A Wearable UHF RFID-Based EEG SystemArtem Dementyev

Department of Electrical EngineeringUniversity of Washington

Seattle, WAEmail: artem.dementyev@gmail.com

Joshua R. SmithDepartment of Computer Science and Engineering

Department of Electrical EngineeringUniversity of Washington

Seattle, WAEmail: jrs@cs.washington.edu

Abstract—The wearable electroencephalogram (EEG) monitor-ing systems are the cornerstone of noninvasive brain-computerinterfaces (BCI) and many medical applications, but state-of-the-art wearable systems are limited by weight, battery life andsize. In this paper we present EEGWISP: am EEG monitoringsystem that is battery-free; is powered by a standard UHF RFIDreader; and uses backscatter to transmit the data using a EPCClass 1 Gen2 protocol. Since EEGWISP does not need batteriesit can be lightweight, miniature and maintenance free for users.We designed a low-power EEG acquisition circuit with 62.6 µAcurrent consumption. For validation, EEG signals were shown bydistinct appearance of 8–12 Hz oscillations (alpha waves) whenwearer’s eyes are closed. EEGWISP can record EEG signals at63 Hz sampling rate, at the distances up to 0.80 m and with 0.1% data loss. With slight modifications, our system can be usedfor other biopotential signals such as ECG.

I. INTRODUCTION

There is a need for a wireless and wearable device toallow long-term ambulatory monitoring of electroencephalo-gram (EEGs), where brain’s electrical activity is measuredwith electrodes placed on the skull [1]. Continuously worn,EEG based, brain-computer interfaces (BCIs) could be used byparalyzed patients to control an external device with visuallyevoked potentials, which are EEG oscillations, with the samefrequency as a flashing visual stimulus [2]. EEG based BCIcould be used for mental-work load detection to alert distractedor drowsy car drivers [3]. Furthermore, there is a clinical needfor continuos EEG monitoring to alert the patient and health-care providers during the onset of an epileptic seizure [4].

There are a number of wireless EEG devices that are pow-ered by a battery. There typically use proprietary or standardwireless protocol to transmit data to a base station [5]. Suchsystems are bulky and can achieve a battery life of only one ortwo days because of a relatively high data rate required by theEEG. For example, one state-of-the-art wireless EEG system,with comparatively low power consumption, uses a customEEG analog acquisition chip with power consumption of 200µW, but has the total power consumption of 4.3 mA, providingonly 30 hour battery life for continuos monitoring [6]. Wirelesscommunications consume most of this energy, suggesting thatit that it is not possible to make a small system with a longbattery life, even by using a 200 µW EEG acquisition chip.

This paper proposes a novel battery-free EEG system calledEEGWISP, that uses a commercial ultra-high frequency (UHF)radio frequency identification (RFID) for energy harvesting

and communications. Since such system does not need batter-ies it can be made small and can work for as long as a suitableradio frequency (RF) energy source is available.

EEG

analog circuit

DRL

Storage

Capacitor

WISP

ADC

MSP430

Reader

PC

RF 900 MHz

RF Backscatter

Impedance

matching

Power

Harvester

Modulator

Voltage

Regulator

Filters

and

amplifiers

Demodulator

Electrode -

Electrode +

Fig. 1: EEGWISP system diagram.

Figure 1 is a block diagram of the EEGWISP, which webuild by adding a custom low-power EEG acquisition circuitto the Wireless Identification and Sensing Platform (WISP),detailed in [7]. The EEGWISP transmits EEG data in a 12-byte EPC (Electronic Product Code). We address the problemsof RF interference and low EEG data integrity by duty cyclingthe reader.

Although we are not aware of any previous work onEEG systems that use RFID for power and communications,similar spaces have been explored, especially for implantableapplications. For example, neuralWISP uses RFID with WISPplatform for energy harvesting and communications to in-vasively record neuronal spike counts inside a brain [8]. Asimilar lightweight RF-powered system that uses backscatterwas developed to collect neuronal activity in flying insects [9].Alternatively, other battery-free energy sources have beenexplored, such as a EEG system that uses thermoelectricenergy from body heat [10].

This paper is structured as follows. Section II and III areabout the EEGWISP’s design, with Section II specifically dis-cussing the EEG acquisition circuit and Section III discussinghow we interfaced the EEG acquisition circuit with RFID.Section IV presents a validation study and studies how thedistance from the reader’s antenna affects the EEGWISP’soperation. Section V provides a conclusion and discussespossible applications. Section VI discusses future work.

IA1 µF

1 MΩ -

+OA

10 kΩ

Vdd/2

Vdd/2

-

+OA

0.1 µF

53.6 kΩ

5.36 kΩTo ADC

To electrode -

To electrode +

2 kΩ

Trimmer

-

+

OA1 MΩ

1 MΩ

100 kΩ

Trimmer

Vdd

-

+

OA

1 MΩ

1 MΩ

Vdd/2

-OA -

OA

5 kΩ

5 kΩ

1000 pF

1000 pF

100 kΩ

10 kΩ To DRL

Electrode

Vdd/2

+

+

+

-

1 2

3

4

5

Fig. 2: Full schematic of the EEG acquisition circuit. Highlighted parts are: 1) High-pass filter 2) Gain stage 3) Low-pass filterwith offset 4) Driven Right Leg (DRL) 5) Reference voltage that is half of the supply voltage (1.8 V).

II. EEG ACQUISITION CIRCUIT

A. Design Requirements

Because EEG signals are typically 1 to 100 µV in am-plitude, sufficient voltage gain is 1000–10,000 V/V (60 -80 dB) [11]. EEG frequencies can go up to 150 Hz, butbecause of the WISP’s limited RAM size, we target the 1to 30 Hz range, which is sufficient for many BCI and medicalapplications. For example the alpha waves, used for validationin this paper are in the 8 to 12 Hz range, and visually evokedpotentials based BCI demonstrated in [2] uses 6 to 14 Hzfrequency range. Additionally, the EEG analog acquisitioncircuit should be sufficiently low power to run from theharvested RF energy. Table I summarizes the EEG acquisitioncircuit’s specifications.

TABLE I: Specifications of the EEG acquisition circuit

IA CMRR 110 dB [12]

IA Input impedance 100 GΩ [12]

Supply voltage 1.8 V

Average current 62.6 µA

IA noise (0.1 to 10 Hz RTI) 1 µVpk−pk [12]

Gain range 1000 - 10,000 V/V (60 - 80 dB)

Low-pass corner 30 Hz

High-pass corner 0.16 Hz

B. Instrumentation Amplifier

The instrumentation amplifier (IA) is directly connected tothe differential electrodes. Because 50 or 60 Hz noise fromthe power lines can saturate the EEG signals, the IA needsmore than 100 dB of common-mode rejection ratio (CMRR)

to attenuate power-line noise. Furthermore, human skin hashigh impedance on the order of tens of kΩ, so the IA requiresmuch higher input impedance to avoid attenuation of theEEG signal. The IA must also have low 1/f noise, since 1/fnoise is inversely proportional to the frequency; 1/f noise ismost severe at the low EEG frequencies. We used a TexasInstruments INA333 IA in our design. Although it is notthe lowest power IA on the market, with 50 µA quiescentcurrent [12], the amplifier offers better CMRR and 1/f noisecharacteristics than lower-power IAs with 40 µA quiescentcurrent such as the Analog Devices AD8236 [13] and theTexas Instruments INA322 [14]. We set the gain of the IAto 11 V/V (21 dB).

C. EEG Filters, Gain, References and DRL

We used the Texas Instruments MCP6044 for all operationalamplifiers (op-amps) because the MCP6044 has quiescentcurrent of 0.6 µA [15], one of the lowest on the market.The amplifier has a suitable gain–bandwidth product of 14kHz, which allows gains up to 470 V/V (53.4 dB) at 30 Hzbandwidth.

Some individual parts of the schematic in Figure 2 arenumbered and are described in detail below.

1) A passive RC high-pass filter with cut-off frequency of0.16 Hz eliminates the DC offset associated with theelectrodes, which could otherwise saturate the circuit.

2) A gain stage allows fine tuning of the gain with atrimmer from 1 to 470 V/V (0 to 53.3 dB), which allowsmatching the amplitude of the amplified EEG signal tothe analog-to-digital converter (ADC) range.

3) An antialiasing filter is a first order active filter witha fixed gain of 10 V/V (20dB). We chose a cut-off

frequency of 30 Hz to satisfy the Nyquist Theorem,because it is half of the analog-to-digital converter’ssampling frequency. We used an adjustable trimmerconnected to the non-inverting op-amp input to eliminatethe offset voltage.

4) We used a driven right leg circuit (DRL) as an activeground for the electrodes. The DRL has its name becausehistorically it was attached to the right leg in electrocar-diography (ECG). In this system, as is typically donewith the EEG, the DRL is attached to the earlobe. TheDRL actively rejects power line noise by implementinga feedback loop that samples the common-mode voltageand injects the current into the ground electrode.

5) We generated a reference voltage (Vdd/2) with a 1 MΩvoltage divider (the high resistance reduces leakagecurrent). To avoid loading, the voltage dividers werefollowed by an op-amp buffer.

Fig. 3: Oscilloscope reading, illustrating duty cycling of thereader: a) The output of the EEG acquistion circuit with a testsine wave. Enlarged plot shows corruption of the test signal,due to RF interference from the reader b) The EEGWISP’sstorage capacitor voltage changes as the reader is duty cycled.

III. INTERFACING THE EEG ACQUISITION CIRCUIT ANDRFID

A. Wireless Identification and Sensing Platform (WISP)

The EEGWISP is based on the WISP [7], a programmableplatform based on the Texas Instruments MSP430 microcon-troller. The WISP uses UHF RFID for communications andpower harvesting, and it uses the standard EPC Gen2 protocoloperating in the 902–928 MHz industrial-scientific-medical(ISM) band. The WISP responds to a query from the readerwith a 12-byte electronic product code (EPC), in which oursystem places the EEG data. WISP-to-reader communicationsare digital backscatter via Frequency Shift Keying (FSK)

modulation of the WISP’s antenna impedance. We used anImpinj Speedway RFID reader to interface with the EEGWISP,with the reader’s transmit power set to +30 dBm and a 6-dBicircularly polarized patch antenna attached.

B. Duty Cycling of RFID Reader

As Figure 3a shows, when the reader is on, the EEG signalsbecome corrupted because of strong RF interference, so weduty cycled the reader and collected EEG signals only whenthe reader was off. We designed the following scheme: thereader stayed on for 10 seconds to charge the EEGWISP’sstorage capacitor and allow the EEGWISP to backscatter thebuffered EEG data; then the reader remained off for 3.5seconds for the EEGWISP to collect EEG data and bufferit to RAM memory. The duty cycling is best illustrated by theEEGWISP’s storage capacitor voltage, shown in Figure 3b.

The 3.5 seconds of off time consists of 0.8 seconds to allowthe EEG acquisition circuit to settle and 2.7 seconds of datacollection. Because the microcontroller has 512 bytes of RAM,it could buffer 175 EEG data points, which equates to 2.7seconds of data at a 63 Hz sampling rate. During this 3.5second interval, the EEGWISP is powered fully by its storagecapacitor.

We chose the 10-second interval experimentally, based onthe time required to charge the EEGWISP’s storage capacitorat 0.8 m distance from the reader. With the current setup, 0.8 mis the farthest distance at which the reader provides enough RFenergy to charge the storage capacitor. At the typical data rateof 0.65 kbps, the EEGWISP needs 4.3 seconds to transmit thearray of buffered EEG data. Note that the data rate includingthe overhead is 7.8 kbps.

We used the following formula to calculate the duty cycleof the EEGWISP:

Tsample

Tsample + Ton + Tsettle=

2.7sec

2.7sec+ 10sec+ 0.8sec= 20%

(1)where Tsample is the interval (in seconds) at which EEG

data is collected, Ton is the time for which the reader is on,and Tsettle is the settling time of the EEG acquisition circuit.

The EEGWISP uses a watchdog timer to fire interruptsthat trigger ADC sampling. The timer is accurate because itruns from an external crystal oscillator. The EEGWISP uses acounter to determine when it needs to sample and store EEGdata. Watchdog timer interrupts always increment the counter,but RF activity resets the counter to zero. ADC samplingoccurs only when the counter reaches 50, indicating that therehas been no RF activity for 50 interrupts (0.8 seconds).

C. Storage Capacitor Sizing

The EEGWISP needs power from a storage capacitor for3.5 seconds with 240 µA average current. We determined thisvalue by averaging the current when the reader was off during5 minutes of EEGWISP’s normal operation. We measured thecurrent with an Agilent U1233A multimeter.

(a) Top of the EEG acquisitionboard with a U.S. Quarter.

(b) Possible usage scenario. EEGWISP is ina white plastic box. Black wires attach tothe DRL and two electrodes, placed underthe white headband.

(c) Bottom of the EEG acquisi-tion board. Two orange rectanglesare the storage capacitors.

(d) EEG acquisition board mounted on the WISP with a U.S.Quarter placed for comparison.

Fig. 4: Components of the EEGWISP

Assuming that the WISP’s power harvester charges thecapacitor to Vc(0) = 5.6 V, and the capacitor is dischargedto the microcontroller’s minimal operating voltage Vc(t) = 1.8V, we calculated the minimum capacitor size as follows:

C =I × t

Vc(0) − Vc(t)=

240µA× 3.5sec

5.6V − 1.8V= 216µF (2)

In practice, higher capacitance is necessary to compensatefor capacitor leakage, and to avoid fully discharging thecapacitor, which will erase the contents of the WISP’s RAM.We used two low-ESR tantalum 6.5 V capacitors in parallelto achieve 320 µF.

0x0F 664 564 467 0x01 0x410002

Data (n) Data (n-1) Data (n-2) CounterSensor

Type Device ID

0x0F 706 664 564 0x02 0x410002

0x0F 609 706 664 0x03 0x410002

EPC

Fig. 5: Example contents of three consecutive 12-byte EPCs.Three redundant samples are highlighted in orange. Data isshown in decimal and other information is in hexademical.Note that the counter always increase by 1 in consecutiveEPCs.

D. Using EPC for Data Transmission

Since EEGWISP operates in a constrained and uncertainenergy environment, sometimes it cannot collect enough RFenergy for a burst to backscatter an EPC. If the EPC protocolis used to transmit the EEG data to the reader withoutmodification, about 10% of the data is lost, the data pointsare often duplicated and received out of order. Tagging eachEPC with a unique counter identifier allows missing EPCs tobe easily identified; we calculated data loss by dividing thenumber of missing counters by the number of counters thatwas sent in each transmission cycle.

Taking advantage of the low energy cost of backscatter com-munications, EEGWISP transmits redundant data to reduce thedata loss from 10% to 0.1%. As Figure 5 shows, each standard12-byte EPC contains three 2-byte data samples: the presentsample (n), the previous sample (n−1), and the sample beforethat (n− 2). In all, each data point is transmitted three times,and the chance of losing a data point is spread over threeindependent EPCs. In addition, each EPC contains a uniquecounter-based identifier, which the receiving PC uses to putdata points in their original order, to eliminate redundant data,and to estimate data loss. We performed data reconstructionin MATLAB, but in the future the reader-side PC can performthis work automatically.

The downside of the EEGWISP’s redundant RF communi-cation scheme is its high data overhead of 92%. A typical solu-

tion to reduce the overhead is to transmit an acknowledgmentfor every EPC, and retransmit the EPC if an acknowledgmentis not received. We avoid this scheme because it would affectthe time the reader needs to stay on to receive the buffer,thereby desynchronizing the reader and the EEGWISP.

E. Circuit Board and Enclosure

As shown in the Figure 4, we designed the 4-layer FR4 EEGacquisition board to fit on the WISP as a daughterboard. Weplaced the device in a 3D printed plastic case, and the insideof the case is lined with a copper tape for RF shielding.

IV. EXPERIMENTAL RESULTS

A. Range from the Reader

To validate the data integrity of the EEGWISP’s EEGreadings, we connected a signal generator with a 2 Hz sinewave, instead of electrodes, through a voltage divider (1.8Vpk−pk to 90 µVpk−pk) to simulate a known EEG signal.We moved the reader’s antenna between 0.1 m and 1 mfrom the EEGWISP in 0.1 m increments, and recorded fiveminutes of data at each interval. The placement of the reader’santenna and the EEGWISP are shown in Figure 6. To make theexperimental conditions as realistic as possible, we conductedour experiments in an office environment and used an antennaas opposed to an attenuator.

As seen in Figure 7, we observed a 20% EEG duty cyclefor distances up to and including 0.5 m. At 0.6 m, 0.7 m and0.8 m, the EEG duty cycle was reduced by half (10% dutycycle); the EEGWISP had only enough energy to collect dataevery other cycle. At 0.9 m and further, we were not able tocollect data because there was not enough harvested energy topower the microcontroller for the 3.5 second interval.

The range of 0.8 m is not inherent, and could be increasedwith an energy-management scheme. The reader could de-cide when to turn on or off, based on the voltage of thestorage capacitors, which the EEGWISP could report. Themicrocontroller firmware could be further optimized for lowerpower consumption by running the watchdog timer only duringsampling interval.

The theoretical maximum range is dictated by the distancefrom the reader at which the storage capacitors can be charged;up to this point there is more incoming RF energy thantotal leakage current. To find the maximum distance at whichthe storage capacitors could be fully charged, we moved theEEGWISP away from the reader while the microcontrollerwas in a sleep mode. We found that RF harvesting couldfully charge the storage capacitors in 32 seconds at 2.6 metersfrom the reader. We calculated the duty cycle to be 7.6%.In practice, the duty cycle at 2.6 meters would be lower,because this calculation does not take into an account theenergy required to backscatter the EPC.

B. Alpha Waves

Alpha waves are 8-12 Hz oscillations in the occipital lobe,located at the back of the skull. There is a significant increasein the amplitude of alpha waves when the eyes are closed.

Fig. 6: Setup used in the range experiments.

Fig. 7: The duty cycle of the EEGWISP at certain distancesaway from the RFID reader

We connected a volunteer to the EEGWISP with silver/silver-chloride (Ag/AgCl) electrodes in locations O2 and Fp2 accord-ing to the 10-20 International System of Electrode Placement,as shown in Figure 8a. We used Ten20 EEG paste from Weaverand Company to decrease the skin–electrode impedance. Asshown in Figure 4b, we used a headband to hold the electrodesin place and connected the DRL electrode to an earlobe.Note that we did not place the EEGWISP directly on theheadband, as shown in Figure 4b; instead it was 0.10 m fromthe volunteer and the EEGWISP was 0.3 m from the reader’santenna. We measured the electrode’s impedance to be 11 kΩand set the gain to 6820 (77 dB). We collected data for 5minutes with eyes closed and 5 minutes with eyes open, thenperformed data analysis in MATLAB.

As shown in Figure 8b and 8c, with eyes closed there is anotable increase in alpha waves, confirmed in both frequencyand time domains for the 2.7 second interval. Because of dutycycling 24 such intervals were collected for open and closedeyes during the 5 minutes of data collection. The average ofthose intervals is shown in Figure 9, demonstrating the sameincrease in alpha waves, when eyes are closed, as a single 2.7second interval. These results indicate that the EEGWISP is

(a) Locations of the electrodes, asseen from the top of the head.

(b) Raw data (c) Frequency content of the raw data in (b)

Fig. 8: Comparison of two 2.7 sec data intervals with eyes open and closed, collected by the EEGWISP.

able to measure EEG signals, and that the 2.7 second timewindow is appropriate for analysis of the alpha waves.

V. CONCLUSION

In this paper we demonstrated that a UHF RFID platformcan be used to collect EEG signals. To our knowledge we arethe first to demonstrate this property. We overcame the maindesign challenges of making a low power EEG acquisitionboard, avoiding RF interference and achieving high dataquality. Detailed system specifications appear in Table II.

The EEGWISP could be attractive for long-term biopoten-tial monitoring such as in BCI or clinical applications, mainlybecause it does not need a battery. It could be worn for anextended time, miniaturized into a band-aid form factor, anddeployed to users with minimal maintenance burden. By in-creasing the duty cycle, decreasing the gain, and adjusting thefilters, the EEGWISP could be used for electrocardiography(ECG or EKG), which measures electrical activity of the heart.

One disadvantage of the RFID based approach is that itrequires an RFID infrastructure, but many places already havethis infrastructure for another purpose, such as the ComputerScience and Engineering building at the University of Wash-ington. The EEGWISP is not as robust as its battery-poweredcounterparts because of the uncertain energy supply, so it ismost useful in situations where the subject is stationary, suchas confined to a bed or a driver’s seat.

VI. FUTURE WORK

User studies will be conducted to better understand thereliability, comfort, and value of the EEGWISP. In this workwe used a standard dipole antenna that was already on theWISP and have not looked into antenna placement and design,which will ultimately determine usability and comfort of thesystem. We will take a closer look at how an antenna can beincorporated into clothing such as a hat or a headband.

Fig. 9: Average spectral density for 5 minutes of EEG datawith eyes open and closed (24 total intervals of 2.7 second)

TABLE II: EEGWISP Specifications.

Resolution 10 bit - 1.76 mV/bit

Sampling frequency 63 Hz

EEG bandwidth 0.16 - 30 Hz

No. of EEG channels 1

EEG duty cycle 20% ( ≤0.5 m); 10% (0.5 m - 0.8 m)

Theoretical max. range 2.6 m

Protocol EPC Class 1 Gen2 RFID

Frequency UHF: 902 - 928 MHz

Battery life Unlimited, with RF energy

Data loss 0.1 %

Backscatter bit rate 0.65 kbit/sec (kbps)

We plan to design system on chip (SoC) EEGWISP tosignificantly improve the performance, such as increase dis-tance from the RFID reader, reduce the size, increase thecommunications rate, and reduce the noise from the readerby using fully differential EEG front-end.

ACKNOWLEDGMENT

The research was supported in part by National ScienceFoundation (NSF) under grant no. 1256082. and in part byNSF grant EEC-1028725. We thank B. Waters, F. Darvas, Y.Zhao, A. Parks, A. Sample, and B. Ransford for the adviceand technical support.

REFERENCES

[1] B. Blankertz, et. al., ”Non-medical uses of BCI Technology,” Frontiersin Neuroscience, vol. 4, no. 198, Dec. 2008.

[2] M. Cheng, X. Gao. S. Gao, D. Xu, ”Design and Implementation of aBrain-Computer Interface With High Transfer Rates,” IEEE Transactionson Biomedical Engineering, vol. 49, no. 10, Oct. 2002.

[3] J. Kohlmorgen, et. al., ”Improving human performance in a real op-erating environment through realtime mental workload detection,” inToward BrainComputer Interfacing, G. Dornhege, J. del R. Millan, T.Hinterberger, D. McFarland, and K.-R. Muller, Eds. Cambridge, MA:MIT Press, 2007, pp. 409-422.

[4] N. Verma, A. Shoeb, J. Bohorquez, J. Dawson, J. Guttag, A. P.Chandrakasan, ”A Micro-Power EEG Acquisition SoC With IntegratedFeature Extraction Processor for a Chronic Seizure Detection System,”IEEE Journal of Solid State Circuits, vol. 45, no. 4, Apr. 2010.

[5] C. Lin, et. al., ”Review of Wireless and Wearable Electroencephalogramsystems and Brain Computer Interfaces,” Gerontology, vol. 56, pp. 112-119.

[6] L. Brown, J. Molengraft, R. F. Yazicioglu, T. Torfs, J. Penders, C. V.Hoof, ”A Low-Power, Wireless, 8-Channel EEG Monitoring Headset,”in Engineering in Medicine and Biology Society (EMBC), 2010 AnnualInternational Conference of the IEEE, pp. 41974200, Aug. 2010.

[7] A. P. Sample, D. W. Yeager, P. S. Powledge, A. V. Mamishev, J. R.Smith, ”Design of an RFID-Based Battery-Free Programmable SensingPlatform,” IEEE Transactions of Instrumentation and Measurement, vol.57, no. 11, pp. 2608-2615, Nov 2008.

[8] D. J. Yeager, J. Holleman, R. Prasad, J. R. Smith, B. P. Otis, ”Neural-WISP: A Wirelessly Powered Neural Interface With 1-m Rang,” IEEETransactions on Biomedical Circuits and Systems, vol. 3, no. 6, Dec.2009.

[9] S. J. Thomas, R. R. Harrison, A. Leonardo, M. S. Reynolds, ”A Battery-Free Multi-Channel Digital Neural/EMG Telemetry System for FlyingInsects,” IEEE BioCAS, pp. 229-232, Nov. 2011

[10] T. Torfs, V. Leonov, R. F. Yazicioglu, R. J. M. Vullers, B. Gyselinckx,”Wearable Autonomous Wireless Electro-encephalography System FullyPowered by Human Body Heat,” IEEE Sensors, pp.1269-1272, Oct.2008.

[11] J. G. Webster, Medical Instrumentation: Application and Design, 2nded. Boston, MA: Houghton Mifflin, 1992.

[12] Texas Instruments, ”Micro-Power (50uA), Zero-Drift, Rail-to-Rail OutInstrumentation Amplifier,” INA333 Datasheet, Oct. 2008.

[13] Analog Devices, ”40 uA Micropower Instrumentation Amplifier withZero Crossover Distortion,” AD8236 Datasheet, May. 2009.

[14] Texas Instruments, ”microPower Singe-Supply CMOS InstrumentationAmplifier,” INA322 Datasheet, Feb. 2006.

[15] Texas Instruments, 600nA, Rail-to-Rail Input/Output Op Amps,MCP6041/2/3/4 datasheet, Feb. 2008.