Biosensors 13 00077 v2 1

biosensors
Article
Highly Sensitive Readout Interface for Real-Time Differential
Precision Measurements with Impedance Biosensors
Sara Neshani 1, *, Kasra Momeni 2 , Degang J. Chen 3 and Nathan M. Neihart 3
1 Electrical and Computer Engineering Department, University of Alabama, Tuscaloosa, AL 35401, USA
2 Mechanical Engineering Department, University of Alabama, Tuscaloosa, AL 35401, USA
3 Electrical Engineering Department, Iowa State University, Ames, IA 50010, USA
* Correspondence: [email protected]
Abstract: Field deployment is critical to developing numerous sensitive impedance transducers.

Precise, cost-effective, and real-time readout units are being sought to interface these sensitive
impedance transducers for various clinical or environmental applications. This paper presents a
general readout method with a detailed design procedure for interfacing impedance transducers
that generate small fractional changes in the impedance characteristics after detection. The emphasis
of the design is obtaining a target response resolution considering the accuracy in real-time. An
entire readout unit with amplification/filtering and real-time data acquisition and processing using a
single microcontroller is proposed. Most important design parameters, such as the signal-to-noise
ratio (SNR), common-mode-to-differential conversion, digitization configuration/speed, and the data
processing method are discussed here. The studied process can be used as a general guideline to
design custom readout units to interface with various developed transducers in the laboratory and
verify the performance for field deployment and commercialization. A single frequency readout unit
with a target 8-bit resolution to interface differentially placed transducers (e.g., bridge configuration)
is designed and implemented. A single MCU is programmed for real-time data acquisition and
sine fitting. The 8-bit resolution is achieved even at low SNR levels of roughly 7 dB by setting the
component values and fitting algorithm parameters with the given methods.
Keywords: real-time; impedance biosensor; low-cost; precision; sensitive; readout; interface

Citation: Neshani, S.; Momeni, K.;
Chen, D.J.; Neihart, N.M. Highly
Sensitive Readout Interface for
Real-Time Differential Precision 1. Introduction
Measurements with Impedance Recently, real-time sensing and detection are revolutionizing many services such as
Biosensors. Biosensors 2023, 13, 77. healthcare, home automation, transportation, etc. Moreover, there is a push to develop
https://doi.org/10.3390/ more field-deployable biosensors for medical diagnostics and environmental monitor-
bios13010077 ing applications. Motivated by this, there has been a surge in the development of new
Received: 15 October 2022 transducers targeting a wide range of biosensing applications for detecting proteins, ions,
Revised: 27 December 2022 temperature, etc.
Accepted: 29 December 2022 Label-free impedance biosensors are famous prototypes for point-of-care real-time
Published: 2 January 2023 applications such as toxins, viruses, whole-cell, bacteria, nucleic acids, detection, or tissue
impedance modeling [1–6]. Two-, three-, or four-electrode transductions are used for
various impedance measurements. In general, the ratio between the voltage across the
reference electrode (RE) and the working electrode (WE) and the current flowing between
Copyright: © 2023 by the authors. the working electrode and the counter electrode (CE) yields the impedance of a specific
Licensee MDPI, Basel, Switzerland. interface (solution electrode) [1,7] or tissue [8], cell culture [9], etc. The variable impedance
This article is an open access article
of the transducer is generally estimated by AC electrochemical impedance spectroscopy
distributed under the terms and
(EIS) [1]. The EIS method measures the medium or interface’s impedance at multiple
conditions of the Creative Commons
frequencies [10,11]. Next, proper fitting techniques estimate the impedance model from
Attribution (CC BY) license (https://
the spectrum [12]. However, for the development of a simple real-time, field-deployable
creativecommons.org/licenses/by/
readout interface, a single frequency AC excitation [1,3] is generally a good alternative
4.0/).
Biosensors 2023, 13, 77. https://doi.org/10.3390/bios13010077 https://www.mdpi.com/journal/biosensors

Biosensors 2023, 13, 77 2 of 25
for applications where the model of the interface or tissue is known a priori and the
quantification of a specific analyte or change in the impedance due to bio-recognition
is desired. One method to quantify the transducer impedance change is measuring the
amplitude and phase of the response signal (current or voltage) with respect to a reference
signal in real-time [13].
However, one challenge in the design of readout interfaces for label-free real-time
single-frequency measurements is that the total change in the impedance of the transducers
is often only a few percent over the entire full range of detection. For example, capacitive
transducers for measuring certain hepatotoxins in water have been presented in [2], where
the total electrode–solution interfacial capacitance changes by only 1% over the entire range
of concentrations. A similar issue is associated with the developed label-free transducers
for detecting Cholera toxins in [14] and Cryptosporidium at lower concentrations [15]. The
impedance phase change at sub-Hz frequency is employed to detect human interleukin-8 in
serum with sub-pg/mL sensitivity [16]. The results indicate a phase change of only 173 m◦
over a baseline impedance phase of −86.6◦ , at the minimum detection level and an overall
sensitivity of 220.4 m◦ /decade (approximately 0.25%). The resistance of the ultra-sensitive
interdigitated electrodes designed in [17] changes by roughly a couple of percent over
the full range of PfHRP2—a malaria biomarker—concentrations in human saliva. The
non-faradaic EIS biosensor for detecting C-reactive protein in a complex medium, such
as human blood, reports a fractional impedance change of approximately 2.5% at the
minimum detection limit [18]. The capacitance of the developed DNA sensors changes by
roughly 3% for the reported concentrations [19]. To complicate matters, avoiding physical
damage to the functional layer and nonlinear distortion effect on the response require that
the magnitude of the applied AC excitation remain small (typically <50 mV) [1,3,7]. The
result is that the absolute value of the transducer output voltage is small, leading to a
full-scale change in the voltage in the order of hundreds of microvolts. Moreover, suppose
we assume that the transducer’s output voltage will be detected with a modest resolution
of 8-bits. In that case, the sensor must be able to detect changes in the transducer’s output
voltage that are less than 1 µV.
The readout process for the abovementioned sensitive transducers is straightforward,
using laboratory-grade bench-top test equipment for characterization, but very sensitive
instrumentation is required for large-scale deployment. As a solution, differential sensing
with a bridge circuit that acts similar to intermediate secondary transduction and a sen-
sitivity booster is proposed in [20]. Two capacitive transducer chips, each with a pair of
electrodes (functionalized WE and chemically non-functional CE providing the electrical
signal path in the solution), are placed in the opposite bridge legs and generate a differential
sensing response. Although with the differential capacitive bridge designed in [20], the
effects of common-mode noise, drift, and temperature variation can be further decreased
and sensitivity enhanced, there is still a need for small, inexpensive, and low-power readout
to reach a particular required resolution and make the commercialization of the developed
transducers in this structure feasible [8,9].
Figure 1 shows the block diagram of the differential sensing unit utilized in [20]
interfaced with high-level generic blocks of a readout system, amplification, filtering,
and digitization. The transducers are designed to sense the change in the transducer
impedance due to bio-recognition with respect to the reference networks differentially.
Amplification/filtering is typically required with the expected weak and noisy sensor
response. For real-time and low-cost operation, an MCU performs both the task of data
acquisition and processing (fitting) and any extra required digital adjustments. The MCU
acquires the response and excitation source signals and measures the amplified response
signal magnitude and differential phase (with respect to the excitation source) in real-time.
Biosensors 2023, 13,
Biosensors 2023, 13, 77
x FOR PEER REVIEW 33of
of 24
25
Figure
Figure 1.1. General
General block
block diagram
diagramofofaadifferential
differentialbio-sensing
bio-sensing unit
unit interfaced
interfaced with
with signal
signal amplifica-
amplification
tion and filtering
and filtering unita and
unit and a microcontroller
microcontroller (MCU)(MCU) for low-cost
for low-cost real-timereal-time operation
operation and fieldand field de-
deployment.
ployment. The transducers are designed to sense the change in the transducer impedance due to
The transducers are designed to sense the change in the transducer impedance due to bio-recognition
bio-recognition with respect to the reference networks differentially. Amplification/filtering is typ-
with respect to the reference networks differentially. Amplification/filtering is typically required with
ically required with the expected weak and noisy sensor response. For real-time and low-cost oper-
the expected
ation, an MCU weak and noisy
performs sensor
both the response. Foracquisition
task of data real-time and low-cost
and operation,
processing an MCU
(fitting) performs
and any extra
both the task of data acquisition
required digital adjustments. and processing (fitting) and any extra required digital adjustments.
Some factors
Some factorslimit
limitthe theperformance
performance of the
of generic
the genericsensorsensor
interface units shown
interface units in Figurein1
shown
if implemented with discrete commercially available components.
Figure 1 if implemented with discrete commercially available components. Typical preci- Typical precision opamps
that opamps
sion would bethat used
wouldin√this application
be used have input referred
in this application have input noise voltages
referred noisethat are onthat
voltages the
order of tens of nV/ Hz. Therefore, to maintain a practical input signal-to-noise ratio
are on the order of tens of nV⁄√Hz. Therefore, to maintain a practical input signal-to-noise
(SNR) greater than 1, with microvolt-level transducer output voltages, the effective noise
ratio (SNR) greater than 1, with microvolt-level transducer output voltages, the effective
bandwidth of the sensor interface circuit could be limited to only a few tens of hertz. Moreover,
noise bandwidth of the sensor interface circuit could be limited to only a few tens of hertz.
approximately 60–80 dB voltage gains will be required to effectively use an ADC for data
Moreover, approximately 60–80 dB voltage gains will be required to effectively use an
acquisition. Extreme amplification and filtering require highly accurate component matching,
ADC for data acquisition. Extreme amplification and filtering require highly accurate
extensive shielding, and careful circuit design. Data acquisition and the specific algorithm
component matching, extensive shielding, and careful circuit design. Data acquisition and
implemented in the MCU for data processing in Figure 1 should be performed simultaneously
the specific algorithm implemented in the MCU for data processing in Figure 1 should be
with no need to transmit or store the raw data in external memory. Therefore, the algorithm
performed simultaneously with no need to transmit or store the raw data in external
processing speed and the ADC sampling rate should be carefully adjusted. Lower complexity,
memory.
power Therefore, and
consumption, the algorithm
overall costprocessing speedgoals
are other design and the ADClimiting
to avoid sampling therate should
potential of
be
thecarefully
sensor. adjusted. Lower complexity, power consumption, and overall cost are other
design goals
This to avoid
paper proposes limiting the potential of
a comprehensive, thedifferential
fully sensor. readout interface design suit-
able for tiny sensor response signals generated differently. Areadout
This paper proposes a comprehensive, fully differential study ofinterface
the design design suit-
tradeoffs
able for tiny sensor response signals generated differently. A
between overall sensor system complexity and performance at low cost, particularly with study of the design tradeoffs
between overall detection
small fractional sensor system complexity
impedance change and performance
(<1%), is given. at lowreadout
Each cost, particularly with
interface block
small
shownfractional
in Figure detection
1 was analyzed impedance change (<
and designed 1%),onisthe
based given. Eachresolution,
required readout interface
overall
block shown in Figure 1 was analyzed and designed based
gain, SNR, etc. It is shown here that extreme amplification and filtering requirements canon the required resolution,
overall gain, the
be met with SNR, etc. Itdesign
careful is shown of ahere that extreme
two-channel amplification
digital acquisitionand andfiltering
processingrequire-
(sine
ments can be met with the careful design of a two-channel digital
fitting) utilizing a single microcontroller. A theoretical design procedure and a practical acquisition and pro-
cessing (sine fitting) utilizing a single microcontroller. A theoretical
discrete implementation example are presented here, targeting mainly real-time, low-cost design procedure and
aoperation
practicalfor discrete implementation
less than 1% full-scaleexample change in arethepresented
transducer’s here,output
targeting mainly
voltage, real-
and an
time, low-cost operation
8-bit characteristic changefor less thanIn1%
resolution. full-scale
Section change
2, design in theconsidering
methods transducer’sthe output
most
voltage,
important and an 8-bit characteristic
performance parameters, changesuch as resolution.
overall gain,Inbandwidth,
Section 2, design methods con-
SNR, common-mode
sidering
rejection theratiomost important
(CMRR), sampling performance
rate, signalparameters, such asand
fitting algorithm, overall gain, implementa-
real-time bandwidth,
SNR, common-mode rejection ratio (CMRR), sampling rate, signal
tion for each of the readout blocks; amplification, filtering, and data acquisition/processing fitting algorithm, and
real-time
are explainedimplementation
in detail. The forimplementation
each of the readout results,blocks;
board amplification,
characterization,filtering, and data
and real-time
acquisition/processing
operation and sensitivity areverification
explained in aredetail. The implementation
presented in Section 3. results, board charac-
terization, and real-time operation and sensitivity verification are presented in Section 3.
2. Materials and Methods
2. Materials and Methods
A reasonable assumption for the differential sensing unit shown in Figure 1 is that
less than a 1% full-scale
A reasonable response
assumption forchange with the bio-recognition
the differential sensing unit shown event inis expected
Figure 1 is before
that
any than
less amplification. The change
a 1% full-scale response in the transducer’s
change impedance (Zt ) event
with the bio-recognition has a linear relationship
is expected before
any amplification. The change in the transducer’s impedance (𝑍𝑡 ) has a linear relationship
Biosensors 2023, 13, 77 4 of 25
with the change in the medium’s concentration. Thus, the differential output voltage of
the sensing unit is linearly related to the transducer fractional impedance change, and the
source voltage (1) shows this linear relationship.
∆Zt
Va − Vb ≈ α V , (1)
Zt AC
where Va − Vb , is the differential output voltage of the sensing unit, α is a proportionality

constant, ∆Z t
Zt , is the transducer fractional impedance change, and VAC , is the AC excitation
source voltage.
Before discussing the amplification stage, it is mandatory to examine some typical
values for the expected differential voltage. Let us consider a typical excitation signal (VAC )
with an AC amplitude of 1–100 mV. Assuming the case with α = 1, and ∆Z t
Zt < 1%, the
maximum change in the amplitude of the bridge output will be less than 1 mV. If the
target resolution for the sensor is set to the typical 8-bits, then the expected least-significant
bit (LSB) of the response will be several microvolts. With the minimum response signal
level knowledge, the proceeding amplification and filtering stages can be designed, and
parameters such as required gain, matching, bandwidth, etc., can be decided. However,
dealing with such small signals requires the careful consideration of parasitic, noise, and
any unwanted interfering signals and trying to minimize such effects and achieve the
dynamic target range; these design considerations are discussed in the following.
2.1. Amplification and Filtering Analysis

Two critical design parameters for the amplification and filtering unit will impact the
accuracy and precision of a fully differential readout circuit for detecting tiny fractional
changes at the output: common-mode-induced differential conversion and noise. The com-
mon mode rejection ratio (CMRR) criteria here is the amplification and filtering interface’s
ability to reject the common mode output voltage (for example, an equal DC signal required
to bias the differential transducers) and amplify the differential output voltage change due
to the transducer impedance change with detection. Failing to reject the common mode
at the output, considering the high required differential gain, will lead to an unpleasant
common mode to differential conversion that not only gives rise to a false detection signal
but also limits the dynamic range. Noise will affect the lower detection limit and shrink the
dynamic range. Various sources causing common-mode to differential conversion and SNR
degradation at the interface’s output before ADC are analyzed in the following section.
Based on this analysis, parameters such as the gain of each stage for a cascaded design,
effective noise bandwidth (ENB) of the filter, and the required matching sensitivity can be
determined and based on the best achievable SNR at the board output, a potential suitable
fitting algorithm in the data acquisition unit can be picked and designed.
2.1.1. Common Mode to Differential Conversion

With the typical available full-scale voltage of the ADCs, A FS , on the order of (1–5 V),
a three-stage amplification and filtering interface is proposed here, as shown in Figure 2.
One stage of amplification and two identical bandpass filtering stages are responsible
for providing a total differential gain that amplifies the initial full-scale differentially
sensed output of roughly 1 mV to match the full-scale range of the ADC. A multiple
feedback bandpass topology is chosen for the fully differential filters [21]. The filter’s
gain, center frequency, and bandwidth can be flexibly tuned with the multiple feedback
topology. Therefore, a total differential gain in the order of 70–80 dB is approximately
required. To estimate the amount of needed common mode rejection, consider the case
with α = 1/2 in (1). With |Va | = |Vb | = Vab,CM equal to tens of millivolts (based on the
excitation source amplitude < 100 mV) and the differential LSB change at the sensing unit
output, |(Va − Vb ) LSB |, will only be several microvolts.
Biosensors 2023,13,
Biosensors2023, 13,77x FOR PEER REVIEW 55 of
of2524
Figure2.2.The
Figure Thethree-stage
three-stageamplification
amplificationand andfiltering
filteringunit,
unit,proceeding
proceedingthe thedifferential
differentialsensing
sensingunit
unit
andsucceeding
and succeedingthe
thedata
dataacquisition
acquisitionand andfitting
fitting unit.
unit. AA fully
fully differential
differential amplification
amplification followed
followed by by
a
a cascaded
cascaded two-stage
two-stage bandpass
bandpass filter
filter provides
provides the the differential
differential gaingain
and and filters
filters the unwanted
the unwanted noisenoise
for
for certain
certain minimum
minimum SNR. SNR. It is critical
It is critical to maintain
to maintain good matching
good matching in the feedback
in the feedback ratios ofratios of the pe-
the peripheral
ripheral components, as shown in (3), to achieve the overall target CMRR.
components, as shown in (3), to achieve the overall target CMRR.
|𝑉ab,CM
V and|(|(𝑉
𝑎𝑏,𝐶𝑀 | and Va𝑎−−V𝑉b𝑏))LSB
𝐿𝑆𝐵||will
willbebeamplified
amplifiedby by
the the common
common mode mode to differential
to differential gain
gaindifferential
and and differential gain, respectively.
gain, respectively. To maintain
To maintain theresolution,
the desired desired resolution,
criteria (2) criteria
should be(2)
satisfied so that the differentially amplified sensing signal at the LSB level is at least twiceat
should be satisfied so that the differentially amplified sensing signal at the LSB level is
least
the twice the interference
interference caused by the caused
commonby the common
mode mode toconversion:
to differential differential conversion:
𝐴𝑣,𝑑𝑖𝑓𝑓 .|(𝑉𝑎 −𝑉𝑏 )𝐿𝑆𝐵 |
Vb ) LSB| | > 2,
Av,di f f𝐴·|(Va −.|𝑉 (2)
𝑣,𝐶𝑀𝑑𝑖𝑓𝑓 𝑎𝑏,𝐶𝑀 > 2, (2)
Av,CMdi f f · Vab,CM
where 𝐴𝑣,𝐶𝑀𝑑𝑖𝑓𝑓 , is the common mode-induced differential gain, 𝐴𝑣,𝑑𝑖𝑓𝑓 , is the differential
where |(𝑉
gain. A 𝑎 − 𝑉f𝑏f),𝐿𝑆𝐵
v,CMdi is |,the
and |𝑉𝑎𝑏,𝐶𝑀 mode-induced
common | are the minimum differential
differential gain, voltage and
Av,di f f , is thethe common
differential
mode voltage at the
gain. |(Va − V ) |, and V output of the sensing unit, respectively. The abovementioned
are the minimum differential voltage and the common criteria
b LSB ab,CM 𝐴𝑣,𝑑𝑖𝑓𝑓
(2) sets
mode the minimum
voltage at the output of 𝐶𝑀𝑅𝑅
for the 𝑇𝑜𝑡𝑎𝑙 =unit,
the sensing 𝐴
, to around
respectively. 100
The dB which, withcriteria
abovementioned 𝐴𝑣,𝑑𝑖𝑓𝑓
A𝑣,𝐶𝑀𝑑𝑖𝑓𝑓
v,di f f
(2)
ofsets the minimum
around 70 − 80 dB, the CMRR
forrequires = Av,CMdi
𝐴𝑣,𝐶𝑀𝑑𝑖𝑓𝑓
theTotal tof be, to
ataround 100to
least −30 dB−20
which, If theArequired
dB.with v,di f f of
f
minimum
around 𝐶𝑀𝑅𝑅
70–80 dB,𝑡𝑜𝑡𝑎𝑙 is notthe
requires reached,
Av,CMdithe
f f target at least −30
to be resolution will −20
tonot bedB.
achieved.
If the required
minimum
• CMRR
Effect of component
total is not reached, the target
matching and tolerance resolution will not be achieved.
• Effect of component
Ideally, the CMRR of matching and tolerance
a fully differential circuit would be infinite. However, in reality,
the CMRR is limited by the CMRR of each stage
Ideally, the CMRR of a fully differential circuit in the circuit
would and by matching
be infinite. However,the in periph-
reality,
eral components. Typical fully differential amplifiers for precision measurements
the CMRR is limited by the CMRR of each stage in the circuit and by matching the peripheral provide
relatively highTypical
components. CMRRfully
values differential 100 dB. However,
of around amplifiers when measurements
for precision configured as a provide
fully dif-
ferential amplification
relatively high CMRR valuesstage, as ofshown
around in100
Figure
dB. 2,However,
the CMRRwhen of each stage is determined
configured as a fully
not only byamplification
differential the CMRR of the Opamp
stage, as shown but
in also
Figure by 2,thethematching
CMRR ofbetween
each stagetheistwo symmet-
determined
𝑅𝐹𝑎1 𝑅𝐹𝑏1 Δ𝛽 Δ𝛽
rical
not onlyfeedback 𝛽𝑎the
ratios of
by the CMRR = Opamp but and 𝛽𝑏by
also = the matching . If 𝛽between
𝑏 = 𝛽 + the 𝛽𝑎 = 𝛽 − ,
and symmetrical
two
𝑅𝐹𝑎1 +𝑅𝐼𝑎1 𝑅𝐹𝑏1 +𝑅𝐼𝑏1 2 2
R Fa1 R Fb1 ∆β ∆β
feedback
therefore,ratios = Δ𝛽
𝛽𝑏 − β𝛽a𝑎 = + R Ia1 𝛽and
R Fa1 and β = R The
𝑏 + 𝛽𝑎b = 2β. + R stage β b = βis+derived
. If CMRR 2 andinβ[22]:
a = β− 2 ,
Fb1 Ib1
therefore, β b − β a = ∆β and β b + β a = 2β. The stage 1CMRR is derived in [22]:
𝐶𝑀𝑅𝑅𝑆𝑡𝑎𝑔𝑒 ≈ 1 Δ𝛽, (3)
+
𝐶𝑀𝑅𝑅1𝑂𝑝𝑎𝑚𝑝 𝛽
CMRRStage ≈ , (3)
where 𝐶𝑀𝑅𝑅𝑆𝑡𝑎𝑔𝑒 , is the effective CMRR ofCMRR
1
eachOpampstage,+ ∆β𝐶𝑀𝑅𝑅
β 𝑂𝑝𝑎𝑚𝑝 is the typical CMRR
Δ𝛽
of any given Opamp, and , is the feedback matching ratio. Therefore, inevitable degra-
𝛽
dation in 𝐶𝑀𝑅𝑅𝑂𝑝𝑎𝑚𝑝 is expected with the worse matching ratio of the feedback net-
works.
Biosensors 2023, 13, 77 6 of 25
where CMRRStage , is the effective CMRR of each stage, CMRROpamp is the typical CMRR of
∆β
any given Opamp, and β , is the feedback matching ratio. Therefore, inevitable degradation
in CMRROpamp is expected with the worse matching ratio of the feedback networks.
For a given stage with a differential voltage gain, Av,dd , R Fb = Av,dd R Ib , the resistance,
R’s tolerance ∆R, is included as R Ib = R ± ∆R, R Fb = Av,dd R ± ∆R. Feedback ratio
matching is related to the component tolerance using the following derivations:
A
v,dd R±∆R A
v,dd R±∆R
− ±2 ∆R
∆β βb − βa ( Av,dd +1) R±2∆R ( Av,dd +1) R±2∆R R
= = Av,dd R±∆R
= , (4)
β β ( Av,dd + 1) ± 2 ∆R
R
( Av,dd +1) R±2∆R
∆β
For β 1,
∆R ∆β ( Av,dd + 1) ∆β ( Av,dd + 1)
± = ≈ , (5)
β 2 1 − ∆β

R β 2
β
where ∆RR , is the percent tolerance of a resistance. Based on (6), a lower component tolerance
is required to maintain a given feedback-matching ratio for a higher differential gain of a
stage. For the design and implementation of a multistage fully differential amplification
unit, as shown in Figure 2; (3), and (5) determine the required degree of component
matching tolerance based on the known differential gain distribution and total CMRR.
• Effect of cascaded stages
The abovementioned design criteria are regarding singular stages; now, given that
several stages participate in the amplification and filtering before the data acquisition, it is
helpful to understand each stage’s contribution to the total CMRR. In [23], the contribution
of each stage, on the total CMRR, of a cascade of 3 differential stages is investigated, and an
approximate formula, (6), is derived.
1 1 1 1
≈ + Av,dd2
+ Av,dd2 Av,dd3
, (6)
CMRR Total CMRRStage1 CMRRSatge2 CMRRSatge3
Av,cc2 Av,cc2 Av,cc3
where, CMRR Total , is the total effective CMRR of the cascaded stages, CMRRStage is the
CMRR contribution of the individual amplification stage, Av,dd , and Av,cc are the differ-
ential to differential and common mode to common mode gains of each of the 3 stages,
respectively. If the output common-mode voltage, VOCM , of the fully differential amplifiers
before ADC, are set externally at A FS /2, to cover the full ADC dynamic range equally, then
the Av,cc for all stages would be 1. Thus, if the two filtering stages are identical, with the
same gains, based on the (6) effect of the CMRR of the first stage on the CMRR Total is domi-
nant compared to higher stages. Therefore, component matching in the first amplification
stage significantly affects the total CMRR and common mode to differential conversion,
and this is specifically important for implementation and debugging.
2.1.2. Noise Analysis

In addition to the common-mode-to-differential conversion, noise is also a primary
limitation on the performance of the sensor readout circuitry. The presence of thermal noise,
if the noise floor is greater than the detection signal around the lower detection limit, will
decrease the dynamic range. Two primary sources of noise are involved in the general
block diagram of the biosensor in Figure 1.; the total noise generated by the electrodes
and differential sensing unit and noise generated by the amplification block. Quantization
noise and sampling clock jitter are also issues, but the quantization noise effect is negligible
if the utilized ADC resolution is higher than the required response target resolution. The
clock jitter gains more importance at high sampling rates (hundreds of MS/s or GS/s)
for higher frequency excitations; this effect will be considered and discussed separately
in sub-Section 2.2.2.
Biosensors 2023, 13, 77 7 of 25
Biosensors 2023, 13, x FOR PEER REVIEW Given the small magnitude of the electrode response on the order of several microvolts 7 of 24
at LSB of detection, the noise of proceeding stages plays a vital role in determining the
minimum expected SNR. The typical input referred voltage noise density for differential
precision amplifiers is several √nV . If a gain of μV
70–80 dB is targeted utilizing the multistage
Hz
amplifiers will be on the order of several
amplification, the total output noise voltage√Hz . Moreover,
density the noise
contributed by theassociated
amplifierswith willthe
be
peripheral resistors and noise
µV from the transducer itself will add to
on the order of several √ . Moreover, the noise associated with the peripheral resistors the total output noise
Hz
density.
and noiseTherefore,
from the considering wide-band
transducer itself will addamplification
to the total stages
outputwithout filtering,
noise density. the min-
Therefore,
imum SNR value before ADC could obtain even less than 0 dB.
considering wide-band amplification stages without filtering, the minimum SNR value However, the amount of
in-band noise can be significantly reduced if bandpass filtering is
before ADC could obtain even less than 0 dB. However, the amount of in-band noise can performed. The quanti-
tative
be analysis of
significantly the expected
reduced noisefiltering
if bandpass level presented
is performed.in the following
The determines
quantitative analysisthe of
method to calculate the required reduction in noise and effective
the expected noise level presented in the following determines the method to calculatenoise bandwidth (ENB)
the
of the filter.
required reduction in noise and effective noise bandwidth (ENB) of the filter.
The noise model
The noise model ofof the
the electrodes
electrodes andand the
the equivalent
equivalent circuitry
circuitry with
with noise
noise sources
sources forfor
the differential sensing unit and amplification stage are required for the
the differential sensing unit and amplification stage are required for the noise analysis. The noise analysis.
The differential
differential bridgebridge
modelmodel utilizing
utilizing capacitive
capacitive transducers
transducers presented
presented inis[20]
in [20] is adopted
adopted here
as the differential sensing unit. The noise model for the differential sensing unitunit
here as the differential sensing unit. The noise model for the differential sensing withwith
the
the noise
noise sources
sources included
included for noise
for noise analysis
analysis is shown
is shown in Figure
in Figure 3a. 3a.
Utilizednoise
Figure 3. Utilized noisemodels
modelsforforthe
thetwo
twoprimary
primary noise
noise contributing
contributing blocks;
blocks; (a) (a)
thethe bridge
bridge as
as the
differential
the sensing
differential unit,unit,
sensing withwith
the series RC model
the series RC modelfor the
for capacitive transducers
the capacitive and and
transducers balancing net-
balancing
works, all the
networks, noise
all the sources
noise in the
sources in bridge are thermal
the bridge typetype
are thermal (b) differential amplification
(b) differential stagestage
amplification with
the associated
with thermal
the associated noise
thermal sources
noise forfor
sources thethe
peripheral
peripheral resistors,
resistors,and
andinput-referred
input-referred current and
current and
voltage noise sources of the Opamp.
voltage noise sources of the Opamp.
In Figure 3a, R𝑅b1 𝑏1,, 𝐶 b1,, 𝑅

C𝑏1 b2 and
R𝑏2 and C 𝐶𝑏2 arethe
b2 are theRC-balancing
RC-balancing networksnetworks made made with
with arrays
of digitally controlled capacitors and resistors to balance the bridge for the AC AC signal
signal path.
path.
𝑅𝑑d resistors
R resistorsareareDC-balancing
DC-balancingresistiveresistivepathspathsthat
thatprovide
provideaa stable
stable DC
DC bias
bias to the working
electrodes and
electrodes andequal
equalDC DCvoltage
voltage at V 𝑉𝑎 and
ata and 𝑉𝑏 . The
Vb . The solution–electrode
solution–electrode interface
interface in thisinstudy
this
study
is is modeled
modeled withRC,
with series series
theRC, the resistive
resistive parts (Rparts
el1 and (𝑅R𝑒𝑙1
el2 )and 𝑅𝑒𝑙2 ) are modeling
are modeling solution
solution resistance
resistance
and and the capacitive
the capacitive parts (Cel1 partsand C(𝐶 ) are
el2𝑒𝑙1 the𝐶interface
and 𝑒𝑙2 ) are the interface capacitance.
capacitance. Gesteland etGesteland
al. show
et al.
in [24]show
that in
the[24] thatofthe
noise noise microelectrode
a metal of a metal microelectrode
can be modeled can beasmodeled
the thermalas the thermal
noise of a
resistance
noise of a in a narrowin
resistance band of frequency,
a narrow band ofwhere the corresponding
frequency, resistance is theresistance
where the corresponding real part
of
is the electrode–solution interface impedance.
real part of the electrode–solution Therefore,
interface the associated
impedance. Therefore,noise sources
the with
associated
the electrodes and balancing resistors in the differential sensing
noise sources with the electrodes and balancing resistors in the differential sensing unit unit is in series with the
is
corresponding
in series with the resistances in
corresponding
√ Figure 3a and
resistances areinall
√ representing
Figure 3a and thermal
are all noise model
representing for the
thermal
resistance,
noise model VnRthe
i.e.,for = resistance,
4kTR, with i.e.,units
𝑉𝑛𝑅of=V/√4𝑘𝑇𝑅, . k is the
Hz with Boltzmann’s
units . 𝑘 is theTBoltz-
of V/√Hz constant, is the
absolute temperature
mann’s constant, 𝑇 isandtheRabsolute
is the corresponding
temperature and resistance.
𝑅 is the Forcorresponding
a perfectly matched and
resistance.
balanced bridge case, with
For a perfectly matched andel1balanced C = C el2 = C
bridge
b1 = Ccase,
b2 = C , R
with el1
X = R
𝐶𝑒𝑙1 = el2 = R = R
𝐶𝑒𝑙2 = 𝐶b1𝑏1 = 𝐶b2 =
𝑏2 = 𝐶𝑋
R X,,
𝑅𝑒𝑙1 = 𝑅𝑒𝑙2 = 𝑅𝑏1 = 𝑅𝑏2 = 𝑅𝑋 , the noise contribution of the differential sensing unit can be
derived based on the model in Figure 3a as follows:
2 2 2 2 2 2
2 2 2 𝑅𝑑 𝐶𝑋 𝜔𝑖𝑛 2 1+𝑅𝑋 𝐶𝑋 𝜔𝑖𝑛 2
𝐸𝑛,𝐵𝑟𝑖𝑑𝑔𝑒 = 𝑉𝑎,𝑛 + 𝑉𝑏,𝑛 = 2 𝜔2 𝑉𝑛𝑅𝑋 + 2 𝜔2 𝑉𝑛𝑅𝑑 , (7)
1+(𝑅𝑋 +𝑅𝑑 )2 𝐶𝑋 𝑖𝑛 1+(𝑅𝑋 +𝑅𝑑 )2 𝐶𝑋 𝑖𝑛
2
where 𝐸𝑛,𝐵𝑟𝑖𝑑𝑔𝑒 , is the output referred noise power density of the bridge, and 𝜔𝑖𝑛 = 2𝜋𝑓𝑖𝑛 ,
is the angular frequency of the operation. The electrodes makeup and solution
Biosensors 2023, 13, 77 8 of 25
the noise contribution of the differential sensing unit can be derived based on the model in
Figure 3a as follows:
2 2 2 R2d CX
2 ω2
in 2 1 + R2X CX
2 ω2
in 2
En,Bridge = Va,n + Vb,n = VnR + VnR , (7)
1 + ( RX + R d ) 2 CX
2 ω2
in
X
1 + ( RX + R d ) 2 CX
2 ω2
in
d
2
where En,Bridge , is the output referred noise power density of the bridge, and ωin = 2π f in ,
is the angular frequency of the operation. The electrodes makeup and solution conduc-
tivity for capacitive transducers are set so that at the frequency of operation, the reactive
part of the interface impedance is dominant, Rel < |1/jCel ωin | → Rel Cel ωin < 1 [1]. The
value of the DC-balancing resistors, Rd , are much larger than the magnitude of the trans-
ducer impedance not to load the functional electrode impedance and decrease sensitivity,
therefore, knowing that R2el Cel
2 ω 2 1 leads to R 2 C2 ω 2 1 [20].
in d el in
Based on the abovementioned considerations, (7) can be simplified as follows:
2 R2d 2 1 2
En,Bridge ≈ 2
VnR + VnR , (8)
( R X + Rd ) X
( R X + R d )2 d
The noise model for a fully differential Opamp with the associated feedback resis-
tors [25,26] is shown in Figure 3b. Again, the resistors have the thermal noise voltage
model in series, and Ein , Iin+ and Iin− are the Opamp input referred voltage and current
noise sources, respectively. The power density
of the
Opamp noise sources
havethe units
f enc f inc
of V2 /Hz and are defined as; Ein
2 = e2 1 +
w f
2
, Iin 2 2
− = Iin + = iw 1 + f . e2w and
i i i
i2w , are the opamp input referred voltage and current white noise powers. f enc and f inc ,
are the voltage and current noise power density corner frequencies. The output referred
2
noise power density of each amplification and filtering stage is En,Stage , For the case of a
i
noiseless excitation source, the total root mean square (rms) noise voltage at the output of
the amplification and filtering board is obtained from (9).
s
Z fH
En,Total = ( G2N1 G2N2 G2N3 En,Bridge
2 + G2N2 G2N3 En,Stage
2
1
+ G2N3 En,Stage
2
2
2
+ En,Stage 3
)d f , (9)
fL
where En,Total is the total rms noise voltage at the board output. Here, it is also assumed
that the output common mode voltage is set externally and adequately filtered. The output
2
referred noise power density of the bridge, En,Bridge , and the output referred noise power
2
density of each amplification and filtering stage, En,Stage , is multiplied by the square of
i
RF i
the noise gain, GNi = 1 + R Ii of the proceeding stages, then summed and integrated
over the bandwidth of the interface (lower, f L , to higher, f H , 3 dB cut off frequency). The
output referred noise power density of each differential amplification and filtering stage is
obtained from (10).
R Fi 2 2

2 2 2 2 2 2 2
En,Stage = G E
Ni ini + R I
Fi ini − + Iini + + 2 VnRF + 2 VnR , (10)
i
i RI i Ii
2 2 fH − fL 1 f enc
where VnR X
= 4kTR X , and VnR = 4kTRd . Assuming
d fL < 1, 2
f in
1, fL 1 and
f inc
fL 1, yields (11). For a 4th-order bandpass filter (consisting of two stages in this design)
ENB = 1.025( f H − f L ) [25,27] Note that if f enc > f L and f inc > f L , the effect of flicker
noise must be included, as well. Based on (11), the ENB of the required overall filtering
to obtain 8-bit resolution at a given expected SNR can be determined using the typical
application values for the bridge components.
Biosensors 2023, 13, 77 9 of 25
v
u 2 2
G2N1 G2N2 G2N3 (Vn2R R R+dRX + Vn2R R +1 RX )+
u
u
d
u X
R D2 d
F
u
√ u 2GN2 GN3 VnR + 2GN2 GN3 R I 1 Vn2R +
2 2 2 2 2
F1 1
En,Total ≈ ENB × u I1
(11)
u

ue2 G2 G2 G2 + G2 G2 + G2 + i2 2G2 G2 R2 + 2G2 R2 + 2R2 +
u w N1 N2 N3 N2 N3 N w N2 N3 F1 N F F3
u R 32 R 23 2
F F
2G2N3 Vn2R + 2G2N3 R I 22 Vn2R + 2Vn2R + 2 R I 33 Vn2R
t
F2 I2 F3 I3
Thus, the minimum expected SNR is computed by knowing the minimum expected
rms differentially sensed and amplified signal with a defined resolution and the total rms
noise signal after filtering. The required ENB can be adjusted accordingly.
 
Av,di f f · (Va − Vb ) LSB,rms
SNRmin = 20 × log , (12)
En,Total
where Av,di f f · (Va − Vb ) LSB,rms is the minimum expected rms differentially sensed and
amplified signal for a given resolution. In practice, however, for low-frequency (<10 kHz)
measurements, the effect of flicker noise cannot be neglected entirely, and other sources of
non-ideality, such as the transducer functional layer instability and environmental noise,
may also add to the estimated total output referred rms noise, and even lower ENB might
be required. There are also practical limitations on realizing filter bandwidth on the order of
10s of hertz. The minimal bandwidth leads to a longer response settling time for step-type
input variations. Sometimes the dynamics of analyte binding are fast, and the readout
interface should be able to follow the relatively fast variations in the response signal.
Therefore, the bandwidth should be set considering the amount of allowable noise and
the fast settling requirement. Although a higher filter bandwidth leads to worse SNR,
proper further digital signal processing techniques with suitable sine-fitting algorithms can
effectively act as an additional filter and even extract the signal information buried in the
noise, and this is described next.
2.2. Real-Time Digitization and Fitting

For any biosensor’s deployment, the transducer’s characteristic change should ulti-
mately be quantified. In general, the response is an electric DC or AC signal that needs
to be acquired in real-time for field deployment. The complex impedance of the solution–
electrode interface can be measured in real-time by extracting the amplitude and phase
data from the amplified transducer’s response signal. While amplitude information can
be obtained using the response signal, the phase should be measured differentially with
respect to some reference signal. In the two-channel data acquisition system, shown in
Figure 4, the ADC alternatively samples the excitation voltage (i.e., the source signal) as a
reference for differential phase measurement and the output of the amplification/filtering
block (i.e., the response signal). To automate this, the digital system must be able to ex-
tract the amplitude and differential phase quantities from the digitized signal using a
specific algorithm and in real-time for field deployment. Simultaneous data acquisition and
real-time processing become feasible using ring buffers without sacrificing memory. The
ADC samples the source and response signals and writes the data into the designated ring
buffer. A sine fitting algorithm implemented within the microcontroller reads the samples
consecutively and applies the fitting algorithm to a specific number of samples as a batch.
block (i.e., the response signal). To automate this, the digital system must be able to extract
the amplitude and differential phase quantities from the digitized signal using a specific
algorithm and in real-time for field deployment. Simultaneous data acquisition and real-
time processing become feasible using ring buffers without sacrificing memory. The ADC
samples the source and response signals and writes the data into the designated ring
Biosensors 2023, 13, 77 10 of 25
buffer. A sine fitting algorithm implemented within the microcontroller reads the samples
consecutively and applies the fitting algorithm to a specific number of samples as a batch.
Digital Feedback
Signal Amplification
Transducer
CE
Reference and Filtering

VAC Network Va MCU
WE
VDC
Transducer
WE Vb
Reference Data Acquisition
Network /Fitting
CE
Digital Feedback
Microcontroller
Feedback
Digital
Source
Ring
ADC Buffer
Response
MUX Balancing
Ring Sine
Feedback
Fitting
Digital
Buffer
Data Acquisition and Fitting
Figure 4. A single microcontroller is utilized to both digitize the source and response signals. The
built-in ADC samples the source and response signals alternatively. The samples are stored in ring
buffers, and MCU computes the amplitude and differential phase information. The results are stored
in a ring buffer for external communication.
The fitting results containing the amplitude and differential phase information are
written in another ring buffer. The MCU uses the results for balancing-related computations
or external communication via UART. The key parameters for the data acquisition and
fitting unit design are the ring buffer length, ADC sampling rate, type of the sine fitting
algorithm, and the number of samples required for each round of fitting, considering the
expected signal and noise levels. Additionally, as the processing occurs in real-time using a
single MCU, it requires a computationally efficient code and algorithm. The design process
is explained in the following.
Sine-fitting algorithms are traditionally seen in ADC testing and characterization [28]
and impedance/frequency response measurements [29].
Researchers have recently recognized the utility of sine-fitting algorithms for the
real-time processing of the transducer output voltage [29,30]. There are many different
approaches to sine-fitting, each with varying degrees of suitability in low-cost, real-time
sensing applications. To design an accurate real-time sine fitting, the designer must set many
parameters such as sampling rate, SNR, ADC resolution, record length, etc. Unfortunately,
there is no general study of important factors for sine fitting of biosensor response in real-
time; thus, the designer must study these tradeoffs for each design. This section will present
a detailed analysis of the various tradeoffs for the abovementioned design parameters
concerning sensing requirements. The results will help the designer pick the proper sine
fitting algorithm based on the available budget, expected noise floor, required dynamic
range, and accuracy with specific DSP hardware capabilities.
In general, sine-fitting algorithms can be classified as either iterative or non-iterative.
Although providing better accuracy in some applications, iterative algorithms, such as the
IEEE standard 4-parameter sine fitting [28], are not the best candidates for real-time imple-
mentation. By nature, the convergence of these algorithms might require multiple iterations,
and data storage requires additional memory usage. Non-iterative algorithms, on the other
hand, offer better real-time solutions, considering their relatively more straightforward
implementation. However, the accuracy of a non-iterative approach with small and noisy
signals and compatibility with low-cost general-purpose microcontroller implementation
needs to be considered for real-time and field deployable applications.
Biosensors 2023, 13, 77 11 of 25
The non-iterative sine parameter extraction algorithm, IEEE standard 3-parameter

sine fit (3PSF) [31], is reviewed for the design requirements in this paper.
The 3-parameter sine parameter extraction is based on the assumption that the ex-
citation source frequency ( f in ) is known, and only the amplitude, initial phase, and DC
offset of each singular channel are estimated. A brief theoretical review of the 3PSF is
available in [31–33]. The real-time implementation of 3PSF in a low-cost microcontroller
can be greatly simplified by using coherent sampling; therefore, simplified 3PSF with
coherent sampling is considered for performance comparison here. Consider a sequence of
N samples (k = 0, 1, . . . N − 1) of a sine wave represented as follows:

f in
y[k] = A cos 2π k + ϕ + DC. (13)
fs
where f s is the sampling rate for both channels (source and response), sine wave frequency
f in , amplitude A, initial phase ϕ, and offset DC. The source and response signals are
derived from the same generator; therefore, they share the same frequency f in . If the ratio
of f in / f s is known, the 3 parameters A, ϕ, and DC can be estimated for each channel,
in the least squares sense, using 3PFS, as shown in [31]. The real-time implementation
of 3PSF in a low-cost microcontroller can be significantly simplified by using coherent
sampling; therefore, simplified 3PSF with coherent sampling is considered for performance
comparison here.
If the estimated signal, ŷ[k] is expressed as follows:

f f
ŷ[k ] = Âc cos 2π in k + Âs sin 2π in k + DCˆ (14)
fs fs
With some simple algebra using the coefficients Âc and Âs for each channel, the
amplitude and phase of y are estimated as in [31].
q
Â = Â2c + Â2s (15)
!
−1 − Âs
ϕ̂ = tan . (16)
Âc
where Â and ϕ̂ are the estimated amplitude and initial phase. Coherent sampling is achieved
when f in / f s = M/N, where M and N are relatively prime integers and represent the total
number of input periods in the record and the total record length, respectively. Under the
assumption of coherent sampling, Âc , Âs and DC ˆ are defined as follows:
 

Âc
 2/N ∑kN=−01 y[k] cos 2π M
N k
 
 Âs  =  2/N N −1 y[k ] sin 2π M k . (17)
∑ k =0 N
ˆ
 
DC
1/N ∑kN=−01 y[k]
For a known input frequency, the given ratio of f in / f s and the ratio of M/N remains
fixed. Hence, a lookup table rather
than a function can be used to compute the sinusoidal
values, cos(2π M M
N k) and sin 2π N k , dramatically reducing the required processing time.
A primary concern, however, remains the effect of uncertainty in the f in / f s ratio, as
well as jitter in the sampling clock, in the presence of very low SNR signals. Therefore,
the effects of the oversampling ratio of the ADC and the SNR of the signal, as well as
clock jitter and computational resource requirements, will be studied before algorithm
implementation. The principal investigated performance metrics in the analysis and
comparisons are the estimated percent amplitude error (e A ), and percent differential phase
(ϕdi f f = ϕresponse − ϕsource ) error (e ϕdi f f ).
Biosensors 2023, 13, 77 12 of 25
2.2.1. Additive White Gaussian Noise

To see the effect of noise on the resolution of the sensor, let us assume the noise, nk , is
additive white Gaussian noise. For a noisy, coherently sampled signal,

f in
y noisy [k] = A cos 2π k + ϕ + DC + nk , (18)
fs
the estimation parameters in (17) are independent and unbiased in the presence
of white Gaussian
noise. In this case, the expected values of the estimation parameters are: E Âc = A
ec = A cos ϕ,
ˆ

E Âs = As = A sin ϕ, and E DC = DC = DC. The covariance matrix, Cnoise , of the 3PSF
e g
with coherent sampling is [32]:
2 2 
N σn 0 0
2 2
Cnoise =  0 N σn 0 . (19)
1 2
0 0 N σn
where σn2 is the variance in the additive white Gaussian noise. The estimated amplitude
parameters, Âc and Âs , in the presence of noise, are statistically analyzed and defined
in [32,33], with their expected values A
ec and Aes , and equal variance 2σn2 /N. The amplitude
and initial phase Â and ϕ̂, are functions of the statistically defined random variables
Âc
and Âs . The mean and variance of a function of two random variables f Âc , Âs , can
be approximately derived based on a Taylor series expansion of the function about the
expected values of the associated random variables A ec , A
es , as shown in [32,34]:
2 2 2
es + 1 ∂ f C11 + 2 ∂ f C12 + ∂ f C22 ,

E f Âc , Âs ≈ f Aec , A (20)
2 ∂ Âc 2 ∂ Âc ∂ Âs ∂ Âs 2
∂f 2 ∂f 2

∂f ∂f
var f Âc , Âs ≈ C11 + 2 C12 + C22 , (21)
∂ Âc ∂ Âc ∂ Âs ∂ Âs

where E f Âc , Âs and var f Âc , Âs are the mean and variance of the function f Âc , Âs .
Using (19)–(21), the mean and variance of the estimated amplitude and initial phase can be
derived as a function of SNR:
2
e + 1 σn ≈ A 1
, var Â ≈ σn2
2

E Â ≈ A e 1+ (22)
N Ae 2N ·SNR N
1
E[ ϕ̂] ≈ ϕ
e, var ( ϕ̂) ≈ (23)
N ·SNR

where E Â , var Â and E[ ϕ̂], var [ ϕ̂] are the mean and variance of the estimated amplitude
and initial phase, respectively. Based on (22), the amplitude estimation is biased with noise
present. When alternate sampling is used, as shown in Figure 4, the initial phase of the
response signal is measured with reference to the source signal, therefore:
h i
var ϕ̂di f f = var ϕ̂response + var [ ϕ̂source ] − 2cov ϕ̂response , ϕ̂source , (24)
h i
where var ϕ̂di f f is the variance of the differential phase, and var ϕ̂response and var [ ϕ̂source ] are
the variance of the initial phase estimation of response and source signals, respectively. When
the source and response signals are corrupted by additive noise, the initial phase estimations
are independent (i.e., cov ϕ̂response , ϕ̂source = 0). From (23) and (24), if the same record length is
assumed for both channels (N), the variance of the differential phase can be derived as:

h i 1 1 1
var ϕ̂di f f ≈ + (25)
N SNRresponse SNRsource
Biosensors 2023, 13, 77 13 of 25
The 3PSF algorithm can effectively reduce the effect of noise on the acquired data
based on (22)–(25). If the ratio of M/N is chosen considering the actual SNR at the lower
limits of detection, better resolution can be achieved.
2.2.2. Sampling Clock Jitter

Sampling clock jitter causes uncertainty in 3PFS estimation results. The uncertainty
caused by sampling clock jitter can be modeled as a normally distributed random variable
αk with zero mean and standard deviation equal to σα [35]. A coherently sampled signal
with jitter is modeled as follows:
y jitter [k] = A cos(2π f in (tk + αk ) + ϕ) + DC = A cos(2π f in tk + θk + ϕ) + DC, (26)
where θk is a normally distributed random variable with zero mean and standard deviation
2π f in σα = σ. In the presence of jitter, the 3PSF is no longer an unbiased estimator for Âc
and Âs , the expected values for the three parameters are as follows:
− σ2
E Âc = Ae 2 cos ϕ, (27)
− σ2
E Âs = − Ae 2 sin ϕ, (28)
ˆ = DC,

E DC (29)
 
A2 2 2
1 − e−σ 1 − 21 e−σ cos 2ϕ 0 0
N 
A2 2 2
Cjitter =  0 1 − e−σ 1 + 21 e−σ cos 2ϕ 0 , (30)
 
 N 
A2 2
0 0 N 1 − 21 e−σ
If the initial phase, ϕ, is assumed to be constant, the covariance matrix of the estimator
in the presence of jitter, Cjitter , is derived in (30). The approximate mean and variance
of the amplitude and initial phase estimation with jitter present, using (27) and (28), are
as follows: σ2
− σ2 A 2 1 − σ2
1 − e−σ

E Â ≈ Ae 2 + e2 + e 2 , (31)
2N 2
A2
2 1 − σ2
1 − e−σ

var Â ≈ 1− e 2 , (32)
N 2
1 2
2 1
E[ ϕ̂] ≈ ϕe , var ( ϕ̂) ≈ 1 − e−σ eσ + , (33)
N 2
In the presence of jitter, the estimations
for the source and output initial phases are
independent, i.e., cov ϕ̂response , ϕ̂source = 0. The differential phase variance is derived
using (24) and (33):

h i 1 −σsource 2

σsource 2 1
−σresponse 2

σresponse 2 1
var ϕ̂di f f ≈ 1−e (e + )+ 1−e (e + ) , (34)
N 2 2
If we examine the mean error in the estimated amplitude using (31) from above:

E Â − A − σ2
lim ε A = = e 2 − 1, (35)
N →∞ A
where ε A is the amplitude estimation relative error. Although it is seen in (35) that the
amplitude mean error will never reach zero even with the largest number of samples for
3PSF, which is also claimed in [35], with an optimized large record length 8-bit detection
resolution for both amplitude and phase is achievable even at non-realistically high jitter
standard deviation of around π radians.
Biosensors 2023, 13, 77 14 of 25
Based on the obtained results, 3PSF can maintain the target 8-bit resolution at jitter
standard deviations, even close to 2π radians, by controlling the record length.
It is worth mentioning that the obtained result for this analysis depends on the source
and sampling frequency. The maximum allowable jitter using an ADC with a resolution
of R bits, and a sine wave input with an amplitude equal to the ADC full-scale and a
frequency of f in , to have a jitter-induced error of less than half LSB is inversely proportional
to (2π f in 2R ) [36]. The maximum allowable jitter, therefore, grows smaller if the sine wave
has an amplitude smaller than full-scale, higher frequency, and with higher resolution
for the ADC. For example, jitter considerations gain more importance for sensors with
hundreds of MHz or GHz level excitation frequencies or when the response signal is not
sufficiently amplified to the ADC’s full-scale range at lower detection limits.
2.2.3. Non-Coherency
A fundamental assumption while simplifying the implementation of the real-time
3PSF algorithm is that the data belong to a coherently sampled sine wave. However,
depending on the source sine wave generator’s accuracy level, the desired source frequency
may deviate from its actual value. The result will be that the record will not contain exactly
M cycles of the input signal, and the readily
hard-coded
lookup table for computing the
sinusoidal values, cos(2π M M
N k) and sin 2π N k . will no longer represent correct samples,
leading to errors in the estimated amplitude and initial phase. The effect of a shift in the
source frequency, ∆ f , can be modeled by assuming a shift in the f in such that [32]:
( Q·δ) f s
∆ f = f − f in = , (36)
N
where Q is the integer part, and δ is the fractional part of the residue. The result is a shift in
the number of periods being sampled. Now, if the cycles in the presence of shift in f in , are
M0 = M + Q·δ, the actual waveform, can be expressed as follows:
M + Q·δ

ydev [k] = A cos 2π k+ϕ , (37)
N
where ydev [k] are the samples of the waveform, including a frequency deviation. Using (15)
and (17), the amplitude is estimated as follows:
 2 1/2
2
M + Q·δ

2 N −1 M
Â( Q·δ) = {  + 2/N ∑ k =0
A cos 2π k + ϕ sin 2π k } (38)
N ∑kN=−01 A cos 2π M+NQ·δ k + ϕ cos 2π M N N
N k
which can be further simplified as a function of ∆ f to be:
∆f ∆f ∆f ∆f ∆ f 1/2

Â(∆ f ) = A[sin c2 N + 2M + sin c2 N + 2sin c N + 2M sin c N cos 2ϕ + 2πN ] , (39)
fs fs fs fs fs
The initial phase estimation, as a function of Q·δ, can be derived using (16) and (17):

−2/N ∑kN=−01 A cos 2π M+NQ·δ k + ϕ sin 2π M

N k
ϕ̂( Q·δ) = tan−1   (40)
N −1 M+ Q·δ M
2/N ∑k=0 A cos 2π N k + ϕ cos 2π N k
which can be further simplified and written as a function of frequency deviation, ∆ f :


tan π f in

∆f

fs
ϕ̂(∆ f ) = tan−1 tan ϕ + πN , (41)
∆

f s tan π + f in
f
fs
Biosensors 2023, 13, 77 15 of 25
With the realistic assumption that the shift in the input frequency is much smaller than
the sampling frequency (∆ f f S ), (39) reveals that the amplitude estimation accuracy
is highly dependent on the record length N. The larger the record length, the greater the
amplitude estimation error for a given ∆ f . On the other hand, if ∆ f f S , (41) simplifies
as follows:
∆f
ϕ̂(∆ f ) ≈ ϕ + πN , (42)
fs
(42) indicates that, with a shift in the input frequency, regardless of the size of the
sample record, there will be a linear phase error associated with the phase estimation. This
error can be effectively mitigated if the output signal phase is measured with respect to a
signal that shares the same frequency generator with the response.
2.2.4. Real-Time Microcontroller Implementation Considerations

For real-time implementation, any algorithm must perform mathematical operations
with the individually taken samples from each channel, store the results in specific variables,
and regularly update them with each incoming sample. The implementation cost is,
therefore, the number of mathematical operations, functions, lookup tables, and the number
of memory positions to hold the variables per fixed amount of data within each recording.
One can estimate the execution time based on the number of clock cycles required by a
specific MCU for each math operation. For the simplified 3PSF with a known excitation
frequency, the values of sin(2π f in tk ) and cos(2π f in tk ) can be pre-computed using a singular
lookup table. For comparison, the non-iterative ellipse fit algorithm [37] requires additional
matrix operations other than the trigonometric functions and square root calculations that
could be implemented with lookup tables. The real-time 3PSF is efficiently implemented
within an MCU, with the computational cost as low as six variables, four multiplications,
six additions, and one lookup table.
3. Results
Provided the theoretical analysis of the critical factors limiting the performance of a
fully differential amplification/filtering and data acquisition/processing board, a readout
interface is designed, fabricated, and tested for target 8-bit impedance sensing resolu-
tion. Board implementation and performance verification procedures are discussed in
this section.
3.1. Differential Amplification and Filtering Implementation

The Block diagram and the implemented boards of the amplification and filtering
units interfacing a differential impedance sensing bridge are shown in Figure 5. A custom
differential impedance sensing bridge with fixed series RC impedances (Zel1 and Zel2 )
mimicking the electrode–solution interface impedance and balancing networks (Zb1 and
Zb2 ) composed of a digitally tunable resistor and capacitor array [20] is interfaced with
the designed readout for performance verification. The DC biasing resistors (Rd = 50 kΩ)
are not shown on the schematic for simplicity but are included on the board, as shown in
Figure 3a. The first fully differential amplification stage is fabricated on the same board as
the differential sensing unit to minimize the interferences caused by wiring and external
connections. Details about the part selection and component values using the proposed
design methodology are explained in the next section.
𝑍𝑏2 ) composed of a digitally tunable resistor and capacitor array [20] is interfaced with the
designed readout for performance verification. The DC biasing resistors (𝑅𝑑 = 50 𝑘Ω) are
not shown on the schematic for simplicity but are included on the board, as shown in
Figure 3a. The first fully differential amplification stage is fabricated on the same board as
the differential sensing unit to minimize the interferences caused by wiring and external
16 of 25
connections. Details about the part selection and component values using the proposed
design methodology are explained in the next section.
Figure
Figure 5. (a) The
5. (a) Theimplemented
implementedamplification
amplification and
and filtering
filtering board
board interfaced
interfaced withwith the differential
the differential im-
pedance sensing
impedance unit:
sensing (b)(b)
unit: TheThe
block diagram
block of the
diagram differential
of the impedance
differential impedancesensing unitunit
sensing and and
the de-
the
veloped readout.
developed readout.The
Thefirst
firstamplification
amplificationstage
stageisisplaced
placedononthe
thesame
sameboard
boardasasthe
thesensing
sensing unit
unit to
to
avoidinterferences
avoid interferenceswith
with the
the small
small sensing
sensing response
response signal
signal caused
caused byby external
external connections.
connections.
3.1.1.
3.1.1. Effect
Effect of
of Component
Component Matching
Matching and Tolerances
Tolerances
The
Thethree-stage
three-stageamplification
amplification is is
configured
configured to provide
to providea total differential
a total gain of
differential 70 dB,
gain of
to
70amplify the sensing
𝑑𝐵, to amplify unit’s unit’s
the sensing full-scale output
full-scale to A FS
output to of
𝐴𝐹𝑆theofADC.
the ADC.The differential
The differential pre-
cision amplifier,
precision Analog
amplifier, Analog Devices
DevicesInc., Wilmington,
Inc., Wilmington,MA, MA,USA,USA,LTC6363
LTC6363isispicked
picked as the the
critical component of the amplification and filtering
critical component of the amplification and filtering unit. unit. The typical CMRR
𝐶𝑀𝑅𝑅𝑂𝑝𝑎𝑚𝑝 ,
Opamp, for for
LTC6363
LTC6363 is is 110
110dB.
dB. To
To maintain total CMRR
maintain a total 𝐶𝑀𝑅𝑅 of of around
around 100 dB for for the readout,
readout, aacom-com-
mon
mon mode-induced
mode-induced differential
differential gain
gain of around −
of around 30 dB
−30 dB is
is required
required based
based on (2). Figure
on (2). Figure 6a 6a
shows
shows thethe percent
percent degradation
degradation in in CMRR
𝐶𝑀𝑅𝑅Stage
𝑆𝑡𝑎𝑔𝑒
/CMRR
/𝐶𝑀𝑅𝑅 Opamp
𝑂𝑝𝑎𝑚𝑝
for
for various
various CMRR
𝐶𝑀𝑅𝑅 Opamp
𝑂𝑝𝑎𝑚𝑝 , , ver-
ver-
sus
sus the
the percent
percent feedback
feedback matching
matching ratio
ratio obtained
obtained from
from (3). Given aa typical
(3). Given typical value
value of of
CMRR
𝐶𝑀𝑅𝑅𝑂𝑝𝑎𝑚𝑝 = 110 dB, feedback matching ratios better than 0.001%
Opamp = 110 dB, feedback matching ratios better than 0.001% are required to ob- are required to obtain
CMRR
atain a 𝐶𝑀𝑅𝑅 of 50 of
Stage 𝑆𝑡𝑎𝑔𝑒 dB.50AdB. stable common-mode
A stable common-mode voltage is provided
voltage for each
is provided differential
for each differ-
difference amplifier using an on-board voltage regulator, Analog
ential difference amplifier using an on-board voltage regulator, Analog Devices Inc.Devices Inc. LT3021, and
A FS
the
Biosensors 2023, 13, x FOR PEER REVIEW voltage
LT3021, andamount is equal
the voltage to half
amount is of the to
equal ADC
halffull-scale
of the ADC reference voltage
full-scale ( 2 =
reference 0.6ofV).
16
voltage 24
Therefore,
(
𝐴𝐹𝑆 considering (6), adjusting the feedback matching ratio of the first
= 0.6 V). Therefore, considering (6), adjusting the feedback matching ratio of the first amplification
2
stage plays a critical role in the total readout CMRR.
amplification stage plays a critical role in the total readout 𝐶𝑀𝑅𝑅.
Figure6.6.(a)
Figure (a)Percent
Percentdegradation
degradationin inthe
theratio
ratioof ofthe
theCMRR
CMRRof ofaafully
fullydifferential
differentialstage
stage(CMRR
(𝐶𝑀𝑅𝑅Stage )
𝑆𝑡𝑎𝑔𝑒 )
and Opamp typical CMRR
and Opamp typical CMRR (𝐶𝑀𝑅𝑅 (CMRR ) for various CMRR
𝑂𝑝𝑎𝑚𝑝 ) for various 𝐶𝑀𝑅𝑅
Opamp , versus percent change in matching
𝑂𝑝𝑎𝑚𝑝 , versus percent change in match-
Opamp
ratios of peripheral
ing ratios resistors
of peripheral constructing
resistors the feedback
constructing network network
the feedback (∆β/β); (b) Percent
(Δ𝛽/𝛽); (b)resistance
Percent tolerance
resistance
(∆R/R) for(Δ𝑅/𝑅)
tolerance variousfor
differential stage gain, Astage
various differential gain, 𝐴percent
v,dd , versus feedback
𝑣,𝑑𝑑 , versus ratiofeedback
percent mismatchratio
(∆β/β).
mismatch
(Δ𝛽/𝛽).
The plots of the required percent component tolerance for various stage gain, 𝐴𝑣,𝑑𝑑 ,
versus percent feedback ratio mismatch using (5), are shown in Figure 6b. Stages with
higher allocated differential gain are less sensitive to component tolerances to obtain a
given feedback matching ratio.
Biosensors 2023, 13, 77 17 of 25
The plots of the required percent component tolerance for various stage gain, Av,dd ,
versus percent feedback ratio mismatch using (5), are shown in Figure 6b. Stages with
higher allocated differential gain are less sensitive to component tolerances to obtain a
given feedback matching ratio.
The peripheral surface-mount resistors and the capacitors of the differential amplifica-
tion and filtering circuit were measured one by one in every purchased batch (containing
at least 10 with 1% tolerance), and the ones with the closest values to obtain the matching
ratio of better than 0.001% were mounted on the boards. The values of the resistors in
the amplification and filtering stages to achieve a gain distribution of 30 × 10.6 × 10.6 are;
R F1 = 300 kΩ, R I1 = 10 kΩ, R F2 = R F3 = 16 kΩ, R I 2 = R I 3 = 750 Ω, Rbb = Rcc = 10 kΩ.
3.1.2. Noise Considerations and Filtering Unit

√
The √ input referred noise voltage, and current of the LTC6363 are 2.9 nV/ Hz and
0.55 pA/ Hz, respectively. With the known gain distribution and resistance values of the
amplification and filtering stages and typical values for the electrode impedance models,
Rel = 200 Ω, Cel = 100√ nF using (8) the output referred noise of the readout board before
ADC is 75.231 µV/ Hz. For an 8-bit resolution and 1% change in the impedance and
VAC = 50 mVrms the LSB of the differentially amplified sensing signal with the gain of
70 dB, is 3.34 mVrms . Considering the fast response settling requirements, by picking an
ENB of 270 Hz, the expected SNR at the minimum response level obtained from (9) is
7~9 dB. A bandwidth of 265 Hz is chosen for the bandpass filter, and with the available
commercial component’s values for the capacitors Cb = Cc = 47.3 nF, the center frequency
of the filter is expected to be approximately 1.03 kHz.
3.2. Real-Time Signal Acquisition and Sine Fitting

With the derived expressions in Section 2.2, the amplitude and phase estimation errors
are good indicators of better noise immunity with various sample record lengths and
specific accuracy requirements. To evaluate the reasonably achievable dynamic range for
the sensor with the expected level of noise, jitter, and frequency accuracy, an optimized
record length for a known M/N is required. Thus, theoretical derivations are verified with
numerical MATLAB simulations. For the subsequent analysis, we assume two sine signals
(source and response) with an equal frequency of 1 kHz, amplitude of 1 V, the relative
phase difference of 45◦ , and equal DC offset of 0.6 V, that are generated using MATLAB. The
sampling rate, f s , and input frequency, f in , are constant and equal to 55 kS/s and 1 kHz,
respectively. The values M and N, are changed for multiple simulations, providing record
lengths of 256, 512, and 1024 samples with 5, 9, and 19 cycles, respectively, to maintain the
fixed M/N ratio.
The ultimate metric of interest for a biosensor is to achieve a particular resolution.
Therefore, the pointed non-idealities and their effect on 3PSF estimation are also examined
here under the resolution (8-bit) context. The resolution of the sensor output is defined,
considering a full-scale voltage at the input of the ADC, A FS , and the full-scale target phase
difference, ϕ FS . One performance criterion given a target of 8-bit resolution is the estimation
mean error and standard deviation remaining within the ± A FS /28 and ± ϕ FS /28 range.
The estimation errors (mean and standard deviation) and resolution lines are normalized to
full-scale values to demonstrate a more generic reference plot. For each non-ideality effect,
the normalized estimation parameters of interest with their percent mean error and percent
standard deviation are shown, and normalized resolution lines are also drawn on the plots
as an indicator of the best achievable resolution with different levels of additive noise, jitter,
and shift in the excitation frequency.
3.2.1. Additive Noise

The expected minimum SNR value at the output of the implemented amplification
and filtering board is 7~9 dB. The analysis in Section 2.2.1 confirmed that an optimized
record length for 3PSF improves both amplitude and differential phase estimation accuracy.
levels of additive noise, jitter, and shift in the excitation frequency.
3.2.1. Additive Noise

The expected minimum SNR value at the output of the implemented amplification
and filtering board is 7~9 dB. The analysis in Section 2.2.1 confirmed that an optimized
18 of 25
record length for 3PSF improves both amplitude and differential phase estimation accu-
racy. To optimize the record length with the expected SNR, the estimation mean and
standard deviation normalized percent errors obtained from the theoretical derivations
To optimize the record length with the expected SNR, the estimation mean and standard
(22)–(25) are simulated using MATLAB. Figure 7 shows the normalized mean amplitude
deviation normalized percent errors obtained from the theoretical derivations (22)–(25) are
and differential phase estimation error and the associated standard deviation for response
simulated using MATLAB. Figure 7 shows the normalized mean amplitude and differential
SNR ranging from −5 to 30 dB and 1000 simulations at each point. The SNR at the
phase estimation error and the associated standard deviation for response SNR ranging
source is fixed and set to 30 dB for the simulations. Apparently, with a larger record
from −5 to 30 dB and 1000 simulations at each point. The SNR at the source is fixed and set
length, lower uncertainty in the estimation is achievable, even for SNR values less than
to 30 dB for the simulations. Apparently, with a larger record length, lower uncertainty in
0 dB.
the estimation is achievable, even for SNR values less than 0 dB.
3PSF(a)
Figure7.7.3PSF
Figure (a)percent
percentmean
meanamplitude
amplitudeand
and(b)
(b)percent
percentmean
meandifferential
differentialphase
phaseestimation
estimationerrors
errors
with their normalized standard deviations (error bars) vs. SNR. Normalized to full-scale resolution
with their normalized standard deviations (error bars) vs. SNR. Normalized to full-scale resolution
linesdemonstrate
lines demonstratethethebest
bestachievable
achievableresolution
resolutionbybyconsidering
consideringthe
thestandard
standarddeviation
deviation(error
(errorbars)
bars)
withthe
with thedegrading
degradingSNR
SNRandandincreasing
increasingrecord
record length
length (N).
(N).
Theresults
The results in
in Figure
Figure 77 show
showthat
thatthe
the8-bit
8-bittarget resolution
target with
resolution SNR
with values
SNR even
values less
even
than 5 dB is achievable with N = 1024.
less than 5 dB is achievable with 𝑁 = 1024.
•• Oversampling
Oversampling
Numerical analysis with varying sampling rates and low SNRs is also carried out
for 3PSF to confirm if a higher sample rate leads to better estimation performance in a
noisy environment. For the numerical simulations, the oversampling ratios range from
4 to 256, and the input SNR is assumed to range from −5 to 35 dB. The simulations for
the 3PSF produce an average percent error within 1% both for the amplitude ratio and the
differential phase for the 256 samples in the record. The produced estimation results show
that with a lower sampling-to-excitation frequency ratio, 3PSF can deliver reliable results
at very low SNRs, with even four samples taken per period.
Although the greater number of samples within a record generates more accurate
results with worse SNR levels, this might limit how fast the results could be produced at
very low (sub-Hz) excitation frequencies. In [38], it is shown that utilizing 3PSF for a sub-Hz
sensor response can produce an impedance estimation variance of 1% while the record
covers only 11% of the whole period. Therefore, 3PSF is flexible for various excitation
frequencies and signal-to-noise ratios. Consequently, it is concluded that at low SNRs, 3PSF
with coherent sampling provides accurate results without oversampling. Therefore, the
only limiting factor on the sampling rate is set by the real-time processing requirements.
3.2.2. Jitter
Based on derivations (31)–(34) for 3PSF in the presence of jitter and numerical sim-
ulations, the mean of the amplitude and differential phase errors with their normalized
standard deviations for 1000 simulations at each point is plotted in Figure 8. The jitter
standard deviation (σ) varies from 0 to 2π for all the simulations.
only limiting factor on the sampling rate is set by the real-time processing requirements.
3.2.2. Jitter
Based on derivations (31)–(34) for 3PSF in the presence of jitter and numerical simu-
lations, the mean of the amplitude and differential phase errors with their normalized
Biosensors 2023, 13, 77 19 of 25
standard deviations for 1000 simulations at each point is plotted in Figure 8. The jitter
standard deviation (𝜎) varies from 0 to 2𝜋 for all the simulations.
3PSF(a)
Figure8.8.3PSF
Figure (a)percent
percentmean
meanamplitude
amplitudeand
and(b) (b)percent
percentmean
meandifferential
differentialphase
phaseestimation
estimationerrors
errors
withtheir
with theirnormalized
normalizedstandard
standarddeviations
deviations(error
(errorbars)
bars)vs.
vs.jitter
jitterstandard
standarddeviation
deviation(𝜎).
(σ).3PFS
3PFScan
can
maintain 8-bit
maintain 8-bit resolution
resolution with
witheven
evenunrealistic jitter
unrealistic standard
jitter standard deviations closeclose
deviations to 2πtoby2𝜋record length
by record
length
controlcontrol at 55sampling
at 55 kS/s kS/s sampling
rate. rate.
Althoughititisisseen
Although seeninin(35)
(35)that
thatthe
theamplitude
amplitudemean
meanerror
errorwill
willnever
neverreach
reachzero
zerofor
for
3PSF [35], with a record length of 1024, the 8-bit detection resolution for both amplitude
3PSF [35], with a record length of 1024, the 8-bit detection resolution for both amplitude
andphase
and phaseisisachievable
achievableeven
evenatatnon-realistically
non-realisticallyhigh
highjitter
jitterstandard
standarddeviation
deviationofofaround
around
π radians.
π radians.
3.2.3.Shift
3.2.3. ShiftininExcitation
ExcitationFrequency
Frequency
Figure 9a shows the normalized amplitude estimation versus the deviation in the
Figure 9a shows the normalized amplitude estimation versus the deviation in the
source frequency ∆ f at various numbers of acquired cycles (M = 5, 9, 19 and 37) both for
source frequency Δ𝑓 at various numbers of acquired cycles (𝑀 = 5, 9,19 and 37) both for
the theoretical derivation (39) and numerical simulation, with f s = 55 kS/s and f in = 1 kHz.
the theoretical derivation (39) and numerical simulation, with 𝑓𝑠 = 55 kS/s and 𝑓𝑖𝑛 =
For a higher numbers of samples, with approximately equal input frequencies and a
1 kHz. For a higher numbers of samples, with approximately equal input frequencies and
constant sampling rate, the fitting accuracy becomes much more sensitive to uncertainty
a constant sampling rate, the fitting accuracy becomes much more sensitive to uncertainty
in the ratio of f in / f s . Figure 9b shows the zoomed-in plot of the amplitude error with
non-coherency for N = 1024, M = 19. For 8-bit detection resolution, a frequency deviation
of approximately 5 Hz can be tolerated, while if the resolution is relaxed to 4-bit, the safe
frequency deviation is raised to 22 Hz. However, with the typical accuracy level of on-chip
sine wave generators utilizing phase-locked loops (PLL) or direct digital synthesis (DDS)
for discrete implementation, at kHz range, this amount of non-coherency is not a concern
for amplitude estimation.
However, at the MHz or GHz range, the accuracy of the generated sine wave is more
of a limiting factor while picking the record length.
The theoretical (41) and numerically simulated initial phase estimation of the source
and response with non-coherency are shown in Figure 10. A deviation in the source
frequency will result in a linear increase in the initial phase, but the slope of this change,
as derived in (41), is equal for both source and response signals, as they share the same
frequency from a mutual source. Therefore, the resultant differential phase estimation
for the alternate sampling scheme will not be affected by frequency deviation for any
record length.
in the ratio of 𝑓𝑖𝑛 /𝑓𝑠 . Figure 9b shows the zoomed-in plot of the amplitude error with non-
coherency for 𝑁 = 1024, 𝑀 = 19. For 8-bit detection resolution, a frequency deviation of
approximately 5 Hz can be tolerated, while if the resolution is relaxed to 4-bit, the safe
frequency deviation is raised to 22 Hz. However, with the typical accuracy level of on-
chip sine wave generators utilizing phase-locked loops (PLL) or direct digital synthesis
Biosensors 2023, 13, 77 20 of 25
(DDS) for discrete implementation, at kHz range, this amount of non-coherency is not a
concern for amplitude estimation.
Figure 9. 3PSF (a) amplitude estimation for a various number of samples within a record (b)
zoomed-in Mean amplitude percent error for N = 1024, M = 19, and normalized percent detection
resolutions vs. source frequency deviation. The plots also validate that the theoretical derivations
for the non-coherency effect follow the numerical simulations accordingly.
and response with non-coherency are shown in Figure 10. A deviation in the source fre-
quency will result in a linear increase in the initial phase, but the slope of this change, as
derived
Figure 9. in (41),
3PSF
Figure 9. 3PSF
(a)isamplitude
(a) equal forestimation
amplitude both source
estimation
for a and
for a response
various numbersignals,
various number of as
of samples they
samplesshare thea(b)
within a record
within same
recordfre-
zoomed-
(b)
quency
in Mean from a
amplitude mutual
percentsource.
error Therefore,
for N = 1024, Mthe
= resultant
19, and differential
normalized phase
percent estimation
detection
zoomed-in Mean amplitude percent error for N = 1024, M = 19, and normalized percent detection for
resolutions
the
vs. alternate
resolutions vs.sampling
source frequency schemedeviation.
deviation.
source frequency will
The not
plotsbe affected
also
The by that
validate
plots also frequency
the that
validate deviation
theoretical for any
derivations
the theoretical record
for the
derivations
length.
non-coherency effect follow
for the non-coherency effectthe numerical
follow simulations
the numerical accordingly.
simulations accordingly.
and response with non-coherency are shown in Figure 10. A deviation in the source fre-
quency will result in a linear increase in the initial phase, but the slope of this change, as
derived in (41), is equal for both source and response signals, as they share the same fre-
quency from a mutual source. Therefore, the resultant differential phase estimation for
the alternate sampling scheme will not be affected by frequency deviation for any record
length.
ϕsource = 45 ◦ = 90◦ .
10.3PSF
Figure 10.
Figure 3PSFinitial phase
initial estimation
phase vs. source
estimation vs.frequency
source deviation
frequencyfordeviation for, ϕ𝜑response
𝑠𝑜𝑢𝑟𝑐𝑒 = 45°,
The
𝜑 plots obtained for N = 1024, M = 19. When the frequency of the source and response signals is generated
𝑟𝑒𝑠𝑝𝑜𝑛𝑠𝑒 = 90°. The plots obtained for N = 1024, M = 19. When the frequency of the source and
from a mutual source, the linear phase estimation error for both has the same slope.
It should be clarified that, as seen in Figure 9a, the estimated amplitude will reach
zero when the shift in the frequency leads to an integer number of acquired cycles differ-
Q
ence (i.e., δ = 0 or when ∆ f = N f s , see (39)), causing a discontinuity in the initial phase
estimation at the same points.
3.3. Board Characterization

The fabricated amplification and filtering board is characterized by measuring the total
Figure 10. 3PSF initial phase estimation vs. source frequency deviation for 𝜑𝑠𝑜𝑢𝑟𝑐𝑒 = 45°,
differential gain and common-mode-induced differential gain to verify the expected board
𝜑𝑟𝑒𝑠𝑝𝑜𝑛𝑠𝑒 = 90°. The plots obtained for N = 1024, M = 19. When the frequency of the source and
total CMRR. For the differential gain measurement, the Audio Precision 2272 instrument is
used to provide a very small differential sinusoidal input signal with a 0.1 mVrms amplitude.
The magnitude of the differential voltage at the output of the filtering board is measured
using a digital multimeter. The differential phase of the board output with reference to the
source signal is measured by an oscilloscope. The gain magnitude and differential phase
are measured and recorded for multiple frequency points at a range from 0 to 2500 Hz.
Common-mode-induced differential gain is characterized with the same method, except
for a 1 Vrms voltage applied to the inputs of the board.
strument is used to provide a very small differential sinusoidal input signal with a
0.1 mVrms amplitude. The magnitude of the differential voltage at the output of the filter-
ing board is measured using a digital multimeter. The differential phase of the board out-
put with reference to the source signal is measured by an oscilloscope. The gain magni-
Biosensors 2023, 13, 77 tude and differential phase are measured and recorded for multiple frequency points 21 ofat25a
range from 0 to 2500 Hz. Common-mode-induced differential gain is characterized with
the same method, except for a 1 Vrms voltage applied to the inputs of the board.
The measurement characterization graphs in Figure 11a show that the interface board
The measurement characterization graphs in Figure 11a show that the interface board can
can achieve a center frequency at 1.02 kHz, a maximum differential gain of 3330.33 𝑣/
achieve a center frequency at 1.02 kHz, a maximum differential gain of 3330.33 v/v corresponding
𝑣 corresponding to 70.44 dB, and a bandwidth of 265 Hz. The maximum common mode
to 70.44 dB, and a bandwidth of 265 Hz. The maximum common mode induced differential
induced
gain showndifferential gain shown
in the measurement in the
graph measurement
of Figure graph
11b is −25.7 dB, of Figure
which 11ba total
yields is −25.7
CMRR dB,
which yields
of 96.14 dB. a total CMRR of 96.14 dB.
Figure 11. The

Figure11. The amplification
amplification and
and filtering
filtering board
board characterization
characterization for
for differential
differentialgain
gainmagnitude
magnitudeand and
differential
differential phase
phase verifies
verifies aa center
center frequency
frequency ofof 1.02
1.02 kHz,
kHz, bandwidth
bandwidth of of 265
265 Hz,
Hz, and
and aa differential
differential
gain magnitude
gain magnitude of 3330.33v/v
of 3330.33 𝑣/𝑣corresponding
corresponding to 70.44dB;
to 70.44 dB;(a).
(a).Maximum
Maximumcommon commonmodemodeinduced
induced
differentialgain
differential gainmagnitude
magnitudeisisverified
verifiedatat−−25.7 dB, which
25.7 dB, which yields
yields aa total
total CMRR
CMRR of 96.14dB
of 96.14 dB;(b).
(b).
3.4.
3.4. Performance
Performance Verification
Verification
The
The amplificationand
amplification and filtering
filtering board
board is
is connected
connected toto the
the Texas
Texas Instruments
Instruments Inc. MSP-
Inc. MSP-
EXP432P401R launchpad shown in Figure 5a for real-time data acquisition
EXP432P401R launchpad shown in Figure 5a for real-time data acquisition and sine and sine fitting.
The utilized key features of the single microcontroller are 48 MHz master clock rate, 14-bit,
1.2 V differential (two channel) ADC. Considering the acquisition of 1024 samples for each
sine fitting operation satisfies 8-bit detection resolution requirements with 7~9 dB expected
minimum SNR. The sampling rate is adjusted at 55 kS/s to accommodate the needed real-
time processing time. For the real-time fitting operation, each sample is multiplied by the
corresponding sin(2π f in tk ) and cos(2π f in tk ) from a lookup table and results consecutively
added to produce the Âs and Âc for each sine fitting.
Real-Time Operation and Sensitivity

The readout sensitivity is tested using the same block diagram of Figure 5b. The test’s target
is to detect a total 1% fractional capacitance change on one of the bridge impedances with an 8-bit
resolution. A symmetrical bridge is configured by replacing Zel1 and Zel2 with the two fixed
equal series RC impedances of 100 nF and 220 Ω. After balancing the differential bridge using
digitally tunable Zb1 and Zb2 networks, the 8-bit sensitivity is tested by additively changing
the capacitance of the Zb1 with 8 pF, 16 pF, 32 pF, 64 pF, 128 pF, 256 pF, 512 pF, and 1 nF
1nF
values. The test mimics an 8-bit detection resolution for a total 1% ( 100 nF ) fractional change of
the capacitance in the symmetrical differential bridge. The real-time response amplitude and
differential phase with reference to the source signal are demonstrated in Figure 12.
differential bridge using digitally tunable 𝑍𝑏1 and 𝑍𝑏2 networks, the 8-bit sensitivity is
tested by additively changing the capacitance of the 𝑍𝑏1 with
8 pF, 16 pF, 32 pF, 64 pF, 128 pF, 256 pF, 512 pF, and 1 nF values. The test mimics an 8-bit
1nF
detection resolution for a total 1% ( ) fractional change of the capacitance in the sym-
100 nF
metrical differential bridge. The real-time response amplitude and differential phase with
Biosensors 2023, 13, 77 22 of 25
reference to the source signal are demonstrated in Figure 12.
Figure 12. Sensitivity to 1% capacitance change with 8-bit resolution and real-time processing
verification for the implementation of the readout board. The obtained amplitude from the real-
time 3PSF for 8 pF, 16 pF, 32 pF, 64 pF, 128 pF, 256 pF, 512 pF and 1 nF capacitance change at
a symmetrical impedance bridge with a fixed capacitance of 100 nF; (a). the obtained real-time
differential phase and the corresponding amplitude can be used to compute actual capacitance
change for a known bridge impedance model (b).
The results show a distinct amplitude change at each consecutive capacitance

change step compared to the initial balanced state amplitude. The change of amplitude
values compared to the initial state at each binary weighted capacitance change is
4 mV, 6 mV, 16.5 mV, 59 mV, 118 mV, 285 mV, 552 mV,and 1142 mV, correspondingly from
the 8-bit to a 1-bit resolution. The 8-bit, 7-bit, and 6-bit amplitude levels are shown in
an inset zoomed-in plot of Figure 12a after a 10-point moving average. The real-time
differential phase is shown in Figure 12b, together with the corresponding amplitude, can
be used to algebraically compute the exact values of the capacitance change with the known
bridge impedance model [20]. The obtained results in Figure 12 reveal that the readout
is sensitive to a 1% capacitance change with an 8-bit resolution. Moreover, the results
confirm that the developed interface can produce the processed response successfully in
real time. The developed readout board could be interfaced with various custom-designed
differential impedance sensing units with known impedance models, and the given design
procedure could be employed for various precision-demanding applications.
Biosensors 2023, 13, 77 23 of 25
4. Discussion
At the lower limits of detection for impedance biosensors, differential sensing and
single frequency measurement with sufficient amplification and filtering are promising
methods for real-time and field deployable implementation. However, when the response
is still small and comparable to the noise level even after amplification, the resolution of
the biosensor is significantly affected by the noise level. Moreover, the clock jitter will
induce additional noise and degrade the SNR. Real-time digitization and processing with
3PSF, in expected low SNRs (even SNR < 0 dB), can lead to a higher resolution and better
noise-immune operation. The algorithm can be implemented within a single MCU to
process the digitized data in real time with an optimized number of samples per fitting to
achieve a specific target resolution. With the assumption of a known operation frequency
for the implementation of 3PSF, it is proven that a non-ideal shift in the source frequency
and sampling clock jitter will affect the estimated response amplitude at any detection
resolution. However, with typical accuracy levels of sine generation and the fact that
both sampling clock and source signals are driven by a single sinusoidal source in most
cases, jitter and source frequency shift will have a minor effect on the detection resolution
both in terms of amplitude and differential phase at the lower sampling rates (i.e., tens
to hundreds of KS/s). The real-time processing of the data obtained from the biosensor
will eliminate the need for data storage and memory requirements and lead to lower
costs for the overall system. A less complicated data processing algorithm with lower
memory requirements, such as 3PSF, facilitates using the same microcontroller for data
processing and other calibration or balancing of the differential system, e.g., the bridge-
based system [20]. More straightforward data processing and less complicated readout
implementation that are compatible to interface with ultra-sensitive transducers are crucial
for the commercialization of cheaper biosensors.
5. Conclusions
A readout interface board suitable for high-precision impedance measurement, par-
ticularly for biosensing applications, is designed and implemented. The provided design
details are first-hand knowledge for researchers in the field of impedance sensors and
biosensors requiring precise measurement with specific resolution and accuracy. Moreover,
the procedure provided here for developing a real-time data-acquisition unit is a guideline
for making custom-designed, low-cost, and real-time digitization and processing units for
numerous sensitive transducers that are currently being characterized with lab instrumen-
tation. Therefore, utilizing the information provided in this paper for less complicated
and yet accurate real-time readout facilitates the deployment of transducers for various in
situ applications.
Author Contributions: Conceptualization, S.N., N.M.N. and D.J.C.; methodology, S.N., N.M.N.
and D.J.C.; software, S.N., K.M. and N.M.N.; validation, S.N.; formal analysis, S.N.; investigation,
S.N.; resources, S.N., N.M.N. and D.J.C.; data curation, S.N.; writing—original draft preparation,
S.N. and N.M.N.; writing—review and editing, S.N., N.M.N., K.M. and D.J.C.; visualization, S.N.,
K.M. and N.M.N.; supervision, N.M.N. and D.J.C.; project administration, S.N., N.M.N. and D.J.C.;
funding acquisition, N.M.N. and D.J.C. All authors have read and agreed to the published version of
the manuscript.
Funding: This research was funded by U.S. Department of Agriculture under Grant 2015-67021-23128
and the University of Alabama.
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement: Not applicable.
Conflicts of Interest: The authors declare no conflict of interest.
Biosensors 2023, 13, 77 24 of 25
References
1. Daniels, J.S.; Pourmand, N. Label-Free Impedance Biosensors: Opportunities and Challenges. Electroanalysis 2007, 19, 1239–1257.
[CrossRef] [PubMed]
2. Loyprasert, S.; Thavarungkul, P.; Asawatreratanakul, P.; Wongkittisuksa, B.; Limsakul, C.; Kanatharana, P. Label-Free Capacitive
Immunosensor for Microcystin-LR Using Self-Assembled Thiourea Monolayer Incorporated with Ag Nanoparticles on Gold
Electrode. Biosens. Bioelectron. 2008, 24, 78–86. [CrossRef] [PubMed]
3. Katz, E.; Willner, I. Probing Biomolecular Interactions at Conductive and Semiconductive Surfaces by Impedance Spectroscopy:
Routes to Impedimetric Immunosensors, DNA-Sensors, and Enzyme Biosensors. Electroanalysis 2003, 15, 913. [CrossRef]
4. Abdelrasoul, G.N.; Anwar, A.; MacKay, S.; Tamura, M.; Shah, M.A.; Khasa, D.P.; Montgomery, R.R.; Ko, A.I.; Chen, J. DNA
Aptamer-Based Non-Faradaic Impedance Biosensor for Detecting E. Coli. Anal. Chim. Acta 2020, 1107, 135–144. [CrossRef]
5. Alsabbagh, K.; Hornung, T.; Voigt, A.; Sadir, S.; Rajabi, T.; Länge, K. Microfluidic Impedance Biosensor Chips Using Sensing
Layers Based on DNA-Based Self-Assembled Monolayers for Label-Free Detection of Proteins. Biosensors 2021, 11, 80. [CrossRef]
[PubMed]
6. Bar-On, L.; Garlando, U.; Sophocleous, M.; Jog, A.; Motto Ros, P.; Sade, N.; Avni, A.; Shacham-Diamand, Y.; Demarchi, D.
Electrical Modelling of In-Vivo Impedance Spectroscopy of Nicotiana Tabacum Plants. Front. Electron. 2021, 2, 753145. [CrossRef]
7. Heinze, J.; Allen, J.; Bard, F.; Larry, F. Faulkner: Electrochemical Methods—Fundamentals and Applications; Wiley: New York, NY,
USA, 1981; Volume 85, pp. 1085–1086. [CrossRef]
8. Baghbani, R.; Moradi, M.H.; Shadmehr, M.B.; Momayez Sanat, Z. A New Bio-Impedance Forceps Sensor for Measuring Electrical
Conductivity of the Biological Tissues. IEEE Sens. J. 2019, 19, 11721–11731. [CrossRef]
9. Kim, S.; Song, H.; Ahn, H.; Kim, T.; Jung, J.; Cho, S.K.; Shin, D.-M.; Choi, J.; Hwang, Y.-H.; Kim, K. A Review of Advanced
Impedance Biosensors with Microfluidic Chips for Single-Cell Analysis. Biosensors 2021, 11, 412. [CrossRef]
10. Magar, H.S.; Hassan, R.Y.A.; Mulchandani, A. Electrochemical Impedance Spectroscopy (EIS): Principles, Construction, and
Biosensing Applications. Sensors 2021, 21, 6578. [CrossRef]
11. Wang, S.; Zhang, J.; Gharbi, O.; Vivier, V.; Gao, M.; Orazem, M.E. Electrochemical Impedance Spectroscopy. Nat. Rev. Methods
Prim. 2021, 1, 41. [CrossRef]
12. Berggren, C.; Bjarnason, B.; Johansson, G. Capacitive Biosensors. Electroanalysis 2001, 13, 173–180. [CrossRef]
13. Manickam, A.; Chevalier, A.; McDermott, M.; Ellington, A.D.; Hassibi, A. A CMOS Electrochemical Impedance Spectroscopy
(EIS) Biosensor Array. IEEE Trans. Biomed. Circuits Syst. 2010, 4, 379–390. [CrossRef] [PubMed]
14. Loyprasert, S.; Hedström, M.; Thavarungkul, P.; Kanatharana, P.; Mattiasson, B. Sub-Attomolar Detection of Cholera Toxin Using
a Label-Free Capacitive Immunosensor. Biosens. Bioelectron. 2010, 25, 1977–1983. [CrossRef] [PubMed]
15. Luka, G.; Samiei, E.; Dehghani, S.; Johnson, T.; Najjaran, H.; Hoorfar, M. Label-Free Capacitive Biosensor for Detection of
Cryptosporidium. Sensors 2019, 19, 258. [CrossRef]
16. Sharma, R.; Deacon, S.E.; Nowak, D.; George, S.E.; Szymonik, M.P.; Tang, A.A.S.; Tomlinson, D.C.; Davies, A.G.; McPherson, M.J.;
Wälti, C. Label-Free Electrochemical Impedance Biosensor to Detect Human Interleukin-8 in Serum with Sub-Pg/Ml Sensitivity.
Biosens. Bioelectron. 2016, 80, 607–613. [CrossRef]
17. Soraya, G.V.; Abeyrathne, C.D.; Buffet, C.; Huynh, D.H.; Uddin, S.M.; Chan, J.; Skafidas, E.; Kwan, P.; Rogerson, S.J. Ultrasensitive
and Label-Free Biosensor for the Detection of Plasmodium Falciparum Histidine-Rich Protein II in Saliva. Sci. Rep. 2019, 9, 17495.
[CrossRef]
18. Tanak, A.S.; Jagannath, B.; Tamrakar, Y.; Muthukumar, S.; Prasad, S. Non-Faradaic Electrochemical Impedimetric Profiling of
Procalcitonin and C-Reactive Protein as a Dual Marker Biosensor for Early Sepsis Detection. Anal. Chim. Acta X 2019, 3, 100029.
[CrossRef]
19. Uygun, Z.O.; Atay, S. Label-Free Highly Sensitive Detection of DNA Approximate Length and Concentration by Impedimetric
CRISPR-DCas9 Based Biosensor Technology. Bioelectrochemistry 2021, 140, 107812. [CrossRef]
20. Neshani, S.; Nyamekye, C.K.A.; Melvin, S.; Smith, E.A.; Chen, D.J.; Neihart, N.M. AC and DC Differential Bridge Structure
Suitable for Electrochemical Interfacial Capacitance Biosensing Applications. Biosensors 2020, 10, 28. [CrossRef]
21. Zumbahlen, H. Basic Linear Design; Analog Devices: Norwood, MA, USA, 2007; ISBN 978-0-916550-28-8.
22. Lokere, K.; Hutchison, T.; Zimmer, G. Precision Matched Resistors Automatically Improve Differential Amplifier CMRR—Here’s
How. Linear Technol. Des. Note 2013, 20, 1023.
23. Pallas-Areny, R.; Webster, J.G. Common Mode Rejection Ratio for Cascaded Differential Amplifier Stages. IEEE Trans. Instrum.
Meas. 1991, 40, 677–681. [CrossRef]
24. Gesteland, R.C.; Howland, B.; Lettvin, J.Y.; Pitts, W.H. Comments on Microelectrodes. Proc. IRE 1959, 47, 1856–1862. [CrossRef]
25. Karki, J. Fully-Differential Amplifiers. Available online: https://www.ti.com/lit/an/sloa054e/sloa054e.pdf?ts=1672562718439&
ref_url=https%253A%252F%252Fwww.google.com%252F (accessed on 15 October 2022).
26. Texas Instruments. Noise Analysis in Operational Amplifier Circuits. In Application Report; Texas Instruments: Dallas, TX,
USA, 2007.
27. Sobering, T.J. Equivalent Noise Bandwidth. 1991, p. 3. Available online: https://www.k-state.edu/edl/docs/pubs/technical-
resources/Technote1.pdf (accessed on 15 October 2022).
28. Pintelon, R.; Schoukens, J. An Improved Sine-Wave Fitting Procedure for Characterizing Data Acquisition Channels. IEEE Trans.
Instrum. Meas. 1996, 45, 588–593. [CrossRef]
Biosensors 2023, 13, 77 25 of 25
29. Ramos, P.M.; Cruz Serra, A. A New Sine-Fitting Algorithm for Accurate Amplitude and Phase Measurements in Two Channel
Acquisition Systems. Measurement 2008, 41, 135–143. [CrossRef]
30. Masi, A.; Danisi, A.; Di Castro, M.; Losito, R. Real-Time High-Precision Reading Algorithm for the Ironless Inductive Position
Sensor. IEEE Trans. Nucl. Sci. 2013, 60, 3661–3668. [CrossRef]
31. IEEE Std 1057-1994; IEEE Standard for Digitizing Waveform Recorders. IEEE: New York, NY, USA, 1994. [CrossRef]
32. Martino, M.; Losito, R.; Masi, A. Analytical Metrological Characterization of the Three-Parameter Sine Fit Algorithm. ISA Trans.
2012, 51, 262–270. [CrossRef]
33. Andersson, T.; Handel, P. IEEE Standard 1057, Crame/Spl Acute/r-Rao Bound and the Parsimony Principle. IEEE Trans. Instrum.
Meas. 2006, 55, 44–53. [CrossRef]
34. Ang, A.H.-S.; Tang, W.H.; Ang, A.H.-S. Probability Concepts in Engineering: Emphasis on Applications in Civil & Environmental
Engineering, 2nd ed.; Wiley: New York, NY, USA, 2007; ISBN 978-0-471-72064-5.
35. Alegria, F.C.; Serra, A.C. Gaussian Jitter-Induced Bias of Sine Wave Amplitude Estimation Using Three-Parameter Sine Fitting.
IEEE Trans. Instrum. Meas. 2010, 59, 2328–2333. [CrossRef]
36. Aperture Jitter Calculator for ADCs. Available online: https://www.maximintegrated.com/en/design/technical-documents/
app-notes/4/4466.html (accessed on 1 July 2022).
37. Ramos, P.M.; Janeiro, F.M.; Tlemcani, M.; Serra, A.C. Recent Developments on Impedance Measurements With DSP-Based
Ellipse-Fitting Algorithms. IEEE Trans. Instrum. Meas. 2009, 58, 1680–1689. [CrossRef]
38. Piasecki, T. Fast Impedance Measurements at Very Low Frequencies Using Curve Fitting Algorithms. Meas. Sci. Technol. 2015,
26, 065002. [CrossRef]
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.

Biosensors 13 00077 v2 1

Uploaded by

Copyright:

Available Formats

Biosensors 13 00077 v2 1

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Biosensors 13 00077 v2 1

Uploaded by

Copyright:

Available Formats

biosensors

Abstract: Field deployment is critical to developing numerous sensitive impedance transducers.

Keywords: real-time; impedance biosensor; low-cost; precision; sensitive; readout; interface

Biosensors 2023, 13, 77. https://doi.org/10.3390/bios13010077 https://www.mdpi.com/journal/biosensors

where Va − Vb , is the differential output voltage of the sensing unit, α is a proportionality

2.1. Amplification and Filtering Analysis

2.1.1. Common Mode to Differential Conversion

2.1.2. Noise Analysis

In Figure 3a, R𝑅b1 𝑏1,, 𝐶 b1,, 𝑅

2.2. Real-Time Digitization and Fitting

Reference and Filtering

Data Acquisition and Fitting

The non-iterative sine parameter extraction algorithm, IEEE standard 3-parameter

2.2.1. Additive White Gaussian Noise

2.2.2. Sampling Clock Jitter

y jitter [k] = A cos(2π f in (tk + αk ) + ϕ) + DC = A cos(2π f in tk + θk + ϕ) + DC, (26)

which can be further simplified as a function of ∆ f to be:

which can be further simplified and written as a function of frequency deviation, ∆ f :

2.2.4. Real-Time Microcontroller Implementation Considerations

3.1. Differential Amplification and Filtering Implementation

3.1.2. Noise Considerations and Filtering Unit

3.2. Real-Time Signal Acquisition and Sine Fitting

3.2.1. Additive Noise

3.2.1. Additive Noise

3.3. Board Characterization

Figure 11. The

Real-Time Operation and Sensitivity

The results show a distinct amplitude change at each consecutive capacitance

You might also like