rate;
%% Start Mutation
for numIter = 1:maxIters
% Natural selection
pop = pop(ind(1:natsel),:); % arrange from best to worst, remove the last natsel number population
chrom = chrom(ind(1:natsel),:); % arrange from best to worst, the last natsel number population
cost = cost(1:natsel); % remove the last natsel number cost
S parameter = S parameter(ind(1:natsel),:); % remove the last natsel number
S parameter
% Mate selection
parents num = 1:natsel;
prob = parents num(natsel:-1:1)/sum(parents num);
odds = [0 cumsum(prob)];
% Generating offspring
% - single point crossover
% temp = ceil(rand*(nvar*nbits-1));
% mask = [ones(1,nvar*nbits-temp) zeros(1,temp)];
% - uniform crossover
r = rand(1,natsel);
parents = zeros(size(pop));
91 offsprings = zeros(size(pop));
mask = round(rand(1,nvar*nbits));
for i = 1:natsel
for j = 1:natsel
if r(i)>=odds(j) && r(i)<=odds(j+1)
parents(i,:) = pop(j,:);
end
end
end
for i = 1:natsel/2
offsprings(((i-1)*2)+1,:) = mask.*parents(((i-1)*2)+1,:) +
not(mask).*parents(((i-1)*2)+2,:);
offsprings(((i-1)*2)+2,:) = mask.*parents(((i-1)*2)+2,:) +
not(mask).*parents(((i-1)*2)+1,:);
end
% - updating population with new offsprings
for i = 1:natsel
pop(natsel+i,:) = offsprings(i,:);
end
% Mutation (key for constraints to work)
rmutnew = rmut/2; % mutation rate is half every iteration
if rmutnew > (1/(nbits*nvar)) && rmutnew >= 0.02 % if mutation rate is larger than 1/length (where length is the length of the chromosome)
rmut = rmutnew; % overwrite with new mutation rate.
92 end
nmutchrom = ceil((nbits*nvar)*rmut); % number of mutations for each chromo- some based on rmut
% - calculating chromosomes (variable values)
test number Local = zeros(1,npop);
test number Global = 0;
poptemp = pop;
for i = 2:npop
pass = 0;
while pass = 1
test number Local(i-1) = test number Local(i-1) + 1;
test number Global = test number Global + 1;
poptemp(i,:) = pop(i,:);
mcolchrom = ceil(rand(1,nmutchrom)*(nbits*nvar));
for j = 1:nmutchrom
poptemp(i,mcolchrom(j))=abs(pop(i,mcolchrom(j))-1);
end
for k = 1:nvar
chrom(i, k) = geometry temp(k, 1) + ( geometry temp(k, 2) - geometry temp(k,
1) ) .* ([2.ˆ(-[1:nbits])] * poptemp(i,( (k-1) *nbits+1):( (k) *nbits))’);
end
pass = Restrant Check no via(chrom(i, :), lengthLimit ext,
lengthLimit int, widthLimit ext, widthLimit int, minTraceWidth, nconstr);
if pass
93 for j = 1:nmutchrom
pop(i,mcolchrom(j))=abs(pop(i,mcolchrom(j))-1);
end
end
end
end
% Calculating cost for new generation (after crossover selection and mutation)
for i = 2:npop
Script writing(tmpPrjFile, tmpScriptFile, tmpDataFile, tmpDesignName, tmpSe- tupName, tmpSweepName, chrom(i, :), geometry var)
runScript = [’cscript’ ’ ’ tmpScriptFile]; system(runScript);
delScriptFile = [’del’ ’ ’ tmpScriptFile]; system(delScriptFile);
run(tmpDataFile);
disp([num2str(i), ’ solutions completed in the ’, num2str(numIter), ’ generation.’]);
tmpCost S11 2p4 = S target(1) - 20*log10(abs(S(1,1,1)));
tmpCost S21 2p4 = S target(2) - 20*log10(abs(S(1,2,1)));
tmpCost S21 4p8 = S target(3) - 20*log10(abs(S(2,2,1)));
tmpCost S22 4p8 = S target(4) - 20*log10(abs(S(2,2,2)));
cost S11 2p4 = tmpCost S11 2p4 * (tmpCost S11 2p4 < 0);
cost S21 2p4 = tmpCost S21 2p4 * (tmpCost S21 2p4 > 0);
cost S21 4p8 = tmpCost S21 4p8 * (tmpCost S21 4p8 > 0);
cost S22 4p8 = tmpCost S22 4p8 * (tmpCost S22 4p8 < 0);
cost(i) = (cost S11 2p4)2ˆ + 2*(cost S21 2p4)2ˆ +
2*(cost S21 4p8)2+ˆ (cost S22 4p8)2;ˆ
94 S parameter(i, :) = [20*log10(abs(S(1,2,1))), (20*log10(abs(S(2,2,1)))),
(20*log10(abs(S(1,1,1)))), (20*log10(abs(S(2,2,2))))];
disp([’The S21 of 2.4GHz is ’, num2str(S parameter(i, 1)),
’ 4.8GHz is ’, num2str(S parameter(i, 2)), ’ S11 of 2.4GHz is ’,
num2str(S parameter(i, 3)), ’ S22 of 4.8GHz is ’,
num2str(S parameter(i, 4)), ’.
The cost is ’, num2str(cost(i)), ’.’]);
disp(’ ’);
t = toc(timeStart);
time = datevec(t./(60*60*24));
disp(time(3:6));
end
[cost, ind] = sort(cost);
% Store parameters again
costbest(numIter+1) = cost(1);
var n cost(:,:,numIter+1) = [chrom(ind(1:npop),:), S parameter(ind(1:npop), :), cost(:)];
disp([num2str(numIter), ’ generation are all simulated.’]);
toc(timeStart);
disp(’ ’);
disp(var n cost(:, :, numIter+1));
temp = var n cost(:, : ,numIter+1);
save([’Result of ’, num2str(numIter),’ generation.txt’],
’temp’, ’-ASCII’, ’-append’);
95 end
hfssRemovePaths(’library/’);
rmpath(’library/’);
for i = 1:(maxIters+1)
temp = var n cost(:,:,i);
save(’Allresults.txt’, ’temp’, ’-ASCII’, ’-append’);
end
var n cost append = dlmread(’Allresults.txt’);
[allcost, ind] = sort(var n cost append(:,nvar+3));
sort var n cost = var n cost append(ind,:); % arrange from best to worst, remove the last natsel population
A.2 Script Writing
function Script writing(tmpPrjFile, tmpScriptFile, tmpDataFile,
tmpDesignName, tmpSetupName, tmpSweepName, chrom, geometry var)
fid = fopen(tmpScriptFile, ’wt’);
hfssOpenProject(fid, tmpPrjFile); % open a particular project file
hfssSetDesign(fid, tmpDesignName); % choose a design within the project file
for i = 1:size(geometry var)
hfssVariableChange(fid, geometry var{i, 1},
chrom(i), geometry var{i, 2}) % change local variables
end
hfssSolveSetup(fid, tmpSetupName); % solve setup
hfssSaveProject(fid, tmpPrjFile, true);
96 hfssExportNetworkData(fid, tmpDataFile, tmpSetupName, tmpSweepName);
hfssCloseActiveProject(fid)
fclose(fid);
end
A.3 Restraint Check
function pass = Restrant Check(chrom, lengthLimit ext, lengthLimit int, width-
Limit ext, widthLimit int, minTraceWidth, nconstr)
test = 0;
% constraint 1
if (chrom(1)+chrom(2)) < lengthLimit ext % lb + lc < lengthLimit ext (Length)
Grond Plane Length = 2*la + lb + lc
test = test + 1;
end
% constraint 2
if (chrom(1)-chrom(3)-chrom(4))>0 % lb - ld - le > 0
test = test + 1;
end
% constraint 3
if (chrom(3)+chrom(2)-chrom(5))>0 % ld + lc - lf > 0
test = test + 1;
end
% constraint 4
97 if (chrom(7)-chrom(6)-2*chrom(8))> 2*minTraceWidth % wc - wb - 2*wd >
2*minTraceWidth
test = test + 1;
end
% constraint 5
if (chrom(7)+2*chrom(9))< widthLimit ext % wc + 2*we < widthLimit ext
test = test + 1;
end
% constraint 6
if (chrom(9)-chrom(10)) > minTraceWidth % we - wf > minTraceWidth
test = test + 1;
end
% constraint 7
if (chrom(6)-2*chrom(11))>0 % wb - 2*viarad > 0
test = test + 1;
end
% constraint 8
if (chrom(12)-chrom(11))>0.1 % holerad - viarad > 0.1mm (4mils)
test = test + 1;
end
% constraint 9
if (chrom(13)+chrom(14)) < lengthLimit int % lb int + lc int < lengthLimit int
(Length) Grond Plane Length = 2*la + lb + lc
test = test + 1;
98 end
% constraint 10
if (chrom(13)-chrom(15)-chrom(16)) > 0 % lb int - ld int - le int > 0
test = test + 1;
end
% constraint 11
if (chrom(15)+chrom(14)-chrom(17)) > 0 % ld int + lc int - lf int > 0
test = test + 1;
end
% constraint 12
if (chrom(19)-chrom(18)-2*chrom(20)) > 2*minTraceWidth % wc int - wb int -
2*wd int > 2*minTraceWidth
test = test + 1;
end
% constraint 13
if (chrom(19)+2*chrom(21))< widthLimit int % wc int +
2*we int < widthLimit int
test = test + 1;
end
% constraint 14
if (chrom(21)-chrom(22)) > minTraceWidth % we int - wf int > minTraceWidth
test = test + 1;
end
% constraint 15
99 if (chrom(18)-2*chrom(23))>0 % wb int - 2*viarad int > 0
test = test + 1;
end
% constraint 16
if (chrom(24)-chrom(23))>0.1 % holerad int - viarad int > 0.1mm (4mils)
test = test + 1;
end
if test == nconstr
pass = 1;
else
pass = 0;
end
end
A.4 Simulation
clc;
clear all;
hfssExePath = ’C:\Program Files\AnsysEM\AnsysEM16.2
\Win64\ansysedt.exe’;
tmpPrjFile = [pwd,’\BJT\ V12\ 32mil.aedt’];
tmpScriptFile = [pwd, ’\miniMacro.vbs’];
tmpDesignName = ’BJT\ SMA\ T1’;
tmpSetupName = ’Setup1’;
tmpSweepName = ’Sweep1’;
100 tmpDataFile = [pwd, ’\tmpData.m’]; addpath(’library/’); hfssIncludePaths(’library/’);
filename = ’Result of 15 generation.txt’; row = 2;
M = dlmread(filename); geometry var = cell(20, 3); geometry var(1, :) = {’lb’, ’mil’, 249.4}; geometry var(2, :) = {’lc’, ’mil’, 101.1}; geometry var(3, :) = {’ld’, ’mil’, 170}; geometry var(4, :) = {’le’, ’mil’, 77.4}; geometry var(5, :) = {’lf’, ’mil’, 198.9}; geometry var(6, :) = {’wb’, ’mil’, 125}; geometry var(7, :) = {’wc’, ’mil’, 231.2}; geometry var(8, :) = {’wd’, ’mil’, 7.92}; geometry var(9, :) = {’we’, ’mil’, 28.59}; geometry var(10, :) = {’wf’, ’mil’, 21.43}; geometry var(11, :) = {’lb int’, ’mil’, 249.4}; geometry var(12, :) = {’lc int’, ’mil’, 101.1}; geometry var(13, :) = {’ld int’, ’mil’, 170}; geometry var(14, :) = {’le int’, ’mil’, 77.4}; geometry var(15, :) = {’lf int’, ’mil’, 198.9}; geometry var(16, :) = {’wb int’, ’mil’, 125}; geometry var(17, :) = {’wc int’, ’mil’, 231.2};
101 geometry var(18, :) = {’wd int’, ’mil’, 7.92};
geometry var(19, :) = {’we int’, ’mil’, 28.59};
geometry var(20, :) = {’wf int’, ’mil’, 21.43};
for i = 1:size(geometry var)
geometry var{i, 3} = M(row, i);
end
geometry var(21, :) = {’move z int’, ’mm’, 3.573};
Script writing(tmpPrjFile, tmpScriptFile, tmpDataFile, tmpDesignName, tmpSe- tupName, tmpSweepName, cell2mat(geometry var(:, 3)), geometry var);
runScript = [’cscript’ ’ ’ tmpScriptFile]; system(runScript);
delScriptFile = [’del’ ’ ’ tmpScriptFile]; system(delScriptFile);
102 Bibliography
[1] B. Cessac and M. Samuelides, “Chapter 1: From neuron to neural networks dynamics,” The European Physical Journal Special Topics, vol. 142, 09 2006. [2] R. R. Harrison, P. T. Watkins, R. J. Kier, R. O. Lovejoy, D. J. Black, B. Greger, and F. Solzbacher, “A low-power integrated circuit for a wireless 100-electrode neural recording system,” IEEE Journal of Solid-State Circuits, vol. 42, pp. 123– 133, Jan 2007. [3] A. M. Sodagar, K. D. Wise, and K. Najafi, “A fully integrated mixed-signal neural processor for implantable multichannel cortical recording,” IEEE Trans- actions on Biomedical Engineering, vol. 54, pp. 1075–1088, June 2007. [4] Y. K. Song, D. A. Borton, S. Park, W. R. Patterson, C. W. Bull, F. Laiwalla, J. Mislow, J. D. Simeral, J. P. Donoghue, and A. V. Nurmikko, “Active micro- electronic neurosensor arrays for implantable brain communication interfaces,” IEEE Transactions on Neural Systems and Rehabilitation Engineering, vol. 17, pp. 339–345, Aug 2009. [5] C. A. Chestek, V. Gilja, P. Nuyujukian, S. I. Ryu, K. V. Shenoy, R. J. Kier, F. Solzbacher, and R. R. Harrison, “Hermesc: Rf wireless low-power neural recording system for freely behaving primates,” in 2008 IEEE International Sym- posium on Circuits and Systems, pp. 1752–1755, May 2008. [6] H.-Y. Chen, J.-S. Wu, B. Hyland, X.-D. Lu, and J. Jin Jason Chen, “A low noise remotely controllable wireless telemetry system for single-unit recording in rats navigating in a vertical maze,” Medical & biological engineering & computing, vol. 46, pp. 833–9, 06 2008. [7] M. Rizk, C. A. Bossetti, T. A. Jochum, S. H. Callender, M. A. L. Nicolelis, D. A. Turner, and P. D. Wolf, “A fully implantable 96-channel neural data acquisition system,” Journal of Neural Engineering, vol. 6, p. 026002, mar 2009. [8] R. R. Harrison, R. J. Kier, C. A. Chestek, V. Gilja, P. Nuyujukian, S. Ryu, B. Greger, F. Solzbacher, and K. V. Shenoy, “Wireless neural recording with single low-power integrated circuit,” IEEE Transactions on Neural Systems and Rehabilitation Engineering, vol. 17, pp. 322–329, Aug 2009.
103 [9] D. J. Yeager, J. Holleman, R. Prasad, J. R. Smith, and B. P. Otis, “Neuralwisp: A wirelessly powered neural interface with 1-m range,” IEEE Transactions on Biomedical Circuits and Systems, vol. 3, pp. 379–387, Dec 2009.
[10] M. Yin, D. A. Borton, J. Aceros, W. R. Patterson, and A. V. Nurmikko, “A 100- channel hermetically sealed implantable device for chronic wireless neurosensing applications,” IEEE Transactions on Biomedical Circuits and Systems, vol. 7, pp. 115–128, April 2013.
[11] L. Song and Y. Rahmat-Samii, “An end-to-end implanted brain–machine in- terface antenna system performance characterizations and development,” IEEE Transactions on Antennas and Propagation, vol. 65, no. 7, pp. 3399–3408, 2017.
[12] D. Seo, R. M. Neely, K. Shen, U. Singhal, E. Alon, J. M. Rabaey, J. M. Carmena, and M. M. Maharbiz, “Wireless recording in the peripheral nervous system with ultrasonic neural dust,” Neuron, vol. 91, pp. 529–539, 2016.
[13] C. W. Lee, A. Kiourti, and J. L. Volakis, “Miniaturized fully passive brain implant for wireless neuropotential acquisition,” IEEE Antennas and Wireless Propagation Letters, vol. 16, pp. 645–648, 2017.
[14] S. Gabriel, R. Lau, and C. Gabriel, “The dielectric properties of biological tissues: Ii. measurements in the frequency range 10 hz to 20 ghz,” Physics in medicine & biology, vol. 41, no. 11, p. 2251, 1996.
[15] C.-Y. Chen, Patch Clamp Technique and Applications, pp. 49–64. New York, NY: Springer New York, 2017.
[16] L. L. Zhang and X. S. Zhao, “Carbon-based materials as supercapacitor elec- trodes.,” Chemical Society reviews, vol. 38 9, pp. 2520–31, 2009.
[17] R. R. Harrison, “A versatile integrated circuit for the acquisition of biopoten- tials,” 2007 IEEE Custom Integrated Circuits Conference, pp. 115–122, 2007.
[18] N. G. Hatsopoulos and J. P. Donoghue, “The science of neural interface sys- tems.,” Annual review of neuroscience, vol. 32, pp. 249–66, 2009.
[19] E. R. Kandel, J. H. Schwartz, and T. M. Jessell, Principles of Neural Science. McGraw-Hill Medical, 4th ed., 2000.
[20] E. N Brown, R. Kass, and P. P Mitra, “Multiple neural spike train data analysis: state-of-the-art and future challenges,” Nature neuroscience, vol. 7, pp. 456–61, 06 2004.
[21] R. R. Harrison, “The design of integrated circuits to observe brain activity,” Proceedings of the IEEE, vol. 96, pp. 1203–1216, July 2008.
104 [22] G. Buzski, C. Anastassiou, and C. Koch, “The origin of extracellular fields and currentseeg, ecog, lfp and spikes,” Nature reviews. Neuroscience, vol. 13, pp. 407– 20, 05 2012.
[23] D. Kipke, W. Shain, G. Buzski, E. Fetz, J. Henderson, J. F Hetke, and G. Schalk, “Advanced neurotechnologies for chronic neural interfaces: New horizons and clinical opportunities,” The Journal of neuroscience : the official journal of the Society for Neuroscience, vol. 28, pp. 11830–8, 12 2008.
[24] G. Schalk and E. C. Leuthardt, “Brain-computer interfaces using electrocortico- graphic signals,” IEEE Reviews in Biomedical Engineering, vol. 4, pp. 140–154, 2011.
[25] L. R. Hochberg, M. Serruya, G. M. Friehs, J. A. Mukand, M. M. Saleh, A. H. Caplan, A. Branner, D. Chen, R. D. Penn, and J. P. Donoghue, “Neuronal ensemble control of prosthetic devices by a human with tetraplegia,” Nature, vol. 442, pp. 164–171, 2006.
[26] G. E. Perlin and K. D. Wise, “Ultra-compact integration for fully-implantable neural microsystems,” in 2009 IEEE 22nd International Conference on Micro Electro Mechanical Systems, pp. 228–231, Jan 2009.
[27] R. R. Harrison, R. J. Kier, B. Greger, F. Solzbacher, C. A. Chestek, V. Gilja, P. Nuyujukian, S. I. Ryu, and K. V. Shenoy, “Wireless neural signal acquisition with single low-power integrated circuit,” in 2008 IEEE International Symposium on Circuits and Systems, pp. 1748–1751, May 2008.
[28] H. N. Schwerdt, W. Xu, S. Shekhar, A. Abbaspour-Tamijani, B. C. Towe, F. A. Miranda, and J. Chae, “A fully passive wireless microsystem for recording of neu- ropotentials using rf backscattering methods,” Journal of Microelectromechanical Systems, vol. 20, no. 5, pp. 1119–1130, 2011.
[29] H. N. Schwerdt, F. A. Miranda, and J. Chae, “A fully passive wireless backscat- tering neurorecording microsystem embedded in dispersive human-head phantom medium,” IEEE Electron Device Letters, vol. 33, pp. 908–910, June 2012.
[30] H. N. Schwerdt, F. A. Miranda, and J. Chae, “Analysis of electromagnetic fields induced in operation of a wireless fully passive backscattering neurorecording microsystem in emulated human head tissue,” IEEE Transactions on Microwave Theory and Techniques, vol. 61, pp. 2170–2176, May 2013.
[31] C. W. Lee, A. Kiourti, J. Chae, and J. L. Volakis, “A high-sensitivity fully passive neurosensing system for wireless brain signal monitoring,” IEEE Transactions on Microwave Theory and Techniques, vol. 63, no. 6, pp. 2060–2068, 2015.
105 [32] A. Kiourti, C. W. Lee, J. Chae, and J. L. Volakis, “A wireless fully passive neu- ral recording device for unobtrusive neuropotential monitoring.,” IEEE Trans. Biomed. Engineering, vol. 63, no. 1, pp. 131–137, 2016.
[33] C. A. Balanis, Antenna Theory: Analysis and Design. New York, NY, USA: Wiley-Interscience, 2005.
[34] U. F. C. Commission et al., “Report and order fcc 96–326,” 1996.
[35] H. N. Schwerdt, F. A. Miranda, and J. Chae, “Wireless fully passive multichan- nel recording of neuropotentials using photo-activated rf backscattering meth- ods,” IEEE Transactions on Microwave Theory and Techniques, vol. 63, no. 9, pp. 2965–2970, 2015.
[36] R. L. Haupt and D. H. Werner, Genetic algorithms in electromagnetics. John Wiley & Sons, 2007.
[37] H. Hori, G. Moretti, A. Rebora, and F. Crovato, “The thickness of human scalp: normal and bald,” Journal of Investigative Dermatology, vol. 58, no. 6, pp. 396– 399, 1972.
[38] F. Merli, B. Fuchs, J. R. Mosig, and A. K. Skrivervik, “The effect of insulat- ing layers on the performance of implanted antennas,” IEEE Transactions on Antennas and propagation, vol. 59, no. 1, pp. 21–31, 2011.
[39] S. Koulouridis, G. Kiziltas, Y. Zhou, D. J. Hansford, and J. L. Volakis, “Polymer– ceramic composites for microwave applications: fabrication and performance as- sessment,” IEEE Transactions on Microwave Theory and Techniques, vol. 54, no. 12, pp. 4202–4208, 2006.
[40] K. Ito, K. Furuya, Y. Okano, and L. Hamada, “Development and characteris- tics of a biological tissue-equivalent phantom for microwaves,” Electronics and Communications in Japan (Part I: Communications), vol. 84, no. 4, pp. 67–77, 2001.
[41] T. Karacolak, R. Cooper, E. S. Unlu, and E. Topsakal, “Dielectric properties of porcine skin tissue and in vivo testing of implantable antennas using pigs as model animals,” IEEE Antennas and Wireless Propagation Letters, vol. 11, pp. 1686–1689, 2012.
[42] A. Young, T. Germon, N. Barnett, A. Manara, and R. Nelson, “Behaviour of near-infrared light in the adult human head: implications for clinical near- infrared spectroscopy.,” British journal of anaesthesia, vol. 84, no. 1, pp. 38–42, 2000.
106 [43] A. N. Bashkatov, E. A. Genina, V. I. Kochubey, and V. V. Tuchin, “Optical properties of human cranial bone in the spectral range from 800 to 2000 nm,” in Saratov Fall Meeting 2005: Optical Technologies in Biophysics and Medicine VII, vol. 6163, p. 616310, International Society for Optics and Photonics, 2006.
[44] A. Bashkatov, E. Genina, V. Kochubey, and V. Tuchin, “Optical properties of human skin, subcutaneous and mucous tissues in the wavelength range from 400 to 2000 nm,” Journal of Physics D: Applied Physics, vol. 38, no. 15, p. 2543, 2005.
[45] J. Kim and Y. Rahmat-Samii, “Implanted antennas inside a human body: Simu- lations, designs, and characterizations,” IEEE Transactions on microwave theory and techniques, vol. 52, no. 8, pp. 1934–1943, 2004.
[46] A. Boulton, G. Baker, and C. Vanderwolf, Neurophysiological Techniques: Ap- plications to Neural Systems. Neurophysiological Techniques, Humana Press, 1990.
[47] D. A. Henze, Z. Borhegyi, J. Csicsvari, A. Mamiya, K. D. Harris, and G. Buzs´aki, “Intracellular features predicted by extracellular recordings in the hippocampus in vivo.,” Journal of neurophysiology, vol. 84 1, pp. 390–400, 2000.
[48] J. R Wilson, R. Clark, U. Banderali, and W. Giles, “Measurement of the mem- brane potential in small cells using patch clamp methods,” Channels (Austin, Tex.), vol. 5, pp. 530–7, 11 2011.
[49] B. R. Masters, “The biomedical engineering handbook, 2nd edition, joseph d. bronzino (editor in chief),” Computers in Biology and Medicine - COMPUT BIOL MED, vol. 31, pp. 525–525, 01 2001.
[50] D. R. Humphrey and E. M. Schmidt, “Extracellular single-unit recording meth- ods,” 2007.
[51] L. A. Geddes, “Historical evolution of circuit models for the electrode-electrolyte interface,” Annals of Biomedical Engineering, vol. 25, p. 1, Jan 1997.
[52] J. Bockris and S. Khan, Surface electrochemistry: a molecular level approach. Plenum, 1993.
[53] M. J. Nelson, P. Pouget, E. A. Nilsen, C. D. Patten, and J. D. Schall, “Review of signal distortion through metal microelectrode recording circuits and filters.,” Journal of neuroscience methods, vol. 169 1, pp. 141–57, 2008.
[54] “FHC electrode description.”
107 [55] A. Bagheri, M. T. Salam, J. L. P. Velazquez, and R. Genov, “Low-frequency noise and offset rejection in dc-coupled neural amplifiers: A review and digitally- assisted design tutorial,” IEEE Transactions on Biomedical Circuits and Sys- tems, vol. 11, pp. 161–176, Feb 2017.
[56] R. R. Harrison, “A low-power, low-noise cmos amplifier for neural recording applications,” in 2002 IEEE International Symposium on Circuits and Systems. Proceedings (Cat. No.02CH37353), vol. 5, pp. V–V, May 2002.
[57] J. Ruiz-Amaya, A. Rodriguez-Perez, and M. Delgado-Restituto, “A low noise amplifier for neural spike recording interfaces,” Sensors, vol. 15, pp. 25313–25335, 09 2015.
[58] C. Moncion, L. Balachandar, S. Bojja Venkatakrishnan, J. J. Riera, and J. Volakis, “Fully-passive wireless implant for neuropotential acquisition: An in vivo validation,” IEEE Journal of Electromagnetics, RF and Microwaves in Medicine and Biology, pp. 1–1, 2019.
[59] J. Blauert and A. Kiourti, “Bio-matched horn: A novel 1-9 ghz on-body an- tenna for low-loss biomedical telemetry with implants,” IEEE Transactions on Antennas and Propagation, pp. 1–1, 2018.
108