Design of a high order Campbelling mode measurement system using open source hardware

HAL Id: cea-02388644

https://hal-cea.archives-ouvertes.fr/cea-02388644

Submitted on 2 Dec 2019

HAL is a multi-disciplinary open access

archive for the deposit and dissemination of

sci-entific research documents, whether they are

pub-lished or not. The documents may come from

teaching and research institutions in France or

abroad, or from public or private research centers.

L’archive ouverte pluridisciplinaire HAL, est

destinée au dépôt et à la diffusion de documents

scientifiques de niveau recherche, publiés ou non,

émanant des établissements d’enseignement et de

recherche français ou étrangers, des laboratoires

publics ou privés.

Design of a high order Campbelling mode measurement

system using open source hardware

G. De Izarra, Zs. Elter, C. Jammes

To cite this version:

G. De Izarra, Zs. Elter, C. Jammes. Design of a high order Campbelling mode measurement

sys-tem using open source hardware. Nuclear Instruments and Methods in Physics Research Section

A: Accelerators, Spectrometers, Detectors and Associated Equipment, Elsevier, 2016, 839, pp.12-22.

�10.1016/j.nima.2016.09.038�. �cea-02388644�

system using open sour e hardware G. deIzarra a , Zs. Elter b, , C.Jammes b a

CEA,DEN,DER,ExperimentalProgramsLaboratory, Cadara he, F-13108Saint-Paul-lez-Duran e,Fran e

b

CEA,DEN,DER,Instrumentation,SensorsandDosimetryLaboratory,Cadara he, F-13108Saint-Paul-lez-Duran e,Fran e

ChalmersUniversityofTe hnology, Departmentof Physi s,DivisionofSubatomi and PlasmaPhysi s,

SE-41296Göteborg,Sweden

Abstra t

Thispaperreviews a new, real-time measurement instrument dedi atedfor online neutron monitoring in nu lear rea tors. The instrument implements thehigher order Campbelling methods and self-monitoring ssion hamber apabilities on an open sour e development board. The board in ludes an CPU/FPGASystem OnaChip.

Thefeasibilityofthemeasurementinstrumentwastestedbothin labora-torywithasignalgeneratorandintheMinerverea tor. Itisshownthatthe instrument provides reliable and robust ount rate estimation over a wide rea torpowerrange.

The ssion hamber failure dete tion ability is also veried, the system is able to identify whether the measured ount rate hange is due to the malfun tionofthedete tororduetothe hangeofneutronux. Theapplied method is based on the hange of the frequen y dependen e of the ssion hamber signalpowerspe tral density, due to the malfun tion. During the experimental veri ation, the onsideredmalfun tion was the hange of the polarization voltage.

Keywords: Highorder Campbelling, FPGA, Measurement system,Count rateestimation

The re ent development of Fren h Sodium- ooled Fast Rea tor (SFR) 2

indu es the need for new neutron ux monitoring systems in order to en-3

han ethesafetyfeatures. Giventhe ongurationofthepooltypeSFR,the 4

neutroninstrumentationisplannedtobesetupintherea torvesselinorder 5

to monitor the neutron ux over a few orders of magnitude [1 ℄. The high 6

temperature ssion hambers arethe best andidates for this purposesin e 7

thetypi altemperature inthevi inityofthe oreisaround500

o

C[2 ℄. This 8

dete tor type was extensively studied in the CEA during the nineties and 9

is still undera tive development. In addition, a resear h eort is urrently 10

done inorderto exploit thewide ux rangemonitoring apabilityof ssion 11

hambers: it was previously shown that theuse of high order Campbelling 12

mode (HOC) provides a linear estimation of the neutron ux over a wide 13

rangeof ount rate bysuppressingtheimpa t ofparasiti noises. 14

Inthisframework,aHOCmeasurementsystemprototypewasdeveloped 15

at the Instrumentation,Sensorsand Dosimetry Laboratory (LDCI)of CEA 16

Cadara he. The main goal of the work is to assess the feasibility and the 17

industrial use of su h measurement system. To omplete the measurement 18

devi e, a ssion hamber malfun tion dete tion module (referred later as 19

"smart-dete tor" apability) wasin luded. 20

In this paper, the implementation of HOC method on an open sour e 21

hardware development boardis detailed. First, the general theory of HOC 22

is dis ussed briey and the most important advantages of its appli ation 23

are highlighted. The theoreti al basis of the ssion hamber malfun tion 24

dete tion is also reviewed. Se ond, the design of the Campbell measure-25

ment system is presented: the preferen e of using a open sour e hardware 26

withaCPU/FPGA hipat theprototypephasearesummarized; TheHOC 27

omputation algorithm isexplained throughdiagrams and lo king gures; 28

The software dedi ated to ontrol the FPGA module is introdu ed. The 29

lastpartofthe paperisdedi atedto thevalidationof themeasurement sys-30

tem through laboratorymeasurements and in- ore experimental ampaigns 31

performedat theMinerve rea tor. 32

2. Theoreti al ba kground 33

2.1. Highorder Campbelling theoreti al ba kground 34

Campbell derived a theorem [3℄ that links the intensity of a shot noise 35

pro ess onsisting of general pulses

f (t)

with an amplitude distribution to 36

andits omplete derivation hasbeen performedin[4 , 5℄: 38

s

0

=

κ

n

hx

n

i

R

f

n

(t)dt

=

k

n

C

n

.

(1)

where

κ

n

is the

n

th order umulant of the signal,

s

0

is the ount rate and 39

hx

n

i

stands for the

n

th order raw moment of the amplitude distribution. 40

Commonly, the methods in whi h

n ≥ 3

are alled higher order methods. 41

Eq. (1) shows that if the pulse shape and the amplitude distribution are 42

known (therefore the alibration oe ient

C

n

is determined), and the u-43

mulant (of any order) of the signal is measured, then the mean ount rate 44

s

0

ofthe signal an be estimated. 45

In the urrent work, the estimation of the umulants is performed by 46

applying unbiased umulant estimators, alled k-statisti s [6℄. If the mea-47

suredsignal onsistsof

N

dis retely sampledvalues

Y

i

,thenthethird order 48

umulant estimatoris givenas: 49

k

3

=

2S

3

1

− 3N S

1

S

2

+ N

2

S

3

N (N − 1)(N − 2)

,

(2)

where

S

1

,

S

2

,

S

3

aretherst,se ondandthirdordersumsofthedatapoints 50 dened as: 51

S

n

=

N

X

i=1

Y

n

i

(3)

Theperforman e ofhigh order Campbelling methods have been intensively 52

studied in [7℄; it was shown that the appli ation of higher order methods 53

su iently suppresses the impa t of various noises. The linearity of the 54

ount rate estimation over a wide ount rate range (from

10

4

to

10

9

ps) 55

have been veried both numeri ally and experimentally. It was highlighted 56

that the appli ation of higher than third order methods do not bring any 57

pra ti aladvantageanda urate ountrateestimation anbea hievedwith 58

the thirdorder Campbellingbased onsignalsamples of afew ms. 59

2.2. Smartdete tor theoreti al ba kground 60

Eq. (1)showsthat any umulant of thesignalmay hange not onlydue 61

to the hange in the ount rate, but also due to the hange in the mean 62

pulseshape or inthe amplitudedistribution. The higherorder methods are 63

parti ularly sensitive to these hanges be ause of the higher exponents in 64

Eq. (1). The hange of the pulse shape and its amplitude may o ur due 65

toa malfun tion,su hastheredu tionof dete torpressureorvoltagesin e 66

onlythe hangeofthe umulant(thereforethe hangeoftheestimated ount 68

rate)willbedete ted,thuswehavetobeableto de idewhetherthis hange 69

o urred due to the hange of the neutron uxaround the dete tor or due 70

to the malfun tion ofthedete tor. 71

Thereforewehavetodeneameasurablequantityofthession hamber 72

signalwhi h issensitive tothepulseshape hange but notto the ount rate 73

hange. As a previous study shows [? ℄, the width of the power spe tral 74

density(PSD) ofthedete torsignal satisesthisrequirement. In thiswork 75

thePSD ofa signal

y(t)

isdened as: 76

P SD(f ) =

F T (y)F T

∗

(y)

T

m

(4)

whereFTstandsfortheFouriertransform,and

T

m

isthemeasurementtime 77

It was shown that by measuring the width of the PSD, one an dete t 78

the hange of the pulse shape due to the leakage of the lling gas. The 79

spe tral widthwasdened as the width at the half maximum of the PSD, 80

it is usually ontained in the 0-20MHz band, whi h is an easily a essible 81

frequen yband withthemoderninstrumentation. 82

3. Design of high order Campbelling measurement system 83

Thedevelopmentof anon-lineneutron monitoring system,whi hmakes 84

use of ssion hamber signals and works in higher order Campbell mode, 85

requires: 86

ˆ Capabilityto onvertandpro essthe signalattheoutputofthe avail-87

ablepre-amplier(between-10and10Vforthetypi alnu lear instru-88

mentation). 89

ˆ Real-time omputation oftherst, se ondand third ordersumof the 90

signal(see Eq.(3) ). 91

ˆ High sampling frequen y in order to resolve the signal onsisting of 92

pulseswithawidthof afew tensof nanose onds. 93

ˆ Ability to pro ess a large amount of data in real-time (given that a 94

timewindowofa fewms hasto be applied for a urate estimations). 95

Sin ethisworkaimsto developaprototypesystemandprovide proofof 96

on ept,two additional requirementshave to be onsidered: 97

mentations. 99

ˆ Largeuser ommunityand preferablyopensour e philosophyinorder 100

tofa ilitate thelearningand to getfastte hni al support. 101

3.1. Hardware sele tion 102

Re ently several single board omputers and system on a hip (SoC) 103

boardsrevolutionizedandfa ilitatedthedevelopmentofdigitalmeasurement 104

instruments. Basedon theabovedened riteria, theRed-Pitaya board was 105

hosen; itwas reatedto providea ustomizablemeasurementsystemwitha 106

generousamountof examplesa ordingtotheopensour e philosophy. Due 107

to itslow footprint, lowpri e andrelatively large user ommunity,it fullls 108

allthe requirementswhi h wereexpe tedat theprototype stage. 109

The Red-Pitaya board is built around a Xilinx Zyn 7010 SoC whi h 110

embedsanFPGAandadual oreArmCPU lo ked at668MHz. The Red-111

PitayaboardhoststwoAnalog-to-DigitalConverters(ADC)andtwo Digital-112

to-AnalogConverters(DAC)whi haredire tly onne tedtotheFPGA.The 113

ADCshaveasamplingfrequen yof125MHzandaresolutionof14bits. The 114

board provides two measurement ranges through jumper positions:

±0.6

V 115

and

±16

V. The great strength of the FPGA is letting to design a ir uit, 116

whi h allows to pro ess the data in line, therefore redu ing the time, and 117

the memory storage for heavy omputations. This makes FPGAs ideal to 118

perform simple omputing patterns on a vast amount of data in real time. 119

These hara teristi s fulll all theabove stated requirements to develop an 120

online neutron monitoring system: the board is able to pro ess the signal 121

atoutputof themost ommon pre-amplier(0,10V fortheCEAPADFand 122

-10,0VfortheCanberraADSpre-ampliersapplied inthiswork),toresolve 123

thession hamberpulses(with a lengthof fewtens ofns), and to perform 124

thereal-timepro essing oflarge amount of data. 125

It hasto be mentioned, that although itmay seem that theCPU ould 126

handleallthedatapro essingneededto al ulatethehighordersumsofthe 127

signal,the real time omputation withtheCPU ouldn't be realizeddue to 128

thehuge amount ofdatatransferand operations(

≈

750[Mop/s℄) itimplies. 129

Therefore,in the neutron monitoring instrument, theFPGAwas dedi ated 130

to the lowlevel, time riti al,redundant operations (namely omputingthe 131

powersums ofthesignalvalues),whereasmore ompli atedoperationswere 132

performedon the CPU. 133

TheFPGAoftheboardis onguredusingVerilog,alowlevelHardware 134

Des riptionLanguage(HDL),andthesoftwaresrunningontheCPUare de-135

veloped in C language. It has to be re ognized that development with a 136

edurallanguage(themain onstraintsaredetailedinthefollowingse tion). 138

Thus,duringthedesignoftheinstrumenttheuseofthelowlevel omputing 139

waskepttoareasonableminimum,inordertofa ilitatethedevelopmentand 140

the maintainability of the devi e while keeping CPU omputing power for 141

prospe tive data pro essing. The following se tions aredevoted to explain 142

theinnerdesign of thenewmeasurement system. 143

3.2. Third order umulant measurementsystem 144

The most time onsuming operations, while omputing the third order 145

umulant estimator

k

3

(Eq.2),arethe omputationsofthesums

S

1

,

S

2

and 146

S

3

(Eq. 3), sin e the higher order power of ea h signal sample is required 147

in real-time. On the other hand, the further operations to ompute the 148

estimator

k

3

based on the sums, has to be done only on e at the end of 149

themeasurementtimeanddon'texhibitanysimple omputationalpatterns. 150

Consequently, the real-time omputation of the sum terms were realized 151

on the FPGA, and, after the transfer of the sums to the CPU, the nal 152

operations to ompute the umulant estimator wereperformedbya ontrol 153

software runningon the CPU. 154

ThedevelopedFPGAmoduleofthemeasurementsystemis omposedof 155

vealgorithmi blo ks. Twoblo ksarededi atedtoreadandwritedatafor 156

thePro essing System (PS,i.e. the omputer partof theSOC). One blo k 157

is in ontrol of themeasurement soft reset. Two blo ks aredire tly related 158

to omputation of thethird order umulant estimator terms. The fun tion 159

andthe realisation of thelasttwo blo ks aredetailedbelow. 160

3.2.1. Unbiased umulant estimation FPGA module 161

Thealgorithmdesignstartswiththedenitionofregisters(variablesused 162

to store data): S1 , S2 and S3 ontain respe tively the sum of single, square 163

and ubi power of the samples for N number of samples. It is favorable to 164

limitthenumberofsamplestoapowerof2tosimplifydivisionbyNinto bit 165

shifting operation. It was shown previously that the proper measurement 166

time for the umulant was between a hundred of mi ro-se onds and a few 167

tens of millise ond [7℄, in terms of number of samples, this orresponds to 168

N

between

2

22

(33.5 ms) and

2

14

(131

µ

s)if thesampling frequen y is 125 169

MHz;Register Nsize wasset to 23bits. 170

Thesizesof S1,S2 andS3have tobe arefully dened aswell, sin etheir 171

sizeshave to be adequately large in order to avoid overows. The develop-172

ment boardprovides two's omplement signed 14 bits samples, an addition 173

of two samples hasto be stored on 15 bits, where as their multipli ation is 174

hasto be36 bits wide be ause of themaximumvalue of N,whi h is

2

22

. S2 176

andS3mustbe50and 64bitswide inorderto ontain respe tively thesum 177

ofsquareand ubi powerof samples. 178

On e ea h used register has the proper size, the algorithm an be de-179

signed. Two important onstraintshave tobetakeninto a ount at thelow 180

level omputing. First,everyoperationinanalgorithmi blo kisperformed 181

inparallelduring a lo kti k; theresults oftheoperations willbeavailable 182

at the end of the lo k ti k whi h prohibits the use of regular pro edural 183

programming. However, bran hing(i.e. onditionaltests)doesnot onsume 184

any omputation time. Se ond, only one operand an be used inan opera-185

tion; it impliesto pipeline the omputation of square and ubi powerover 186

several lo kti ks. 187

Fig. 1 summarizes the implemented algorithm: the main algorithmi 188

blo k ontainsaloopdedi atedtothe omputationofthesums. Toprovidea 189

betterunderstandingofthetimeoperationofthealgorithm,Fig. 2illustrates 190

is lo king diagramfor

N = 4

. 191

Atea h lo kti k,thedatastream omingfromtheADC(ad _a )isstored 192

into two temporary registers (single ontains the ADC value and double

←

֓

193

ontainsthesquareoftheADCvalue),whi hareusedtopipelinethepower 194

omputation. Inthesametime,ad _a isaddedto S1 ,doubleto S2andtriple 195

toS3 (triple isalso atemporary register dedi atedto storethe ubi power 196

ofsamples, itisfed withtheresultof themultipli ation single*double ). 197

A sample ounter (sample ) isin remented at ea h y le to keep tra k of 198

theamountofsamplespro essedsin ethestartofthe urrent measurement. 199

Whensample isequal toN ,the ompletion pro essand thetransferofS1,S2, 200

S3 to registers a essible by thePS (namely S1mem , S2mem , S3mem ) begins. To 201

transferthe sumstothe areafromwhi hthedata transfertowardstheCPU 202

takespla e,intermediateregisters(namelyS1inter ,S2inter ,S3inter )haveto 203

beusedto storethesums duetothepipelining. Thereasonisthatthenal 204

valuesofS2andS3willbeavailablerespe tivelyoneandtwo lo kti ksafter 205

S1 ,however,thedatatransferhastobedone inone lo k y leinordernot 206

to mixoldand newdatainregisterswhere theCPU hasa ess. 207

To a hieve the transfer, rst the ag data_transfer responsible for data 208

transferfromintermediateregistersto thememory areaa essiblebyCPUs 209

is set to false. In the same lo k ti k when the sample is equal to

N

, the 210

ag spe ifying the availability of S1 (S1ready ) is set to true, while the ags 211

indi atingtheavailabilityofS2andS3 (S2readyand S3ready )aresetto false . 212

In the next lo k ti k, a test S1ready==true allows S1 to be transferred 213

to the intermediate register S1inter , while the omputation of S1 restarts. 214

ti k. During the following ti k, a test S2ready ==true allows S2 to be opied 216

to S2inter and its omputation restarts, while S3ready isalso set to true . In 217

the next lo k ti k, the test S3ready==true is veried and the opy of S3

←

֓

218

to S3interm isperformed;a register whi hkeepstra kof measurement time 219

(time_stamp) isalso in remented and theag data_transfer is reset to true

←

֓

220

. All the sums an now be opied from intermediate registers to the ones 221

a essible by the PS (the algorithmi blo k responsible for this transfer is 222

summarizedinFig.3)),whiletheregisters ontaining thesums,S1 ,S2 ,S3are 223

alreadylled withthedataof thenewmeasurement. 224

3.2.2. Control software of the High Order Campbelling module 225

A software waswritten inahighlevelprogramming language to ontrol 226

theFPGAmodule. Itsmaintaskistoreadthedataprovided bytheFPGA 227

at the designated memory addresses, to onstru t the estimator

k3

and to 228

make itavailable forfurther pro essing. Atthis phase, two onstraints have 229

to be onsidered to fulll the requirements of real-time monitoring: rst, 230

the software must be apable to onstru t the estimator in a time window 231

shorter than the duration of the measurement without being slowed down 232

bythepost-pro essingofdata. Se ond,itshouldnotmissanymeasurement 233

andshouldstore ea h resultprodu ed inthememory forfurtherpro essing. 234

Inordertofullltheserequirements,thesoftwaredesignusestwothreads, 235

whi h allow to take advantage of thedual ore ar hite ture. Onetransfers 236

the available measurement into the memory of the PS: it reads the time 237

stamp omputedbytheFPGA,andifthistimeislargerthan theonestored 238

in omputer memory,

S1

,

S2

and

S3

are transferred and

k3

estimation is 239

onstru ted. These ond thread isresponsiblefor heavy and slow pro esses 240

su h as printing and saving the umulant estimator. It has only a ess to 241

the data provided by the measurement thread. In order not to lose data, a 242

FIFO (First In First Out) pile an be used by the measurement thread to 243

storethetermsofthe umulant. Boththreadsaredetailedthroughdiagrams 244

available inFig. 4and Fig. 5. 245

Figure1: Thealgorithm for the omputation ofthe terms ofthe estimator

k3

:

S1

,

S2

and

S3

.

For the sake of simpli ity, the number of samples per measurement is set to

N = 4

. Thesquares represent thestateofregistersat ea hti k whilethearrows summarizethe operation performed during the ti k. Bubbles at the topof the diagram are related to agsfornishingthe omputationandinitiatingmemorytransfer.

Themeasurementprototypewas ompletedwithsmartdete tor apabil-247

itiesinordertodete t the hange inthewidthofthepowerspe traldensity, 248

whi h indi ates a possible malfun tion duringoperation. 249

As it was shown previously [8 ℄, it is satisfa tory to monitor the power 250

spe tral density on the s time s ale. In order to minimize the ne essary 251

hange intheFPGAmodulesdevelopedfor thehigherorder umulant om-252

putations, inthe smartdete tor module, theFPGA ( ontrolled by thePS) 253

isresponsibleonlyforre ordingtherawdata. The omplexdatapro essing, 254

su has omputingthePSD and determiningits width,isdone on theCPU 255

withspe i C routinesfromtheGSL (GNUS ienti Library) [9 ℄. 256

3.3.1. Smart dete tor FPGA module 257

The hardware part of the smart dete tor module onsist of a pun tual 258

rawdatare orderwhi hmakesuseoftheblo kRAMavailableontheFPGA; 259

the Artix-7 have 60 blo ks of 36 kB RAM whi h an have a limited set of 260

onguration [10 ℄: Sin e the data oming from the ADC are 14 bits wide, 261

only

2

16

data points(0.524 ms) an be stored inthememory. 262

TheFPGAmoduleis onstru tedfromthreealgorithmi blo ksavailable 263

in Fig. 6: the rst one stores ADC data into a memory buer, while the 264

se ondandthirdonedealrespe tively withdatatransfertothePSandwith 265

messagedispat hing throughthe smartdete tor module. 266

For the sake of simpli ity, it was de ided to use the memory transfer 267

algorithm in luded in the Red-Pitaya proje t and a serialised ar hite ture 268

to move databuer fromthe FPGA to the PS: The PS sends thememory 269

address to be read and the memory transfer ode blo k makes ready the 270

relateddata(buff [yraw_read_addr℄) for aneventual read ommand. 271

Even with a limited signal length and slow memory transfer (

≈

12.5 272

Mdata/s),this ar hite ture is suitable for thession hamber failure dete -273

tion. Nevertheless,inthefuture, itispossibletoimprove thismodule: sin e 274

the granularityofmemory transferis32bits, transferringmorethan one14 275

bitsofuseful dataina lo kti k ould speedupthetransfertothePSbya 276

fa tor2. 277

3.3.2. Control software of the smart dete tor module 278

The ontrol softwareofthe ampbell measurementsystemwasextended 279

inorder toin lude the smartdete tor apabilities. Thegeneral ar hite ture 280

remains the same: two threads are running on the two ores of the CPU. 281

One is dedi ated to the measurement and only does lightweight pro essing 282

while the other is related to heavy data pro essing (as shown in Fig.7). A 283

the smartdete tormodule. 285

The measurement thread was adapted to transfer signal segments: one 286

agissharedwiththedatapro essingthread,psd_ ompute_flag isusedto no-287

tifytheFPGAthatthesignalbuerhastobelled. Then,themeasurement 288

thread he ks periodi ally ifthe data buer is ready. On e it is ready, the 289

datais transferred. When a omplete signal segment has been transferred, 290

the measurement threadindi ates itwith psd_ ompute_flag. 291

Thedatapro essingthread isin hargeof PSD omputation. Whenthe 292

buer datais ready to be pro essed, the thread pro eeds to the PSD om-293

putation after requestinga newsignal segment. Thespe trumis omputed 294

using the Bartlett's method. When the the amount of omputed spe tra 295

rea hes MAX_PSD, the average spe trum is omputed, and the width of the 296

spe trumisestimated. Finally,the mean spe trumissaved on thedisk. 297

4. Experimental validation 298

The measurement system prototype was tested through several experi-299

ments. During the development phase, experiments inlaboratory were per-300

formedto he kwhethertheFPGAmodules arewellimplemented. Finally, 301

the measurement system was onne ted to ssion hambers and tested in 302

theMinerve rea tor underreal working onditions. The measurementsand 303

the obtained resultsare detailedinthe following se tions. 304

4.1. Laboratory validation 305

4.1.1. Validationof the HOC measurementsystem 306

The validation of theHOC measurement systemwas done intwo steps. 307

Inthe rststep, theproper omputation andtransferof

S1

,

S2

and

S3

was 308

he ked by repla ing theADC input databy a onstant value of 2. It was 309

veried that for N samples, the omputed sums

S

1

,

S

2

and

S

3

are equal to 310

2N

,

4N

and

8N

respe tively. 311

In the se ond step, thea ura yof thethird order umulant estimation 312

was tested with Poisson pulse trains simulated by a pulse train generator. 313

The pulse trains onsisted of exponential damped pulses with a width of 314

around 100 ns and a random normally distributed amplitude. The ount 315

rates were varying between

4 · 10

5

to

4 · 10

7

/s. The pulse trains were 316

loaded (inthe properformat) and played bya Tektronix AWG5012 signal 317

generator. Thedatasetswere0.128se ondlong,withasamplingtimeof8ns 318

andthepulseamplitudesele tedtobe onsistentwiththeoutputofatypi al 319

ssion hambermeasurement hain(3%ofthe

±16

V rangewereused). For 320

surementthread ommuni ateswiththeFPGAsmartdete tormodule,eveniftherequest foranewmeasurmentissentthroughthesharedvariablepsd_ ompute_flagbythedata pro essingthread.)

uploading to thesignal generatorand ompared with theestimation of the 322

measurement system. Theresults aresummarized intheTable 1. 323

The measuredthird order umulants are lose to the omputed ones. A 324

maximum of 3.5 %of relative overestimation was found during these tests. 325

Thedis repan yismostprobablyduetotheele troni stransferfun tionand 326

thetrun ation made by theADC.Investigations hasshownthat the trans-327

ferfun tion results a slight reshaping, thereforethe measured estimation is 328

slightly higher thenthe omputed umulant. Nevertheless, inpra ti al sit-329

uations the alibration methodology will inherently take into a ount the 330

reshaping, hen ethe ountrate will not be overestimated. 331

. rate( /s)

k3

( omputed)

k3

a

(measurement)

k3

b

(measurement)

µ

s 4

·10

6

2.84

·10

−

5

(2.90

±

0.01)

·10

−

5

(2.91

±

0.19)

·10

−

5

4

·10

7

5.70

·10

−

5

(5.85

±

0.02)

·10

−

5

(5.86

±

0.36)

·10

−

5

4

·10

6

3.22

·10

−

5

(3.33

±

0.02)

·10

−

5

(3.37

±

0.27)

·10

−

5

4

·10

5

1.15

·10

−

5

(1.19

±

0.02)

·10

−

5

(1.27

±

0.27)

·10

−

5

Table1: Computedandmeasuredthirdorderestimatorsforpulsetrainsatvarious ount rates.

k3

a

referstotheestimationbasedon34mssamplesand

k3

b

referstotheestimation basedon262

µ

ssamples.

4.1.2. Smart dete tor module validation 332

In order to test the smart dete tor module, and thePSD measurement 333

apabilities of thesystem, several pulse trains witha length of 0.128s were 334

simulated,andplayedwiththesignalgenerator. Thetrains ontained Gaus-335

sian shaped pulses with a mean ount rate of

10

6

/s. The width (i.e. the 336

standarddeviation of the Gaussian)ofthe pulseswere hanged (5 ns, 10ns 337

and15 ns were onsidered). 338

The obtained power spe tral densities are available in Fig.8. As it an 339

be seen, the spe trum shape is hara teristi of the pulse shape. The line 340

entered around 1 MHz and its harmoni s are artifa ts due to thefa t the 341

same 0.128 s signal was played periodi ally. The fun tioning of the smart 342

dete tormodulewasappropriate. 343

4.2. In rea tor validation 344

Finally,themeasurementdevi ewastestedduringanexperimental am-345

paignat the Minerve fa ilityof CEA[11 ℄. 346

Several setups were realized in order to assess the ompatibility of the 347

devi e with the standard nu lear instrumentation. During the ampaign, 348

two pre-ampliers were used: the ADS, manufa tured by Canberra and 349

the PADF designed by the CEA instrumentation and ele troni s labora-350

tory. Thesepre-ampliers are dierent in their output voltage and transfer 351

fun tion. 352

In order to over a wide ount rate range, two type of ssion hambers 353

weretestedduringtheexperiments: theCFUL01(arelativelylarge hamber, 354

whi h ontains 1g of U235; its pulse shape is around 80 ns wide) and the 355

CFUR(arathersmall hamber,whi h ontains10

µg

ofU235;itspulseshape 356

isaround20nswide). Atthesameneutronux,theCFUR hamberresults 357

5orders of magnitudelower ount rate than theCFUL01. 358

TheCFUL01 hamberwaslo atedinthesurroundingofthedriverzone, 359

whereastheCFURwasinstalled inthe enterof therea tor. 360

Duringthe ampaign, thefollowing experimentswereperformedinorder 361

to assessvarious aspe tsof themeasurement devi e: 362

ˆ CumulantestimationwiththeCFUL01andthePADFatvariouspower 363

levels: to assess thelinearityofthe measurement system. 364

ˆ Cumulant estimationwiththeCFURandthePADF atvarious power 365

ˆ Pulsetrain re ordingwiththeCFURand thePADFat various power 367

levels: to alibratetheHOCsystem(inordertoretrievethe ountrate 368

with the higher order method), and to estimate the ount rate with 369

pulse ounting algorithms. 370

ˆ PSD measurement withthe CFUL01 andthe ADSpre-amplier with 371

variousbiasvoltages: tosimulateadete torfailureandtomeasurethe 372

hange of the spe tralwidth. 373

4.2.1. Third order umulantmeasurements, CFUL01/PADF 374

Thethirdorder umulant hasbeenestimatedat rea tor powersbetween 375

10Wand 80Wwith the CFUL01, based on 33mstime windows. Both the 376

±0.6

V and the

±16

V input ranges have been used, in order to assess the 377

linearitywithbothranges. 378

The signal saturates at 30 W rea tor power, when measured with low 379

voltagerange. Therefore,onlytwo measurementsweredone withthis range 380

(at10 Wand 20 W).The umulant overpower ratiosare

(4.04 ± 0.36) 10

6

381 (a.u.).W

−1

and

(4.11 ± 0.28) 10

6

(a.u.).W

−1

for the10Wand 20Wpower, 382

respe tively. Although, in the future a better resolution of the power is 383

neededto drawdeeper on lusion, the good agreement oftheratios implies 384

thatthe behavioris linear. 385

Withthe

±16

Vrangethewholepowerrangewas overed. Theobtained 386

umulant estimations arepresentedinFig.9. The measuredthird order u-387

mulantshowslinearitywiththerea tor power. Thedeparturefromlinearity 388

is lower than 1.6 %, whi h is the result of the random error of the estima-389

tion. In orderto estimatethe ount rate, the measurement hain hasto be 390

alibrated. The alibration, through applying the methodology des ribed 391

in[12 ℄ (namely, to evaluate the oe ient

C

n

inEq. (1)by measuring the 392

meanpulseandthepulseamplitudedistributionatlowpower),wasplanned 393

to bedone during the post-pro essingof re orded signalsamples. Unfortu-394

nately,forthis purposethesignalwasre ordedwiththehighvoltagerange, 395

whi h was not appropriate to dis riminate properly the single pulses from 396

thenoise. Inthe future, whenfurther rea tor time anbeobtained for sim-397

ilar measurement purpose, the alibration is going to be repeated with the 398

low voltage range as well. In order to avoid similar problems, also a new 399

alibrationpro edureisunderdevelopment,whi h anbeperformedduring 400

thereal-timeoperation,and doesnot requirepost-pro essing. Nevertheless, 401

the measurement with the CFUL01 was still valuable to assess the linear 402

umulant estimation ofthemeasurement devi e. 403

wn

x









n



n







nD

y

p

2

)

2

4

T

nn  

±042(Tn n0(cnD( p("

l−

%±0d2(−n nc2nD( p

D'p±('f%

)

0

nD

1

(+

p

)

6)))

−)

4

−(6-−)

4

2-−)

4

2(6-−)

4

0-−)

4

./nD"p

)

2)

4)

T)

d)

−))

Figure9: Thirdorder umulantre ordedwiththe

±16

Vrangeasafun tionoftherea tor power.

4.2.2. Third order umulantmeasurements, CFUR/PADF 404

Measurements with the CFUR hamber were performed only with the 405

low voltage range, sin e the ount rate of the signal was expe ted to be 406

rather lowonthe powerrangeof Minerve. 407

Themeasurementsweredone at40Wand80Wpowerlevel. Theresults 408

are summarized in Table 2. The umulant estimation based on one 33ms 409

long sample results a high standard deviation, whi h is expe ted sin e at 410

these ount ratesonlyfewpulsesappearduringone sample,andthenumber 411

of observed pulses is un ertain. Nevertheless, the expe ted value of the 412

umulant estimatorwasbasedon

1900

,33ms longsignalsamples,therefore 413

theoverall deviationof theestimatedmean umulant islessthan

2.3%

. (In 414

omparison,whentherea torwasall ontrol rodsdownandestimation was 415

based only on the noise inthe system, the estimated third order umulant 416

appeared to be

350 ± 50

, whi h is lessthan the deviation of the umulant 417

estimation for the ssion hamber signal). The mean umulant over the 418

powerratios showgoodagreement, whi h implieslinear behaviour. For the 419

estimated ount rates (dis ussedlater),thedeviation referstothe

400 · 0.52

420

mssample, not onlyto one sample. 421

To alibrate the ssion hamber through the methodology presented 422

ounting.

Power(W)

k

3

(a.u.)

hk

3

i/P

(a.u.).W

−

1

HOC ountrate /s Ref. ountrate .s 40 (1.18

±

0.39)

10

4

295

±

12 3430

±

354 (2861

±

118) 80 (2.38

±

0.56)

10

4

297

±

10 6918

±

686 (6103

±

171)

in Ref. [12 ℄, the raw signal re order module dedi ated to smart dete tor 423

wasused, witha minimalist ontrol software. Several signal segments were 424

re orded at 80Wand thepulseswere isolated duringpost-pro essing in or-425

derto determine the alibration oe ient in (1) . From the measurements 426

nearly

2700

pulses wereisolated, whi h allows to have a eptable statisti s. 427

Themean pulse shape and theamplitude distribution of thepulses is illus-428

trated inFig. 10. Asone an see, thedynami ofthe measurement system 429

allows to dis riminate the pulses from the noise and the resolution is ne 430

enough to observe even the urrent boun ing ba k from the able (a small 431

bump following the main pulse). The prototype is apable of working as 432

a raw signal re order as well. The estimated alibration oe ient for the 433

thirdorder is: 434

C

clas

= 3.44 ± 0.3 (a.u.).s/c

(5)

The alibration haslarge un ertainty, whi h showsthe disadvantage ofthis 435

methodology. As it was highlighted in Ref. [12 ℄ aswell, for the high order 436

methods an empiri al alibration may be favorable. Su h alibration isnot 437

plausible for the traditional Campbelling method, due to the linearity gap 438

between pulse ounting methods and these ondorder Campbelling. 439

Inordertoobtainareferen e ountrate, 400signalsegmentsre ordedat 440

area torpowerof80Wand400 re ordedat apower of40 Wwereanalyzed 441

with pulse ounting method as well. The estimated ount rates obtained 442

bypulse ounting, and the ones omputed fromthe alibrated higher order 443

Campbelling are in luded in Table 2. It has to be highlighted again that 444

the standard deviation refersto the

400 · 0.52

ms longsignal sample for the 445

ountingalgorithmresultwhileitisrelatedtothe

1900·33

mslongsignalfor 446

the alibrated HOC. The standard deviation ofan estimation based onone 447

0.52mssampleismu hhigherforthepulse ount,butthegoalwastodene 448

the referen e(i.e. the real) ount rateof the hamber at thesepowers. The 449

goodagreementoftheestimated ountratesshowthattheresultsmeasured 450

71 













u

)

)d

211

1

211

311

411

511

611

T11

  ud

21

s

_2)621

s

_321

s

_3)621

s

_421

s

_4)621

s

_521

s

3l 













l

al

ill

ial

tll

  cm my

l

illl

tlll

dlll

(lll

Figure10: Left: meanpulse shape omputedwithsingle pulses oming from the80W measurementdata. Right: amplitudedistributionofpulses omputedondatasetsre orded at80W.

bythesystemarephysi ally orre t. Theresultsalsoimplythatmonitoring 451

atreallylow ountrates(intheorderof

10

3

ps)ispossiblewithhigherorder 452

Campbelling,butlongermeasurementsarene essary(nevertheless,thesame 453

holdsfor pulse mode measurementsaswell). 454

4.2.3. PSD measurements, CFUL01/ADS 455

When the tests of the smartdete tor module were performed, only the 456

ADSpreamplier were available, whi h allowed to verifythat the devi e is 457

apable towork withother instruments aswell. 458

The ADS pre-amplier and a CFUL01 hamber were used to test the 459

smartdete tor module. Using theCFUL01 wasadvantageous for this pur-460

pose, sin e it has a higher ount rate, therefore its power spe tral density 461

an measuredmore a uratelyduringreal timeoperation. 462

In the urrent experimental work, the hange of the pulse shape was 463

a hieved by hanging the ssion hamber voltage in the saturation regime. 464

The in rease of the voltage has similar ee ts on the pulse width as the 465

de reaseofthe gaspressure, butitissimpler toa hieveduringthe measure-466

ment. 467

Measurements were taken at a onstant rea tor power of20W withthe 468

voltage hangedbetween600Vand850V.Forea happliedvoltage,thePSD 469

was onstru tedbyusing4000datasetsof0.52mslongsignals. Thelowv ari-470

hD−eeD

hD−1e

hD3eeD

hD2ee

hD21e

H











DA

u



um

e

(e

.

a(e

.

.(e

.

)(e

.

1(e

.

−(e

.

x !"#DAs%m

e

1(e

−

_(e

3

_(u1(e

3

_a(e

3

x





D&



'

D

D

'



(D



)







DA

s

%m

3ua(e

−

3u.(e

−

3u)(e

−

3u1(e

−

3u−(e

−

3u3(e

−

*+*!D,*-DAm

−ee

−1e

3ee

31e

2ee

21e

Figure11: Left: PSD as afun tionof the appliedvoltage. Right: Spe tral widthas a fun tionofthepolarisationvoltage.

an eofthespe traestimatedfromthisamountofdata,allowstodistinguish 471

hange inthe spe tralassmall as50kHz. Themeasured PSD and the esti-472

mated spe tral width are presented in Fig.11. The slight os illation of the 473

PSDisanartifa tdueto theapplied able. Asexpe ted, thespe tralwidth 474

in reases with the in rease of the applied voltage, and it saturates at high 475

voltages. Thereasonofthesaturationofthespe tralwidthisthesaturation 476

ofele trondrift velo ityinargon-nitrogen mixturesat highredu edele tri 477

elds[13℄. 478

Therefore, the proof of on ept of the smart dete tor is validated: it 479

is possible to dete t a hange of mean pulse shape from investigating the 480

powerspe traldensity,whereasthemeasurementnoiseandthelowfrequen y 481

lteringof thesystem have negligible inuen e on thedetermination of the 482

spe tralwidth. 483

Although,inthe urrent experiment,thetimeneededto re ordand pro-484

ess the 4000 datasets isapproximately 300s due to theslowdatatransfer, 485

thisalreadyallowstotestinevery 5minuteswhetherthe hamber malfun -486

tions. However, the pro essing time of the smart dete tor prototype ould 487

be redu ed by a fa tor of 3 by using optimised FFT routines, and in the 488

future even faster tests an be a hieved for the industrial appli ation by 489

implementingthesame methodon board withhigherperforman e. 490

Aninnovativemeasurementsystemprototypeforrealtimeneutron mon-492

itoringwas presentedand validatedthrough this paper. Theprototype was 493

built using an open sour e CPU/FPGA devi e with ADC on board. Su h 494

ar hite ture has a several advantages: time riti al, simple operations an 495

be performedon the FPGA, immediately after re ordingthe data, whereas 496

omplexandheavydatapro essing anbeperformedontheCPU.The sim-497

pli ity of the hosen board (Red-Pitaya) allowed fast and straightforward 498

development. 499

The main purpose was to prove the feasibility of a real time neutron 500

uxmonitoring systemusingthe thirdorder Campbell mode. Thismethod 501

suppressesthe impa tofnoiseandprovideswide rangeof operation. Inthis 502

work it was shown thatthe method is even apable to work at ount rates 503

aslowas

10

3

ps. 504

In the work, the on ept of ssion hamber failure dete tion was also 505

in luded. Theself monitoring apabilityof thesystemis basedon dete ted 506

the hange inthe widthofthepower spe traldensityofthe signal. 507

The paper provides detailed des ription of the implemented FPGA al-508

gorithms and the ontrol software running on the CPU. All the hallenges 509

and solutions were highlighted in order to serve as a tutorial for similar 510

developments. 511

Thereliabilityofthe on eptsandtherobustnessofthedevi ewastested 512

through an experimental ampaign at the Minerve rea tor. The linear re-513

sponse and the real time operation of the devi e was veried over a wide 514

powerrange. Through the alibration ofthesystemthe physi al validity of 515

the measured results wasassessed. The self monitoring apability wasalso 516

tested,thesystemis apableto dete tthe hangeinthevoltage setbetween 517

theele trodesof the hamber. 518

Sin e the alibration of the systemis rather elaborate, a simpler, auto-519

mati and realtime alibration pro edureisunder development. 520

For industrial usage, the next step is going to be the implementation 521

of the same on epts on a board whi h has higher performan e inorder to 522

a hieve faster self monitoring apability. 523

A knowledgment 524

ThisstudywaspartlysupportedbytheCEAINSNUandTECNAProje ts 525

andbytheSwedishResear hCoun il(GrantNo.B0774801). Thisstudyalso 526

ofsodium ooledfastrea tors between Chalmersand CEA. 528

[1℄ C.Jammes,N.Chapoutier,P.Filliatre,J.P.Jeannot,F.Jadot,D. Ver-529

rier, A.-C. S holer, B. Bernardin, Neutron ux monitoring system of 530

the Fren h GEN-IVSFR: Assessment of diverse solutions for in-vessel 531

dete tor installation, Nu lear Engineering and Design 270 (2014) 272 532

282. 533

[2℄ C. Jammes,P.Filliatre, B. Geslot,T. Domene h, S.Normand, Assess-534

ment ofthehightemperature ssion hamberte hnology for thefren h 535

fast rea torprogram, IEEE Transa tionson Nu learS ien e 59(2012) 536

13511359. 537

[3℄ N. R. Campbell, V. J. Fran is, A theory of valve and ir uit noise, 538

Journal of the Institution of Ele tri al Engineers-Part III: Radio and 539

Communi ation Engineering. 540

[4℄ I. Lux, A.Baranyai, Higher order ampbellte hniques for neutronux 541

measurement, Nu lear Instruments and Methods in Physi s Resear h 542

202. 543

[5℄ L.Pál, I. Pázsit, Zs.Elter, Comments on thesto hasti hara teristi s 544

ofssion hambersignals,Nu learInstrumentsandMethodsinPhysi s 545

Resear h Se tion A: A elerators, Spe trometers, Dete tors and Asso-546

iated Equipment 763(2014) 4452. 547

[6℄ E.Parzen, Sto hasti Pro esses,Classi sinAppliedMathemati s,1999. 548

[7℄ Zs. Elter, M. Bakkali, C. Jammes, I. Pázsit, Performan e of Higher 549

Order Campbell methods, Part I: review and numeri al onvergen e 550

study, Nu lear Instruments and Methods in Physi s Resear h Se tion 551

A: A elerators, Spe trometers, Dete tors and Asso iated Equipment 552

821 (2016) 66 72. doi:http://dx.doi.org/10.1016/j.nima.201 6.03.0 23. 553

[8℄ Zs.Elter,P.Filliatre,G.deIzarra,I.Pázsit,C.Jammes,Self-monitoring 554

ssion hamber: theoreti al groundwork,in: Physor onferen e, 2016. 555

[9℄ M.Galassi, J.Davies,J.Theiler,B.Gough,al.,GNUS ienti Library 556

Referen e Manual(3rdEd.). 557

[10℄ Xilinx, 7SeriesFPGAs MemoryResour es,2014. 558

physi sexperimentsonzeropowerrea tors, in: D.Ca u i(Ed.), Hand-560

bookof Nu learEngineering, Springer US, 2010,pp.20532184. 561

[12℄ Zs.Elter,G.deIzarra,P.F.C.Jammes,I.Pázsit,Performan eofHigher 562

Order Campbell methods, Part II: alibrationand experimental appli-563

ation,Nu learInstruments&MethodsinPhysi sResear h,Se tionA: 564

A elerators,Spe trometers,Dete tors,andAsso iatedEquipmentBeen 565

submitted. 566

[13℄ G.Haddad,Driftvelo ityofele tronsinnitrogen-argonmixtures,Aust. 567

J.Phys.36 (1983)297303. 568