This paper has been accepted for publication by 2017

IEEE 18th Workshop on Control and Modeling for Power

Electronics (COMPEL).

Personal use is permitted, but republication/redistribution

requires IEEE permission.

DOI: 10.1109/COMPEL.2017.8013316

Citation: S. Talukder, S. Park, and J. Rivas-Davila, "A

portable electrostatic precipitator to reduce respiratory

death in rural environments," in 2017 IEEE 18th Workshop

on Control and Modeling for Power Electronics

(COMPEL), Jul. 2017.

IEEE Xplore URL:

http://ieeexplore.ieee.org/document/8013316/

A Portable Electrostatic Precipitator to ReduceRespiratory Death in Rural Environments

Sabera Talukder1,2, Sanghyeon Park1 & Juan Rivas-Davila1

Department of Electrical Engineering1 & Biology2

Stanford UniversityStanford, CA 94305

Email: sabera@stanford.edu

Abstract—This paper presents the design methodology of aminiaturized, portable, and low power electrostatic precipitator(ESP) to reduce indoor air pollution (IAP) from rural cookstoves.Decreasing the concentration of aerosol particles helps reducethe incidence of respiratory tract infections that can lead todisease and even death. We trace existing technologies andstandards to combat IAP in cookstoves, design an ESP thatallows precipitation at lower voltages via an electrode designthat utilizes sub 5 kV voltages, and explore how rapidly pulsedconverters decrease the power consumption of the circuit.

I. INTRODUCTION

Fig. 1: Indoor air pollution in Nepal [1].

It is estimated that 4.3 million people die every year fromillnesses attributable to cooking fume IAP [2], Fig. 1. IAPis characterized by many particle sizes, but many malignantparticles are sub 10 µm in diameter because they can penetratedeep into the lungs and alveoli [3], Fig. 2. In order to addressthe cookstove emissions of sub 10 µm particles, the GlobalAlliance for Clean Cookstoves (GACC) and InternationalOrganization for Standardization (ISO) characterized emissionbased tier requirements for clean cookstoves. The various tierlevels and emission requirements, based on high power andlow power systems, can be found in Table I. Among thevarious cookstove classifications, tier 4 is the best performingand requires emissions less than 1 mg/min/L for 2.5 µm

Fig. 2: Particles less than 10 µm in diameter enter our lungs.Image taken from [3].

TABLE I: Cookstove tiers based on 2.5 µm particle emissions,where tier 4 is the best performing. The unit [mg/MJd] denotesmilligrams per megajoule delivered to the pot. We will onlybe focusing on the low power column because that is whereour ESP will operate. Data taken from [4].

High power PM2.5 Low power PM2.5

[mg/MJd] [mg/min/L]

Tier 0 > 979 > 8Tier 1 ≤ 979 ≤ 8Tier 2 ≤ 386 ≤ 4Tier 3 ≤ 168 ≤ 2Tier 4 ≤ 41 ≤ 1

diameter particles [4]. In the developing world, childhoodpneumonia is unfortunately tightly correlated with the amountof cook-stove smoke emitted, Fig. 3. Childhood pneumoniacauses around one million deaths per year in children underthe age of five [5], and demonstrates the deadly consequencesof inhaling smoke particles.

Although ESPs are primarily used to reduce pollution emis-978-1-5090-1815-4/16/$31.00 c©2017 IEEE

×1

×2

×3

0 100 200 300

Ris

k o

f C

hil

d P

neu

mo

nia

PM2.5 Exposures from Indoor Emissions (g/m³)

WHO air quality

annual guideline:

10 g/m³

WHO air quality

annual interim target 1:

35 g/m³

Tier 4 Tier 3 Tier 2 Tier 1

Fig. 3: Risk of childhood pneumonia linked to the cookstovetier of 2.5 µm particle emissions. Reproduced from [5].

sions in industrial settings [6], ESPs once optimized, could bea viable low power solution to address the problem of IAP indeveloping countries. ESPs establish a large electric field, andas particles flow through this field they are negatively chargedon the electrode, and then via electrostatic forces, collect onthe discharge surface.

We will discuss the ESP mechanical design, and an ESPcircuit design which operates in the 2.7 kV to 3.7 kV range,namely a 6.78 MHz class-Φ2 inverter and a class DE-rectifier.Our goal is to explore a design methodology for emissionsreducing ESPs by generating corona at low voltages viaour electrode design, and evaluate a high frequency circuittopology that can power the ESP.

II. MECHANICAL DESIGN

The three main ways to structurally modify an ESP areexplained in subsections II-A, II-B, and II-C. Keeping theseparameters of (1) distance between the electrode and dischargesurface, (2) shape of the discharge surface, and (3) shape ofthe electrode in mind we tried multiple different types of ESPconfigurations: hexagonal, linear, and cylindrical. All of theESP designs were approximately 200 mm tall to ensure thatthe resulting system would be lightweight and portable in ruralenvironments. Our final cylindrical design was able to achievecorona and precipitation at relatively lower voltages becauseof our electrode design.

A. Distance between the Electrode and the Discharge Surface

In our electrode design, we used 11 mm separation betweenthe electrode and the discharge surface. This distance betweenthe electrode and discharge surface was meant to maintaina small cylindrical ESP that can be compatible with smallrural cooking stoves. Moreover, smaller separations create astronger corona that seems to enhance the collection of finerparticles [7].

(a)

(b)

Fig. 4: (a) The electrode has length = 20 cm and diameter= 33 mm. (b)The discharge electrode is a frayed wire halfcylinder.

B. Shape of the Discharge Surface

The discharge surface area needs to be greater than theelectrode area because with our ESP configuration the smokecollects on the discharge surface. Our discharge surface is ahalf cylinder with a diameter of 57 mm, Fig. 4b. In additionto the larger surface area requirement, the discharge surfaceand electrode must be far enough away so that arcing doesnot occur, but close enough so that precipitation occurs.

C. Shape of the Electrode

Our literature review indicated that the best electrode shapeis barbed metal and smaller barb distances are optimal forprecipitating finer particles [7]. For efficient precipitation ofsmall particles with low voltages, we created a frayed electrodeusing litz wire 675/48 AWG, Fig. 4a and Fig. 4b. We achievesignificant particle precipitation in the 2.7 kV to 3.7 kVthanks to our electrode design. In addition, because of safetyconsiderations we decided to design a voltage converter thatoperated within this range of voltages.

III. DC RESONANT HIGH VOLTAGE CONVERTER

A. 6.78 MHz, 3.7 kV, 7-stage Class-DE Rectifier Design

The 6.78 MHz rectifier, introduced in [8], Fig. 5, was usedto creates a 3.7 kV dc output for the ESP. The circuit com-prises seven stages of a resonant class-DE rectifier connectedstacked in series. Since CB<t,b><1−7> act as dc blocks atthe frequency of operation, all the diode junction capacitancesCj,<t,b><1−7> appear in parallel with the resonant inductorLr and the extra capacitance Cextra to constitute a parallelLC tank. When the ac input vo,inv resonates the LC tank,

vo,inv Lr Cextra

RESP,eq

iinLs

Cj,b1

Cj,t1CB,t1

(Co1)Dt1Db1

CB,b1

CB,b8

Cj,b2

Cj,t2CB,t2

(Co2)Dt2Db2

CB,b2

Cj,b7

Cj,t7CB,t7

(Co7)Dt7Db7

CB,b7

vout

+

node X

Rrect,eq

Fig. 5: 3.7 kV 6.78 MHz 7-stage class-DE resonant rectifier.

Fig. 6: Simulated rectifier output voltage vout waveform thatreaches a steady state at 3.8 kV after 0.8 ms of operation.

the voltage of node X is amplified and subsequently rectifiedby diodes D<t,b><1−7>. The rectified output voltages add inseries and drive the load RESP,eq that represents the ESPdevice. A single class DE rectifier stage needs a discreteoutput capacitance to maintain the output voltage constantduring operation. But when stacked, the blocking capacitorsCB<b><2−7> can also take the role of output capacitors, as isevidently shown in Fig. 5 The elimination of output capacitorsCo<1−7> improves the transient response of the system since,and significantly reduces the part count and total energystored in the circuit, as there are fewer capacitors to charge.Specifically, removing the output capacitors works becausefor any n-th rectifier stage, CB,bn and CB,b(n+1) are alreadyproviding a low impedance path for high frequency ripplecurrents. The simulated transient waveform output voltagevout, Fig. 6, settles to 3.7 kV after 1 ms with a 80 Vpp

6.78 MHz input.

TABLE II: 3.7 kV, 6.78 MHz, 7-stage class-DE rectifiercomponent values.

Component(s) Valuevo,inv 80 Vpp, 6.78 MHzLs 2590 nHLr 360 nH

Cextra 1.5 nFCB<t,b><1−7> 1 nFD<t,b><1−7> STPSC406B

RESP,eq 680 MΩ

0 V

2 V

4 V

6 V

8 V

10 V

1

(bottom)

2 3 4 5 6 7

(top)

Pea

k-t

o-p

eak

Vo

ltag

e R

ipp

le

Individual Rectifier Stage

Cout = 1 nF

Cout = 0 nF

Fig. 7: The simulated output voltage ripple for each stage withand without output capacitors.

Table II lists component values that are used in this rectifier.The charge stored in this un-optimized iteration of the rectifieris 14.8µC, which is still less than 45µC (maximum amount ofcharge allowed in the body [9]). This charge can be loweredby reducing the 1 nF dc blocking capacitors CB<t,b><1−7>,at the cost of a slower vout transient.

The simulated voltage ripple for each rectifier stage withand without the output capacitors, Co<1−7> in Fig. 5, can beseen in Fig. 7.

Each stage nominally produces 540 V and the diodes’voltage rating is 600 V (although breakdown really occursat 650 V), therefore, as long as the each stage’s ripplevoltage does not exceed 110 V (650 V - 540 V) the circuit’sfunctionality will be maintained. Since the maximum ripplevoltage seen without output capacitors is roughly 9 V, for stage1 and 7 in Fig. 7, we can safely remove the output capacitorsfrom our circuit. Removing the output capacitors leads toa somewhat faster charging and discharging time constant,which can be seen in Fig. 8. With the output capacitors thesimulated converging time is 0.9 ms versus 0.8 ms.

In addition, as the corona current during precipitation isvery low, we found that it was convenient to pulse the powersupply of the precipitator with relatively low duty cycles. Atthe repetition rates evaluated, the electrode voltage does notdecay to zero. This suggests that we can easily reduce thepower rating of the design rectifier in a future iteration, oradjust the repetition rate to adjust for various stove sizes andsmoke flows.

Fig. 8: The cyan curve is the simulated output voltage with1 nF output cap on each stage, and the red curve is withoutthe output capacitors.

Vin LMRCextra

CB,inv

Rrect,eq

iinLF

CMRSignal gen.

Ls

Q

vo,inv

Fig. 9: 6.78 MHz 36 V-to-80 Vpp class-Φ2 resonant inverter.

B. 6.78 MHz Class-Φ2 Inverter Design

In the 6.78 MHz, 36 V input class-Φ2 inverter [10], Fig. 9,LMR and CMR work as a second harmonics trap and, whenproperly tuned, result in a trapezoidal waveform at the tran-sistor drain node. Fig. 10 shows the 80 Vpp output voltagewaveform. Table III lists the inverter component values.

C. DC-DC Converter Implementation

We built a 36 V-to-3.7 kV dc-dc converter to drive theESP. The rectifier generates a 3.7 kV output, shown in thered curve, Fig. 11c. The converter consumes 700 mW power,but to reduce power consumption, we pulsed the circuit bydriving the inverter for 500 switching cycles out of a 10 msperiod (equivalent to 0.74 % on-off duty ratio). During the offperiod, the output voltage drops from 3.7 kV to 2.7 kV beforethe next pulsing cycle begins. In addition to reducing the powerconsumption pulsing enables the surface potential to decreaseduring the voltage off period which improves the collectionefficiency of high resistivity dusts [11] while improving smoketo flow through the ESP. The design in Fig. 11 is advantageouswhen the output power is larger than several watts and thussystem efficiency is important. On the other hand, where theoutput power is low and efficiency is not a major concern,conventional approaches such as Cockcroft-Walton generatorsor high-voltage flyback converters are effective.

Fig. 10: Simulated waveform of the inverter output voltagevo,inv that drives the rectifier.

TABLE III: 6.78 MHz class-Φ2 inverter component values.

Component(s) ValueVin 36 V

duty cycle 0.39LF 90 nHLMR 390 nHCMR 350 pFCextra 1700 pFCB,inv 4700 pF

Ls 2590 nHRrect,eq 50 Ω

Q FDMC86116LZ

IV. EXPERIMENTAL RESULTS

A. Indoor Test

The ISO does not select a single laboratory protocol to de-termine cookstove performance [4], therefore our experimentalset up consisted of a fan running at 5 V blowing a glycerinand water smoke mixture through the gap between the ESPelectrode and the discharge plate, Fig. 12b and Fig. 12c. Inaddition, the full resonant converter connected to the ESPelectrode and the discharge plate can be seen in Fig. 12a.Once the fan was on we switched the ESP device on and offfor 1 min periods for 9 min and recorded the 2.5 µm and 10µm concentrations from the particle meter every 5 sec.

The temporal changes in PM2.5 and PM10 levels during theexperiment swing from the sensor saturation (999.9 µg/m3)down to lower than 50 µg/m3. The swings have latency onthe fall and rise for both 2.5 µm and 10 µm particles. 2.5 µmrising latency: 17.63 s, 2.5 µm falling latency: 18.74 s, 10 µmrising latency: 8.68 s, 10 µm falling latency: 10.23 s. Whentaking these latencies into account, the particle concentrationsare in sync with the ESP switching cycle, Fig. 13a.

Fig. 13b shows the rectifier output voltage waveform thatdrives the ESP. The voltage reaches 3.7 kV when the converteroperates, and decays to 2.7 kV during the converter off periodbefore the converter turns on again. Fig. 13c compares thevertical view of the ESP device when it is off (Fig. 13cleft) and on (Fig. 13c right). The difference in the amountof smoke escaping the ESP device indicates its effectivenessat removing aerosol particles. This is corroborated by the data;when accounting for the latency times, the ESP reduces 2.5 µm

(a)

(b)

(c)

Fig. 11: The 36 V-to-3.7 kV, 6.78 MHz resonant dc-dc con-verter has a total power consumption of 700 mW. (a) The class-Φ2 inverter [10] converts 36 V dc input to 60 Vpp ac output.(b) The seven-stage-stacked class-DE rectifiers generates aDC output of 540 V each and 3.7 kV in total. (c) Voltagewaveforms consisting of: 3.7 kV DC output voltage, red curveat the top in 500 V/div scale; 6.78 MHz 60 Vpp SC outputvoltage from the inverter, blue curve and the bottom curve atthe bottom in 50 V/div scale.

(a)

(b)

ESP device

Particlemeter

Rectifier

Inverter

Smoke in

Smoke out sampling

tube

(c)

Fig. 12: Experimental setup to test the ESPs effectiveness.(a) The 6.78 MHz, 36 VDC-to-3.7 KVDC resonant converterdriving the ESP electrode and discharge plate. (b) The ex-perimental setup design. (c) The experimental setup showingthe sampling tube and the smoke flow direction (bright greenarrows). For demonstration purposes, the ESP is a half cylinderattached on a glass pane.

particle emission to 2.5 µg/m3/min, and reduces the 10 µmparticle emission to 5.68 µg/m3/min.

B. Outdoor Test

To reiterate the ISO does not select a single laboratoryprotocol to determine cookstove performance [4], thereforeour experimental set up consisted of the design shown inFig.14a and Fig.14b. This experiment was very similar to theindoor tests except real fire wood was used for the smoke. Theexperiments all used approximately 1.0g of wood and 1.0g ofpaper for the fire. In addition, in the outdoor experiments theESP was either turned on for the duration of the experiment orturned off for the duration of the experiment; in other wordsthe outdoor experiments did not pulse the ESP like the indoorexperiments did.

Fig. 15a shows the particulate matter levels in µg/m3 versustime in seconds. The orange traces correspond to when theESP was off and the fire was burning, while the black tracescorrespond to when the ESP was on and the fire was burning.In both traces two peaks are present. The orange dashed tracecorresponds to 2.5 µm particles while the solid orange tracecorresponds to 10 µm particles. Both of these traces rise fairlyrapidly and then stay high (with the 10 µm particle trace evenmaxing out the particle meter) and then eventually falling. Theblack dotted line corresponds to 2.5 µm particles while thesolid black trace corresponds to 10 µm particles. The averagevalues of particulate matter levels were as follows: ESP on2.5 µm (dotted black) = 56 µg/m3, ESP on 10 µm (solid black)= 60 µg/m3, ESP off 2.5 µm (dashed orange) = 228 µg/m3,ESP off 10 µm (solid orange) = 270 µg/m3. This data showsthat the average particulate matter values are lower when theESP is turned on demonstrating that particles from typical firecan be collected by the ESP.

Fig. 15b shows the integral of the curves in Fig. 15a plusthe Biolite stove which is one of the stoves listed on theGACC website. The Biolite stove is a product that attemptsto reduce the amount of smoke produced by using a fan toblow air onto the fire to improve combustion of materials andtherefore reduce the amount of smoke produced. This productdemonstrates that there is a desire in the marketplace to combatthe issue of IAP. Based on the total mass of particulate matter,an ESP with only one cylinder is roughly the same as theBiolite. This demonstrates that there is promise in lightweightand portable ESP technology.

V. CONCLUSION

In this paper we explored a design methodology for emis-sions reducing ESPs that are miniaturized, portable, and canreduce respiratory disease deaths. These goals were addressedby our electrode design that reduced precipitation voltage,and pulsed system free of output capacitors that reducedpower consumption and balanced the need for smoke flow andcollection of fine particles. The experiments were performedwith half and a full cylinder ESPs and are on par with otherGACC listed technologies for battling IAP. In conclusion, theminiaturized ESP is an attempt to create a portable smoke

(a)

(b)

(c)

Fig. 13: Experimental results. (a) The particle meter readoutsof PM2.5 and PM10 particle concentration levels. The lightregion indicates that the ESP is on. (b) Measured waveformsof the inverter drain voltage (green curve at the top; 50 V/div),inverter input current (yellow curve in the middle; 1 A/div),and the rectifier output voltage (red curve at the bottom;1 kV/div). (c) Visual demonstration of the ESP in operation;In the left picture the ESP is on, in the right picture the ESPis off.

(a) (b) (c)

Fig. 14: Experimental setup for the outdoor test. (a) The schematic of the setup. The fire is made with roughly 1.0 g of woodand 1.0 g of paper and is surrounded by a large metal encasement to keep smoke encased. The ESP sits on a hole on top ofthe metal encasement so that all the smoke generated by the fire flows through the ESP. The chimney and the sampling tubedirect a fraction of the smoke to the particle meter. (b) The picture of the setup. (c) The waveform of the dc-dc converteroutput voltage that is applied to the ESP. We operate the circuit in a burst mode by running it for 500 cycles every 5 msperiod, thereby reducing the total power consumption down to 2 W. The voltage reaches 3.5 kV during the circuit operationand drops to 2.1 kV before the next bursting cycle begins (voltages that appear on the figure are scaled by half due to thelimit on the oscilloscope attenuation ratio).

0

200

400

600

800

1000

0 30 60 90 120 150 180 210 240 270

Par

ticu

late

Mat

ter

Lev

els

[

g/m

3]

Time [s]

PM 2.5

PM 10

PM 2.5, ESP ON

PM 10, ESP ON

(a)

0

20

40

60

80

Biolite fan ON ESP ON Both OFF

To

tal

Mas

s o

f P

arti

cula

te

Mat

ter

[arb

itra

ry u

nit

] PM 2.5

PM 10

(b)

Fig. 15: Experimental results of the outdoor test. (a) Fig. 15ashows the particulate matter levels in µg/m3 versus time inseconds. (b) Fig. 15b shows the integral of the curves inFig. 15a plus the Biolite stove.

filter that can reduce the amount of particles that enter ourlungs. The hope is that by reducing the amount of smoke thatenters our lungs fewer people will fall victim to preventablerespiratory diseases.

REFERENCES

[1] Worldbank. Indoor air pollution in nepal. [Online]. Available:http://factsreports.revues.org/docannexe/image/3851/img-1.jpg

[2] WHO, “Household air pollution and health,” World Health OrganizationFact Sheet, 2012.

[3] T. Hertzberg and P. Blomqvist, “Particles from res - a screening ofcommon materials found in buildings,” FIRE AND MATERIALS, vol. 27,no. N/A, pp. 295 – 314, 2003.

[4] ISO, “Guidelines for evaluating cookstove performance,” The Inter-national Organization for Standardization, during the InternationalWorkshop Agreement, pp. 1 – 20, 2012.

[5] GACC, “Measuring progress during phase i: Building on the iwa interimguidelines,” N/A, vol. N/A, no. N/A, pp. 1 – 6, 2012.

[6] N. Grass, W. Hartmann, and M. Klockner, “Application of different typesof high-voltage supplies on industrial electrostatic precipitators,” IEEETransactions on Industry Applications, vol. 40, no. 6, pp. 1513 – 1520,2004.

[7] J. Miller, B. Hoferer, and A. Schwab, “The impact of corona electrodeconfiguration on electrostatic precipitator performance,” Journal of Elec-trostatic, vol. 44, no. 1, pp. 67–75, 1998.

[8] L. Raymond, W. Liang, L. Gu, and J. R. Davila, “13.56 mhz high voltagemulti-level resonant dc-dc converter,” in 2015 IEEE 16th Workshop onControl and Modeling for Power Electronics (COMPEL), July 2015, pp.1–8.

[9] IEC, “Safety requirements for electrical equipment for measurement,control, and laboratory use - part 1: General requirements,” InternationalElectrotechnical Commission, vol. 3, no. N/A, p. N/A, 2012.

[10] J. M. Rivas, Y. Han, O. Leitermann, A. D. Sagneri, and D. J. Perreault,“A high-frequency resonant inverter topology with low-voltage stress,”IEEE Transactions on Power Electronics, vol. 23, no. 4, pp. 1759–1771,July 2008.

[11] A. Mizuno, “Electrostatic precipitation,” IEEE Transactionson Di-electrics and Electrical Insulation, vol. 7, no. 5, pp. 615 – 624, 2000.