1

An Aquatic Wireless Biosensor for Electric OrganDischarge with an Integrated Analog Front End

Wei Tang Member, IEEE, Paul M. Furth Senior Member, IEEE, Venkat Harish NammiGaurav Panwar, Vicente Ibarra, Xiaochen Tang, Graciela A. Unguez, and Satyajayant Misra

Abstract—This paper presents a novel wireless underwaterdata acquisition system for sensing Electric Organ Discharge(EOD) signals generated from the weakly electric fish S. macru-rus. Variation in frequency and amplitude of the EOD signals areof interest by biologists to study behavioral and environmentaleffects on electric organ cells. In order to record the EOD signals,a miniature wearable wireless sensing system is designed for thetarget fish. The system consists of a custom-designed integratedAnalog Front-End (AFE), an ATmega328p microcontroller unit(MCU) and a wireless transmitter (TX), as well as a battery witha power management module. In order to save TX power, thewireless sensor only transmits the calculated frequency and am-plitude information of the EOD signal based on a proposed gain-feedback control method. A receiver attached to a host computerreceives the wirelessly transmitted EOD data to perform furtheranalysis. A wearable waterproof backpack for the fish is designedto house the wireless sensor and battery. The overall systemhas been successfully tested in a clinical experiment with theweakly electric fish. The AFE integrated circuit, which includesa novel rail-to-rail dynamic comparator, is fabricated in a 0.18-µm CMOS process. The total area of the core circuit is 1.0 mm2.The LNPA achieved a Noise Efficiency Factor (NEF) of 1.8.The AFE power consumption is 2.2 µW from a single +1.8 Vsupply, whereas the overall system power consumption is 15 mW.The proposed gain-feedback control structure for frequency andamplitude measurement saves the transmitter data rate by 3,756times compared with sending the raw data.

Index Terms—Analog Front-End (AFE), Electric Organ Dis-charge (EOD), Underwater Wireless Biosensor, Dynamic Com-parator, Rail-to-rail Comparator.

I. INTRODUCTION

THIS paper introduces a novel sensor structure designedfor a wireless underwater biosensor. In recent years,

wireless sensor and data acquisition systems have been widelyused in biological research applications [1]–[8]. The primarygoal of such systems is to sense and analyze biological signalswhile the subject animal can move freely. To achieve thisgoal, the wireless biosensor system generally consists of anAnalog Front-End (AFE), which includes amplifiers and anAnalog-to-Digital Converter (ADC), a Digital Signal Process-ing (DSP) module, and a wireless transceiver or transmitter(TX). The design challenges of such systems include low-power consumption of the overall system, low-noise design in

This work is sponsored by U.S. NIH grant 1SC1GM092297-01A1 and U.S.National Science Foundation grant 1248109, 1408019 and 1652944. Copyright(c) 2018 IEEE. Personal use of this material is permitted. However, permissionto use this material for any other purposes must be obtained from the IEEE.

All authors are with New Mexico State University, USA. Wei Tang, VenkatHarish Nammi, Paul M. Furth, Vicente Ibarra, and Xiaochen Tang are with theKlipsch School of Electrical and Computer Engineering. Gaurav Panwar andSatyajayant Misra are with the Department of Computer Science. Graciela A.Unguez is with the Department of Biology.

Correspondence should be addressed to: Wei Tang, 1125 Frenger Mall, LasCruces, New Mexico 88003 USA. E-mail: wtang@nmsu.edu

AFE Module DSP

RF

TX

Electrodes

near the skin

of the fish

Fig. 1. General architecture of EOD recording system.

the AFE, hardware-friendly DSP algorithms, and the minia-turized physical size and weight of the sensor device. Besidesthese typical design considerations, extra design considerationsmay be included in special applications, such as deploying thesensor in an aquatic environment, as illustrated in Fig. 1.

Aquatic wireless sensors [9]–[13] play an important rolein facilitating non-invasive long-term studies of underwaterbiological processes and animal behaviors. One such behaviorof great interest to biologists is the production of electricalfields by electric fish. The long-tailed knifefish Sternopygusmacrurus, like all electric fish, produces an electric organdischarge (EOD) that is essential for navigation, communica-tion, and mating [14]. The EOD of S. macrurus is the outputby the muscle-derived electrogenic cells called electrocytes,which are activated continuously throughout the life of theanimal [14], [15]. Characterization of the EOD of S. macrurushas been largely studied with fish in captivity. However,relatively little is known about environmental factors, such aslight intensity, temperature, and chemistry of aquaria water,that may affect the fish’s EOD during a 24-hr period or longer.Moreover, functional studies on the temporal recovery of theEOD during the transdifferentiation of skeletal muscle cellsinto electrocytes during tail regeneration after amputation isnot known [15].

Currently, biologists still lack an understanding of hownerve-dependent electrical activity regulates the regenerationof the striated muscle cell phenotype. This is a fundamentalproblem in biology with high implications to neuro-muscularrepair and restoration following injury or disease – a researcharea in muscle biology and regeneration fields that is under-studied. For instance, the adult S. macrurus regenerates its tailafter amputation through formation of a blastema beneath theepidermis at the wound site. Regeneration of all tissue types,including the spinal cord, skeleton, dermis, skeletal muscleand EO can occur after repetitive tail amputations. However,full restoration of all amputated muscle fibers and the completetransdifferentiation of muscle cells into EO cells does not takeplace in the absence of electrical activity. So in vivo studiesmerit the pursuit of discovering the mechanisms that link

2

electrical activity to regulation of distinct skeletal muscle bi-ology in this highly regenerative aquatic vertebrate. However,this requires instrumentation that can manipulate and recordelectrical stimulation during regeneration after tail amputation.To date, there is no experimental hardware for underwaterchronic electrical stimulation and recording. Therefore, thecapability to record EOD amplitude and frequency underdifferent experimental conditions while the fish moves freelywould greatly enrich ongoing cellular and molecular studies.

To meet such requirements, aquatic sensors have been de-signed and implemented to collect data for biological studies.In [16], a wireless sensing framework was designed andimplemented to record the EOD signal using off-the-shelf-components. Due to the size and power consumption of thesensor, the device can only be deployed near the fish nest,which limits the accuracy of sensing and suffers from signalloss if and when the fish swims away from the nest. Theamplitude measurement is also not reliable since the distancebetween the sensor and fish constantly varies. In [17], com-pressive sensing was applied using the senor in [16]. In [18], astimulator circuit was built for chronic in vivo stimulation witha wearable backpack for the fish. Though the subject fish canmove freely with the stimulator backpack, sensing the electricfield relies on the sensor in the fish nest [16]. In these aquaticprototypes, major design challenges beyond a typical biosensorinclude the waterproof capsule for the sensor, underwaternoise, wireless signal attenuation, and ground signal offset.

To solve the above problems, in this paper, we designedand tested an aquatic wireless biosensor, as shown in Fig. 1.The major innovations and contributions of the work include:(1) the prototype design and implementation of the firstcomplete aquatic wireless biosensor in a clinical environmentthat remotely measures EOD frequency and amplitude forbiology studies, (2) the system design of gain-feedback controlbetween the AFE circuit and microcontroller to achieve reli-able frequency measurement while calculating the amplitude,which saves wireless power by avoiding transmitting the rawwaveform, and (3) a rail-to-rail dynamic clocked comparatorwith zero static power consumption. This design can be used asa reference design for other aquatic wireless biosensors. Thepaper is organized as follows: Section II introduces systemdesign considerations. Section III describes circuit-level andsystem-level implementation details. Section IV presents theexperimental setup and results. Section V provides discussionand conclusion.

II. SYSTEM DESIGN CONSIDERATIONS

S. macrurus is a weakly electric fish that generates anelectric organ discharge (EOD) using synchronized dischargesof spinal electromotoneurons via activation of their targetelectrocytes, which make up the electric organ. The EOD is thestereotypical electric organ pulses generated by electric fishesthat have a species-specific frequency and rhythm. The EODis generated in the pacemaker neurons located in the brainstemof the fish and relayed to the electric organ to generatethe electrical field around the fish. The EOD signal has anamplitude of 1-2 mV when measured by near-field electrodesplaced within 5 cm of the fish. The primary frequency of theEOD signal ranges between 50-200 Hz [14]. The frequency

LDOs

AFE ATmega

328p

RF

TX

AVDDMVDD

Host

Computer

RF RX

3.7 - 4.2 V

1.8 V 3.0 V

LiPo

Fig. 2. Overall setup for recording EOD signals from a weakly electric fish.

and amplitude of the EOD signal vary between individualfish and are affected by behavioral and environmental factors.Therefore, in order to study the relationship between the EODsignal and other biology topics, scientists are expecting wear-able biosensors that can be attached to the fish and wirelesslytransmit the EOD signal’s amplitude and primary frequencywhile the subject fish moves freely. Design considerationsof this automatic data acquisition system, therefore, includeaccuracy, weight, size, and battery lifetime of the wirelesssensor, as well as reliability and bio-compatibility.

The system building blocks and experimental setup areillustrated in Fig. 2. This system consists of the non-invasivewireless sensor attached to the subject fish and a host computerwith a wireless receiver collecting the sensed data. The sensorcontains the custom-designed AFE chip, an 8-bit ATmega328pmicrocontroller unit (MCU), a radio frequency (RF) transmit-ter (TX), and two low-dropout voltage regulators (LDOs). Thewhole system is powered by a rechargeable lithium polymer(LiPo) battery with an output voltage in the range of 3.7-4.2 V.LDOs step down the battery output voltage to clean suppliesof 1.8 V and 3.0 V. 1.8 V is used to power the AFE, while 3 Vis used to power the MCU and the TX module. All the abovecircuits and battery are placed on a 1.5”× 1” printed circuitboard (PCB) and encapsulated in a plastic 4”× 2” waterproofcase. The EOD sensing probe electrodes are placed near thetail of the fish to sense the near-field electric field.

The system measures the EOD frequency and amplitudeusing zero-cross detection and a gain-feedback control method.While in operation, the AFE continuously senses the analogEOD signal and performs analog amplification, frequencyfiltering, DC offset cancellation, and analog-to-digital con-version. The MCU receives the digital data from the AFEand measures the frequency of the EOD signal using a zero-crossing detection algorithm: the MCU monitors the ADC out-put and compares it with a pre-defined threshold value duringa certain time period, and counts the number of times theADC output crosses that pre-defined threshold value. Althoughthis method is simple and saves computing power comparedto performing a Fast Fourier Transform, it is susceptible toinput signal noise. Therefore, in order to better implementthis method, the ADC input signal should maintain a highamplitude, i.e., a peak-to-peak swing that is nearly rail-to-rail. This can be achieved by increasing the gain in the pre-amplifier. Nevertheless, if the gain of the pre-amplifier istoo high, the ADC input signal would saturate and then theamplitude of the subject EOD signal cannot be accuratelymeasured. To solve this problem, in this design we added avariable gain amplifier (VGA) stage between the pre-amplifierand the ADC. The gain of the VGA is controlled by the MCU

3

Low Noise Pre-Amplifier

Variable Gain Amplifier

8-bit SAR ADC

VOP_PRE

VON_PRE

VOP_VGA

VON_VGA

D7

D6

D0

EOD+

EOD-

CLK RST

S7 S1

Microcontroller

Gain-feedback Control

Transmitter

Analog Front End Chip

Power Management

Battery

Fig. 3. Sensor system architecture.

via digital gain-feedback control logic. The MCU monitors theADC output data and increases the gain of the VGA if theADC input amplitude decreases, or vice versa. This functionguarantees that the ADC input amplitude is nearly rail-to-rail, but not saturated. By using this method, the frequencyof the signal can be reliably measured using the zero-crossingmethod and without saturating the ADC. The amplitude ofthe EOD signal is calculated from both the VGA gain and theADC output.

The system transmits both the frequency and amplitudedata using a short-range wireless module. The MCU firstcombines the data into a serial format and then organizesthe data into packets. The wireless TX module transmits thedata packets using Amplitude-Shift-Keying (ASK). Finally, theASK receiver decodes the data and saves it on the computerfor further analysis. The system only transmits frequency andamplitude data instead of the raw EOD waveform in order toreduce the data rate and, hence, the power of the RF block.Details of each circuit building block are described in thefollowing section.

III. SENSOR ARCHITECTURE AND IMPLEMENTATION

The proposed EOD sensor is based on our previous workwhich employed discrete off-the-shelf components, as pub-lished in [16]. In this work, we update the design using anintegrated AFE chip and the gain-feedback control architec-ture. Fig. 3 shows the proposed AFE and gain-feedback controlbuilding blocks. The AFE includes a Low-Noise Pre-Amplifier(LNPA), followed by a Variable Gain Amplifier (VGA) andan 8-bit Successive Approximation Register Analog-to-DigitalConverter (SAR ADC) that incorporates a novel rail-to-raildynamic comparator. The microcontroller takes data from theADC and measures the amplitude and the primary frequencyof the EOD signal. The microcontroller also adjusts the gain ofthe VGA based on the instance amplitude of the EOD signal.The calculated frequency and amplitude data are then sentwirelessly by the transmitter unit. The following subsectionsdetail the circuit and system level design and implementation.

A. Low-Noise Pre-Amplifier (LNPA)

The first stage of the integrated AFE chip is the Low-NoisePre-Amplifier (LNPA). The overall design goal of the LNPAis to optimize trade-offs between gain, bandwidth, noise, andpower consumption for the target EOD signal. The design isbased on [19] with modified parameters for a 0.18-µm CMOSprocess. The schematic of the LNPA is shown in Fig. 4. Dueto the noisy aquatic environment, the LNPA adopts a fully-differential design to improve the Common-Mode RejectionRatio (CMRR) and Power Supply Rejection Ratio (PSRR).Fully-differential design is particularly important with lowsupply voltages, as the signal swing is effectively doubled.Large DC offsets from the electrode-tissue interface are re-jected by AC coupling the input signal through input capacitorCIN . The input-referred noise of the LNPA is approximatelyequal to the input-referred noise of operational amplifier A1.The gain of the LNPA is set to 100. Considering the expectedfrequency range of EOD signals (50-200 Hz), the low cut-offfrequency is set to 10 Hz and the high cut-off frequency is setto 2 kHz.

VCM

EOD+ VON_PRE

VOP_PRE

CIN

CIN

CFB

CFB

PR

PR

A1

EOD-

M1 M2

M3 M4M5 M6

M12 M13 M14 M15

M9 M10

M7 M8

VDD

VSS

VCMFB

VCMFB

VIPVIN

VOPVON

VCM

VB

VOP

VONCC CC

CCM

CCM

PRCM

VIP

VIN

MR1 MR2

PRCM2IB2IB2IB2IB

M11

IB

VOP

VON

Fig. 4. Low-Noise Pre-Amplifier (LNPA).

B. Variable Gain Amplifier (VGA)

A variable gain amplifier adds robustness to the AFE byincreasing its dynamic range. Fig. 5 shows the schematicof the Variable Gain Amplifier (VGA), which consists of afully-differential amplifier in a closed-loop configuration. Thearchitecture is similar to that used in the LNPA of Fig. 4; keydifferences are the op-amp uses an NMOS input differentialpair and the input capacitors and compensation capacitors areprogrammable.

4

VCM

VOP_PRE

VON_PRE

VOP_VGA

VON_VGA

CFB

CFB

CIN2

CIN7

PR

PR

A2

S7

S2

S1 S7-

CIN1S1

CIN2

CIN7S7

S2

CIN1S1

Fig. 5. Variable Gain Amplifier (VGA) with digital select lines S1 − S7.

The midband gain of the VGA is CIN/CFB , where CFB

is fixed and equal to 500 fF and CIN is implemented as abinary-weighted capacitor bank from 500 fF to 32 pF, in orderto adjust the VGA gain from 0 dB to 36 dB in steps of 6 dB,which corresponds to a factor of 2. The main idea of having avariable input capacitor instead of variable feedback capacitoris to maintain a constant low cut-off frequency. The inputcapacitor of the VGA blocks any DC offsets generated at theoutput of the LNPA. The op-amp used in the VGA is a two-stage Miller-compensated op-amp. This topology is chosenbecause of its simplicity and ability to operate at low voltagesupplies with a wide output swing. The transistor sizes andother design values for the LNPA and VGA are summarizedin Table I.

Parameter LNPA (A1) VGA (A2)M1,2 [µm/µm] 100/1 100/1M3,4 [µm/µm] 110/50 10/1M5−10 [µm/µm] 10/1 10/1M11 [µm/µm] 5/1 5/1

M12−15 [µm/µm] 10/1 10/1MR1,R2 [µm/µm] 2/45 2/45

CIN [pF] 50 0.5–32CFB [pF] 0.5 0.5CC [pF] 1 2.5–160CCM [pF] 1 1IB [nA] 50 80

TABLE ILNPA AND VGA PARAMETER VALUES

C. 8-bit Successive Approximation Register (SAR) ADCAfter amplifying and conditioning an input noisy signal

from the fish through the LNPA and VGA, it is digitized byan Analog-to-Digital Converter (ADC). This conversion helpsin further analysis of the EOD signal, as digital data is lesssusceptible to external noise. The Successive ApproximationRegister (SAR) ADC is preferred over other ADC architec-tures in this particular low-speed application because of itsmoderate resolution and good power efficiency.

The proposed SAR ADC is based on the design in [20],with modifications made in the control logic and comparator.

M1 M2

M4 M5

M3

M9M8M7

M6

M10

M11

M13M12

M14

M15

M16 M17

M19 M20

M18

M22M21

M25 M26M23M24

M27

M28

M30M29 M34M33

M32

M31

VDD

VDD

VSS

VSS

CLK

CLK

CLK

CLK CLK

VINNVINP

VOP

A

D

B

C

AB

CD

VON

CLK CLK

CLK

m=3 m=3

m=3 m=3

Fig. 6. Proposed rail-to-rail dynamic comparator.

A novel rail-to-rail dynamic comparator helps achieve zerostatic power in the ADC. For this application, 8-bit resolutionand 10 kHz sampling rate are selected.

D. Rail-to-Rail Dynamic Comparator

Fig. 6 shows the proposed rail-to-rail dynamic comparatorthat was designed for this application with two main features:1) rail-to-rail input range and 2) fully dynamic operation,resulting in no static power loss. The inputs to the comparatorare nodes VINP and VINN , which are the sampled outputs ofthe VGA and which are also attached to the DAC outputs ofthe binary-weighted capacitor arrays. Since 1 LSB is equal to1.8 V/256, the DAC output voltage can go as low as 7 mVand as high as 1.793 V. As such, a rail-to-rail comparatoris required in this particular SAR ADC architecture. Theproposed design is based on the current-controlled latch senseamplifier of [21], which was analyzed and optimized in [22].More recently, this sense amplifier has been adapted for lowsupply voltage in [23] and reduced delay in [24]. In our work,a low-power sense latch originally intended for high-speedSRAM is modified to achieve rail-to-rail input range in a low-speed bio-sensor application.

As shown in Fig. 6, rail-to-rail input common-mode rangeis achieved using parallel NMOS and PMOS input differentialstages. The NMOS and PMOS structures are complementarywith inputs VINP and VINN and outputs VOP _CMP andVON_CMP tied across each half of the circuit. The top half ofthe comparator with the NMOS differential pair and NANDSR-latch operates when the input common-mode voltage ishigher than the NMOS threshold voltage. The bottom half ofthe comparator with the PMOS differential pair and NOR SR-latch comes into the picture when the input common-modevoltage is below VDD by the PMOS threshold voltage. Forinputs near VDD/2, both structures operate in parallel.

5

Referring to the top half of Fig. 6, when input CLKC islow, the comparator is reset and nodes A and B are bothset to VDD. In this state, the outputs of the NAND SR-latch VOP _CMP and VON_CMP are in the hold state, thatis, they are holding the value of the previous comparison. Incomplementary fashion, referring to the bottom half of Fig. 6,when CLKC is high, nodes D and C are reset to 0 V, placingthe outputs of the NOR SR-latch in the hold state. Now, atthe rising edge of CLKC , let us assume that VINP is greaterthan VINN . In that case, in the top half of the circuit, node Agoes low, pulling VOP _CMP to VDD through transistor M16

and, through feedback transistor M11, VON_CMP is pulled toVSS . Similarly, in the bottom half of the circuit, at the fallingedge of CLKC , node D goes high, pulling VON_CMP to VSS

through transistor M29 and, through feedback transistor M32,VOP _CMP is pulled to VDD.

In order to prevent contention current in the case thatone differential pair makes a decision faster than the other,feedforward transistors M12, M16, M29 and M33 are sizedwith multiplicity (m) equal to three. Suppose that VINP isgreater than VINN , but that the input common-mode voltageis close to VDD, such that the NMOS differential pair makes afast decision and the PMOS differential pair makes no decisionat all. In this case, after the rising edge of CLKC , node Agoes low quickly, whereas node D remains in the reset state.Node A going low creates the possibility of contention currentbetween transistors M16 and M34, if we further assume thatoutput VON was being held from the last comparison in thehigh state. By oversizing transistor M16, VOP is pulled highthrough M16 and, via feedback transistors M11 and M30, VON

is pulled low.The zero-static power rail-to-rail input CM comparator

would find application in a clocked (discrete-time) pulse-widthmodulator, a clocked envelope-tracking power converter, or asa general-purpose input block that could be used for any CMrange. However, it may be not an optimal choice for the SARADC in this system. Since the SAR ADC follows the design in[20], either a PMOS or NMOS input Strong-Arm Capacitordesign would have been sufficient to handle the rail-to-railinput. Though the rail-to-rail dynamic comparator increasessystem power and area, it serves as a testing circuit for futureoptimized ADC architecture of the integrated EOD sensor.

E. Microcontroller and Wireless Interface

The ATmega328p microncontroller unit (MCU) is used toadjust the gain-feedback control signal to the VGA, computeamplitude and frequency, and organize the data into packetsfor the wireless transmitter. To each data packet, preamblebits "10101010" are added, which translate to HEX 0xAA.The preamble helps the receiver identify the beginning of adata packet. The system applies off-the-shelf radio transmittermodules QAM-TX2 and receiver QAM-RX2 as the radiointerface. At the receiver side, the QAM-RX2 module isconnected to a laptop through USB. The serial output receivedfrom the UART is recorded on the laptop, which runs a Pythonscript to read the serial data from the USB. An error detectionalgorithm is implemented in the laptop using the checksumof the data packet. If the checksum doesn’t match, or the

LNPA

VGA

ADC

1000 µm

1000 µ

m

(a) (b)

Fig. 7. (a) Micrograph and (b) Layout of the analog front-end IC for EODsignal recording system.

detected frequency is out of a pre-defined range, the systemabandons the packet and displays an error message. Otherwise,the Python script records the amplitude, frequency, and timinginformation to a text file.

In this EOD measurement application, since the frequencyand amplitude do not change suddenly, we adopted an errordetection algorithm, which includes both the checksum and apre-defined frequency range, to abandon faulty data. This errordetection algorithm does not affect the experiment if the errorrate of the packet is not too high. In our experiment, the errorrate of the packet is approximately 4% of the total readings,which does not affect recordings of frequency and amplitude.

The gain-feedback control in this design saves system powerby reducing the transmitter throughput. This is because, usingthe gain-feedback method, the MCU can calculate the fre-quency and amplitude of the EOD signal even when the EODsignal has a small amplitude. In this design, the frequency andamplitude of the signal are transmitted every three seconds.Within each data packet, there are only 64 bits: 32 bits forheader synchronization, 16 bits for data, followed by the 8-bit checksum, then another 8 bits for trailing synchronization,with bit pattern "10101010". Therefore the total data rateis 21.3 bits per second(bps). Compared to the method ofsending raw data, which would be at least 80 kbps (= 8 bits× 10 kHz sampling rate), the proposed method can reduceradio throughput by 3,756 times. As presented in the nextsection, transmitting power dominates the overall sensor powercost; therefore, the proposed method, while increasing theprocessing power, can greatly reduce the overall system power.

IV. EXPERIMENTAL RESULTS

Experiments are performed to evaluate the performance ofthe custom-designed AFE front-end integrated circuit and tovalidate the entire wireless aquatic biosensor system. The AFEchip is implemented using Global Foundry 0.18-µm CMOStechnology. A micrograph of the chip is shown in Fig. 7with circuit building blocks highlighted. The chip contains theLow-Noise Pre-Amplifier (LNPA), Variable Gain Amplifier(VGA), and Successive Approximation Register Analog-to-Digital Converter (SAR ADC). The overall silicon footprintis 1.5 x 1.5 mm2; the core circuit area is 1 x 1 mm2.

6

EOD+

EOD-

VBATT

(3.7 – 4.2 V)

MVDD (3.0 V)

DATA

AVDD (1.8 V)

Right Leg Driver (RLD) Circuit

LT6005 Op-amps

LT1763

LDOLP5907

LDOLT6930-8.00

Oscillator

AnalogFront-End

ATmega328p

MCU

D7

D6

D0

RLD

+

-

QAM-TX2AM

Transmitter Module

CLK

8 MHz

100 kHz

Fig. 8. Block-level schematic of PCB.

(a) (b)

Fig. 9. Aquatic wireless biosensor (a) PCB and (b) packaging.

The AFE chip, power management circuits, microcon-troller (MCU), and the radio transmitter module are placedon a 1.5”× 1” printed circuit board (PCB). A block-levelschematic of the PCB is shown in Fig. 8. The schematicincludes a right-leg driver (RLD) circuit [25], which consistsof an instrumentation amplifier and feedback amplifier. Thegoal of the RLD circuit is to (a) extract the common-modesignal between inputs EOD+ and EOD− and (b) drive thatcommon-mode signal towards ground. The RLD electrode isplaced in the center of the fish, whereas the EOD electrodesare placed on either side of the tail. An 8 MHz oscillatorfeeds the MCU and the MCU generates a 100 kHz clock forthe SAR ADC. The PCB and battery are encapsulated in a4”× 2” waterproof plastic case, as illustrated in Fig. 9.

A. AFE PerformanceThe AFE integrated circuit was designed to operate from

a single 1.8 V power supply voltage. The simulated andmeasured frequency response of the LNPA is shown in Fig. 10.The measured closed-loop gain is 37.5 dB. The measuredhigh cut-off frequency is 1.25 kHz while the low cut-offfrequency is 14.6 Hz. The total current consumption, includingthe biasing branch, is 450 nA.

Fig. 11 shows the simulated and measured output-referrednoise power spectrum density of the LNPA. In Fig. 11, themeasured noise between 1 Hz and 10 Hz is significantly

10−1

100

101

102

103

104

105

10

15

20

25

30

35

40

Frequency (Hz)

Gai

n (d

B)

(a)

(b)

Fig. 10. (a) Simulated and (b) measured closed-loop gain of the LNPA.

10−1

100

101

102

103

104

105

0

20

40

60

80

Frequency (Hz)O

utpu

t Vol

tage

Noi

se (µ

V/S

qrt(

Hz)

)

(a)

(b)

Fig. 11. (a) Simulated and (b) measured output-referred noise of the LNPA.

higher than in simulation. While in the closed-loop gain plot(Fig. 10), the slope of the measured gain is slightly steeperbetween 1 Hz and 10 Hz compared to simulation. Thesediscrepancies are most likely due to inaccurate noise modelingof the pseudo-resistor in the pre-amplifier. Since the pseudo-resistor is operating in the cut-off region in order to create avery large resistance in the order of 1010 Ohms, it may bevery difficult to measure the true resistance and therefore theactual model is difficult to be established [19]. In addition,due to process, voltage, and temperature (PVT) variations,even normal resistors could experience variations in its valuesbetween the typical model value and the actual measuredvalues. Additional error may come from the measurementinstrument, since the number of sample points is small whenmeasuring low frequencies.

A well-known figure-of-merit used in front-end amplifiers isthe Noise Efficiency Factor (NEF) as it assesses the trade-offamong a few important parameters such as noise, bandwidth,and total power. NEF is defined as

NEF = Vin,RMS

√2 · ITOT

π · UT · 4kT ·BW(1)

where Vin,RMS is the input-referred noise, ITOT is the totalcurrent consumption of the LNPA, BW is the bandwidthof the amplifier and other parameters are as conventionallydefined. The measured NEF of the LNPA is 1.8. Table IIsummarizes the measured results of the proposed LNPA andcompares it with similar designs in the literature which focuson amplifying weak bio-potential signals.

The gain of the VGA can be programmed by adjusting theinput capacitance of the binary-weighted capacitor array. Themeasured frequency responses are shown in Fig. 12. The gain

7

Parameter [19] [26] [27] [28] [29] [30] This workYear 2003 2007 2012 2015 2018 2018 2018

Process 1.5µm 0.5µm 130nm 65nm 0.18µm 0.18µm 0.18µmVDD (V) ±2.5 1 1 0.6 1 0.8 1.8

IQ(nA) 1600 805 1250 1.7 250 5000 450High Frequency 7.2kHz 4.7kHz 8.5kHz 370Hz 10kHz 130Hz 1.25kHzLow Frequency 0.025 0.3Hz 0.4Hz 1.5Hz 250Hz 0.6Hz 14.6Hz

Gain (dB) 39.5 36.1 40.5 32 25.6 56 38.2Vni,rms 2.2µV 3.6µV 3.2µV 26µV 5.5µV 0.16µV 2.45µV

Area(mm2) 0.16 0.046 0.047 0.105 0.02 0.03 0.23NEF 4.0 1.8 20.3 2.1 1.07 13.7 1.8

TABLE IIPERFORMANCE COMPARISON WITH OTHER BIO-POTENTIAL LNPAS.

100

101

102

103

104

−20

−10

0

10

20

30

40

Frequency (Hz)

Gai

n (d

B)

32x

16x

8x

4x

2x

1x

64x

Fig. 12. Measured closed-loop gain of the VGA.

Specification This workTechnology 0.18µm

Sampling Rate(kS/s) 11Resolution (bit) 8

ENOB (bit) 7.4Power (µW) 2.2

TABLE IIIPERFORMANCE SUMMARY OF THE AFE CHIP.

can be programmed from 36 dB down to 0 dB in steps of6 dB using digital control signals. The measured high cut-offfrequency is 3-4 kHz and the low cut-off frequency is 2 Hz.

The measured ADC spurious-free dynamic range is 54.2 dBat a fundamental frequency of 190.7 Hz and sampling rateof 11.11 kHz, as shown in Fig. 13. The measured THD is46.5 dB, while the measured SNR is a much larger 88.4 dB.As such, measured SNDR = 46.5 dB. Therefore, the EffectiveNumber of Bits (ENOB = (SNDR – 1.76) / 6.02) is 7.4. Theoverall AFE performance is summarized in Table III.

B. EOD Recording ResultsIn the EOD sensing experiment, the target S. macrurus is

a freshwater species of knifefish native to South America andwas obtained commercially from Segrest Farms (Gibsonton,FL, USA). Adult fish, 20-35 cm in length, were housedindividually in 56–75 liter aerated aquaria maintained at 25–28 ◦C and were fed three times weekly. The overall system hasbeen implemented in a free-swimming aquarium environmentto test its underwater performance. The EOD sensing probe

0 2 4 6 8 100

10

20

30

40

50

60

70

80

90

100

110

Frequency (kHz)

Am

plitu

de (d

B)

Fundamental frequency = 190.7

SNR = 88.4SFDR = 54.2

THD= 46.5Harmonics

Noise ßoor

Fig. 13. Measured FFT of the ADC at 11.11 kHz sampling rate.

electrodes are placed near the tail of the fish to collect near-field electric field. The probe electrodes are 0.5 mm 316LVMstainless steel. The electrodes are fixed on the fish usingflexible elastic bands. A small LiPo battery with 250 mAhcapacity provides power for the whole wireless sensor. A 3D-printed backpack is applied to host the circuits and the battery.An annotated picture of the fish carrying the backpack isshown in Fig. 14.

Fig. 15 presents the recorded frequency and amplitude dataof the EOD signal collected by the proposed system over atime period of 15 hours. When obtaining the data of Fig. 15,since the sensor was attached close to the target fish, theamplitude of the EOD signal was quite stable, such thatit was not necessary to implement automatic gain-feedback.In a more challenging environment, automatic gain-controlcan be achieved via digital feedback signals from the MCUto select one of 7 gain settings in the VGA. The motionartifact is not an issue in this experiment for the followingreasons: (1) S. macrurus’s behavior in our aquaria tanks ischaracterized by prolonged stationary positions at bottom oftanks, either uncovered or covered (i.e., inside a PVC tube)and (2) the tank space assigned to fish during this experimentsis minimized to deter excessive movement. Moreover, studiesperformed to investigate how the EOD of an electric fishis affected by chronic changes in the environment and/orafter surgical or pharmacological manipulations are optimal

8

FISH

PCB MOUNTED

ON FISH RF RECEIVER

Fig. 14. PCB mounted on the fish in the aquarium.

when the EOD signal is stable with minimal noise. Hence,to minimize frequency data independent or unrelated to thefishâAZs EOD, we implemented the error detection algorithmthat automatically removes data when it is out of range of thegenerated fish EOD.

The total power consumption of the wireless EOD signal is15 mW, of which the TX module consumes 9 mW, the MCUconsumes 6 mW, and the custom-designed AFE integratedcircuit consumes 2.2 µW.

This design provides a prototype of the aquatic biosensorplatform. Individual building blocks, particularly the TX mod-ule and MCU, have room for optimization. For example, thesuper-low power ultra-wideband radio transmitter of [31] canbe used to replace the QAM-TX2 transmitter to reduce theTX power from 10 µJ/bit to as low as 1 nJ/bit. Moreover, acustom digital integrated circuit can replace the MCU to saveprocessing power.

In order to ensure that the receiver is able to detect thetransmitted QAM pulses, in this preliminary experiment, weallowed the TX module to transmit a digital 1 at full powerwithout controlling the duty-cycle. On the other hand, usingQAM, a digital 0 is transmitted with 0 power. The sensorsystem is able to tolerate full TX power and meet the goalof 15 hours of continuous recording because it only transmitsamplitude and frequency.

V. DISCUSSION AND CONCLUSION

In the reported experiment, the received data are only thefrequency and amplitude; as such, we cannot fully recoverthe raw signal. This is arguably the main disadvantage of theproposed method. However, the most salient information forthe biologist is available. Now, if the user wanted to receive aportion of the raw waveform, a possible solution is to modifythe sensor so that it transmits a portion of raw data whenrequired, so that other types of analysis can be performed,such as FFT. Another disadvantage of the proposed system isthe accuracy of the recorded amplitude and frequency, whichis degraded by in-band noise. Adding a measure of in-bandnoise to this sensor would allow one to roughly estimate theaccuracy in the measurement.

Fig. 15. Measured frequency and amplitude of the EOD signal of the fish.

A novel low power aquatic wireless biosensor has been pre-sented. The system applies gain-feedback control to measurethe amplitude and frequency of the electric organ dischargesignal of a weakly electric fish. A custom-designed analogfront-end integrated circuit with a novel rail-to-rail dynamiccomparator has been applied in the system with a very lowNoise Efficiency Factor of 1.8. A data collection system withradio and software interface has been developed to store thesensed data. The whole wireless sensor has been encapsulatedin a plastic box and attached to the subject fish, and has beenclinically tested to collect amplitude and frequency data ofEOD for biological studies. This system provides a referencedesign for aquatic wireless biosensors.

Wireless power dominates the system power consumption.In order to reduce radio power, an MCU was incorporatedto process the raw data so that the wireless transmitter onlysends amplitude and frequency data instead of the raw EODwaveform. The signal processing applies a zero-crossing detec-tion method, which is supported by the gain-feedback controlarchitecture. In future designs, the MCU can be replaced by adigital integrated circuit in order to save system power. A morepower-efficient wireless radio module can also be applied tofurther reduce power consumption.

REFERENCES

[1] V. N. Anisimov, J. A. Herbst, A. N. Abramchuk, A. V. Latanov, R. H. R.Hahnloser, and A. L. Vyssotski, “Reconstruction of vocal interactionsin a group of small songbirds,” Nature Methods, vol. 11, pp. 1135 EP–, 09 2014. [Online]. Available: http://dx.doi.org/10.1038/nmeth.3114

[2] R. R. Harrison, P. T. Watkins, R. J. Kier, R. O. Lovejoy, D. J. Black,B. Greger, and F. Solzbacher, “A low-power integrated circuit for awireless 100-electrode neural recording system,” IEEE Journal of Solid-State Circuits, vol. 42, no. 1, pp. 123–133, Jan 2007.

[3] E. Culurciello, W. Tang, E. J.-H. Park, B. Goldstein, D. Kim, andP. Weerakoon, Biomedical Circuits and Systems. Lulu. com, 2013.

[4] D. Kim, B. Goldstein, W. Tang, F. J. Sigworth, and E. Culurciello,“Noise analysis and performance comparison of low current mea-surement systems for biomedical applications,” IEEE Transactions onBiomedical Circuits and Systems, vol. 7, no. 1, pp. 52–62, 2013.

[5] E. Greenwald, M. Mollazadeh, C. Hu, W. Tang, E. Culurciello, andN. Thakor, “A VLSI neural monitoring system with ultra-widebandtelemetry for awake behaving subjects,” IEEE Transactions on Biomed-ical Circuits and Systems, vol. 5, no. 2, pp. 112–119, 2011.

9

[6] W. Tang, A. Osman, D. Kim, B. Goldstein, C. Huang, B. Martini,V. A. Pieribone, and E. Culurciello, “Continuous time level crossingsampling ADC for bio-potential recording systems,” IEEE Transactionson Circuits and Systems I: Regular Papers, vol. 60, no. 6, pp. 1407–1418, 2013.

[7] V. Ibarra, M. Araya-Salas, Y.-p. Tang, C. Park, A. Hyde, T. F. Wright,and W. Tang, “An RFID based smart feeder for hummingbirds,” Sensors,vol. 15, no. 12, pp. 31 751–31 761, 2015.

[8] X. Tang, Q. Hu, and W. Tang, “Delta-sigma encoder for low-power wire-less bio-sensors using ultrawideband impulse radio,” IEEE Transactionson Circuits and Systems II: Express Briefs, no. 7, pp. 747–751, August2016.

[9] M. Asadnia, A. G. P. Kottapalli, Z. Shen, J. Miao, and M. Triantafyllou,“Flexible and surface-mountable piezoelectric sensor arrays for under-water sensing in marine vehicles,” IEEE Sensors Journal, vol. 13, no. 10,pp. 3918–3925, Oct 2013.

[10] L. Margheri, B. Mazzolai, M. Cianchetti, P. Dario, and C. Laschi, “Toolsand methods for experimental in-vivo measurement and biomechanicalcharacterization of an octopus vulgaris arm,” in 2009 Annual Interna-tional Conference of the IEEE Engineering in Medicine and BiologySociety, Sept 2009, pp. 7196–7199.

[11] A. Y. Zhou, T. J. Zajdel, M. A. TerAvest, and M. M. Maharbiz, “Aminiaturized monitoring system for electrochemical biosensing usingshewanella oneidensis in environmental applications,” in 2015 37thAnnual International Conference of the IEEE Engineering in Medicineand Biology Society (EMBC), Aug 2015, pp. 7518–7521.

[12] L. Coelho, C. Pereira, J. Mendes, J. M. M. M. de Almeida, B. Kovacs,K. Balogh, T. Borges, and P. A. S. Jorge, “New developments on fibreoptic colorimetrie sensors for dissolved CO2 in aquatic environments,”in OCEANS 2017 - Aberdeen, June 2017, pp. 1–5.

[13] G. Saviozzi, E. Buselli, C. Stefanini, C. Laschi, and P. Dario, “A multi-depth sensorised micro sampling system,” in OCEANS 2015 - Genova,May 2015, pp. 1–7.

[14] M. V. L. Bennett, “Electric organs,” Fish Physiology, vol. 5, (Hoar, W.S. & Randall, D. J., eds), pp. 347–491, 1971.

[15] G. A. Unguez and I. H. Zakon, “Phenotypic conversion of distinctmuscle fiber populations to electrocytes in a weakly electric fish,”Journal of Comparative Neurology, vol. 399, pp. 20–34, 1998.

[16] M. Harris, E. Salazar, R. Guth, V. Nawathe, M. Sharifi, W. Tang, andS. Misra, “Wireless sensing framework for long-term measurements ofelectric organ discharge,” in Biomedical Circuits and Systems Confer-ence (BioCAS), 2013 IEEE. IEEE, 2013, pp. 53–56.

[17] H. Al-Azzawi, H. Huang, S. Misra, and W. Tang, “On using compressedsensing for efficient transmission & storage of electric organ discharge,”in Circuits and Systems (ISCAS), 2014 IEEE International Symposiumon. IEEE, 2014, pp. 1616–1619.

[18] G. Unguez, C. Duran, D. Valles-Rosales, M. Harris, E. Salazar, M. Mc-Dowell, and W. Tang, “3D-printed wearable backpack stimulator forchronic in vivo aquatic stimulation,” in Engineering in Medicine andBiology Society (EMBC), 2015 37th Annual International Conferenceof the IEEE. IEEE, 2015, pp. 2147–2150.

[19] R. R. Harrison and C. Charles, “A low-power low-noise CMOS amplifierfor neural recording applications,” IEEE Journal of Solid-State Circuits,vol. 38, no. 6, pp. 958–965, June 2003.

[20] C.-C. Liu, S.-J. Chang, G.-Y. Huang, and Y.-Z. Lin, “A 10-bit 50-MS/s SAR ADC with a monotonic capacitor switching procedure,” IEEEJournal of Solid-State Circuits, vol. 45, no. 4, pp. 731–740, 2010.

[21] T. Kobayashi, K. Nogami, T. Shirotori, and Y. Fujimoto, “A current-controlled latch sense amplifier and a static power-saving input buffer forlow-power architecture,” IEEE Journal of Solid-State Circuits, vol. 28,no. 4, pp. 523–527, Apr. 1993.

[22] B. Wicht, T. Nirschl, and D. Schmitt-Landsiedel, “Yield and speedoptimization of a latch-type voltage sense amplifier,” IEEE Journal ofSolid-State Circuits, vol. 39, no. 7, pp. 1148–1158, July 2004.

[23] B. Goll and H. Zimmerman, “A comparator with reduced delay time in65-nm CMOS for supply voltages down to 0.65 V,” IEEE Transactionson Circuits and Systems–II: Express Briefs, vol. 56, no. 11, pp. 810–814,Nov. 2009.

[24] S. Babayan-Mashhadi and R. Lotfi, “Analysis and design of a low-voltage low-power double-tail comparator,” IEEE Transactions on VLSISystems, vol. 22, no. 2, pp. 343–352, Feb. 2014.

[25] M. Harris, E. Salazar, R. Güth, V. Nawathe, M. Sharifi, W. Tang, andS. Misra, “Wireless sensing framework for long-term measurements ofelectric organ discharge,” in 2013 IEEE Biomedical Circuits and SystemsConference (BioCAS). IEEE, 2013, pp. 53–56.

[26] J. Holleman and B. Otis, “A sub-microwatt low-noise amplifier for neuralrecording,” in 2007 29th Annual International Conference of the IEEE

Engineering in Medicine and Biology Society, Aug 2007, pp. 3930–3933.

[27] F. Zhang, J. Holleman, and B. P. Otis, “Design of ultra-low powerbiopotential amplifiers for biosignal acquisition applications,” IEEETransactions on Biomedical Circuits and Systems, vol. 6, no. 4, pp.344–355, Aug 2012.

[28] P. Harpe, H. Gao, R. van Dommele, E. Cantatore, and A. van Roermund,“A 3nW signal-acquisition IC integrating an amplifier with 2.1 NEF anda 1.5fJ/conv-step ADC,” in 2015 IEEE International Solid-State CircuitsConference - (ISSCC) Digest of Technical Papers, Feb 2015, pp. 1–3.

[29] L. Shen, N. Lu, and N. Sun, “A 1-v 0.25- µW inverter stacking amplifierwith 1.07 noise efficiency factor,” IEEE Journal of Solid-State Circuits,vol. 53, no. 3, pp. 896–905, March 2018.

[30] H. Wang and P. P. Mercier, “A current-mode capacitively-coupled chop-per instrumentation amplifier for biopotential recording with resistive orcapacitive electrodes,” IEEE Transactions on Circuits and Systems II:Express Briefs, vol. 65, no. 6, pp. 699–703, June 2018.

[31] Q. Hu, X. Tang, and W. Tang, “Integrated asynchronous ultrawidebandimpulse radio with intrinsic clock and data recovery,” IEEE Microwaveand Wireless Components Letters, vol. 27, no. 4, pp. 416–418, 2017.