null  User manual
Analog Integrated Circuit and System Design
for a Compact, Low-Power Cochlear Implant
Wannaya Ngamkham
Analog Integrated Circuit and System Design for a Compact,
Low-Power Cochlear Implant
Proefschrift
ter verkrijging van de graad van doctor
aan de Technische Universiteit Delft,
op gezag van de Rector Magnificus prof. ir. K.C.A.M. Luyben,
voorzitter van het College voor Promoties,
in het openbaar te verdedigen op 7 december 2015 om 12:30 uur
door
Wannaya NGAMKHAM
Master of Electrical Engineering
Mahanakorn University of Technology
geboren te Nong Prue, Chonburi, Thailand
This dissertation has been approved by the
promotor: Prof.dr.ir. W.A. Serdijn
Prof.dr.ir. J.H.M. Frijns
Composition of the doctoral commitee:
Rector Magnificus
Prof.dr.ir. W.A. Serdijn
Prof.dr.ir. J.H.M. Frijns
Delft University of Technology, chairman
Delft University of Technology (EWI), promotor
Leiden University Medical Centre, promotor
Independent membes:
Dr. P. Boyle
Dr.dr.ir. J.J. Briaire
Prof.dr.ir. G.G.E. Gielen
Prof.dr.ir. R. Dekker
Prof.dr.ir. P.J. French
Advanced Bionics Corporation, UK
Leiden University Medical Centre
Katholieke Universiteit Leuven, Belgium
Delft University of Technology (EWI)
Delft University of Technology (EWI)
The reseach project, Smart Cochlear Implants (Smac-It) was financially
supported by the Dutch Technology Foundation (STW).
ISBN
Copyright©2015 by Wannaya NGAMKHAM
All rights reserved. No part of this publication may be reproduced, stored
in a retrieval system, or transmitted in any form or by any means without
the pior written permission of the copyright owner.
Printed by Ipskamp drukkers, The Netherlands
To my father Suthep Ngamkham
Summary
Cochlear Implants (CIs) are prosthetic devices that restore hearing in profoundly deaf patients by bypassing the damaged parts of the inner ear and
directly stimulating the remaining auditory nerve fibers in the cochlea with
electrical pulses. This thesis describs the electronic circuit design of various
modules for application in CIs in order to save area, reduce power consumption and ultimately move towards a fully implantable CI.
To enhance the perception of tonal languages (such as Thai and Chinese)
and music, an effort to realize the speech processor in a CI that imitates
the inner hair cells and the auditory nerve behaviour more precisely should
be made. According to recent physiological experiments, the envelope and
phase of speech signals are required to enhance the perceptive capability of
a CI implanted patient. The design of an analog complex gammatone filter
is introduced in order to extract both envelope and phase information of the
incoming speech signals as well as to emulate the basilar membrane behavior. A subthreshold Gm − C circuit topology is selected in order to verify
the feasibility of the complex gammatone filter at very low power operation.
Several speech encoding strategies like continuous time interleaved sampling
(CIS), race-to-spike asynchronous interleaved sampling (AIS), phase-locking
zero-crossing detection (PL-ZCD) and phase-locking peak-picking (PL-PP)
are studied and compared in order to find a compact analog speech processor
that allows for full implantation and is able to convey both time and frequency components of the incoming speech to a set of electrical pulse stimuli.
A comparison of the input and reconstructed speech signals in terms of correlation factor and hardware complexity pointed out that a PL-PP strategy
provides a compact solution for the CI electronic hardware design since this
strategy does not require a high precision envelope detector. A subthreshold
i
CMOS peak-instant detector to be used in a PL-PP CI processor has been
designed. Circuit simulations, using AMIS 0.35 μm technology, show that
the proposed detector can be operated from a 1.2 V supply and consumes
less than 1 μW static power for detecting a 5 kHz input signal. The output
signal of the detector together with the input signal amplitude (the output
of the band-pass of each channel) is expected to be used as control parameters in a stimulator for apical cochlear electrodes.
To design stimulators that are implanted inside the body, there are very
strict requirements on the size and power consumption. Therefore, it is important to convey as much charge as possible into the tissue while using an
as low as possible supply voltage to minimize power consumption. A novel
method for maximizing the charge transfer for constant current neural stimulators has been presented. This concept requires a few additional current
branches to form two feedback loops to increase the output resistance of a
MOS current mirror circuit that requires only one effective drain-source voltage drop. The main benefit we achieve for neural stimulation is the larger
amount of charge that can be conveyed to the stimulation electrode. In
other words, for the same amount of charge required, the supply voltage can
be reduced. Also, a compact programmable biphasic stimulator for cochlear
implants has been designed by using the the above concept and implemented
in AMS 0.18 μm high-voltage CMOS IC technology, using an active chip area
of only 0.042 mm2 . Measurement results show that a proper charge balance
of the anodic and cathodic stimulation phases is achieved and a dc blocking
capacitor can be omitted. The resulting reduction in the required area enables many stimulation channels on a single die.
As the work laid out in this thesis produced only stand-alone modules, future work should focus on combining all these modules together to form an
analog CI processor suitable for a fully implantable cochlear implant.
ii
Samenvatting
Cochleaire Implantaten (CIs) zijn protheses die het gehoor herstellen bij
volledig dove patienten door middel van het omzeilen van de beschadigde delen van het binnenoor en de overgebleven gehoorzenuwen in het slakkenhuis
direct te stimuleren met elektrische pulsen. Dit proefschrift beschrijft het
circuit ontwerp van verschillende modulen welke kunnen worden toegepast
in CIs zodat ruimte kan worden bespaard, het vermogensverbruik kan worden gereduceerd en dat er uiteindelijk kan worden gewerkt in de richting van
een volledig implanteerbaar CI.
Om de waarneming van tonale talen (zoals Thais en Chinees) en muziek te
verbeteren zal er een inspanning moeten worden geleverd om de spraakprocessor, welke verantwoordelijk is voor het imiteren van de binnenste haarcellen en het gedrag van de gehoorzenuw, preciezer te maken. Uit recente
fysiologische experimenten blijkt dat de omhullende en fase van spraaksignalen nodig zijn om de spraakverstaanbaarheid van een CI geïmplanteerde
patiënt te verbeteren. Het ontwerp van een analoog complexe gammatoon
filter is voorgesteld om zowel de omhullende als de faseinformatie uit het
binnenkomende spraaksignaal te extraheren en het gedrag van het basilaire
membraan te imiteren. Een subthreshold Gm −C circuittopologie is gekozen
om de geschiktheid van een complex gammatoon filter welke functioneert op
heel erg lage vermogens te onderzoeken.
Verschillende spraakcoderingsalgoritmes zoals “continuous time interleaved
sampling (CIS)”, “race-to-spike asynchronous interleaved sampling (AIS)”,
“phase-locking zero-crossing detection (PL-ZCD)” en “phase-locking peakpicking (PL-PP)” zijn bestudeerd en vergeleken om tot een compacte analoog
spraakprocessor te komen die volledig kan worden geïmplanteerd en tegelijkertijd zowel de tijd- als frequentiecomponenten van het binnenkomende
iii
signaal kan omzetten in een set elektrische stimulatie pulsen. Aan de hand
van een vergelijking van de ingangs en de gereconstrueerde spraaksignalen,
kijkend naar de correlatiefactor en hardware complexiteit, geeft een PL-PP
strategie een compacte oplossing voor het CI elektrische hardwareontwerp
omdat deze strategie geen hogeprecisie omhullendedetector nodig heeft. Een
subthreshold CMOS peak-instant detector welke kan worden gebruikt in
een PL-PP CI processor is ontworpen. Circuitsimulaties in AMIS 0,35 μm
technologie laten zien dat voor de voorgestelde detector een 1,2 V voeding
kan worden gebruikt en dat deze een statisch vermogensverbruik heeft van
minder dan 1 μW voor het detecteren van een 5 kHz ingangssignaal. Het
uitgangssignaal van de detector, samen met de amplitude van het ingangssignaal (de uitgang van het banddoorlaatfilter van elk kanaal) kunnen worden gebruikt als aanstuurparameters in een stimulator voor apicale cochleaire
elektroden.
Stimulatoren die ontworpen zijn voor implantatie in het lichaam moeten aan
hele strenge eisen voldoen op het gebied van afmetingen en vermogensverbruik. Daarom is het belangrijk om zo veel mogelijk lading in het weefsel
te brengen terwijl er zo min mogelijk voedingsspanning wordt gebruikt zodat het vermogens verbruik zo laag mogelijk wordt gehouden. Een nieuw
uitgevonden methode om de ladingsoverdracht van constantestroom stimulatoren te maximaliseren is gepresenteerd. Dit concept maakt gebruik van
een aantal extra aftakkingen voor het vormen van twee stroomterugkoppelingen zodat de uitgangsweerstand van de MOS stroomspiegel wordt verhoogd
en slechts een effectieve drain-source spanningsval heeft. Het grootste voordeel dat we hiermee bereiken is de grotere hoeveelheid lading die naar de
stimulatieelektrode kan worden gebracht. In andere woorden, voor dezelfde
hoeveelheid lading die nodig is kan een lagere voedingsspanning worden gebruikt. Aan de hand van bovenstaand concept is er een compacte programmeerbare bifasische stimulator voor cochleaire implantaten ontworpen. Deze
is geïmplementeerd in AMS 0,18 μm hoogspannings CMOS IC technologie en
gebruikt een actief chip oppervlak van slechts 0, 042 mm2 . Meetresultaten
laten een juiste ladingsbalancering van de anodische en kathodische stimulatiefasen zien waardoor een DC blokkeercapaciteit kan worden weggelaten.
De resulterende reductie in de benodigde oppervlakte maakt het mogelijk
om vele kanalen op een enkele chip te plaatsen.
Omdat het werk in dit proefschrift alleen uit losse modules bestaat, zou
iv
toekomstig onderzoek moeten worden gericht op het combineren van al deze
modules om een CI processor te formeren die geschikt is om toe te passen
in een compleet implanteerbaar cochleair implantaat.
v
Contents
Summary
i
Samenvatting
iii
1 Introduction
1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.2 Challenges of cochlear implants from an electronic design perspective . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.3 Objective and scope of the thesis . . . . . . . . . . . . . . .
1.3.1 Ultra low-power analog speech processor design . . .
1.3.2 Stimulator design . . . . . . . . . . . . . . . . . . . .
1.4 Thesis outline . . . . . . . . . . . . . . . . . . . . . . . . .
2 Background information and signal processing strategies for
cochlear implants
2.1 The peripheral auditory system . . . . . . . . . . . . . . . .
2.2 Hearing loss . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.3 A brief history of cochlear implants . . . . . . . . . . . . . .
2.4 Cochlear implant system overview . . . . . . . . . . . . . . .
2.5 Signal processing in cochlear implants . . . . . . . . . . . . .
2.6 Review and comparison of the existing speech processing strategies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.6.1 Continuous Interleaved Sampling (CIS) . . . . . . . .
2.6.2 Zero-Crossing Detection . . . . . . . . . . . . . . . .
2.6.3 Peak-Picking technique . . . . . . . . . . . . . . . . .
2.6.4 Race-to-Spike Asynchronous Interleaved Sampling . . .
2.7 System estimation and evaluation . . . . . . . . . . . . . . .
2.8 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . .
vii
1
1
4
4
4
5
6
9
9
14
14
15
17
21
21
24
25
25
26
30
3 An
3.1
3.2
3.3
3.4
analog complex gammatone filter
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . .
Temperal envelope and fine structure . . . . . . . . . . . .
The gammatone auditory filter bank . . . . . . . . . . . .
Gammatone filter topology design . . . . . . . . . . . . . .
3.4.1 Laplace transform of the complex gammatone filter
3.4.2 Orthonormal state space representation . . . . . . .
3.5 Circuit design of the complex gammatone filter . . . . . .
3.6 Circuit Simulations . . . . . . . . . . . . . . . . . . . . . .
3.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . .
4 An ultra low-power peak instant detector for a peak
cochlear implant processor
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . .
4.2 Peak-Picking strategy . . . . . . . . . . . . . . . . . .
4.3 The circuit of the current mode peak-instant detector
4.3.1 Concept . . . . . . . . . . . . . . . . . . . . .
4.3.2 Circuit design . . . . . . . . . . . . . . . . . .
4.4 Simulation results . . . . . . . . . . . . . . . . . . . .
4.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
picking
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
5 A charge balanced biphasic stimulator for cochlear implants
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.2 Constant current stimulation . . . . . . . . . . . . . . . . . .
5.3 Electrode-tissue interface model . . . . . . . . . . . . . . . .
5.4 Design concepts . . . . . . . . . . . . . . . . . . . . . . . . .
5.4.1 High output resistance current source . . . . . . . . .
5.4.2 Proposed current source . . . . . . . . . . . . . . . .
5.5 Circuit implementation . . . . . . . . . . . . . . . . . . . . .
5.5.1 High-voltage and low-voltage DAC configuration . . .
5.5.2 Differential amplifier Av . . . . . . . . . . . . . . . .
5.5.3 Switch array . . . . . . . . . . . . . . . . . . . . . . .
5.6 Measurement results . . . . . . . . . . . . . . . . . . . . . .
5.6.1 Output DC characteristics . . . . . . . . . . . . . . .
5.6.2 Biphasic stimulation . . . . . . . . . . . . . . . . . .
5.6.3 Charge error . . . . . . . . . . . . . . . . . . . . . . .
5.6.4 Current and power efficiency . . . . . . . . . . . . . .
5.6.5 Multichannel operation . . . . . . . . . . . . . . . . .
viii
33
33
34
35
37
38
39
43
47
50
51
51
52
54
54
55
57
61
63
64
65
66
67
67
69
70
71
72
74
74
75
77
82
82
83
5.7
Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . .
83
6 Conclusions and future work
6.1 General conclusions . . . . . . . . . . . . . . . . . . . . . . .
6.2 List of scientific contributions . . . . . . . . . . . . . . . . .
6.3 Future research directions . . . . . . . . . . . . . . . . . . .
85
85
87
88
References
91
Acknowledgements
101
List of Publications
105
About the Author
107
ix
Chapter 1
Introduction
1.1
Motivation
Cochlear implants (CIs) are commonly accepted as therapeutic devices for
clinical use and have restored hearing to more than 320,000 profoundly deaf
people all around the world by the middle of 2013 [1]. CI users can participate in normal conversation and other situations where the sound environment is relatively clean. However their hearing performance drops dramatically in complex sound environments, causing poor appreciation of music
and inability to converse in crowded rooms [2]. One of the major remaining problems with CIs is an individual difference of outcomes. Many users
achieve a good score in hearing ability, while other users obtain little benefit [3, 4]. There is much in the world of electrical hearing that is not yet
well understood. New research directions in the field of CI implantation are
increasingly diverse and interdisciplinary in order to solve the perceptional
limitation which may be because of the design of the hardware, the interface,
the electrode, the method to convey the sound information into the brain or
the brain itself.
Conventional CIs rely on an external unit comprising a microphone and
sound processor to pick up and encode sound. For programmability, cochlear
implant processing is done mostly in the digital domain by digitizing the output of a microphone front-end and feeding it to a digital signal processor
(DSP) as is shown in Fig. 1.1a. Although digital technology in the external
sound processor has the advantage of being more flexible to modifications
through software, the power consumed by the microphone front-end, analog
1
2
CHAPTER 1. INTRODUCTION
to digital converter (ADC), and DSP are high, approximately 5 mW. Additionally, the power consumption used for stimulation can range from 1 mW
to 10 mW, depending on the electrode-tissue impedance and the stimulation strategy [5, 6]. This results in the need of a big battery. The power
consumption is always a key issue as it is directly related to the size of the
CI battery and thereby to the overall size of the CI itself. It would be very
convenient for a CI user if the device could be made completely invisible.
There are some interesting reasons motivating the need for the development
of CIs towards a fully implantable device as the next generation CIs.
The first reason is that a conventional CI creates inconvenience during activities in the CI user’s daily life. Users gave their opinion about the external
part, worn behind the ear [7–10]. It is big and looks not nice thereby affecting their self-confidence. The external part needs to be turned off or
removed while sleeping or engaging in rough sports, etc. In these situations,
the CI user is not able to hear anything. Moreover, having every part of
the device implanted in the body, it becomes significantly easier to use for
children. Second, having the microphone implanted in the ear canal may
help improving perception as the CI user can make use of the directional
“amplification” provided by the external pinna, while also reducing noise
from wind, an effect observed from “in-the-canal” hearing aids. In addition,
the data rate limitation between the implanted part and the external processor can be solved. The higher temporal resolution can be used without
compromising the number of active channels. The positive result on CI performance is not proven clinically, it is inferred from vocoder simulations and
experiments with auditory chimerae [11–13].
A fully implantable CI is still in the research phase and it needs more investigation [14–20]. The development of an implantable microphone, rechargeable battery and low-power sound processor is still a challenge. The microphone must be space efficient, low noise and have sufficient sensitivity. It is
either implanted underneath the skin in the ear canal so that the user can
benefit from directional cues or attached to the middle ear so that sound
signals can be captured from the vibration of the tympanic membrane or the
ossicles (the three tiny bones in the middle ear). The rechargeable battery
must be small, safe and have a long life. The sound processor and stimulator
must be efficient to minimize power consumption and at least small enough
for implantation with conventional surgery.
3
1.1. MOTIVATION
ADC
Converter
DSP
Processor
To modulator
or stimulator
Band-pass
filter
Subband
Processing block
Small ADC
Band-pass
filter
Subband
Processing block
Small ADC
Band-pass
filter
Subband
Processing block
Small ADC
Band-pass
filter
Subband
Processing block
Small ADC
To DSP
Figure 1.1: (a) Audio front-end architecture consisting of an ADC converter
and DSP processor (b) Audio front-end architecture consisting of an array
of bandpass filters, subband signal processing and small ADCs.
Technology allows more miniaturization than ever before; advancements
in microelectronics/MEMS and battery technology allow for smaller device
size and longer lifetime; computer architecture innovations allow for smarter
system-level design. With the help of these advances it is feasible to improve
the cochlear implant in terms of reliability, power consumption, sound perception, convenience and reduced visibility of external parts.
4
CHAPTER 1. INTRODUCTION
1.2
Challenges of cochlear implants from an
electronic design perspective
For cochlear implants, power consumption is always a key issue as it is directly related to the size of the CI battery and thereby to the overall size
of the CI itself. Apart from battery size, also the number of external (nonintegrated) components is a major concern, as they contribute to the size of
the CI and affect its reliability. For fully implantable CIs, even the size of
the chip can become a limiting factor, something which is not often the case
in the field of integrated circuit design. In this thesis, power consumption
and size of the (integrated) electronic circuitry and speech processing performance are the major design aspects addressed.
For a fully implantable CI, placing an analog signal processing block immediately after the audio sensor gives an advantage in power saving. For
instance, analog circuitry can be used for frequency decomposition. More
signal processing, envelope, peak, and zero-crossing detection etc., could be
performed with analog circuits on the subband signals before they are recombined or sent through individual, smaller ADCs, as is illustrated in Fig.
1.1b. Consequently, the overall system can have a smaller digital processing
block than was previously required (Fig. 1.1a) [21].
1.3
Objective and scope of the thesis
This thesis aims to develop an improved cochlear implant to become fully
implanted, area efficient and consuming the least amount of power. The
low-power electronic circuit design of various CI modules and the selection
of an appropriate speech coding strategy required for a fully implantable
cochlear implant will be introduced. The two major contributions of this
thesis will be described in the following subsections.
1.3.1
Ultra low-power analog speech processor design
• To convey both envelope and fine structure (phase) to the stimulator circuit, thereby enhancing the perception of tonal languages, i.e.,
1.3. OBJECTIVE AND SCOPE OF THE THESIS
5
Thai and Chinese, and music, we follow a design methodology that
is both bio-mimetic and employs extremely compact low-power circuits for realizing the analog CI speech processor. A continuous-time
filter structure that simultaneously imitates the basilar membrane behaviour (providing a gammatone impulse response) and extracts envelope and phase information from the incoming speech signals have
been developed [22]. This filter structure is called complex gammatone
filter, which can be realized by a network of subthreshold transistors
and capacitors which requires a very low power consumption (a few
micro-watts per channel).
• A comparison of speech processing strategies for the design of an ultra low-power analog CI has been presented [23]. A comparison of
the input and reconstructed signals in terms of correlation factor and
hardware complexity pointed out that a phase-locking peak-picking
strategy (PL-PP) provides a compact solution for the CI electronic
hardware design since this strategy does not require a high precision
envelope detector.
• The design of a subthreshold CMOS peak-instant detector (PID) to
be used in the analog CI speech processor has been introduced [24].
The detector is formed by a nano-power sample and hold amplifier
(SHA) and a voltage comparator to perform the detection of occurrences of maximum and minimum values of the input. The proposed
detector can be operated from a 1.2 V supply and consumes less than
1 μW static power for detecting a 5 kHz input signal (viz., the maximum frequency of the processor). This PID extracts amplitudes at
the relevant moments that the peaks occur which is equivalent to the
amplitude extracted by conducting a Hilbert transformation but in
this case a power consumption of less than 1 micro-watt per channel
is obtained. The output signals (amplitude and time) of the PID will
be used as control parameters in the stimulator.
1.3.2
Stimulator design
• A novel circuit technique for the stimulator that requires less supply
voltage and maintains precise stimulation current in a biphasic scheme
by using a double-loop feedback topology has been proposed [25]. The
circuit has a higher power efficiency than commonly used stimulator
6
CHAPTER 1. INTRODUCTION
circuits due to a lower voltage drop accross the current generation
transistor.
• A compact biphasic programmable stimulator chip for cochlear implants has been presented [26]. The proposed circuit has been implemented in AMS 0.18 μm high-voltage CMOS technology. Its active
chip area is very small, only 0.04 mm2 [27]. The charge mismatch was
found to be well below the safety limits.
1.4
Thesis outline
The remainder of this thesis is organized as follows:
Chapter 2 explains the auditory system, a brief history and a system
overview of cochlear implants. Additionally, the signal processing in cochlear
implants is also described. Four types of speech processing strategies are
compared and evaluated in order to find an appropriate strategy for a fully
implantable cochlear implant.
Chapter 3 presents the design and circuit simulation of an analog complex
gammatone filter using CMOS transistors operating in their subthreshold
region. It will be shown that a continuous-time complex filter can be implemented by employing a state space orthonormal ladder structure and a
subthreshold Gm − C circuit topology.
Chapter 4 presents an ultra low-power, robust, compact, discrete-time
peak-instant detector (PID) for application in an analog cochlear implant
speech processor. A design methodology based on a subthreshold sample
and hold circuit and a comparator is used to obtain a peak-instant detector
having a very low power consumption and compact hardware implementation.
Chapter 5 shows the design of a compact programmable biphasic stimulator
for cochlear implants. By using a double-loop negative feedback topology,
a single power supply and a switch array the circuit requires less supply
voltage and maintains precise charge balance in a biphasic scheme. The
output current is set by scaling a reference current using a two-stage binaryweighted transistor DAC configuration (3 bits high-voltage transistor DAC
1.4. THESIS OUTLINE
7
and 4 bits low-voltage transistor DAC). By employing two stages the speed
of the stimulation pulses is improved and the area of the circuit can be minimized. The measurement results are also reported.
Chapter 6 summarizes the thesis and discusses the possibilities for future
work.
Chapter 2
Background information and
signal processing strategies for
cochlear implants
This chapter begins with describing the human peripheral auditory system
in Section 2.1, followed by the explanation of hearing loss in Section 2.2.
Section 2.3 gives a brief history of cochlear implants (CIs). A cochlear
implant system overview will be described in Section 2.4. In order to find a
compact analog sound processor that allows for full implantation and is able
to convey both time and frequency components of the incoming sound to a
set of electrical pulse stimuli, signal processing strategies will be considered
and discussed in Section 2.5. Then, four types of speech encoding strategies
are studied, compared and evaluated in Section 2.6 and 2.7, respectively.
Finally, Section 2.8 summarizes this chapter.
2.1
The peripheral auditory system
In order to perceive sound, the auditory system must complete three tasks.
First, it must deliver acoustic signals to the middle ear; second, it must convert the acoustic signals from pressure changes into electrical signals; and
third, it must transmit these electrical signals to the auditory nerve and
then to the brain where the electrical signals are converted into a meaningful perception.
9
10
CHAPTER 2. BACKGROUND INFORMATION AND SIGNAL PROCESSING
STRATEGIES FOR COCHLEAR IMPLANTS
Figure 2.1: The structure of the ear (reproduced with permission from OpenStax
College, Anatomy & Physiology, © Jan 5, 2015).
In normal hearing, the human ear can be divided into three components,
the outer ear, the middle ear and the inner ear as shown in Fig. 2.1 [28].
The outer ear consists of the visible part of the ear or auricle, the ear canal
(the external auditory canal) and the tympanic membrane or eardrum. It
is responsible for gathering sound waves and guiding them to the eardrum.
The middle ear is an air-filled chamber containing three interlocking small
bones called ossicles (malleus, incus and stapes). The function of the middle
ear is to transmit the vibrations caused by sound stimulation at the tympanic membrane to the inner ear. The ossicles act as amplitude control and
impedance matching between the external ear and the inner ear. They also
prevent damage to the inner ear caused by very loud sounds. The inner
ear is fluid-filled and contains the cochlea where the auditory perception is
located together with the vestibular organ, which is responsible for keeping
the body balanced.
Sound waves are gathering into the ear canal by the auricle and consequently
hitting the tympanic membrane causing it to vibrate. These vibrations are
guided from the tympanic membrane to the cochlea by the malleus, incus
and stapes. The stapes is attached to the oval window which is located at
2.1. THE PERIPHERAL AUDITORY SYSTEM
11
the base of the cochlea.
The cochlea has a spiral shape similar to the shell of a snail. It is approximately 30 mm long from the base (where the oval window and the
round window are located) to the apex (the top the spiral) and 2 mm diameter. In the cochlea, there are three chambers: the scala vestibuli which
joints the oval window, the scala tympani which ends the round window and
the scala media (cochlear duct) as shown in Fig. 2.2. Inside the cochlear
duct, the organ of Corti is located. The organ of Corti contains hair cells,
and is located on top of the basilar membrane and covered by the tectorial
membrane.
The movement of the stapes in response to sound pushes the oval window
in and out. This action causes the fluid in the scala vestibuli and the scala
tympani to vibrate. The fluid vibration in the scala vestibuli and the scala
tympani brings the basilar membrane into motion. The motion is described
as a travelling wave. The maximum amplitude of the travelling wave that
occurs at each position on the basilar membrane depends on the frequency
of the sound. High-frequency sounds cause greatest amplitude of the travelling wave near the base of the membrane, and low frequencies cause greatest
amplitude of the travelling wave near the apex. This means the basilar
membrane acts like a series of band-pass filter. Each point on the membrane responds to the sound at a different centre frequency. Fig. 2.3 shows
the tonotopic mapping of the basilar membrane.
The organ of Corti contains up to 5 rows of outer hair cells which run along
the basilar membrane. The row closest to the center of the cochlea consists
of the inner hair cells. They transform mechanical energy into electrical
signals. When the basilar membrane vibrates, the inner hair cells located
at the position where the travelling wave occurs respond. The hair bundle
of the inner hair cells bends. This results in patterns of electrical impulses
which are sent to the cochlear nerve [30]. The information coming from
the cochlear nerve is integrated by the central auditory nervous system and
relayed to the brain to translate it into the meaning of sound.
12
CHAPTER 2. BACKGROUND INFORMATION AND SIGNAL PROCESSING
STRATEGIES FOR COCHLEAR IMPLANTS
Figure 2.2: Crossection of the cochlea (By courtesy of Encyclopaedia Britannica,
Inc., copyright 1997; used with permission).
2.1. THE PERIPHERAL AUDITORY SYSTEM
Figure 2.3: Tonotopic mapping of the basilar membrane [29].
13
14
2.2
CHAPTER 2. BACKGROUND INFORMATION AND SIGNAL PROCESSING
STRATEGIES FOR COCHLEAR IMPLANTS
Hearing loss
In normal hearing, humans can comfortably detect sounds in a wide range
of sound pressure levels (from 0 to 100 decibel (dB SPL)). A person who
cannot detect sounds up till 70 dB SPL is considered to have severe-toprofound hearing loss. The common cause of such a profound hearing loss
is damage to or complete destruction of the hair cells. There is no pathway
to generate and transmit electrical signals to the auditory nerve. The hair
cells can be damaged by certain diseases (e.g., rubella, meningitis, Meniere’s
disease), aging, congenital disorders, prolonged exposure to loud sound or
drugs like streptomycin [30–32]. The deafness due to damaged hair cells,
as long as the auditory nerve is still present, can be treated by cochlear
implants.
2.3
A brief history of cochlear implants
A cochlear implant is an electronic medical device that bypasses the damaged parts of the inner ear and directly stimulates the remaining auditory
nerve in the cochlea with electrical signals. The journey of cochlear implants
started when Alessandor Volta (1745-1872), an Italian physicist, inserted the
metal rods of a 50 V battery into his own ear. He received a shock in his
head and after some moments he heard a bubbling or crackling sound. He
did not repeat this experiment because he believed it might be dangerous
due to the shock in his brain [33]. This was the first report of using electrical signals to stimulate the auditory nerve. In 1880, Alexander Graham Bell
founded the Volta laboratory. He used his profits from the laboratory for research that relates to deafness. Research laboratories of the Bell Telephone
System conducted early research in hearing and speech that forms the theoretical foundation needed for the later success of the cochlear implant [34,35].
In 1957, Djourono and Eyries reported successful hearing using electric stimulation [36–38]. They used a wire to directly stimulate the auditory nerve
of a deaf patient with an electrical current. The patient showed improved
lip-reading capabilities. This observation gave an impulse to the search for a
treatment of profound deafness. In 1961, House, Urban and Doyle implanted
a gold electrode insulated with silicon rubber in the scala tympani of two
deaf patients for nearly three weeks. Both patients reported useful hearing
2.4. COCHLEAR IMPLANT SYSTEM OVERVIEW
15
with electric stimulation. They described that loudness changed with the
level of stimulation and the pitch changed with variation in the rate of stimulation [39]. In 1964, Blair Simmons at Stanford Medical School placed a
six stainless-steel electrode array through the vestibule and directly into the
modiolus in a profoundly deaf patient. The patient showed an ability to discriminate a pitch at a different stimulation position and rate [40]. In 1978,
Graeme Clark in Australia implanted a 20-electrode array in two deaf patients. Other similar experimental efforts included Chouard in France [41],
Eddington in Utah [42], and Hochmair in Austria [43].
In 1972, the first single-channel electrode array was introduced by House
and 3M. By 1975, 13 patients in the United States had functioning, singlechannel CIs. The National Institutes of Health (NIH)-funded study on these
patients reported that the patients were able to hear some environmental
sounds and gained some improvements in lipreading scores, but were unable
to understand speech [44].
In 1980, systems using multiple channels and multiple stimulation sites
in the cochlea were developed and these systems supported significantly
higher levels of speech reception than the single-channel device. Since that
time, research has been dedicated to improve the design of the implant system [35, 39, 45].
In 1984, the Food and Drug Administration (FDA) in United States approved the cochlear implant for adults.
Currently there are three major cochlear implant manufacturers including
Advanced Bionics Corporation, USA, Med-El Corporation, Austria and
Cochlear Corporation, Australia [35].
2.4
Cochlear implant system overview
There are two types of cochlear implants, a modern commercial cochlear
implant and a totally implantable cochear implant. A modern commercial
cochlear implant system consists of two main components as shown in Fig.
2.4 [46]: the external component, which is worn on the outer ear, and the
internal component, the implant, which delivers stimuli to the hearing nerve.
16
CHAPTER 2. BACKGROUND INFORMATION AND SIGNAL PROCESSING
STRATEGIES FOR COCHLEAR IMPLANTS
Figure 2.4: A modern comercial cochlear implant system (Picture courtesy of
Advanced Bionics).
Cochlear implants bypass the damaged part of the ear; sound is captured
by a microphone (1) and offered to the sound processor (2). The sound processor converts the captured sound into detailed digital information. The
magnetic headpiece (3) transmits the digital signals to the internal implant
under the skin. The implant (4) turns the received digital information into
electrical information that travels down the electrode array (5) to the auditory nerve. The auditory nerve sends impulses to the brain, where they are
interpreted as sound. A typically modern cochlear implant system has the
functional blocks as shown in Fig. 2.5. An external unit, also known as the
sound processor consists of a digital signal processing (DSP) unit, a power
amplifier and an RF transmitter for the transmission of power and stimulus
information through the skin. The internal unit consists of a RF receiver to
demodulate and reconstruct the digital signal consisting of encoded sound
data. Then, the microcontroller decodes the sound data and offers this to
the stimulator. The stimulator applies stimuli to the electrode arrays. The
readout system reads out the neural response.
17
2.5. SIGNAL PROCESSING IN COCHLEAR IMPLANTS
Outside the body
Micro
phone
Inside the body
RF link
Digital sound
processor
Modulator +
Amplifier
Receiver
(demodulator,
power, data
clk recovery)
Power
Data
Clk
Digital
Controller
Read out
system
Microstimulator
Electrode Array
Figure 2.5: Functional block diagram of a modern commercial cochlear implant system
Another type of implant, which is still in the research phase, is a totally
implantable cochlear implant, as shown in Fig. 2.6. In this type, everything
is embeded inside the body and the RF coil is used only when an external
control and battery charging system is needed. Fig. 2.7. shows the functional block diagram of a totally implantable cochlear implant. In order to
pick up sound, a microphone is anchored within the auditory canal [19] or
an accelerometer is attached to the middle ear bone structure to convert
bone vibration into an electrical signal representing the original acoustic information, thus acting as a microphone [47]. A sound processor of this type
can be implemented by using analog or digital methods which consume as
little power as possible.
2.5
Signal processing in cochlear implants
A sound processor plays an important role in the development of different
techniques for deriving electrical stimuli from the sound signal. In multichannel stimulation this goal has been achieved by the invention of a speech processing strategy called ‘Continuous Interleaved Sampling, (CIS)’ [48] which
roughly emulates the behavior of the basilar membrane and inner and outer
hair cells, and successfully prevents simultaneous interactions between electrodes using fixed-rate interleaved amplitude-modulated stimuli. CIS has
been employed as a default processor in several commercially available CI
18
CHAPTER 2. BACKGROUND INFORMATION AND SIGNAL PROCESSING
STRATEGIES FOR COCHLEAR IMPLANTS
Figure 2.6: A totally implantable cochlear implant system (Reproduced with
permission from University of Utah, Case Western Reserve University).
19
2.5. SIGNAL PROCESSING IN COCHLEAR IMPLANTS
Outside
the body
RF link
Inside the body
Power
management
and control
unit
Microphone
sensor and
interface
Digital/
Analog sound
processor
Microstimulator
Electrode Array
Figure 2.7: Functional block diagram of a totally implantable cochlear implant system.
devices produced by different manufacturers, i.e., MED-EL GmbH, Cochlear
Ltd., Advanced Bionics Corp., and the results obtained from clinical experiments have shown to offer reliable understanding of sentences in quiet
environments but poor results are obtained for simple melodies. In typical
noisy environments, the patients (CI recipients) are still having difficulties
to understand both sentences and melodies [49].
These indications imply that the temporal fine structure (TFS, fast varying components of the sound) is not being conveyed to the brain. To gain
the perception of tonal languages and music, an effort of realizing a speech
processing strategy that imitates the inner hair cells and the auditory nerve
behavior more precisely should be considered. For this reason, the ‘Hilbert
Transform, (HT)’ has been introduced in the CI processor, to extract the
temporal envelope, instantaneous frequency and phase, and thereby several
strategies that convey the TFS have been introduced [50].
Fig. 2.8 shows a general block diagram that can be used to describe all
strategies that are considered in this chapter. The processor comprises three
layers of operation. At Layer-1, indicated by the white background boxes,
the incoming sound is pre-emphasized by either (linearly/non-linearly) am-
20
CHAPTER 2. BACKGROUND INFORMATION AND SIGNAL PROCESSING
STRATEGIES FOR COCHLEAR IMPLANTS
Envelope
Ch 1
Band-pass
filter
Sound
signal
Ch 2
Envelope
.
.
.
.
.
.
Preemphasis
.
.
.
Envelope
Ch N
Figure 2.8: General block diagram for envelope-based speech processing.
plifying or filtering or both before entering the bank of band-pass filters
(BPFs). This mechanism is adapted from the role of the outer hair cells that
map the wide range of the incoming sound pressure onto the limited dynamic
range of the ear. The BPF bank roughly mimics the basilar membrane behavior by decomposing the signal into a limited number (N) of frequency
bands (channels). The signal strength of each channel will be extracted in
the form of the temporal envelope (which roughly emulates the role of the
inner hair cell) and then modulated with the generated pulse trains to further stimulate the nerve fibers. These are common for all envelope-based
processors.
The study in [51] shows that spiking patterns depend somewhat on input
frequency so that Layer-2 (indicated by the gray boxes) is introduced. Particular features (frequency, phase, TFS) of the output waveform of each
channel will be detected and combined with the envelope to define the suitable stimulation pulse features. At this layer, the pulses generated from each
channel are independent from each other and the stochastic spiking behavior
of the auditory nerve is ignored. Layer-3 (represented by the dotted lines)
is therefore added to include this phenomenon by somehow conditioning
the features detected from different channels to create a stimulation pat-
2.6. REVIEW AND COMPARISON OF THE EXISTING SPEECH PROCESSING
STRATEGIES
21
tern that avoids electrode interaction and preserves the relevant extracted
features. Note that the attempt to convey all the features of the incoming
signal to the stimulation electrode is based on the assumption that the brain
can interpret this information, but in practice there are factors that deviate
from what really happens along the auditory pathway. So the number of
layers (system complexity) does not guarantee the quality of perception in
real patients [52] but serves only as a first order estimation.
This thesis aims to design a totally implantable CI by using an analog
sound processor. We thus explore some processing strategies that do not
require the computationally intensive and power exhausting HT. We try to
optimally balance the quality of the sound that can be conveyed via a set
of pulse trains to the stimulation electrodes and reduce the hardware complexity of the processors. The study covers widely recognized continuous
time interleaved sampling (CIS) and strategies to convey the temporal fine
structure (TFS), including race-to-spike asynchronous interleaved sampling
(AIS), phase-locking (PL) using zero-crossing detection (ZCD), and PL using a peak-picking (PP) technique. To estimate the performances of the four
systems, a spike-based reconstruction algorithm is employed to retrieve the
original sounds after being processed by different strategies. The correlation
factors between the reconstructed and original signals imply that strategies
conveying TFS outperform CIS. Among them, the peak picking technique
combines good performance with great compactness since envelope detectors
are not required. The details will be described in the next section.
2.6
2.6.1
Review and comparison of the existing
speech processing strategies
Continuous Interleaved Sampling (CIS)
From the default setting of several CI models [53], it can be said that CIS
is the most successful strategy. Fig. 2.9 shows the block diagram of the CIS
strategy. CIS uses only the 1st layer of operation. The temporal envelope
from each band is extracted by either half-wave or full-wave rectification followed by a low-pass filter. The envelope is then logarithmically compressed
to match the widely varying acoustic amplitudes to the narrow electric dynamic range. The compressed envelope amplitude modulates a fixed rate
biphasic carrier, whose rate can vary from several hundreds to several thou-
22
CHAPTER 2. BACKGROUND INFORMATION AND SIGNAL PROCESSING
STRATEGIES FOR COCHLEAR IMPLANTS
envelope
Ch 1
Low-pass
filter
Rectifier
Band-pass
filter
Sound
signal
Ch 2
envelope
Low-pass
filter
Rectifier
Preemphasis
.
.
.
Ch N
Compression
.
.
.
Rectifier
envelope
Low-pass
filter
Compression
.
.
.
Compression
Figure 2.9: Block diagram of the continuous interleaved sampling (CIS)
strategy.
sands per second. To avoid simultaneous electrical field interference the
biphasic carriers are time interleaved between the bands so that no simultaneous stimulation occurs between the bands at any time [35].
There is some evidence that quality of speech perception obtained from
a CIS processor strongly depends on the precision of the extracted envelopes [50, 54]. Accordingly, an attempt to replace the simple envelope
detector (ED) comprising a rectifier and a low-pass filter (LPF) by a HT
based ED is of interest. This issue needs to be carefully considered for an
analog processor since in order to perform the HT, a high complexity of
constituting electronic circuitry is unavoidable [22]. Fig. 2.10(a) shows a
fraction of the speech signal from the word ‘die’ after 4th -order Butterworth
BP filtering with a center frequency of 150 Hz. The envelopes are extracted
by a simple ED with 200 Hz LPF cutoff frequency (the dashed line of Figs.
2.10(b)-(c) and (e)) and by the HT-ED (the dotted line of Figs. 2.10(b)-(e)).
The positive pulse train generated within the CIS processor is represented
by the solid line in Fig. 2.10(b). In this case, we can clearly see that the
accuracy of the amplitude of the pulses highly depends on the accuracy of
the ED. Also, it is hardly possible that the brain can recognize frequency,
2.6. REVIEW AND COMPARISON OF THE EXISTING SPEECH PROCESSING
STRATEGIES
(a) Input signal
0.08
0.04
0
-0.04
-0.08
0.1
(b) CIS
Normalized Amplitude
0
0.1
(c) PL-ZCD
0
0.1
(d) PL-PP
0
0.1
(e) AIS
0
0.35
0.36
0.37
0.38
0.39
0.4
0.41
0.42
0.43
0.44
Time (s)
Input signal
LPF envelope
Hilbert envelope
Stimulation pulses
Figure 2.10: Waveforms obtained from different strategies.
23
CHAPTER 2. BACKGROUND INFORMATION AND SIGNAL PROCESSING
STRATEGIES FOR COCHLEAR IMPLANTS
24
Ch 1
Band-pass
filter
Sound
signal
Ch 2
Preemphasis
Compression
Envelope
Zero-crossing
detector
Compression
Envelope
Zero-crossing
detector
.
.
.
Envelope
Ch N
.
.
.
.
.
.
Compression
Zero-crossing
detector
Figure 2.11: Block diagram of the phase-locking zero-crossing detection (PLZCD) strategy.
phase and TFS from the fixed timing interval of the pulse train.
2.6.2
Zero-Crossing Detection
In this strategy, the 2nd layer is put on top of the 1st layer to introduce
a phase locking amplitude modulated pulse train. Fig. 2.11 shows the
block diagram of a phase-locking zero-crossing detection strategy (PL-ZCD)
or a simulated phase-locking stimulation (SPLS) strategy [55]. The signal
after each band-pass filter goes through 2 signal pathways: the envelope
extraction and the phase extraction. At the moment that the input signal
crosses zero from negative to positive values, the pulse is generated and will
be modulated with the momentary value of the envelope at that moment to
create the stimulation pulse train. As we can see from the blue line of Fig.
2.10(c), the amplitudes of the pulse are also defined by the quality of the
ED but the real-time period of the fundamental frequency (F0) can only be
roughly encoded. This processor thus requires high precision zero crossing
instant and high accuracy envelope detectors.
2.6. REVIEW AND COMPARISON OF THE EXISTING SPEECH PROCESSING
STRATEGIES
25
Ch 1
Band-pass
filter
Sound
signal
Peak-timing
detector
Ch 2
Preemphasis
Peak-timing
detector
Ch N
Peak-timing
detector
Figure 2.12: Block diagram of the phase-locking peak-picking (PL-PP) strategy.
2.6.3
Peak-Picking technique
This strategy also contains two layers of operations (1 and 2). But instead of
detecting the zero-crossing moments to create the phase-locked pulse train,
the occurrences of peaks in the input signal are detected [56]. The block
diagram of this strategy is shown in Fig. 2.12. There are two main features
different from the PL-ZCD. First, the number of peaks detected is higher
than the number of zero crossing moments which can be seen from the blue
lines in Figs. 2.10(c) and (d) during 0.35s < t < 0.37s. This implies that
more instantaneous frequency information other than F0 can be conveyed
to the stimulation electrodes. Second, as we can see from the peaks that
always touch the Hilbert envelope, the BPF output signal and the detected
peaking moments can be used to generate the stimulation pulses directly
without the need for ED. For this reason, the precision of the stimulation
pulse amplitudes is relayed to be dependent on the precision of a peakinstant detector.
2.6.4 Race-to-Spike Asynchronous Interleaved Sampling
In this case the 3rd layer is introduced. It has been proposed in [57] that
to achieve the stochastic stimulation behavior, the gray boxes of Fig. 2.8
26
CHAPTER 2. BACKGROUND INFORMATION AND SIGNAL PROCESSING
STRATEGIES FOR COCHLEAR IMPLANTS
are replaced by half-wave rectifier circuits as shown in Fig. 2.13. Then, at
particular repetitive time instants, the amplitudes from all channels will be
sent to a winner takes all (WTA) network letting only the strongest amplitude pass to enable the pulse generator. To avoid successive stimulation
within one channel that violates the bio-realism spiking behavior [58], additional circuit blocks are inserted to create an inhibition. At the moment
that a stimulation pulse is being generated, there will be a signal created
and applied to inhibit the signal from the half-wave rectifier (within the
channel that is being stimulated), so that it will not be processed by the
WTA network at the next time step. Even if the signal strength of that
channel is highest, it will be ignored. Within this processor, the amplitude
of each pulse is still specified by the ED of each channel but the location
of the stimulated electrode is defined by the strength of the signal at that
moment. The pulse waveform obtained from this processor is shown by the
solid line in Fig. 2.10(e). To some extent, encoding sound using this strategy
can emulate the random spiking behavior of the normal auditory nerve fiber
and the perception of music is expected. It is unfortunate that the system
is very complicated requiring two more additional circuits blocks.
2.7
System estimation and evaluation
MATLAB was used to simulate all encoding strategies. Three kinds of
sounds were picked up for the simulation with a sampling frequency of 11,025
Hz including the word ‘die’, the sentence ‘the discrete Fourier transform
of a real value signal is conjugate symmetric’ and the song phrase ‘Hallelujah’ from Handel’s Messiah. An 8 channel 4th -order Butterworth BPF
bank is used for all strategies with center frequencies ranging from 150 Hz
to 4,000 Hz arranged according to the equivalent rectangular bandwidth
(ERB) scheme [59]. Each envelope detector is formed by a full-wave rectifier followed by a 4th -order Chebyshev LPF with 200 Hz cutoff frequency.
The envelope detector is applied for all processors except PL-PP since this
strategy does not need one.
The stimulation pulses obtained from Channels 1 to 8 of all strategies (the
solid line as shown in Fig. 2.10 is only from the 1st channel) are collected for
reconstruction using the spike-based technique (in this context, spike refers
to the pulse signal). This technique has its foundation in prior neurophysiology work showing that the original analog waveform can be accurately
27
2.7. SYSTEM ESTIMATION AND EVALUATION
x
x
Sound
signal
x
Preemphasis
8 channals
Filter Bank
x
Envelope
x
x
x
x
+
-
+
-
+
-
+
-
+
-
+
Half Wave
Rectifier
Race to spike
(find channel
is larger than
Others)
-
+
-
+
-
The inhibition
signals
Figure 2.13: Block diagram of the asynchronous interleaved sampling (AIS)
strategy.
28
CHAPTER 2. BACKGROUND INFORMATION AND SIGNAL PROCESSING
STRATEGIES FOR COCHLEAR IMPLANTS
Random
noise
Stimulation
pulses ch.1
Band-pass
Filter 1
Stimulation
pulses ch.2
Band-pass
Filter 2
.
.
.
Reconstructed
sound
.
.
.
Stimulation
pulses ch.N
.
.
.
Band-pass
Filter N
Figure 2.14: Spike-based reconstruction.
reconstructed from a spiking waveform [60]. We therefore use this technique
for the signal reconstruction. Fig. 2.14 shows a block diagram of this reconstruction technique. The stimulation pulses of each channel are multiplied
by uniformly distributed random noise before injecting into the BPF with
the same center frequency as in the processor.
The resulting signals from all channels were added to produce the output
sound. To exemplify the reconstructed waveforms, Figs. 2.15 and 2.16 show
the reconstructed sounds of the word “die” from the CIS and PL-PP strategies, respectively. The original and reconstructed signals are represented
by green and blue lines, respectively. Roughly, it is visible that the reconstructed signal from PL-PP is closer to its origin than that of CIS. The
correlation coefficient (r) between the original signal and the reconstructed
signal was used to estimate the quality of the signals encoded from different strategies. The correlation coefficient is computed from the following
equation
n P
Xi − X Yi − Y
s
(2.1)
r = s i=0
2 P
2 ,
n n P
Xi − X
Yi − Y
i=0
i=0
where Xi ,Yi , X and Y are the original signal, the reconstructed signal and
the mean values of Xi and Yi , respectively. The correlation coefficient varies
in the range of -1 to 1, where 1 indicates a perfect positive correlation, -1 the
29
2.7. SYSTEM ESTIMATION AND EVALUATION
0.6
Normalized amplitude
0.4
0.2
0
-0.2
-0.4
-0.6
-0.8
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Time (s)
Figure 2.15: Reconstructed waveform of the word “die” from the CIS.
0.6
Normalized amplitude
0.4
0.2
0
-0.2
-0.4
-0.6
-0.8
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Time (s)
Figure 2.16: Reconstructed waveform of the word “die” from the PL-PP.
opposite and 0 indicates there is no correlation. The resulting correlation
coefficients obtained from different strategies are shown in Table 2.1. It is
clear that CIS performs worst of all. Besides, within the results from CIS,
the values of r depend on the complexity of the original sounds. The highest value of r = 0.11 is from the simple word (single tone) and the lowest
r = 0.02 is from the song which contains several tones that CIS could not
capture.
Among the PL strategies, as expected from the coding mechanism that conveys more instantaneous information without loss from the non-ideality of
30
CHAPTER 2. BACKGROUND INFORMATION AND SIGNAL PROCESSING
STRATEGIES FOR COCHLEAR IMPLANTS
Table 2.1: Correlation for different strategies
Strategy
CIS
PL-ZCD
PL-PP
AIS
“die”
0.11
0.36
0.47
0.49
Sentence
0.05
0.14
0.25
0.38
‘hallelujah’
0.02
0.50
0.59
0.52
the ED (see Fig. 2.10(d)) the PL-PP provides a better value of r than PLZCD for all cases. The race-to-spike AIS gives the best values of r for less
complicated sounds (word and sentence) but for multitone sounds (song),
the highest value of r is given by the PL-PP.
2.8
Conclusions
In order to understand how the ear functions, the anatomy of the ear has
been described. Defects in the ear are mainly caused by the destruction
or abscence of hair cells. These defects are causing hearing loss. In an
attempt to solve this hearing loss, cochlear implants are developed. For understanding the working of cochlear implants, a brief history and overview
of cochlear implants has been discussed. Despite the potential of the device,
users are still not capable to hear melody or tonal languages. Reasearchers
have developed several signal processing strategies in order to improve the
cochlear implants performance. The system complexities and quality of the
reconstructed signals from different signal processing strategies have been investigated and compared. Targeting the design of a fully implantable analog
CI with an ability of tone recognition, the PL-PP provides the best solution,
both in terms of compactness and correlation factors. Since the information
of frequency, phase and TFS cannot be conveyed to the stimulation electrodes by CIS, it is really hard to believe that the brain can recognize any
tone without proper input information. CIS is therefore removed from our
consideration. It is true that the numbers of the correlation factor cannot
100% guarantee the quality of hearing perception in real CI recipients. Still,
we are optimistic that the brain can interpret multi-tone sounds from the
fast varying information conveyed to the stimulation electrodes by the rest of
the strategies and that, after long term training, the patients would be able
to recognize tonal languages and melodies. Combining the aforementioned
2.8. CONCLUSIONS
31
facts with the feasibility of building ultra low-power analog hardware, the
PL-PP has proven itself as the most suitable analog sound processor.
Chapter 3
An analog complex gammatone
filter
This chapter describes the design of an analog complex gammatone filter
in order to extract both envelope and phase information of the incoming
signals as well as to emulate the basilar membrane spectral selectivity. To
synthesize the filter, the transfer functions are mapped onto an orthonormal
ladder structure which provides good dynamic range, minimum sensitivity
to component variations and high sparsity. A subthreshold Gm − C filter is
selected to realize the filter in order to verify the feasibility of the complex
gammatone filter at very low power operation.
3.1
Introduction
To realize the spectral analysis in an analog speech processor, band-pass
filter designs based on 2nd -order filters in the form of log-domain [35,61] and
Gm − C filters [15] using CMOS circuits operating in weak inversion have
been reported. These filter circuits are successful in terms of power consumption, but lack operation that is analogous to a real cochlea. Besides, in
conventional speech processors [15, 35, 61], the envelopes are extracted by a
full-wave rectifier and a low-pass filter that may provide compact hardware
implementation, yet their high frequency information is corrupted [50, 54].
It was found in [62] that, observing a cochlear nucleus after electrical stimulation, a gammatone function could closely describe the resulting cochlear
impulse responses. As a consequence, the gammatone filter has been pop33
34
CHAPTER 3. AN ANALOG COMPLEX GAMMATONE FILTER
ularly used in cochlear modeling [63] and speech recognition [64]. Also it
has been suggested in [50, 54] that, in order to preserve the high frequency
information of speech signals, the Hilbert transform should be employed
instead of the simple rectifier combined with a low-pass filter. Recently,
partially driven by the motivation mentioned above, a realization of a gammatone filter has been introduced [65]. The design is based on a class-AB
log domain circuit using signal splitting and cascaded class-A biquad sections. The filter successfully emulates the pseudo-resonance behavior of the
basilar membranes and provides a very high dynamic range of 120 dB. This
paves the way for high performance bio-inspired analog filter design for new
generations of cochlear implants.
In order to develop further, this chapter introduces a bio-realism of the
cochlear channels by combining the gammatone impulse response with the
’Hilbert Transform (HT)’ within a compact frequency selective circuit. The
design methodology starts with Laplace transforming the gammatone function into two band-pass transfer functions which represent the real and imaginary signal of the complex gammatone filter.
3.2
Temperal envelope and fine structure
During the past several years, encoding of the temperal fine structure cue
in cochlear implants has received much attention [66–71]. The temporal
fine structure contributes mainly to auditory object information whereas
the temporal envelope contributes to speech intelligibility [72]. In [13], it
has also been recognized that the temporal fine structure is more important
for music perception than the temporal envelope.
In 1912 David Hilbert showed that signals can be decomposed into slowly
varying envelopes that modulate high frequency carriers [73]. Applying the
Hilbert transform to an incoming speech signal, which is considered a real
signal, xre (t), results in an imaginary signal xim (t) which can be combined
to create the analytic signal
s (t) = xre (t) + xim (t) .
(3.1)
35
Amplitude
Amplitude
Amplitude
3.3. THE GAMMATONE AUDITORY FILTER BANK
(a) Signal, s(t)
2
0
-2
(b) Envelope
a (t ) 
2
0
2
x re2 (t )  xim
(t )
(c) cos  (t )
 xim (t ) 

 xre (t ) 
-2
 (t )  tan 1 
1
0
-1
0.02
0.03
0.04
0.05
Time, s
0.06
0.07
0.08
Figure 3.1: Decomposition of a signal using the Hilbert transformation.
(a) original waveform, (b) the Hilbert envelope, and (c) the Hilbert fine
structure.
Temporal envelope and fine structure (phase information) are respectively
defined by
q
(3.2)
a (t) = x2re (t) + x2im (t),
and cos (φ (t)), where
φ (t) = tan−1
xim (t)
xre (t)
!
(3.3)
(3.3) is the phase of the analytic signal. An example of such a decomposition
is presented in Fig. 3.1
3.3
The gammatone auditory filter bank
The name gammatone (or Ã-tone) was given by Aertsen and Johannesma
after observing its impulse response which consists of a gamma-distribution
envelope times a sinusoidal tone [65, 74]. Its popularity, within the auditory
modeling community, results from its ability to provide an appropriately
shaped ‘pseudo-resonant’ frequency transfer function that can be used to
reasonably match physiologically measured responses [65]. The gammatone
36
Arbitrary Units
CHAPTER 3. AN ANALOG COMPLEX GAMMATONE FILTER
0.4
(a)
0.2
0
Arbitrary Units
Arbitrary Units
(b)
1
0
-1
0.5
(c)
0
-0.5
0
1
2
3
4
5
Time, s
6
7
8
9
10
Figure 3.2: The components of a gammatone filter impulse response: (a)
the gamma-distribution envelope, (b) the sinusoidal tone, and (c) the gammatone impulse response.
filter function describes the impulse responses of the mammalian cochlea
and is defined by
g(t) = at(n−1) e(2πbt) cos (2πfc t + φ) ; (t > 0) .
(3.4)
The parameter n is the order of the filter, b is the bandwidth of the filter,
fc is the centre frequency of the filter, a is used as an arbitrary factor in
the filter response and it is typically chosen to make the peak gain equal
unity and φ is the starting phase. The gammatone impulse response with
its components is shown in Fig. 3.2.
The impulse response of the gammatone filter function provides an excellent fit to the impulse responses obtained from cats with the revcor technique [75, 76]. When n is 4 and b is 1.019 times the Equivalent Rectangular
Bandwidth (ERB), (4) can represent the human auditory filter [77].
The ERB is a psychoacoustic measure of the bandwidth of the auditory filter at each point along the cochlea. In this work Glasberg and Moore’s [78]
37
3.4. GAMMATONE FILTER TOPOLOGY DESIGN
Table 3.1: ERBs and center frequency of 16 cochlear channels
Channels
ERB (Hz)
fc (Hz)
1
40.89
150.0
2
50.55
239.4
3
62.48
350.0
4
77.23
486.7
5
95.46
655.6
6
118.00
864.4
7
145.86
1122.5
8
180.30
1441.6
9
222.87
1835.9
10
275.49
2323.4
11
340.53
2926.0
12
420.93
3670.9
13
520.31
4591.6
14
643.16
5729.7
15
795.01
7136.5
16
982.71
8875.5
parameters have been adopted to calculate the ERB which is given as
!
fc
ERB = 24.7 4.37
+1 .
1000
(3.5)
The fc at each channel was determined using an expression derived in [79]
for implementation of the Patterson-Holdsworth cochlear filter bank based
on ERBs, shown as
fc = − (EarQ + minBW ) + · · ·
(lowf req+(EarQ+minBW ))
fs
T
)
numChannels
e((1:numChannels) ×−log( 2 +(EarQ+minBW ))+log
fs
×
+ (EarQ + minBW )
2
!
(3.6)
in which EarQ is the asymptotic filter quality at large frequencies, minBW
is the minimum bandwidth for low frequency channels, numChannels is
the number of channels, lowf req is a low frequency and fs is the sampling
rate. An example of the centre frequencies of 16 cochlear channels and their
ERB is shown in Table 3.1. The parameters in this example were set as
follows: EarQ=9.26 and minBW =24.7 (Glasberg and Moore’s parameters),
numChannels=16, lowf req=150 Hz and fs =22050 Hz. Together, (3.4),
(3.5) and (3.6) define the gammatone auditory filter bank.
3.4
Gammatone filter topology design
In order to implement the analog complex gammatone filter circuit, the gammatone functions in the time domain need to be transformed to the complex
frequency domain (s-domain) using the Laplace transform and then these
transfer functions need to be implemented by a state space orthonormal ladder structure. Using this structure, the subsequent circuit implementation
can be performed easily. The flowchart of the gammatone filter disign is
shown in Fig. 3.3. Details of the first two steps will be described in the next
subsections.
38
CHAPTER 3. AN ANALOG COMPLEX GAMMATONE FILTER
m
Figure 3.3: Gammatone filter design flowchart.
3.4.1
Laplace transform of the complex gammatone
filter
A complex gammatone filter can be written as
gc (t) = at(n−1) e−2πbt ejωt .
(3.7)
By setting a=1, n=4 and φ=0, (3.7) can be modified for a complex tone as
gc (t) = gre (t) + gim (t)
= t3 e−2πbt cos (ωt) + jt3 e−2πbt sin (ωt) ,
(3.8)
where gre (t) and gim (t) are the real and imaginary gammatone filter function,
respectively.
Converting (3.8) into the frequency domain using the Laplace transform, we
obtain
s+B
3
d
2
2
Nre (s)
(s+B) +ω
Gre (s) =
=−
,
(3.9)
Dre (s)
ds
and
Gim (s) =
Nim (s)
=−
Dim (s)
d3
B
(s+B)2 +ω 2
ds
,
(3.10)
where B = 2πb = 2π (1.019ERB) and ω = 2πfc . Gre (s) and Gim (s) are
3.4. GAMMATONE FILTER TOPOLOGY DESIGN
39
the transfer functions of the real and imaginary signals, respectively. Note
that the denominators of both transfer functions are the same, facilitating
a compact hardware implementation, which will be illustrated shortly.
To exemplify the case of a centre frequency of 1 kHz and an ERB of 132.64
Hz, we have normalized the transfer function as
Nre (s) = −1.05 × 1012 s4 − 3.566 × 1015 s3 + 2.441 × 1020 s2
+4.197 × 1023 s − 1.457 × 1027 ,
Nim (s) = −2.638 × 1016 s3 − 6.722 × 1019 s2 + 9.845 × 1023 s
+8.684 × 1026 and
D (s) = s8 + 6.794 × 103 s7 + 1.781 × 108 s6 + 8.389 × 1011 s5
+1.11 × 1016 s4 + 3.372 × 1019 s3 + 2.878 × 1023 s2
+4.413 × 1026 s + 2.611 × 1030 .
(3.11)
Fig. 3.4 shows the impulse responses including the envelope (simulated using
MATLAB) of the real and imaginary transfer function in (3.11). Magnitude
responses of 16 cochlear channel real transfer functions by using the parameters in Table 3.1 are shown in Fig. 3.5.
3.4.2
Orthonormal state space representation
In order to implement the complex gammatone filter, both real and imaginary transfer functions are mapped onto a state space orthonormal ladder
structure. When designing high-order filters, it is very desirable to concentrate on circuits that are less sensitive to component variations. The Dynamic range-Sparsity-Sensitivity figure-of-merit (DRSS) in [80] shows that
the orthonormal ladder structure presents the best performance (minimum
DRSS) compared to the other state space description such as Optimal Dynamic Range, Biquad, Schur and Canonical. The common form of a state
space description in the Laplace domain is given by
sX(s) = AX(s) + BU (s)
Y (s) = CX(s) + DU (s),
40
CHAPTER 3. AN ANALOG COMPLEX GAMMATONE FILTER
400
real output
imaginary output
envelope
300
Amplitude
200
100
0
-100
-200
-300
-400
0
2
6
4
8
10
12
Time, ms
16
14
18
20
Figure 3.4: Impulse responses including the envelope (simulated using MATLAB) of the real and imaginary transfer function in (3.11).
0
Magnitude, dB
-50
-100
-150 1
10
2
10
3
10
4
10
5
10
Frequency, Hz
Figure 3.5: Magnitude responses of 16 cochlear channel real transfer functions.
41
3.4. GAMMATONE FILTER TOPOLOGY DESIGN
where U (s) and Y (s) are the input and output signals of the system, respectively. The vector variable X(s) represents the state of the system. A, B,
and C are the coefficients that weigh the state variable, input and output,
respectively. D is the coefficient that weigh the system inputs. For many
physical systems the matrix D is the null matrix, which also holds for the
orthonormal ladder structure. The state space description, viz., A, B and
C matrices of this structure are given by [81]

0

 −α1
A =









CT =










..
.
0
0
c1
c2
c3
..
.
α1
0
α2
−α2 0
..
.
···
α3
..
.
· · · −αn−2
···
0
0
···
..
.
0
−αn−1







,
.. 
. 


αn−1 
B=
αn
0
0
0
..
.









 0
 q
αn
π












and






.



cn−1 
(3.12)
cn
The coefficients, α are defined as
s
αi =
αn =
1
1 ≤ i < n,
xi xi+1
1
i = n,
xn
(3.13)
where xi is the reactive component of a singly terminated LC ladder structure. In order to calculate the values of the reactive elements xi , one needs
to apply a continued fraction expansion on the denominator of the transfer
function. In order to implement the numerator N (s) of the transfer function,
the proper C vector must be obtained. Using the intermediate function Fi ,
the coefficients ci can be written as
N (s) = c1 · F1 + c2 · F2 + · · · + cn · Fn ,
where
(3.14)
42
CHAPTER 3. AN ANALOG COMPLEX GAMMATONE FILTER
-a1
XN-2
...
a2
-aN-2
ò
ò
-aN-3
a1
S
...
X2
aN-2
-aN-1
(a)
X1
X2
U(s)
c1
...
XN
S
c2
XN
S a
N-1
ò
ò
S
S
ò
X1
S
aN
 aN 
bN  

  
1
2
Y(s)
cN
(b)
Figure 3.6: Block diagram of an orthonormal ladder filter [81], (a) Leapfrog
structure and (b) Output summing stage.
x1
· N (0)
π
s
=
F1
α1
1
(sFi−1 + αi−2 Fi−2 )
=
αi−1
r
F1 =
F2
Fi
3 ≤ i ≤ n.
(3.15)
The proper C vector is found as the multiplying coefficients required to create the desired numerator [82] .
Fig. 3.6 shows a block diagram of a general orthonormal ladder filter [81].
As shown in the block diagram, the filter output is obtained from a linear
combination of the outputs of all integrators.
The state space description, viz., the A, B, and C matrices of the example
transfer function (3.11) are given by







Are,im = 






0
6201
0
0
0
0
0
0
−6201
0
778.5
0
0
0
0
0
0
−778.5
0
6093
0
0
0
0
0
0
−6093
0
1608
0
0
0
0
0
0
−1608
0
5788
0
0
0
0
0
0
−5788
0
3350
0
0
0
0
0
0
−3350
0
7391
0
0
0
0
0
0
−7391 −6794







,






43
3.5. CIRCUIT DESIGN OF THE COMPLEX GAMMATONE FILTER
Cre =
h
Cim =
−39.4 10.99 4.96 −0.324 −0.158 0 0 0
h
10.95 39.63 −1.03 −2.46 0 0 0 0
i
i
,
,
and
BT =
h
0 0 0 0 0 0 0 46.5
i
.
Fig. 3.7 shows the block diagram and the impulse responses of this example
transfer function. The A matrix determines the denominator of both real
and imaginary transfer functions. A scaling factor for the whole filter is
defined by matrix B. To realize the real and imaginary outputs, different
parameters in matrices Cre and Cim will be used. Therefore, in a practical
implementation (a Gm − C structure in this case, See Fig. 3.9 in Section
3.5), we can save area and power since all the time constants whose realizations consume most of the total chip area and power can be shared. Unlike
other realizations that double hardware to realize real and imaginary transfer functions [83], in this work, only two different sets of transconductors are
required.
3.5
Circuit design of the complex gammatone filter
The circuit design of the filter is based on the Gm −C integrator approach using identical simple subthreshold differential transconductors as main building blocks shown in Fig. 3.8. The voltage to current relationship of the
transconductor is given by
Iout = IB tanh
Vd
,
2nUT
(3.16)
where n and UT are the subthreshold slope factor and thermal voltage, respectively [84].
The small signal transconductance, gm = IB /2nUT can be found from the
first term of the Taylor series expansion of (3.16). Fig. 3.9 shows the topology of the complex gammatone filter corresponding to the above state-space
representation. The integrator time constants in the A matrix are defined
CHAPTER 3. AN ANALOG COMPLEX GAMMATONE FILTER
44
U(S)
46.5
1
s
-6794
7391
1
s
-7391
3350
Impulse responses
1
s
-3350
5788
1
s
-5788
-0.153
1608
1
s
-1608
-0.324
6093
1
s
-6093
-2.461
4.833
778.5
-1.029
1
s
-778.5
10.71
6021
39.63
1
s
-6021
+
+
+
+
+
-39.48
+
+
+
+
10.95
Figure 3.7: The blockdiagram and the impulse responses of the example transfer function (3.11).
Amplitude
Yre(s)
Yim(s)
45
3.5. CIRCUIT DESIGN OF THE COMPLEX GAMMATONE FILTER
VDD
IB
Vd
Iout
VDD
IB
(b)
-
+
Iout
IB
Vi
Vo
gm
C
Vo ( s ) g m

Vi ( s ) Cs
(c)
(a)
Figure 3.8: Subthreshold Tanh transconductor (a) circuit (b) symbol and
(c) the Gm − C integrator.
by (τ = C/gm ) where C = Ci = 20pF. The bias current of each transconductor is set according to the coefficients in the matrices. The centre frequency
and the gain of the filter can be varied by scaling the bias currents of the
transconductors of matrices A, B, and C, respectively. The parameter Aij
is implemented by the corresponding Gm − C integrator with bias current
IAij , defined by
IAij = 2nUT Ci Aij .
(3.17)
The state-space vector B is realized by the first row from the top of Fig.
3.9. The current IBi1 is related to the parameter Bi1 by
IBi1 = 2nUT Ci Bi1 .
(3.18)
In the orthonormal case, only one non-zero parameter of the B vector is
present (B81 ). Consequently, IBi1 = IB81 . The bias current vector ICi1 ,
which is controlled by the vector C, is defined as
ICi1 = 2nUT Ci1 .
(3.19)
The imaginary output is obtained by adding only the C vector of the imaginary transfer function (Cim ), as shown in Fig. 3.9.
CHAPTER 3. AN ANALOG COMPLEX GAMMATONE FILTER
46
-A
21
+
IA21
IA12
+A12
-
IA32
- A32
+
C1
+
-
IA23
A23
IA43
- A43
+
C2
+
-
IA34
A34
IA54
- A54
+
C3
Integrator section : A matrix
IA45
A45
A65
IA65
+
-
IA56
A56
IA76
- A76
+
C5
vin
IA67
A67
IA87
- A87
C6
+
-
+
Input section : B vector
+
-
+
C4
+
-
+
-
IB81
B8
IA78
A78
IA88
C8
- A88
+
C7
IC11
-
+
-
+
IL2
vout(imaginary)
IL1
vout(real)
Summation section :
Cre vector
Cre3
IC13
Cre2
IC12
- Cre1
+
+
-
-
+
IC14
IC15
- Cre4
+
Cim2
IS12
Cim1
IS11
- Cre5
+
-
+
+
-
IS13
IS14
- Cim3
+
- Cim4
+
Summation section :
Cim vector
Figure 3.9: Circuit of the gammatone filter.
47
3.6. CIRCUIT SIMULATIONS
2
1.5
real output
imaginary output
Amplitude, mV
1
0.5
0
-0.5
-1
-1.5
-2
0
2
4
6
8
10
12
Time, ms
14
16
18
20
Figure 3.10: Simulated impulse response of the example transfer function,
(3.11).
3.6
Circuit Simulations
The concept of the complex gammatone filter was verified in Cadence using
RF spectre and AMIS 0.35 µm CMOS technology. The dimensions equal
W/L=6µm/6µm and 6µm/2µm for the PMOS differential pair and all transistors in the cascoded current mirrors, respectively. Supply voltage VDD =
2V and the common mode voltage reference was set at 1 V. The quiescent
power consumption equals 4.71 µW. Fig. 3.10 shows the impulse response of
the filter at 1 kHz centre frequency by applying a positive pulse signal, with
an amplitude of 10 mV and a pulse width equal to 100 µs. The common
mode signal has been removed for clarity. The frequency responses of the
filter are shown in Fig. 3.11 and 3.12 for the real and imaginary outputs, respectively. It is clear that, in the pass-band, the results are close to the ideal
case. Fortunately, errors induced from non-idealities of the transconductor
occurred mainly in the stop-bands and do not harm the filter’s functionality.
The integrated output noise power over the range from 15 Hz to 15 kHz is
2µV2 and the noise power spectral density (PSD) is shown in Fig. 3.13. Fig.
3.14 shows the total harmonic distortion (THD) of the real and imaginary
output, when sinusoidal inputs signals, ranging from 2 mV to 34 mV, are
applied at the centre frequency of 1 kHz.
48
Magnitude, dB
CHAPTER 3. AN ANALOG COMPLEX GAMMATONE FILTER
0
Ideal response
Simulated response
-100
-200 1
10
2
10
3
10
4
10
5
10
Phase, degree
500
0
-500
-1000 1
10
2
10
3
10
Frequency, Hz
4
10
5
10
Magnitude, dB
Figure 3.11: Frequency responses of the filter’s real output.
0
Ideal response
Simulated response
-100
Phase, degree
-200 1
10
2
10
3
10
4
10
5
10
500
0
-500
-1000 1
10
2
10
3
10
Frequency, Hz
4
10
5
10
Figure 3.12: Frequency responses of the filter’s imaginary output.
49
3.6. CIRCUIT SIMULATIONS
14
Output noise, nV2/Hz
12
10
8
6
4
2
0
1
10
2
10
3
10
Frequency, Hz
4
5
10
10
Figure 3.13: Output noise (PSD).
1.4
real output
imaginary output
1.2
THD, %
1
0.8
0.6
0.4
0.2
0
2
6
10
22
14
18
Input voltage amplitude, mV
Figure 3.14: Total harmonic distortion.
26
30
34
50
3.7
CHAPTER 3. AN ANALOG COMPLEX GAMMATONE FILTER
Conclusions
The theory and design of a complex gammatone filter for cochlear implants
has been introduced in this chapter. The key features for speech intelligibility
including temporal envelope and phase information can be extracted from
a compact Gm − C poly-phase band-pass filter circuit. The gammatone
impulse response is first transformed into the complex frequency (s-) domain
and the resulting 8th -order transfer function is subsequently mapped onto a
state-space description of an orthonormal ladder filter. Using this approach,
the real and imaginary transfer functions that share the same denominator
can be extracted using two different C matrices. This results in a compact
filter structure.
Chapter 4
An ultra low-power peak
instant detector for a peak
picking cochlear implant
processor
Targeting the design of a fully implantable analog cochlear implant with
an ability of tone recognition, a comparison of the input and reconstructed
signals in terms of correlation factor and hardware complexity in Chapter 2
pointed out that a phase-locking peak-picking strategy (PL-PP) provides a
compact solution for the fully implantable cochlear implant electronic hardware design. This chapter presents the design of a subthreshold CMOS
peak-instant detector (PID) to be used in a PL-PP cochlear implant processor.
4.1
Introduction
Trying to allow a cochlear implant (CI) processor to convey phase and temporal fine structure (TFS) in addition to the envelope of the speech signal to
the auditory nerve fibers, several speech processing strategies have been proposed to be used in the CI processor including phase-locking zero-crossing
detection (PL-ZCD) [55], asynchronous interleaved sampling (AIS) [57], and
PL-PP [56] strategies. They are based on the idea of imitating either phase
locking [51] or random firing [85] of the spike train in the auditory nerve fiber.
51
52
CHAPTER 4. AN ULTRA LOW-POWER PEAK INSTANT DETECTOR FOR A PEAK
PICKING COCHLEAR IMPLANT PROCESSOR
In the normal mammalian auditory nerve fibers, the spike trains synchronize with the stimulus waveform periodicity up to 5 kHz [51]. Beyond that
frequency range, the spike trains are generated randomly [85]. These mechanisms are missing in the conventional CIS strategy since the pulse stimulation rate is fixed [48]. To gain the perception of tonal languages and music,
an effort of realizing the cochlear implant processor that imitates the inner
hair cells and the auditory nerve behavior more precisely is necessary. One
way to achieve this is introducing the Hilbert Transform (HT) to the CI
processor to extract temporal envelope, instantaneous frequency and phase,
and thereby the TFS [50] and try to convey them to the stimulation electrodes. Although extraction is possible, conveying all of the information to
the brain via electrical pulse trains is still a challenge that remains. Besides,
performing the HT imposes a large computational cost for both digital [86]
and analog [22] processors.
The comparative study in Chapter 2 [23] suggests that the PL-PP strategy provides a compact solution to partially convey the TFS suitable for an
ultra low-power analog cochlear implant processor. This chapter designs a
low-power peak-instant detector for supporting the PL-PP strategy to be
incorporated together with the continuous interleaved sampling (CIS) strategy within an ultra low-power analog cochlear implant processor. To operate
the entire CMOS circuit in weak inversion region for the purpose of very low
power consumption, the switched current (SI) technique is applied to let us
benefit from the mismatch insensitive feature of the SI memory cell.
4.2
Peak-Picking strategy
Fig. 4.1 shows a block diagram of the PL-PP speech processing strategy.
It comprises four main elements; a preemphasis block, a band pass filter
(BPF) bank, a number of peak-instant detectors (PIDs), and modulators.
The incoming sound is pre-emphasized by non-linearly amplifying it before
entering the bank of BPFs. This mechanism is adapted from the role of the
outer hair cells that map the wide range of the incoming sound pressure onto
the limited dynamic range of the ear. The BPF bank roughly mimics the
basilar membrane behavior by decomposing the signal into a limited number
(N) of frequency bands (channels). The peaks of the signal coming from the
53
4.2. PEAK-PICKING STRATEGY
BPF bank
Ch 1
Peak-instant
detector
Sound
signal
Preemphasis
Ch 2
Peak-instant
detector
Ch N
Peak-instant
detector
Figure 4.1: PL-PP speech processing block diagram.
BPF of each channel will be extracted by the PID enabling the modulator to
perform multiplication of a rectangular pulse signal and the peak amplitude.
This results in a set of rectangular stimulation pulses of which the amplitude
is defined by the peak of the signal of each channel and the pulse frequency
changes according to the speech signal of each channel.
Fig. 4.2 shows a fraction of the speech signal from the word ‘die’ after
4th -order Butterworth BP filtering with a center frequency of 150 Hz (Ch.
1 of the block diagram in Fig. 4.1). The signal obtained from Ch. 1 is
represented by the dashed line. The HT (which is not part of the PL-PP
strategy) is applied to the dashed line signal and the envelope is extracted
and shown by the dotted line. The resulting pulses taken from the output
of the modulator are represented by the solid line. As we can see from the
peaks that always touch the Hilbert envelope, the BPF output signal and the
detected peaking moments can be used to generate the stimulation pulses
directly without the need for a very precise envelope detector. The precision
requirement of the stimulation pulse amplitudes is relayed onto the precision
of a PID instead.
54
CHAPTER 4. AN ULTRA LOW-POWER PEAK INSTANT DETECTOR FOR A PEAK
PICKING COCHLEAR IMPLANT PROCESSOR
Norm. Amplitude
1
0
-1
0.35
0.36
0.37
0.38
0.39
0.40
0.41
Time (s)
Hilbert envelope
Ch.1 signal
Stimulation pulses
Figure 4.2: Waveforms obtained from PL-PP strategy.
4.3
4.3.1
The circuit of the current mode peakinstant detector
Concept
The basic idea of the proposed PID is shown by the block diagram in Fig.
4.3a [87, 88]. Input signal xt is split into two signal paths. First, it goes to
the sample and hold amplifier, SHA, (with a unity gain) generating a half
delayed signal, xtd . Second, it goes to the summing node. The sample and
hold time period are controlled by a clock signal (Clk) with 50% duty cycle
indicated as the middle trace of Figures. 4.3b and 4.3c. The subtracted
result of xt and xtd at the holding period will change its polarity when xt
reaches its maximum and minimum values. For this reason, the comparator
can decide on its logical output yt according to its input sign reversing moment.
Since the incoming signal is random, this concept gives us two extreme cases
of delay time (assuming the comparator is ideal). Firstly, the minimum delay
time (td ) occurs when the falling edge of the clock signal is located exactly
at the peak of xt (See Fig. 4.3b). In this case, the detected yt will not be
delayed. Secondly, the maximum delay time occurs when the rising edge
55
4.3. THE CIRCUIT OF THE CURRENT MODE PEAK-INSTANT DETECTOR
xt
SHA
xtd
Comp
yt
(a)
xtd
xtd
xt
xt
Clk
Clk
yt
td = 0
(b)
yt
td = 0.5T
(c)
Figure 4.3: Peak-instant detection concept (a) operational block diagram
(b) best case detection and (c) worst case detection.
of the clock is located exactly at the peak. In this case, td becomes a half
period of the clock signal. In practice, the charge injection error and output
noise of the SHA and minimum detectable input signal and delay time of
the comparator introduce additional errors. In circuit level design, the architecture of the SHA needs to be insensitive to the switch charge injection
error and the resolution and speed of the comparator need to be sufficiently
good to decide on its logical output within a half clock period.
4.3.2
Circuit design
Fig. 4.4 shows a macro-model of the proposed PID. It comprises a fully
differential SI-SHA and a voltage comparator. The PID is controlled by two
non-overlapping clock phases S1 (sampling) and S2 (holding). When the set
of switches S1 turns on, the differential input current will be converted into
voltages across CH . In the next phase, S1 turns off and the set of switches
S2 turns on. Both identical CH ’s will memorize the voltages across them
producing a constant differential current via transconductor Gm . The memorized current will be compared with the input current and converted into a
56
CHAPTER 4. AN ULTRA LOW-POWER PEAK INSTANT DETECTOR FOR A PEAK
PICKING COCHLEAR IMPLANT PROCESSOR
S1
S2
CH
Iin+
S1
Av
Iin-
S2
Gm
Vout
S1
CH
Clk S2
S2
Figure 4.4: Macromodel of the SI peak-instant detector.
differential voltage at the input nodes of the comparator. The comparator
will make a decision within this phase and generate an output logical voltage
Vout . The clock control signal is applied to allow the comparator to operate
only at the holding period otherwise the logical output remains. Due to the
large loop gain providing by voltage amplifier Av , voltages across switches S1
are forced to be fixed at the input common mode level. This leads to signal
independent charge injection errors after the (practical MOS) switches are
turned off and thus will be cancelled out by the differential operation at the
input of the comparator [89].
The sub-circuits used to realize all active elements in Fig. 4.4 are shown in
Fig. 4.5. Av is formed by the circuit of Fig. 4.5a and its output commonmode level VC2 is controlled by the common-mode feedback (CMFB) circuit
depicted in Fig. 4.5b. Gm is realized by the circuit in Fig. 4.5c and its
output common-mode level VC1 is controlled by the CMFB circuit in Fig.
4.5d. Stability of the feedback loop can be maintained by setting a fixed
ratio of bias currents IB2 and IB1 and a value of CH that needs to be bigger
than the parasitic capacitances present at the input nodes of the Av and
those of Gm when the loop is closed [89]. We satisfied this condition by
setting IB2 = 2.2IB1 = 220 nA and realizing CH by NMOS capacitors biased
in their strong inversion region. We thus set supply voltage VDD = 1.2 V,
common-mode voltages VC1 = 1 V and VC2 = 0.2 V. All the transistors in the
entire circuit are working in weak inversion and the parasitic capacitances
are smaller compared to those of MOS transistors in the strong inversion
region for the same device size.
57
4.4. SIMULATION RESULTS
Table 4.1: Transistor dimensions
MOSFET
M1 , M 4 , M 6 , M 9
M2 , M3 , M5 , M7 ,M8 , M10
NMOS MH (CH )
M11
M12
M13 , M14 , M15 ,M16
W [μm]
24
3
11
6
3
0.5
L [μm]
6
3
11
3
3
0.35
The comparator is realized by the circuit shown in Fig. 4.6. It is composed of
a differential input stage cascaded by a chain of CMOS inverters to enhance
the overall gain. Input parasitic capacitors Cin+ and Cin− are employed to
memorize the input voltage for keeping the output voltage at the same value
for the whole sampling period (switches S2 are turned off). Note that the
comparator employed in this work is a simple high gain open-loop amplifier
which does not provide high speed and high sensitivity. Better results can
be expected from using more sophisticated comparator circuitry, if needed.
4.4
Simulation results
The PID circuit has been designed to be implemented in AMIS 0.35 μm
CMOS process technology. The bias current of the comparator is set to
IB2 = 50 nA. The total bias current becomes 720 nA (excluding that of the
bias generator circuit). This results in a static power consumption of 864
nW. Dimensions of the MOS transistors used are listed in Table 4.1. The
transistors are largely sized to alleviate the mismatch problem of MOS transistors in weak inversion.
Fig. 4.7 demonstrates the transient response of the proposed PID circuit
in the worst case detection (see Fig. 4.3c) for a sinusoidal differential input
current with an amplitude of 80 nA, a 5 kHz frequency and a 100 kS/s sampling frequency. The input current and the holding current are shown on
the top by the dotted and solid lines, respectively. There are large transient
glitches appearing at the beginning of the holding phases but they do not
affect the corrected circuit operation. In the middle graph, we can see that
58
CHAPTER 4. AN ULTRA LOW-POWER PEAK INSTANT DETECTOR FOR A PEAK
PICKING COCHLEAR IMPLANT PROCESSOR
VDD
Vcmf Vcmf
VDD
3
Vcmf
5
3
IB2
Vi+
Vo+
Vi−
1
Vo−
Vo+
4
Vc2
kIB2
2
(a)
Vo−
4
kIB2
(b)
VDD
VDD
kIB1
7
Vi+
Vi−
Vo+
6
Vo−
Vo+
kIB1
Vc1
9
9
Vo−
IB1
8
Vcmf
Vcmf
(c)
8
10
Vcmf
(d)
Figure 4.5: Sub-circuits of the SHA (a) transconductor Gm , (b) CMFB
circuit for the Gm , (c) voltage amplifier Av and (d) CMFB circuit for the
Av .
59
4.4. SIMULATION RESULTS
VDD
12
VDD
12
15
13
Vout
Vin+
Vin−
11
14
Cin+
IB3
16
Cin−
Figure 4.6: Comparator circuit
Current, nA
100
Iin
0
-100
1.2
Voltage, V
Ihold
Vin-
1
0.8
Vin+
Voltage, V
Vout
1
0.5
0
0
100
200
Time (µs)
300
400
Figure 4.7: The transient response of the proposed PID circuit.
60
CHAPTER 4. AN ULTRA LOW-POWER PEAK INSTANT DETECTOR FOR A PEAK
PICKING COCHLEAR IMPLANT PROCESSOR
Figure 4.8: Delay time versus input current amplitude.
the voltages at the input nodes of the comparator are swinging up and down
crossing each other within the first holding period after the peak occurred.
This operation is consistent with the theory explained in Section 4.3.1 but,
as we can see from the output waveform Vout shown in the bottom graph,
the comparator produces an additional time delay. Also it can be seen that
the delay time for the negative peak is slightly shorter.
In Fig. 4.8, it is indicated that the delay times of the PID circuit for both
positive and negative peaks depend on the input amplitude. For very small
input amplitudes less than 50 nA the proposed circuit gives a delay time
bigger than 10µs which is 5% of the period of the input signal. This is due
to limited resolution of the comparator. For the range of input amplitude
of 50 nA to 100 nA, the delay time remains less than 5% of the input signal period. The delay time goes up again for the input amplitude higher
than 100 nA. This is not because of a limitation of the comparator but of
the SHA. As the internal voltage swings at the input of Gm go too high, the
charge injection error cannot be completely cancelled out leading to a wrong
decision of the comparator.
Since the mismatch in weak inversion is worse than in strong inversion, a
Monte Carlo simulation has been performed to verify the circuit operation.
For the same condition of the transient response shown in Fig. 4.7, with 300
runs, it gives the mean values (x) and standard deviation (σv) of 8.8 μs and
2.16 μs for the positive peak and x = 8.2 μs and σv = 2.4 μs for the negative
4.5. CONCLUSIONS
61
one. These numbers indicate that the delays spread around 5% of the input
signal’s period.
4.5
Conclusions
An ultra low-power PID designed for a PL-PP BE processor has been presented. The instants detected are delayed within less than one clock period
even if the transistors’ mismatch is taken into account. Either the rising
or the falling edges of the output signal together with the input signal amplitude are expected to be used as control parameters in a stimulator for
cochlear apical electrodes which operates in the frequency range of 300 Hz 5 kHz.
Chapter 5
A charge balanced biphasic
stimulator for cochlear
implants
The output signal of the filter at each channel (Chapter 3) serves as the
input of the peak-picking instant detector (Chapter 4). Either the rising or
the falling edges of the output signal of the detector together with the input
signal amplitude (the output of the filter) are expected to be used as control
parameters in a stimulator for cochlear apical electrodes. To design stimulators that are implanted inside the body, there are very strict requirements
on the size and power consumption of the device. Therefore it is important
to be able to convey as much charge as possible into the tissue while using
an as low as possible voltage supply to minimize power consumption.
This chapter presents a compact programmable biphasic stimulator for cochlear
implants. By employing double-loop negative feedback, the output impedance
of the current generator is increased, while maximizing the voltage compliance of the output transistor. To make the stimulator circuit compact, the
stimulation current is set by scaling a reference current using a two stage
binary-weighted transistor DAC (comprising a 3 bit high-voltage transistor
DAC and a 4 bit low-voltage transistor DAC). With this structure the power
consumption and the area of the circuit can be minimized.
63
64
CHAPTER 5. A CHARGE BALANCED BIPHASIC STIMULATOR FOR COCHLEAR
IMPLANTS
5.1
Introduction
Since the stimulator is implanted inside the body, a small size and a low
power consumption are critical requirements, especially if a large number of
channels is preferred. Moreover, the circuit must be able to provide chargebalanced stimulation in order to prevent tissue damage [90]. A current mode
stimulator seems to be an attractive method because the amount of charge
injected into the tissue can easily be defined by the current amplitude and
the duration of the pulses.
Several current mode stimulators have been reported thus far. Stimulators based on a current mirror circuit have been widely used [57, 91–99]. To
maintain constant current stimulation, wide-swing and regulated cascode
current mirror topologies are used but these limit the voltage compliance.
Moreover, when using a dual supply with two current sources to create the
stimulator, additional circuitry to match the two current sources is needed
to ensure charge cancellation [92–94].
A voltage-controlled resistor based implementation has been presented in
[95] to achieve a high voltage compliance but it needs additional circuitry
to reduce non-linearity. A blocking capacitor free stimulator using dynamic
current matching is a useful idea to reduce the size of the implant and preserve a charge error less than 6 pC [57]. However, a dual supply is used and
additional circuitry is needed in this method.
Another issue is related to making the output current programmable. In
many cases a digital-to-analog converter (DAC) is used to generate a programmable reference current [91–93], [96], [97]. Then, a current mirror replicating or scaling the reference current is used to provide the stimulation
current. All these system blocks consume power and area. It has therefore
been suggested to combine the DAC function into the output current stage
in order to reduce the complexity and minimize the silicon area and power
consumption [98]. This work has adopted this suggestion and tackles several
other drawbacks mentioned in the previous paragraph.
In this chapter, a current mode, biphasic neural stimulator for application in
cochlear implants is presented. It uses a compact stimulator circuit, avoids
the use of external blocking capacitors by achieving a good charge balance
5.2. CONSTANT CURRENT STIMULATION
65
and thereby allows for an increase in the number of stimulator channels. By
using a double-loop negative feedback topology, the output impedance of
the current source can be maximized while only one effective drain-source
voltage drop (Veff ) is required. This means that more voltage headroom at
the tissue is achieved and more charge can be conveyed into the tissue.
5.2
Constant current stimulation
The principle of electrical stimulation is to activate neural cells by injecting
a particular amount of charge into the tissue, either by using a constant
voltage or a constant current. Since stimulation comes down to lifting the
potential of the tissue either above or below a particular threshold and the
tissue can be modeled as being merely capacitive in nature, stimulation entails inserting a particular amount of charge into the tissue. This makes
charge the most fundamental quantity for neural stimulation. When charge
is built up at the electrodes, electrolysis will start to occur at the interface,
generating toxic by-products [90]. These toxic by-products will damage the
tissue and therefore it is important to remove any charge built up at the
electrodes after neuronal recruitment. Charge cancellation can be conveniently achieved by the stimulation scheme called constant current biphasic
stimulation. The pulse of current is typically made up of equal-sized negative and positive phases, with the goal of delivering no net charge through
the electrode at the end of the pulse [100].
In Fig. 5.1(a) the circuit principle for constant current stimulation is depicted. The stimulator uses a single supply and a single current source
stimulation scheme. The switches allow the current to reverse its direction
in between the two stimulation phases S1 and S2 . The advantage of using
switches over a topology using positive and negative current sources is that
only a positive supply voltage is needed. To ensure charge cancellation, it is
important that the current through the tissue remains constant during both
stimulation phases. To maintain a constant current through the tissue, it
is important to have a current source with an output impedance as high as
possible.
The principle of constant current biphasic stimulation can be described using Fig. 5.1(b). The stimulation pattern is controlled by the switch array
66
CHAPTER 5. A CHARGE BALANCED BIPHASIC STIMULATOR FOR COCHLEAR
IMPLANTS
VDD
tc
Istim
S2
ti
ta
Istim
S1
-Istim
S1
A
ZL
S3
S2
(b)
C
Rf
S1
(a)
S2
ZL
Rs
(c)
Cdl
Figure 5.1: (a) Single supply stimulation scheme, (b) Biphasic stimulation
waveform and (c) Electrode-tissue interface model.
comprising switches S1 , S2 , and S3 . When switches S1 are closed (during
tc ) the current flows from A to C. When switches S2 are closed and S1 are
opened (during ta ) the current reverses its direction. An inter-phase delay
(ti ) is added between the stimulation phases when switches S1 and S2 are
opened. Switch S3 is used to short circuit and thus passively discharge the
tissue. The advantage of using a switch array for performing both anodic
and cathodic current injection is that only a single voltage supply is needed.
Also, since only one source is used, the currents are easily matched during
both phases.
5.3
Electrode-tissue interface model
The load (ZL ) of the stimulator comprises the tissue and the electrodetissue interface which can be modeled in the electrical domain. In this work
a simple model is used as shown in Fig. 5.1(c) [101]. Rs corresponds to the
resistance of the electrode leads and the tissue. In practice the resistance of
the electrolyte (the tissue) will be much higher and is therefore the dominant
component. Cdl and Rf correspond to the interface between the electrode
5.4. DESIGN CONCEPTS
67
and the tissue. Cdl is the double-layer capacitance and Rf is the Faradaic
resistance. For simplify, we assume that Rf remains large (>1 MΩ) during
the stimulation pulse (< 100 μs), and can therefore be neglected. In cochlear
implants, a typical RC series impedance presented by an electrode can be
as high as 10 kΩ and 1 nF.
5.4
5.4.1
Design concepts
High output resistance current source
A straightforward implementation of a current driver uses a current mirror,
which is often cascoded in order to have a sufficiently high output resistance [102]. Fig. 5.2(a) shows the simplest version of a PMOS cascode
current mirror. A small reference current, IDAC , generated from a digital
to analog converter, is applied through diode connected transistors M1 and
M4 . Subsequently, the current is scaled up by a factor n to become the
stimulation current, Istim , flowing through transistors M2 , M3 , and load ZL .
In this case, the output resistance equals gm3 ro3 ro2 , where
ro2 and ro3 are
q
the output resistance of M2 and M3 , respectively, gm3 = 2ID3 Kp W/L is the
transconductance of M3 , where ID3 is the drain current through M3 , Kp is
the intrinsic transconductance, and W and L are the width and length of
M3 , respectively. However, the minimum required voltage across the current
source (voltage drop, Vd ) becomes one source-gate voltage (VSG ) plus one
effective source-drain voltage (Veff ). This limits the voltage headroom (VL )
and the amount of charge that can be conveyed to the tissue (load).
To increase VL , a current mirror employing active feedback to boost the
output impedance can be used. In Fig. 5.2(b), a high-gain amplifier, Av , is
applied to make the drain voltage of M2 equal to the drain voltage of M1 .
The same biasing condition makes Istim n times IDAC . The output resistance of the active feedback current generator is given by Av gm3 ro3 ro2 . This
causes Vd to become Veff3 + Veff2 . Due to the fact that the output resistance
of this mirror is higher than in the previous cases but requires less Vd , this
cascoded structure is popularly used in neural stimulation [92], [103].
68
CHAPTER 5. A CHARGE BALANCED BIPHASIC STIMULATOR FOR COCHLEAR
IMPLANTS
VDD
M1
VDD
1:n
M2
1:n
M1
M2
Vd
Vd
M4
Av
M3
Vbias
Istim
M4
M3
Istim
VL
VL
ZL
IDAC
ZL
IDAC
(b)
(a)
Figure 5.2: Cascode current sources (a) simple cascode (b) regulated cascade.
DD
m
d
v
stim
e
L
L
DAC
Figure 5.3: Proposed current source.
69
5.4. DESIGN CONCEPTS
5.4.2
Proposed current source
In order to allow for an even higher VL , we proposed a high output impedance
current source that requires Vd to be only a single Veff [25], [26]. The concept of the proposed current source is shown in Fig. 5.3. It contains two
feedback loops. The first local one is used for high precision down scaling
of Istim (to Istim /n). Transistor M2 generates Istim flowing through ZL . The
gate terminal of transistor M1 is connected to the gate terminal of M2 to
accurately scale down the current flowing through M3 to summing node A.
Based on the same principle as used for the current mirror of Fig. 5.2(b),
Av is used to force the drain terminals of M1 and M2 to be equal, resulting
in a very precisely copied current Istim /n flowing into node A. This current
will be compared with IDAC and an error current, Ie , equal to
Ie =
Istim
− IDAC ,
n
(5.1)
flows into transimpedance amplifier Zm converting the very small Ie into the
voltage needed at the gate of M2 to produce Istim . Ie will be forced to be
zero by the large loop gain of the second global feedback loop creating a
relationship of
Istim = nIDAC .
(5.2)
We can simplify the feedback block diagram of the proposed circuit as shown
in Fig. 5.4. The loop gain of the system can be found to be
LG =
Gm2 Zm
,
n
(5.3)
where Gm2 is the transconductance gain of transistor M2 .
To maintain the desired current given by (5.2), LG needs to be as large as
possible. Since Gm2 is limited by the values of Istim and the dimension of M2
and n is preferred to be high (10-100) to keep the total power consumption
low, a large Zm becomes the main factor that defines the accuracy of the
proposed circuit.
The simulated output current versus the voltage headroom of the three current source designs VL , the simple cascode (Fig. 5.2(a)), the regulated cascode (Fig. 5.2(b)), and the proposed circuit (Fig. 5.3) is shown in Fig. 5.5.
70
CHAPTER 5. A CHARGE BALANCED BIPHASIC STIMULATOR FOR COCHLEAR
IMPLANTS
ref
stim
e
m
m2
fb
1
n
Figure 5.4: Feedback block diagram of the proposed current source.
AMS 0.18µm high-voltage technology parameters were used for circuit simulations. The simulation is performed using the same transistor dimensions
for all three designs. A high voltage supply (>10 V) is needed to accommodate the maximum current required (1 mA) through the maximum load
expected (< 20 kΩ). VDD is set at 18 V and IDAC is 50 µA, yielding Istim =1
mA with a scaling factor of n = 20. Ideal op-amps with a gain Av = 200 are
used for the circuits in Figs. 5.2(b) and Fig. 5.3. VL was varied from 0 to
18 V with a 0.5 V step size by using an ideal voltage source, as we can see
the proposed current source achieves a larger voltage headroom than those
of the others (see Fig.5.5). This verifies that the proposed current source
can inject more charge into the tissue for the same supply voltage.
5.5
Circuit implementation
From the concept as described in the previous section, we have derived a
programmable biphasic stimulator circuit for cochlear implants. The design
aims to support the flexible electrode array as developed in the SMACIt (Smart cochlear implants) project [104]. In order to reduce the size and
parasitic capacitances of the stimulator output transistor (M2 ), the proposed
current source was modified as shown in Fig. 5.6. Istim is created by scaling a
reference current (Iref ) by scaling factors n and m of M7 and M2 , respectively.
Thereby Istim becomes
Istim = n · m · Iref .
(5.4)
To make n and m programmable M7 and M2 are implemented using a binary
weighted DAC scheme. The circuit diagram of the implemented stimulator
circuit is shown in Fig. 5.7(a). This requires the use of high-voltage (HV)
71
5.5. CIRCUIT IMPLEMENTATION
1.2
Stimulation current (mA)
1
0.8
0.6
Proposed
Regulated cascode
Simple cascode
0.4
0.2
0
0
2
4
6
8
10
Voltage headroom (V)
12
14
16
18
Figure 5.5: A comparison between the output characteristics of the current
sources.
transistors (indicated by the thick drain terminal) combined with low-voltage
(LV) transistors. To minimize the area occupied by the circuit, the number
and size of HV transistors used should be as small as possible. The design
of the individual sub-circuits will be described in the next subsections.
5.5.1
High-voltage and low-voltage DAC configuration
In order to create a 10 μA resolution for a 1 mA full-scale stimulation current, a 7-bit resolution is required. The silicon area of the circuit can be
minimized when using two stages in cascade, a HV DAC and a LV DAC.
The number of bits in the HV DAC should be as small as possible. This will
reduce the number of (large) HV transistors resulting in a smaller area as
well as a lower parasitic capacitance. However, a certain minimum equivalent transistor size is needed to be able to supply the maximum stimulation
current. In our design 3 bits for the HV DAC was found to be optimal.
The remaining 4 bits can be implemented using LV transistors. These transistors are much smaller, making the area contribution negligible compared
to the area occupied by the HV DAC. The reference current was chosen to
72
CHAPTER 5. A CHARGE BALANCED BIPHASIC STIMULATOR FOR COCHLEAR
IMPLANTS
VDD
m
Zm
1
2
Vd
Av
If
Istim
ZL
VDD
3
Ie
Iref
nIref
VL
7
n
6
Figure 5.6: Modification of the proposed current source.
be Iref = 10 μA. By enabling one or more transistors in the binary weighted
DACs (using transistor switches), Istim can be made programmable using
the following relation :
Istim =
2
X
u=0
!
au 2u
3
X
!
al 2l Iref ,
(5.5)
l=0
in which u and l are the bit-numbers of the enabled HV transistors M2 and
LV transistors M7 , respectively. In this way the LV DAC can generate a
current in steps of 10 μA from 10 μA to 150 μA. The HV DAC can scale this
current with a factor 1 up to 7, resulting in a maximum stimulation current
of 1.05 mA.
5.5.2
Differential amplifier Av
Amplifier Av is used in the feedback loop to control the drain voltage of
M1 . It is implemented using a standard differential amplifier (using HV
transistors) with an active load as depicted in Fig. 5.7(b). An offset voltage
source, Voff , is needed at the output of the amplifier to bias the gate of M3
properly. It has been implemented using a diode connected LV transistor
chain and current sources Is = 10 μA. The minimum common mode input
voltage that the amplifier can handle is about 3 V because of the biasing of
M6
M6
1:1
1
S3
(a)
M6
M6
Ie
M5
M4
3
S2
2
n=1
a
m=1
M7
4
S1
S2
n=2
n=8
Low Voltage DAC
n=4
Av
m=4
Switch array
RL
CL
S1 Istim
m=2
M2
High Voltage DAC
nIref
If
M3
b
M1
a
M13
Is
VDD
M12
M11
g
M13
VDD
M12
(b)
M11 V
off
c
e
g
M14
M15
If
M3
b
VDD
M15
e
M16
c
M14
Figure 5.7: Principle circuit diagram of the implemented programmable biphasic stimulator circuit.
M6
M6
Amplifier
Zm
M5
Iref
M5
VDD
M4
VDD
5.5. CIRCUIT IMPLEMENTATION
73
74
CHAPTER 5. A CHARGE BALANCED BIPHASIC STIMULATOR FOR COCHLEAR
IMPLANTS
Table 5.1: Transistor dimensions
MOSFET
M1 , M2 (m=1), M4 , M5 , M11 , and M12
M6 , M7 , and M16 (n=1)
M13
M3 , M14 , and M15
W [μm]
5
0.5
5
5
L [μm]
5
0.18
0.4
0.6
the LV DAC. When Va < 3 V an error is introduced in the output current
because VDS,M2 6= VDS,M1 . However this error is small because |VDS,M2 | |VDS,M1 − VDS,M2 |.
5.5.3
Switch array
The implemented switch array is shown in Fig. 5.8. The upper switches can
be implemented using PMOS HV transistors. The minimum gate length
(1.5 μm) permitted by the process was chosen to provide low on-resistance.
A gate width of 50 μm was chosen. The lower switches are implemented
using NMOS HV transistors. The gate lengths and widths were chosen 1.5
μm and 15 μm, respectively. It should be noted that at node n the voltage
with respect to ground can become negative because of the charging of CL
and the subsequent reversed current direction during the charge cancellation
phase. Therefore a substrate isolated Schottky diode is placed in series to
prevent leakage between the substrate and the drain of the transistor. Switch
S3 is implemented using back-to-back PMOS transistors with their source
terminals biased at VDD . The back-to-back configuration is necessary in
order to allow for current flow in both directions. Finally, a standard crosscoupled level shifter is used to convert a LV control signal into a HV control
signal,Vcontrol [105]. The dimensions of the transistors are indicated in Table
5.1 and were selected to suit their application in fully implantable cochlear
implants.
5.6
Measurement results
The stimulator circuit has been implemented in AMS 0.18 μm HV CMOS
technology. The active area is approximately 200 μm × 210 μm. The layout
75
5.6. MEASUREMENT RESULTS
VDD
Istim
8
Vcontrol
S1
S2
10
VDD
Vcontrol
RL
S3
n
5
0.6
50
0.6
Istim
9
CL
15
1.3
Vcontrol
sub!
d
S1
S2
Figure 5.8: Switch array.
capture and the micrograph of the chip are depicted in Fig. 5.9(a) and Fig.
5.9(b), respectively. From the layout it can be seen that the HV transistors
dominate the area. The area was minimized by implementing multiple HV
transistors in the same deep n-well whenever possible, e.g. in the HV DAC.
The measurement setup is presented in Fig. 5.10. An external 18 V supply
is used for the main circuit. A 5 V supply is used for powering the pad ring
of the chip. The control of the HV and LV DACs were set by DIP switches.
Iref and Is were set at 10 μA by a Keithley 6430 Sub-Femtoamp source meter. Finally, 5 digital outputs from an Arduino UNO microcontroller board
were used in order to control the timing and current direction of the switch
array.
5.6.1
Output DC characteristics
First the accuracy of the stimulation current was measured. For this, the
load was chosen to be a single resistor RL =10 kΩ without a capacitor. The
output current was measured using a Keithley 6430 Sub-Femtoamp source
meter. Fig. 5.11 shows the measured output current versus full-scale digital
input code for the positive and negative stimulation direction, respectively.
CHAPTER 5. A CHARGE BALANCED BIPHASIC STIMULATOR FOR COCHLEAR
IMPLANTS
212 m
212 m
Differential amplifier and Voff
High-voltage DAC
and switch control
200 m
Low-voltage DAC
and switch control
Switch array
and level shifter
(b)
(a)
Figure 5.9: (a) Layout capture and (b) micrograph.
Keithley 6430 Sub-Femtoamp
current source
5
RPull-up
Is
Iref
5
18
7
Load
Out 1
5 6
HV DAC
3 4
Cochlear Stimulator
Chip
LV DAC
1 2
ON
76
S1
7-bit stimulation current setup
S2
S1
S2
RL
Out 2
CL
S3
Switch array controller
(Arduino UNO R3
microcontroller)
Figure 5.10: Measurement setup.
5.6. MEASUREMENT RESULTS
77
The current values in both positive and negative directions are almost identical, which is required for proper charge balancing. The percentage of
current mismatch is 0.02% and 0.3% at the minimum and maximum output
current, respectively. The current mismatch mostly occurred at large output currents. This is because of the effect from the threshold voltage of the
Schottky diode at the positive direction.
The measured output current versus output voltage at several stimulation
current levels is shown in Fig. 5.12. The output voltage was varied from
0 to 18 V with a 0.5 V step size by using a Keithley 6430 Sub-Femtoamp
source meter as a voltage source. The voltage headroom is about 12.5 V
and 16.5 V at maximum and minimum stimulation current, respectively.
This value is mainly limited due to the relatively small dimensions of the
HV DAC transistors. The voltage headroom can be increased by increasing
the width of the HV DAC transistors. Moreover, the voltage headroom also
depends on the voltage across the switch array. This voltage is related to
the dimensions of the HV switch transistors in order to allow the maximum
stimulation current to pass through. In this design it also depends on the
voltage across the Schottky diode. This diode has a small (0.4 V) voltage
drop.
The output resistance calculated from the measurement results plotted in
Fig. 5.12 is about 33 MΩ and 500 kΩ at Istim =10 μA and 1.05 mA, respectively.
5.6.2
Biphasic stimulation
Next, the chip was measured while providing a programmable biphasic stimulation current in the range of Istim =10 μA (minimum) to 1.05mA (maximum). The digital outputs from the microcontroller board are used to
control the timing of the biphasic waveform. The pulse widths of ta , ti and
tc were set to 50 μs, and the total cycle time to 600 μs. The measured output voltages for Istim = 50 μA and 1.05 mA across a 10 kΩ resistive load are
shown in Fig. 5.13.
Fig. 5.14, Fig. 5.15, and Fig. 5.16 show the measured biphasic output voltages for three amplitudes of Istim across a 1kΩ+10nF (minimum), 10kΩ+1nF
78
CHAPTER 5. A CHARGE BALANCED BIPHASIC STIMULATOR FOR COCHLEAR
IMPLANTS
Positive
Negative
Figure 5.11: Output current in the positive and negative stimulation direction.
10
0
10 11 12 13 14 15 16 17
Figure 5.12: Measured output characteristic of the stimulator.
79
Amplitude (V)
4
Amplitude (V)
10
Amplitude (V)
5.6. MEASUREMENT RESULTS
10
Istim=10 A
2
0
1.02
1.04
1.06
Istim=500 A
5
0
-5
Istim=1.05mA
0
-10
1
1.2
1.4
1.6
Time (ms)
1.8
2
Figure 5.13: Biphasic output voltage.
(maximum), and 10kΩ+10nF (typical) load, respectively. These measurement results show that the stimulator can be used with a wide range of
electrode-tissue impedances. Note that for the minimum load (Fig. 5.14)
the output voltage at Istim =10 μA is not shown because it is very small. In
Fig. 5.15, the stimulation time has changed to 10 μs in order to prevent voltage clipping. The spikes due to switching, and consequently settling of the
stimulator current sources, (see the magnified output voltage in Fig. 5.15
and Fig. 5.16) do not contribute to significant charge mismatch, as will be
discussed in the next section.
In order to test the stimulator in a realistic situation, the load was changed
to a CI electrode array in 0.9% saline solution. Clarion HiFocus Cochlear
electrodes were used. These are platinum iridium electrodes, produced by
Advanced Bionics, used for studies in animal cochleae and each electrode has
an area of approximately 0.2 mm2 . Fig. 5.17 shows the measured output
voltage across the two electrode sites at Istim = 500 μA, and 1.05 mA. As can
be seen from Fig. 5.17, the circuit works as expected. The output voltage at
the end of each stimulation cycle remains constant and goes to zero without
creating any voltage accumulation.
CHAPTER 5. A CHARGE BALANCED BIPHASIC STIMULATOR FOR COCHLEAR
IMPLANTS
Amplitude (V)
0.3
Istim=500 A
0.2
0.1
0
Amplitude (V)
-0.1
0.6
Istim=1.05mA
0.4
0.2
0
-0.2
1.5
Time (ms)
1
2
Figure 5.14: Output voltages for a 1kΩ+10nF load.
Amplitude (V) Amplitude (V) Amplitude (V)
80
0.5
Istim
0
-0.5
2
1.14
1.16
1.18
Istim
0
-2
2
Istim
0
-2
Figure 5.15: Output voltages for a 10kΩ+1nF load.
81
Amplitude (V)
4
Amplitude (V)
10
Amplitude (V)
5.6. MEASUREMENT RESULTS
10
Istim=10 A
2
0
1.02
1.04
1.06
Istim=500 A
5
0
-5
Istim=1.05mA
0
-10
1
1.2
1.4
1.6
Time (ms)
1.8
2
Amplitude (V)
Amplitude (V)
Figure 5.16: Output voltages for a 10kΩ+10nF load.
2
Istim=500 A
0
-2
Istim=1.05mA
2
0
-2
1
1.2
1.4
1.6
Time (ms)
1.8
2
Figure 5.17: Output voltages for CI electrode load in 0.9% saline solution.
82
CHAPTER 5. A CHARGE BALANCED BIPHASIC STIMULATOR FOR COCHLEAR
IMPLANTS
Table 5.2: Charge error and dc current error for RL =10 kΩ, CL =10 nF.
Stimulation current
10μA (50μs)
500μA(50μs)
1.05mA (10μs)
5.6.3
Charge error
0.50pC
0.45pC
1pC
DC current error
0.83nA
0.75nA
1.60nA
Charge error
The residual voltage (Vresidual ) at the end of the stimulation cycle has been
measured to determine the remaining charge imbalance. It was measured by
connecting an instrumentation amplifier (AD826) in parallel with CL . The
output of the amplifier was connected to a 20-bit analog to digital converter
(ADC) card (APPLICOS model ATX7006). A shielding enclosure (ground
connected Faraday cage) was used to reduce the amount of noise picked up
from the environment. A computer running APPLICOS ATXView software
was used to acquire, store and analyze the data. The average DC offset
voltage and noise in the data acquisition hardware was measured before
each residual voltage measurement and subsequently subtracted from the acquired data. The charge error can be calculated by multiplying the measured
residual voltage at the end of stimulation cycle with the capacitive load value
(Qerror = CL Vresidual ). The dc current error is subsequently determined by
dividing the charge error by the stimulation cycle time (Idc = Qerror /tstim ).
Table 5.2 shows the computed charge errors and dc current errors for several
values of Istim at 10kΩ+10nF load. The pulse width was set to 50 μs and
the stimulation cycle to 600μs. For Istim =1.05 mA the pulse widths tc and
ta were chosen to be 10 μs to prevent clipping of Istim . The results show
that the charge error and DC current error stay well below the safety limits.
The specified industry limit on current mismatch in cochlear implants is 25
nA [106].
5.6.4
Current and power efficiency
For the maximum output current (Istim =1.05 mA) through the maximum
load (RL = 10 kΩ) the current efficiency (defined by the ratio of the load
current and the supply current) is 87% as shown in Fig. 5.18. The maximum
power efficiency is found to be 61%. The power consumption is dominated by
the bias sources in the differential amplifier (30 μA) and the current through
83
5.7. CONCLUSIONS
Figure 5.18: Current efficiency.
the DACs, depending on the number of bits enabled in the LV DAC (ranging
from 40∼158 μA). The measured quiescent current is 210 μA which is limited
by the biasing of M3 . Note that all these bias sources can be switched off
when stimulation is not active, yielding very low static power consumption.
5.6.5
Multichannel operation
The proposed stimulator can be used for multichannel stimulation as shown
in Fig. 5.19. The reference circuit provides biasing quantities for the LV
DAC, HV DAC and amplifier Zm , which can be shared among stimulation
channels. The control logic for setting the 7 bit current amplitude and 3 bit
current direction/pulse width are received from a programmable controller.
5.7
Conclusions
A compact programmable biphasic stimulator chip for cochlear implants has
been presented in this chapter. A double loop negative feedback topology
was employed to increase the output impedance. The circuit can deliver
stimulation amplitudes in the range of 10 μA∼1.05 mA for a wide range of
electrode-tissue impedances: RL =1kΩ∼10kΩ, CL =1nF∼10nF. The current
error (< 1.6 nA) was found to be well below the safety limits. It consumes
84
CHAPTER 5. A CHARGE BALANCED BIPHASIC STIMULATOR FOR COCHLEAR
IMPLANTS
4
3
4
3
4
3
4
3
4
3
4
3
Vref /Iref
HV DAC
LV DAC
Stimulator
core
E0
E1
Switch array
Vref /Iref
HV DAC
LV DAC
Stimulator
core
E2
E3
Switch array
Vref /Iref
HV DAC
LV DAC
Stimulator
core
En-1
En
Switch array
Figure 5.19: Multichannel stimulator.
a very small chip area (0.042 mm2 ) allowing for many stimulation channels
on a single die.
Chapter 6
Conclusions and future work
This thesis describes the electronic circuit design of various modules for
application in CIs in order to save area, consume less power and ultimately
moving towards a totally implantable CI. Besides that, the design process
also considers how to convey correct information to the cochlea by means of
a speech processing strategy which is suitable for a small and low-power CI.
6.1
General conclusions
Miniaturizing area and reducing power consumption of cochlear prosthetic
devices is strongly required for full implantation. In Chapter 2, several
speech encoding strategies were studied and compared in order to develop
a compact speech processor that allows for full implantation and is able to
convey both time and frequency components of the incoming speech to a set
of electrical pulse stimuli. The study covers the widely recognized continuous time interleaved sampling (CIS) algorithm and strategies that convey
the temporal fine structure (TFS) including race-to-spike asynchronous interleaved sampling (AIS), phaselocking (PL) using zero-crossing detection
(ZCD), and PL using a peak-picking (PP) technique. To estimate the performance of the four systems, a spike-based reconstruction algorithm has
been employed to retrieve the original sounds after being processed by these
strategies. The correlation factors between the reconstructed and original
signals imply that strategies conveying TFS outperform CIS. Among them,
the peak picking technique combines good performance with great compactness since envelope detectors are not required.
85
86
CHAPTER 6. CONCLUSIONS AND FUTURE WORK
According to physiological experiments, the envelope and phase of speech
signals are required to enhance the perceptive capability of a cochlear implant processor. In Chapter 3, the design of an analog complex gammatone
filter is introduced in order to extract both envelope and phase information
of the incoming speech signals as well as to emulate the basilar membrane
spectral selectivity. The gammatone impulse response is first transformed
into the frequency domain and the resulting 8th -order transfer function is
subsequently mapped onto a state-space description of an orthonormal ladder filter. Using this approach, the real and imaginary transfer functions
that share the same denominator can be extracted using two different C
matrices. This results in a compact filter structure. The proposed filter is
designed using Gm −C integrators and subthreshold CMOS devices in AMIS
0.35 μm technology. Simulation results using Cadence RF Spectre confirm
the design principle and ultra low power operation.
Chapter 4 demonstrates the design of a subthreshold CMOS peak-instant
detector (PID) to be used in an analog CI speech processor based on the
result from Chapter 2. The detector is formed by a nano-power sample and
hold amplifier (SHA) and a voltage comparator to perform the detection
of occurrences of maximum and minimum values of the input signal. The
proposed detector can be operated from a 1.2 V supply and consumes less
than 1 μW static power for detecting a 5 kHz input signal. This PID extracts amplitudes at the relevant moments that the peaks occur which is
equivalent to the amplitude extracted by Hilbert transformation but in this
case a power consumption of less than 1 micro-watt per channel is obtained.
The output signals (amplitude and time) of the PID can be used as control
parameters in a stimulator.
Chapter 5 discusses a novel method to maximize the charge transfer for
constant current neural stimulators. By employing double-loop negative
feedback, the output resistance of a MOS current mirror circuit is increased
and the circuit requires only one effective drain-source voltage drop. The
proposed circuit requires a few additional current branches to form two feedback loops. With its compact structure, the main benefit we achieve for
neural stimulation is the larger amount of charge that can be conveyed to
the stimulation electrode. In other words, for the same amount of charge required, the supply voltage can be reduced. From this concept, we have also
designed a compact programmable biphasic stimulator for cochlear implants.
6.2. LIST OF SCIENTIFIC CONTRIBUTIONS
87
To make the stimulator circuit compact, the stimulation current is set by
scaling a reference current using a two stage binary-weighted transistor DAC
(comprising a 3 bit high-voltage transistor DAC and a 4 bit low-voltage transistor DAC). With this structure the power consumption and the area of the
circuit can be minimized. The current error was found to be well below the
safety limits. It consumes a very small chip area (0.042 mm2 ) allowing for
many stimulation channels on a single die.
6.2
List of scientific contributions
The outcomes of the research work in this thesis can be summarized as
follows:
• A subthreshold Gm − C complex gammatone filter has been designed
and verified by circuit simulations in AMIS 0.35 μm technology. The
filter provides both real and imaginary outputs that can be used to
extract both envelope and phase information of the incoming speech
signals.
• An ultra low-power, robust, compact, discrete-time peak instant detector (PID) has been designed to be implemented in AMIS 0.35 μm
technology to support the PL-PP CI processor. This design employs
a weak-inversion sample and hold circuit and a comparator in order
to extract amplitudes at the relevant moments that the peaks occur
which is equivalent to the envelope extracted by Hilbert transformation. Moreover, in this case the proposed detector can be operated
from a 1.2 V supply and consumes less than 1 μW static power for
detecting a 5 kHz input signal.
• A least-voltage drop high output resistance current source for neural stimulation has been introduced and verified by circuit simulation
in AMIS 0.35 μm technology. This concept requires a few additional
current branches to form two feedback loops to increase the output resistance of a MOS current mirror circuit that requires only one effective
drain-source voltage drop. With its compact structure, the proposed
circuit is suitable as a current generator for neural stimulation.
• A compact programmable biphasic stimulator for cochlear implants
has been designed. The proposed circuit has been implemented in AMS
88
CHAPTER 6. CONCLUSIONS AND FUTURE WORK
0.18 μm high-voltage CMOS IC technology, using an active chip area
of about 0.042 mm2 . Measurement results show that proper charge
balance of the anodic and cathodic stimulation phases is achieved and
a dc blocking capacitor can be omitted. The resulting reduction in the
required area makes the proposed system suitable for a large number
of channels on a single die.
6.3
Future research directions
The following items can be studied and/or developed further:
• Starting from the complex waveforms obtained from the complex gammatone filter in Chapter 3, we can realize the envelope information by
taking the square root of the sum of the real output squared and the
imaginary output squared. We can obtain the phase information by
computing the arctangent of the ratio of the imaginary and the real
outputs. Both envelope and phase information are useful for realizing
a speech processing strategy that contains TFS. However, additional
circuitry to implement these nonlinear functions will cause a higher
power consumption to the system.
• The PID in Chapter 4 requires the output from the BPF at each
channel for processing. So, if we do not require a specific BPF characteristic like that of a gammatone filter for realizing a speech processing
strategy, this detector can be implemented with another type of BPF,
for example the nanopower 4th -order BPF presented in [107]. A very
compact and low-power system can be expected.
• As several electronic CI modules in this thesis have only been used in
a stand-alone fasion, future work should combine all modules together
and implement them in the same technology, including the design of
the logic controller for the system. Then, an evaluation of the area and
power consumption using a multichannel topology can be performed.
• The analog CI processor presented in this thesis can be combined with
a readout system that is able to read out the neural response already
during stimulus and artifact [108]. This makes the system complete
6.3. FUTURE RESEARCH DIRECTIONS
89
with a neural recording system that can perform superior data analysis
and adjust the stimulation strategy in a closed-loop fashion.
• A single component failure analysis of the stimulator circuit should
be considered. A monitoring circuit can be added in order to make
sure that the stimulator is switched off when a component failure takes
place causing unbalanced stimuli and/or dc currents being applied to
the tissue.
References
[1] E. Strauss, “Lasker foundation,” Tech. Rep., 2013.
[2] N. R. Petersona, D. B. Pisonia, and R. T. Miyamotoa, “Cochlear implants and
spoken language processing abilities: Review and assessment of the literature,”
Restorative Neurology and Neuroscience, vol. 28, pp. 237–250, 2010.
[3] B. Krueger, G. Joseph, U. Rost, A. Strauss-Schier, T. Lenarz, and A. Buechner,
“Performance groups in adult cochlear implant users: speech perception results
from 1984 until today,” Otology and Neurotology, vol. 29, pp. 509–512, 2008.
[4] B. S. Wilson, M. F. Dorman, M. G. Woldorff, and D. L. Tucci, “Cochlear implants
matching the prosthesis to the brain and facilitating desired plastic changes in brain
function,” Progress in Brain Research, vol. 194, pp. 117–129, 2011.
[5] D. W. Graham, “A biologically inspired front end for audio signal processing using programmable analog circuitry,” Ph.D. dissertation, School of Electrical and
Computer Engineering Georgia Institute of Technology, 2006.
[6] J. J. Sit and R. Sarpeshkar, “A cochlear-implant processor for encoding music and
lowering stimulation power,” IEEE Pervasive Computing, vol. 7, no. 1, pp. 40–48,
Jan. 2008.
[7] “Will cochlear implants soon be internal?” May 23, 2014. [Online]. Available:
http://www.signhealth.org.uk/will-cochlear-implants-soon-be-internal/
[8] “Invisible
cochlear
implant,”
Dec.
2011.
[Online].
Available:
http://www.hearingaidforums.com/showthread.php?11699-InvisibleCochlear-Implant
[9] K. Bouton, “Cochlear implants are miraculous and maddening,” pdf file download,
Feb. 19, 2013. [Online]. Available: http://www.bloomberg.com/news/articles/201302-18/cochlear-implants-are-miraculous-and-maddening
[10] W. Germanovix and C. Toumazou, “Design of a micropower current-mode logdomain analog cochlear implant,” IEEE Transactions on Circuits and Systems II:
Analog and Digital Signal Processing, vol. 47, no. 10, pp. 1023–1046, Oct. 2000.
[11] R. V. Shannon, F. G. Zeng, V. Kamath, J. Wygonski, and M. Ekelid, “Speech
recognition with primarily temporal cues,” Science, vol. 270, no. 5234, pp. 303–304,
Oct. 1995.
91
92
REFERENCES
[12] L. M. Friesen, R. V. Shannon, D. Baskent, and Y. B. Wang, “Speech recognition
in noise as a function of the number of spectral channels : Comparison of acoustic
hearing and cochlear implants,” Journal of the Acoustical Society of America, vol.
110, no. 2, pp. 1150–1163, 2001.
[13] Z. M. Smith, B. Delgutte, and A. Oxenham, “Chimaeric sounds reveal dichotomies
in auditory perception,” Nature, vol. 416, pp. 87–90, 2002.
[14] R. J. S. Briggs, H. C. Eder, P. M. Seligman, R. S. C. Cowan, K. L. Plant, J. Dalton, D. K. Money, and J. F. Patrick, “Initial clinical experience with a totally
implantable cochlear implant research device,” Otology and Neurotology, vol. 29,
no. 2, pp. 114–119, 2008.
[15] R. Sarpeshkar, C. Salthouse, J. J. Sit, M. Baker, S. Zhak, T. T. Lu, L. Turicchia,
and S. Balster, “An ultra-low-power programmable analog bionic ear processor,”
IEEE Transactions on Biomedical Engineering, vol. 52, no. 4, pp. 711–727, April
2005.
[16] J. Georgiou and C. Toumazou, “A 126 uW cochlear chip for a totally implantable
system,” IEEE Journal of Solid-State Circuits, vol. 40, no. 2, pp. 430–443, Feb.
2005.
[17] H. P. Zenner, A. Limberger, J. Baumann, G. Reischl, I. M. Zalaman, P. S. Mauz,
R. W. Sweetow, P. K. Plinkert, R. Zimmermann, I. Baumann, H. D. Maddalena,
H. Leysieffer, and M. M. Maassen, “Phase III results with a totally implantable
piezoelectric middle ear implant: speech audiometry, spatial hearing and psychosocial adjustment,” Acta Otolaryngol, vol. 124, pp. 155–164, 2004.
[18] J. W. Bauman and H. Leysieffer, “Totally implanted hearing system,” U.S. Patent,
Patent no. US 6,807,445 B2, Oct. 19 2004.
[19] P. G. Berrang, H. V. Bluger, S. D. Jarvin, and A. J. Lupin, “Totally implantable
cochlear prosthesis,” U.S. Patent, Patent no. US 6,648,914 B2, Nov. 18 2003.
[20] K. Kroll, I. A. Grant, and E. Javel, “The envoy totally implantable hearing system,
st. croix medical,” Trends in Amplification, vol. 6, pp. 73–80, 2002.
[21] D. W. Graham, P. D. Smith, R. Ellis, R. Chawla, and P. E. Hasler, “A programmable bandpass array using floating-gate elements,” in Proc. IEEE International Symposium on Circuits and Systems (ISCAS), vol. 1, Vancouver, Canada,
May 2004, pp. I–97–I–100.
[22] W. Ngamkham, C. Sawigun, S. Hiseni, and W. A. Serdijn, “Analog complex gammatone filter for cochlear implant channels,” in Proc. IEEE International Symposium
on Circuits and Systems (ISCAS), Paris, France, May 2010, pp. 969–972.
[23] C. Sawigun, W. Ngamkham, and W. A. Serdijn, “Comparison of speech processing
strategies for the design of and ultra low-power analog bionic ear,” in Proc. IEEE
Engineering in Medicine and Biology Society (EMBC), Buenos Aires, Argentina,
2010, pp. 1374–1377.
REFERENCES
93
[24] C. Sawigun, W. Ngamkham, and W. A. Serdijn, “An ultra low-power peak-instant
detector for a peak picking cochlear implant processor,” in Proc. IEEE Biomedical
Circuits and Systems Conference (BioCAS), Nov. 2010, pp. 222–225.
[25] C. Sawigun, W. Ngamkham, and W. A. Serdijn, “A least-voltage drop high output
resistance current source for neural stimulation,” in Proc. IEEE Biomedical Circuits
and Systems Conference (BioCAS), Paphos, Cyprus, Nov. 2010, pp. 110–113.
[26] W. Ngamkham, C. Sawigun, and W. A. Serdijn, “Biphasic stimulator circuit for a
wide range of electrode-tissue impedance dedicated to cochlear implants,” in Proc.
IEEE International Symposium on Circuits and Systems (ISCAS), Seoul, Korea,
May 2012, pp. 1083–1086.
[27] W. Ngamkham, M. N. van Dongen, W. A. Serdijn, C. J. Bes, J. J. Briaire, and
J. H. M. Frijns, “A 0.042 mm2 programmable biphasic stimulator for cochlear implants suitable for a large number of channels,” in arXiv.org, arXiv:1502.00549, Jan.
2015.
[28] OpenStax College, Anatomy & Physiology, OpenStax CNX, Jan 5, 2015
http://cnx.org/contents/[email protected]@7.21.
[29] J. Ashmore, “The mechanics of hearing,” pdf file download, May
31, 2011. [Online]. Available: http://www.open.edu/openlearn/science-mathstechnology/science/biology/hearing/content-section-3.3
[30] R.
Fettiplace,
“The
transformations
of
sound
stimuli
into
electrical
signals,”
pdf
file
download,
May
31,
2011.
[Online].
Available:
http://www.open.edu/openlearn/science-mathstechnology/science/biology/hearing/content-section-6.2
[31] P. Loizou, “Introduction to cochlear implants,” IEEE Engineering in Medicine and
Biology Magazine, vol. 18, no. 1, pp. 32–42, Jan 1999.
[32] B. S. Wilson and M. F. Dorman, “Cochlear implants: current designs and future
possibilities,” Journal of Rehabilitation Research & Development, vol. 45, no. 5, pp.
695–730, 2008.
[33] A. Volta, “On the electricity excited by mere contact of conducting substances of
different kinds,” Royal Society Philosophical Transactions, vol. 90, pp. 403–431,
1800.
[34] H. Fletcher, Speech and Hearing in Communication, 2nd ed.
Nostrand, 1953.
New York: D. Van
[35] F. G. Zeng, S. Rebscher, W. Harrison, X. Sun, and H. Feng, “Cochlear implants:
System design, integration, and evaluation,” IEEE review in Biomedical Engineering, vol. 1, pp. pp.115–142, 2008.
[36] A. Djourno and C. Eyries, “Auditory prosthesis by means of a distant electrical
stimulation of the sensory nerve with the use of an indwelt coiling,” La Presse
medicale, vol. 65, p. 1417, 1957.
94
REFERENCES
[37] A. Djourno, C. Eyries, and P.Vallancien, “Preliminary attempts of electrical excitation of the auditory nerve in man, by permanently inserted micro-apparatus,”
Bull. Acad. Nat. Med., vol. 141, pp. 481–483, 1957.
[38] A. Djourno, C. Eyries, and B. Vallancien, “Electric excitation of the cochlear nerve
in man by induction at a distance with the aid of microcoil included in the fixture,”
C. R. Seances Soc. Biol. Fil., vol. 151, pp. 423–425, 1957.
[39] C. J. Brown, A. Geers, B. Herrmann, K. I. Kirk, J. B. Tomblin, S. Waltzman, R. Levinson, G. Linn, and S. Brannen, “Cochlear implants,” the
American Speech-Language-Hearing Association’s (ASHA), Tech. Rep., 2003,
http://www.asha.org/policy/TR2004-00041/#sec1.1.
[40] J. J. Briaire, “Cochlear implants from model to patients,” Ph.D. dissertation, Leiden
University Medical center, 2008.
[41] C. H. Chouard, “The surgical rehabilitation of total deafness with the multichannel
cochlear implant. indications and results,” Audiology, vol. 19, pp. 137–145, 1980.
[42] D. K. Eddington, W. H. Dobelle, D. E. Brackmann, M. G. Mladevosky, and J. L.
Parkin, “Auditory prosthesis research with multiple channel intracochlear stimulation in man,” Ann. Otol. Rhinol. Laryngol., vol. 87, pp. 1–39, 1978.
[43] I. J. Hochmair-Desoyer, E. S. Hochmair, K. Burian, and R. E. Fischer, “Four years
of experience with cochlear prostheses,” Med. Prog. Technol., vol. 8, pp. 107–119,
1981.
[44] S. B. Waltzman and J. T. Roland, Cochlear implants, 2nd ed. New York: Thieme
medical, 2011, ch. History of the cochlear implant, pp. 1–10.
[45] B. Wilson and M. F. Dorman, “Cochlear implants: A remarkable past and a brilliant
future,” Hearing Research, vol. 242, pp. 3–21, Aug 2008.
[46] “How hearing with a cochlear implant system works,” 2015. [Online]. Available:
http://www.advancedbionics.com/com/en/your_journey/what_is_a_cochlearimplantsystem.html
[47] D. Young, M. Zurcher, M. Semaan, C. Megerian, and W. Ko, “Mems capacitive
accelerometer-based middle ear microphone,” IEEE Transactions on Biomedical
Engineering, vol. 59, no. 12, pp. 3283–3292, Dec. 2012.
[48] B. S. Wilson, C. C. Finley, D. T. Lawson, R. D. Wolford, D. K. Eddington, and
W. M. Rabinowitz, “Better speech recognition with cochlear implants,” Nature, vol.
352, pp. 236–238, 1991.
[49] Y. Y. Kong, R. Cruz, J. A. Jones, and F. G. Zeng, “Music perception with temporal
cues in acoustic and electric hearing,” Ear and Hearing, vol. 25, no. 2, pp. 173–185,
2004.
[50] F. G. Zeng, “Trends in cochlear implants,” Trends in Amplification, vol. 8, pp. 1–34,
l 2004.
REFERENCES
95
[51] I. Tasaki, “Nerve impulses in individual auditory nerve fibers of guinea pig,” Journal
of Neurophysiology, no. 17, pp. 97–122, 1954.
[52] P. Schleich, “Pulsatile cochlear implant stimulation strategy,” U.S. Patent, Patent
Application Publication, US20090264961A1, Oct. 22 2009.
[53] T. Nguyen, S. Zupancic, and D. Lie, “Engineering challenges in cochlear implants
design and practice,” IEEE Circuits and Systems Magazine, vol. 12, no. 4, pp.
47–55, Fourth quarter 2012.
[54] K. Nie, A. Barco, and F. G. Zeng, “Spectral and temporal cues in cochlear implant
speech perception,” Ear and Hearing, vol. 27, pp. 208–217, 2006.
[55] J. Chen, X. Wu, L. Li, and H. Chi, “Simulated phase-locking stimulation: An
improved speech processing strategy for cochlear implants,” Journal for Oto-RhinoLaryngology, vol. 71, pp. 221–227, 2009.
[56] B. S. Wilson, D. T. Lawson, C. C. Finley, and M. Zerbi, “Speech processors for
auditory prostheses,” NIH project N01-DC-9-2401 National Institutes of Health,
10th Quarterly Progress Report, 1991.
[57] J. J. Sit, A. M. Simonson, A. J. Oxenham, M. A. Faltys, and R. Sarpeshkar,
“A low-power asynchronous interleaved sampling algorithm for cochlear implants
that encodes envelope and phase information,” IEEE Transactions on Biomedical
Engineering, vol. 54, pp. 138–149, Jan. 2007.
[58] M. White, M. Merzenich, and J. Gardi, “Multichannel cochlear implants: Channel
interactions and processor design,” Otolaryngology, vol. 110, pp. 493–501, 1984.
[59] B. R. Glasberg and B. C. J. Moorei, “Derivation of auditory filter shapes from
notched-noise data,” Hearing Research, vol. 47, pp. 103–108, 1990.
[60] R. Wessel, C. Koch, and F. Gabbiani, “Coding of time-varying electric field amplitude modulations in a wave-type electric fish,” Hearing Research, vol. 75, pp.
2280–2293, 1996.
[61] W. Germanovix and C. Toumazou, “Design of a micropower current-mode logdomain analog cochlear implant,” IEEE Transactions on Circuits and Systems II:
Analog and Digital Signal Processing, vol. 47, no. 10, pp. 1023–1046, Oct 2000.
[62] P. Johannesma, “The pre-response stimulus ensemble of neuron in the cochlear
nucleus,” in Proceedings of the IPO Symposium of Hearing Theory, Eindhoven,
The Netherlands, 1972, pp. 58–69.
[63] J. L. Flanagan, “Models for approximating basilar membrane displacement - part
II. effects of middle-ear transmission and some relations between subjective and
physiological behaviour,” Bell System Technical Journal, vol. 41, pp. 959–1009,
May 1960.
[64] K. Ishizuka and N. Miyazaki, “Speech feature extraction method representing periodicity and aperiodicity in sub bands for robust speech recognition,” IEEE International conference on Acoustics, Speech and Signal Processing, vol. 1, pp. 141–144,
May 2004.
96
REFERENCES
[65] A. G. Katsiamis, E. M. Drakakis, and R. F. Lyon, “A biomimetic, 4.5 uW, 120+
dB, log-domain cochlea channel with AGC,” IEEE Journal of Solid-State Circuits,
vol. 44, no. 3, pp. 1006–1022, 2009.
[66] K. Nie, G. Stickney, and F. G. Zeng, “Encoding frequency modulation to improve
cochlear implant performance in noise,” IEEE Transactions on Biomedical Engineering, vol. 52, no. 1, pp. 64–73, Jan. 2005.
[67] C. S. Throckmorton, M. S. Kucukoglu, J. J. Remus, and L. M. Collins, “Acoustic
model investigation of a multiple carrier frequency algorithm for encoding fine frequency structure: Implications for cochlear implants,” Hearing Research, vol. 218,
pp. 30–42, 2006.
[68] C. Arnoldner, D. Riss, M. Brunner, M. Durisin, W. D. Baumgartner, and J. S.
Hamzavi, “Speech and music perception with the new fine structure speech coding
strategy: preliminary results,” Acta Oto-laryngologica, vol. 127, no. 12, pp. 1298–
1303, 2007.
[69] R. Schatzer, A. Krenmayr, D. K. K. Au, M. Kals, and C. Zierhofer, “Temporal fine
structure in cochlear implants: Preliminary speech perception results in cantonesespeaking implant users,” Acta Oto-laryngologica, vol. 130, no. 9, pp. 1031–1039,
2010.
[70] C. Fei and Z. Yuan-Ting, “Zerocrossing-based nonuniform sampling to deliver lowfrequency fine structure cue for cochlear implant,” Digital Signal Processing, vol. 21,
no. 3, pp. 427–432, 2011.
[71] X. Li, K. Nie, N. S. Imennov, J. T. Rubinstein, and L. E. Atlas, “Improved perception of music with a harmonic based algorithm for cochlear implants,” IEEE
Transactions on Neural Systems and Rehabilitation Engineering, vol. 21, no. 4, pp.
684–694, July 2013.
[72] F. G. Zeng, “Role of temporal fine structure in speech perception,” The Journal of
the Acoustical Society of America, vol. 123, no. 5, p. 3710, 2008.
[73] D. Hilbert, Grundzuge einer allgemeinen theorie der linearen integralgleichungen.
Leipzig, B. G. Teubner, 1912.
[74] A. M. H. J. Aertsen and P. I. M. Johannesma, “Spectrotemporal receptive fields of
auditory neurons in the grassfrog I: characterization of tonal and natural stimuli,”
Biological Cybernetics, vol. 38, no. 4, pp. 223–234, 1980.
[75] E. de Boer and H. R. de Jongh, “On cochlear encoding: Potentialities and limitations of the reverse-correlation technique,” The Journal of the Acoustical Society of
America, vol. 63, no. 1, pp. 115–135, 1978.
[76] L. Carney and C. Yin, “Temporal coding of resonances by low-frequency auditory
nerve fibers: single-fiber responses and a population model,” Journal of neurophysiology, vol. 60, no. 5, pp. 1653–1677, 1988.
[77] R. D. Patterson, K. Robinson, J. Holdsworth, D. Mckeown, C. Zhang, and M. Allerhand, “Complex sounds and auditory images,” in Proc. 9th Int. Symp. Hearing
Audit., Physiol. Perception, 1992, pp. 429–446.
REFERENCES
97
[78] B. R. Glasberg and B. C. J. Moore, “Derivation of auditory filter shapes from
notched-noise data,” Hearing research, vol. 47, no. 1, pp. 103–138, 1990.
[79] M. Slaney, “An efficient implementation of the patterson-holdsworth auditory filter
bank,” Apple Comput. Tech. Rep.# 35, 1993.
[80] S. A. P. Haddad, “Ultra low-power biomedical signal processing: An analog wavelet
filter approach for pacemakers.” Ph.D. dissertation, Delft University of Technology,
2006.
[81] D. A. Johns, W. M. Snelgrove, and A. S. Sedra, “Orthonormal ladder filters,” IEEE
Transactions on Circuits and Systems, vol. 36, no. 3, pp. 337–343, March 1989.
[82] S. A. P. Haddad and W. A. Serdijn, Ultra Low-Power Biomedical Signal Processing:
An Analog Wavelet Filter Approach for Pacemakers, 1st ed. The Netherlands:
Springer Netherlands, 2009, ch. Optimal State Space Descriptions, pp. 75–94.
[83] X. Zhang and Y. Shinada, “A second-order active bandpass filter with complex
coefficients and its applications to the hilbert transform,” Electronics and Communication in Japan, vol. 79, no. 6, pp. 13–22, March 1996.
[84] E. A. Vittoz and J. Fellrath, “CMOS analog integrated circuits based on weak
inversion operation,” IEEE Journal of Solid-State Circuits, vol. 12, no. 3, pp. 224–
231, 1977.
[85] N. Y. S. Kiang, T. Watanabe, E. C. Thomas, and L. F. Andclark, Discharge patterns
of single fibers in the cats auditory nerve. Cambridge and Mass.: M.I.T Press, 1965.
[86] S. K. Padala and K. M. M. Prabhu, “Systolic arrays for the discrete Hilbert transform,” in Proc. IEE Circuits Devices Systems, vol. 144, no. 5, Paris, France, May
1997, pp. 259–264.
[87] K. Koli and K. Halonen, “Low voltage mos-transistor only precision current peak
detector with signal independent discharge time,” in Proc. IEEE International Symposium on Circuit and Systems (ISCAS), Hong Kong, June 1997, pp. 1992–1995.
[88] R. Dlugosz and I. Iniewski, “High precision analog peak detector for xray imaging
applications,” IET Electronics Letters, vol. 43, no. 8, pp. 440–441, 2007.
[89] C. Sawigun and W. A. Serdijn, “A 24 nW, 0.65 V 74 dB SNDR 83 dB-DR class-AB
current-mode sample and hold circuit,” in Proc. IEEE International Symposium on
Circuits and Systems (ISCAS), Paris, France, May 2010, pp. 3132–3135.
[90] J. C. Lilly, J. R. Hughes, J. E. C. Alvord, and T. W. Galkin, “Brief, noninjurious
electric waveforms for stimulation of the brain,” Science, vol. 121, pp. 468–469,
1955.
[91] S. C. DeMarco, W. Liu, P. R. Singh, G. Lazzi, M. S. Humayun, and J. D. Weiland,
“An arbitrary waveform stimulus circuit for visual prosthesis using a low-area multibias DAC,” IEEE Journal of Solid-State Circuits, vol. 38, no. 10, pp. 1679–1690,
Oct. 2003.
98
REFERENCES
[92] M. Sivaprakasem, W. Lui, M. S. Humayun, and D. J. Weiland, “A variable range biphasic current stimulus driver circuitry for an implantable retinal prosthesis device,”
IEEE Journal of Solid-State Circuits, vol. 40, no. 3, pp. 763–771, Mar. 2005.
[93] E. K. F. Lee and A. Lam, “A matching technique for biphasic stimulation pulse,”
in Proc. IEEE International Symposuim on Circuit and Systems, New Orleans, LA,
May 2007, pp. 817–820.
[94] M. Ortmanns, N. Unger, A. Rocke, M. Gehrke, and H. J. Tietdke, “A 0.1 mm2
digitally programmable nerve stimulation pad cell with high voltage capability for
a retinal implant,” in Proc. IEEE International Solid-State Circuits Conference,
pp. 89–98, Feb. 2006.
[95] M. Ghovanloo and K. Najafi, “A compact large voltage compliance high output
impedance programmable current source for implantable microstimulators,” IEEE
Transactions on Biomedical Engineering, vol. 52, pp. 97–105, Jan. 2005.
[96] M. Ghorbel, M. Samet, A. B. Hamida, and J. Tomas, “A 16-electrode fully integrated and versatile CMOS microstimulator dedicated to cochlear implant,” Journal
of Applied Sciences, vol. 6, no. 15, pp. 2978–2990, Sep. 2006.
[97] L. Zeng, X. Yi, S. Lu, Y. Lou, J. Jiang, H. Qu, N. Lan, and G. Wang, “Design of a
high voltage stimulator chip for a stroke rehabilitation system,” in Engineering in
Medicine and Biology Society (EMBC), 2013 35th Annual International Conference
of the IEEE, July 2013, pp. 834–837.
[98] X. Liu, A. Demosthenous, and N. Donaldson, “An integrated implantable stimulator that is fail-safe without off-chip blocking-capacitors,” IEEE Transactions on
Biomedical Circuits and Systems, vol. 2, no. 3, pp. 231–244, Sep. 2008.
[99] H. Chun, Y. Yang, and T. Lehmann, “Safety ensuring retinal prosthesis with precise
charge balance and low power consumption,” IEEE Transactions on Biomedical
Circuits and Systems, vol. 8, no. 1, pp. 108–118, Feb. 2014.
[100] J. J. Sit and R. Sarpeshkar, “A low-power blocking-capacitor-free charge-balanced
electrode-stimulator chip with less than 6 nA DC error for 1 mA full-scale stimulation,” IEEE Transactions on Biomedical Circuits and Systems, vol. 1, no. 3, pp.
172–183, Sept 2007.
[101] E. McAdams and J. Jossinet, “Nonlinear transient response of electrode-electrolyte
interfaces,” Medical and Biological Engineering and Computing, vol. 38, pp. 427–
432, 2000.
[102] K. E. Jones and R. A. Normann, “An advanced demultiplexing system for physiological stimulation,” IEEE Transactions on Biomedical Engineering, vol. 44, no. 12,
pp. 1210–1220, Dec. 1997.
[103] E. Noorsal, K. Sooksood, H. Xu, R. Hornig, J. Becker, and M. Ortmanns, “A neural
stimulator frontend with high-voltage compliance and programmable pulse shape
for epiretinal implants,” IEEE Journal of Solid-State Circuits, vol. 47, no. 1, pp.
244–256, Jan. 2012.
REFERENCES
99
[104] N. Lawand, W. Ngamkham, G. Nazarian, P. French, W. Serdijn, G.N.Gaydadjiev,
J. Briaire, and J. Frijns, “An improved system approach towards future cochlear
implants,” in Proc. IEEE Engineering in Medicine and Biology Society Conference
(EMBC), July 2013, pp. 5163–5166.
[105] C. Ping-Yuan and Y. Chien-Cheng, “A voltage level converter circuit design with
low power consumption,” in The 6th IEEE International Conference on ASIC, Oct.
2005, pp. 24–27.
[106] C. Q. Huang, R. K. Shepherd, P. M. Carter, P. M. Seligman, and B. Tabor, “Electrical stimulation of the auditory nerve: Direct current measurement in vivo,” IEEE
Transactions on Biomedical Engineering, vol. 46, pp. 461–470, Apr. 1999.
[107] C. Sawigun, W. Ngamkham, and W. A. Serdijn, “A 0.5 V, 2 nW, 55 dB-DR,
fourth-order bandpass filter using single branch biquads: An efficient design for
FoM enhancement,” Microelectronics Journal, vol. 45, pp. 367–374, April 2014.
[108] C. J. Bes, C. Sawigun, and W. A. Serdijn, “An additive instantaneously companding readout system for cochlear implants,” in Proc. IEEE Biomedical Circuits and
Systems Conference (BioCAS), Paphos, Cyprus, Nov. 2010, pp. 3–5.
Acknowledgements
Certainly, this thesis would not have been possible without the support and
understanding of many people. Therefore, I would like to thank all of them,
including those who I may not have a chance to mention here.
First of all, I would like to thank Prof. Wouter A. Serdijn, my supervisor, for
his enthusiastic guidance and support. Thanks for giving me an opportunity
to have a great experience in an international research group. I feel like I am
one of the family members in the Section Bioelectronics and not only being
a Ph.D. student. You support us in all kind of activities including a great
time of practising for and performing at the ELCA music festival. You show
to me that you see the value of every student. You always remind us: “don’t
forget to balance your life”. I sincerely appreciate you for all the support,
encouragement, patience and trust in me throughout these years and I am
very proud to be a part of the Section Bioelectronics under your supervision.
I would like to express my gratitude to the project leaders of the SMAC-It
project, Prof.dr. Paddy French, Prof. John Long, Prof.dr.ir. G. N. Gaydadjiev, Prof.dr.ir. Johan Frijns, and Dr.ir. Jeroen Briaire for your support and
giving me your valuable time, useful advice and discussions for the project.
Special thanks go to my project participants, Nishant Lawand, and Ghazaleh Nazarian for the interesting discussions and the help to each other to
improve this work.
I thank the Dutch Technology Foundation (STW) for the financial support
for this Ph.D. project.
I would also like to thank the members of my graduation committee, who
101
have taken the effort to read my thesis. Thank you all for your time.
I am indebted to Chutham Sawigun who encouraged me to apply for this
project. Thank you very much for your caring, company and being a great
colleague. I have learned many things from you, not only in scientific research but also about politics and personal life.
I thank Marion de Vlieger who always is very helpful and kind to me. Thanks
very much for arranging all kind of things for me since the begining of my
Ph.D. studies until the end, cheering me up and encouraging me to be strong
when I sometimes faced difficulties. I also received many help from Loek
van Schie on solving computer problems, Wil Straver and Atef Akhnoukh
for preparing my chip for measurement. Thank you very much for your help.
And particularly, Ali Kaichouhi who spent so much time teaching me about
the software for lay-out and PCB, helping me on Cadence problems and chip
bonding. I am also grateful to him.
I wish to express my gratitude to all former students and colleagues in
the Section Bioelectronics: Vaibhav Maheshvari, Chema García del Pozo,
Dr. Reza Lotfi, Vincent Bleeker, Michiel Grashuis, Salim Alahdab, Gaurav
Mishra, Duan Zhao, Marijn van Dongen, Senad Hiseni, Mark Stoopman,
Yongjia Li, Yao Liu, Andre Mansano, Gustavo Martins, Jose Ángel Miguel
Díaz, Menno Vastenholt, Wu Chi Wing, Hossein Tajeddin, Jory Koolwijk,
Sander Fondse and others I may have forgotten. I will never forget the
great time during our group meetings and all activities we joined. Thank
you all very much for your friendship and support. I also want to thank
the people that I learned to know on the 18th floor, EWI building: Sumit
Bagga, Wanghua Wu, Yi Zhao, Mina Danesh, Wei Liat Chan, Nitz Saputra, Leonardo Vera, Yanyu Jin, Seyed Morteza Alavi, Akshay Visweswaran,
Augusto Ximenes, Sandro Binsfeld Ferreira, Gerasimos Vlachogiannakis,
Satoshi Malotaux and many others I may have forgotten. Thank you all
very much for your friendship and support.
My Ph.D. journey might not have begun without the support from Mahanakorn University of Technology (MUT), Thailand, and all of my colleagues at MUT who I share a great friendship with until now. I would
like to thank them all. Special thanks to Dr. Suchada Jongsatitsathaporn
for your kindness sending many letters and poscards form Thailand to me.
102
Good luck in your life. I must send my gratitude to Dr. Benjalug Namfa and
her husband, Dirak Namfa, for their trust and support for my study abroad.
I will never forget both of you. I would also to send my gratitude to my
former teachers. Prof. Wiwat Kiranon, Dr. Panom Petchjatuporn, Chaiboon Kangsajian, Miles Tretheway and Dr. Phitchaphat Bowonrattanaset
who always encouraged and pushed me to pass through my Ph.D. studies.
I also want to thank all my good friends in Thailand: Pattaraporn Yaprom,
Saowanee Arsakit, Phasuk Atipunyavises, Sasicha Kamphueng, Nattaporn
Sangkornchiputt, Amornrat Kawinteeraphap and Narumol Kiatwarin (from
Switzerland). Thank you all for your help and keeping in touch with me. My
friends from the Netherlands should also be mentioned: Luca Giangrande
and Susanna Valenti, “Rainbow” band members (Rainbow, Florrie, Sophia
and Anne), Nertnil, Kris and Jan Linskens, Sunisa Staszewski. Thank you
all for a warm friendship and support.
I would like to thank the Bes family: Kees, Zilla, Dick, David, Daan, Bart
and also Sabina Trocha, Signe Weber Carlsen and Lies for their support and
warm friendship.
Kind support from my parents and family members also contributed to my
Ph.D. life. I am indebted to them.
My Ph.D. journey might not have come to an end without my lovely husband, Cees-Jeroen Bes. From being a colleague, a close friend until now my
husband, you are so supportive and helpful. Thank you very much for your
love, understanding, and being everything to me.
Finally, my special thanks go to my father, Suthep Ngamkham who passed
away while I was studying in the Netherlands. You always collected all
materials and documents that I got from school and university except this
thesis. You have always been the power behind my progress. I love you and
miss you every day!
Wannaya Ngamkham
103
List of Publications
1. W. Ngamkham, M. N. van Dongen, W. A. Serdijn C. J. Bes, J. J. Briaire, and J.
H. M. Frijns, “A 0.042 mm2 Programmable Biphasic Stimulator for Cochlear Implants Suitable for a Large Number of Channels,” in arXiv.org, arXiv:1502.00549,
Jan. 2015.
2. C. Sawigun, W. Ngamkham and W. A. Serdijn, “A 0.5 V, 2 nW, 55 dB-DR,
Fourth-Order BPF Using Single Branch Biquads: An Efficient Design for FoM
Enhancement” Microelectronics Journal, vol. 45, pp.367-374, April 2014.
3. N. S. Lawand, W. Ngamkham, G. Nazarian, P. J. French, W. A. Serdijn, G. N.
Gaydadjiev, “An improved system approach towards future cochlear implants, in
Proc. IEEE Engineering in Medicine and Biology Society (EMBC), Osaka, Japan,
July 2013.
4. W. Ngamkham, M. N. van Dongen and W. A. Serdijn,“A Charge Balanced
Biphasic Stimulator for Cochlear Implants,” invited presentation, BME 2013, Egmond
aan Zee, the Netherlands, Jan. 24-25, 2013.
5. C. Sawigun, W. Ngamkham, and W. A. Serdijn, “A 2.6 nW, 0.5 V, 52 dB-DR,
4th-order Gm-C BPF: Moving Closer to the FoM’s Fundamental Limit,” in Proc.
IEEE International Symposium on Circuits and Systems (ISCAS), Seoul, South
Korea, May 20-23, 2012, pp. 656-659.
6. W. Ngamkham, M. N. van Dongen, and W. A. Serdijn, “Biphasic Stimulator
Circuit for a Wide Range of Electrode-Tissue Impedance Dedicated to Cochlear
Implants,” in Proc. IEEE International Symposium on Circuits and Systems (ISCAS), Seoul, South Korea, May 20 - 23, 2012, pp. 1083-1086.
7. W. Ngamkham, Marijn van Dongen, Wouter A. Serdijn: An Implementation of
Least-Voltage Drop Stimulator Circuit for Cochlear Implants, Book of Abstracts,
ICT.Open, Rotterdam, the Netherlands, October 22-23, 2012.
8. W. Ngamkham, C. Sawigun and W. A. Serdijn, “The design of an ultra lowpower analog bionic ear,” invited presentation, BME 2011, Egmond aan Zee, the
Netherlands, Jan. 20, 2011.
9. C. Sawigun, W. Ngamkham, and W. A. Serdijn, “A least-voltage drop high
output resistance current source for neural stimulation,” in Proc. IEEE Biomedical
Circuits and Systems Conference (BioCAS), Paphos, Cyprus, Nov. 2010, pp. 110113.
105
10. C. Sawigun, W. Ngamkham, and W. A. Serdijn, “An ultra low-power peakinstant detector for a peak picking cochlear implant processor,” in Proc. IEEE
Biomedical Circuits and Systems Conference (BioCAS), Paphos, Cyprus, Nov.
2010, pp. 222-225.
11. C. Sawigun, W. Ngamkham, and W. A. Serdijn, “Comparison of speech processing strategies for the design of and ultra low-power analog bionic ear,” in Proc.
IEEE Engineering in Medicine and Biology Society (EMBC), Buenos Aires, Argentina, 2010, pp. 1374-1377.
12. W.Ngamkham, C.Sawigun, S.Hiseni, and W.A.Serdijn, “Analog complex gammatone filter for cochlear implant channels,” in Proc. IEEE International Symposium on Circuits and Systems (ISCAS), Paris, France, May 2010, pp. 969-972.
13. S. Hiseni, C. Sawigun, W. Ngamkham, and W. A. Serdijn, “A Compact, NanoPower CMOS Action Potential Detector”, In Proc. IEEE Biomedical Circuits and
Systems Conference (BioCAS), Beijing, China, Nov. 26-28 2009.
106
About the Author
Wannaya Ngamkham was born in Nong Prue, Thailand on August 23rd ,
1972. She received the B. Eng and M. Eng. both in electrical engineering
from Mahanakorn University of Technology (MUT), Bangkok, Thailand,
in 1995 and 2001, respectively. From 1995-2008 she was a lecturer at the
Department of Control Systems and Instrumentation, MUT. In November
2008, she joined the Section Bioelectronics of the Department of Microelectronics, Faculty of Electrical Engineering, Mathematics, and Computer
Science, Delft University of Technology (TUDelft), the Netherlands, where
she started her research towards her Ph.D. degree. Her research is part of
the STW funded SMAC-It project (Smart Cochlear Implants). Her research
interests include low power analog integrated circuits and system designs for
totally implantable cochlear implants.
107
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF

advertisement