Many Ways Lead to the Goal—Possibilities of Autonomous and Infrastructure-Based Indoor Positioning

elec onics A icle Many Ways Lead o he Goal—Possibili ies o Au onomous and In as uc u e-Based Indoo Posi ioning Hossein Shoush a i 1,*, Thomas Willemsen 2and Ha ald S e nbe g 1   Ci a ion: Shoush a i, H.; Willemsen, T.; S e nbe g, H. Many Ways Lead o he Goal— Possibili ies o Au onomous and In as uc u e-Based Indoo Posi ioning. Elec onics 2021,10, 397. h ps://doi.o g/10.3390/ elec onics10040397 Academic Edi o : Raed A. Abd-Alhameed Recei ed: 14 Decembe 2020 Accep ed: 2 Feb ua y 2021 Published: 5 Feb ua y 2021 Publishe ’s No e: MDPI s ays neu al wi h ega d o ju isdic ional claims in published maps and ins i u ional a il- ia ions. Copy igh : © 2021 by he au ho s. Licensee MDPI, Basel, Swi ze land. This a icle is an open access a icle dis ibu ed unde he e ms and condi ions o he C ea i e Commons A ibu ion (CC BY) license (h ps:// c ea i ecommons.o g/licenses/by/ 4.0/). 1Ha enCi y Uni e si y, 20457 Hambu g, Ge many; ha ald.s e nbe [email p o ec ed] 2Hochschule Neub andenbu g, 17033 Neub andenbu g, Ge many; [email p o ec ed] *Co espondence: hossein.shoush a i@hcu-hambu g.de Abs ac : The e a e many ways o na iga e in Global Na iga ion Sa elli e Sys em-(GNSS) shaded a eas. Reliable indoo pedes ian na iga ion has been a cen al aim o echnology esea che s in ecen yea s; howe e , he e s ill exis open challenges equi ing e-examina ion and e alua ion. In his pape , a no el da ase is used o e alua e common app oaches o au onomous and in as uc u e- based posi ioning me hods. The au onomous a ian is he mos cos -e ec i e ealiza ion; howe e , ealiza ions using he eal es da a demons a e ha he use o only au onomous solu ions canno always p o ide a obus solu ion. The e o e, co ec ion h ough he use o in as uc u e-based posi ion es ima ion based on sma phone echnology is discussed. This app oach in okes he minimum cos when using exis ing in as uc u e, whe eby Pedes ian Dead Reckoning (PDR) o ms he basis o he au onomous posi ion es ima ion. Realiza ions wi h Pa icle Fil e s (PF) and a opological app oach a e p esen ed and discussed. Floo plans and ou ing g aphs a e used, in his case, o suppo PDR posi ioning. The esul s show ha he posi ioning model loses s abili y a e a gi en pe iod o ime. Fi h Gene a ion (5G) mobile ne wo ks can enable his ea u e, as well as a massi e numbe o use-cases, which would bene i om use posi ion da a. The e o e, a usion concep o PDR and 5G is p esen ed, he bene i o which is demons a ed using he simula ed da a. Subsequen ly, he i s implemen a ion o PDR wi h 5G posi ioning using PF is ca ied ou . Keywo ds: indoo na iga ion; au onomous; in as uc u e; pa icle il e ; pedes ian dead eckoning; usion 5G; ine ial senso s 1. In oduc ion The de elopmen o eliable indoo posi ioning based on Mic o-Elec o-Mechanical Senso s (MEMS) has been a cen al aim o echnology esea che s in ecen yea s. No only a e such senso s cos and ene gy e ec i e, bu hey also co espond o Ma k Weise ’s ision [ 1 ]; in ha hey wo k o e e yone, anywhe e. Thus, he objec i e is o de elop applica ions ha a e as embedded as possible. Indeed, e e y sma phone— ca ied a ound by almos e e yone oday—can be coun ed as a unique sou ce o such an applica ion [2]. Ano he possibili y would be 5G New Radio (NR) ne wo ks, which a e expec ed o enable highly accu a e posi ioning as an ideal suppo o senso -based au onomous app oaches. They will be a ailable o he as majo i y o he popula ion, as is planned o 2027 in he U.K., o example [ 3 ]. The high dependency be ween sma phones and 5G signals seems a pe ec ma ch. Mo eo e , wi eless adio-based communica ions ha e he in insic p oblem o signal un eliabili y, due o mul ipa h e ec s and No Line-o -Sigh signal condi ions, pa icula ly in indoo en i onmen s [ 4 ], which can be imp o ed by using he abo emen ioned senso s. Ine ial measu emen uni s (IMU) a e such senso s, which a e no highly a ec ed by en i onmen al condi ions; hus, many mode n posi ioning echniques ely on hei use as key senso s. The aim is o a ge hund eds o loca ion-based se ices while keeping humans as a cen al pa o he sys em, using sensing echnologies such as wea able senso s, which Elec onics 2021,10, 397. h ps://doi.o g/10.3390/elec onics10040397 h ps://www.mdpi.com/jou nal/elec onics Elec onics 2021,10, 397 2 o 17 ha e been ending in applica ion [ 5 ]. T acking he human pose by using i ual eali y headse s, o example, is ano he o many examples o such applica ions. Many sma objec s, such as wa ches and shoes, can also di ec ly bene i pedes ian acking echniques, especially in GNSS-shaded a eas whe e he e is no guiding e e ence. To imp o e localiza ion accu acy, p io wo ks ha e gene alized and used o he sou ces which u ilize desc ip i e obse a ions. The obse a ions om di e en senso s, such as ision ea u es [ 6 , 7 ], poin clouds [ 8 , 9 ] om came as, and Lase Iden i ica ion De ec- ion and Ranging (LIDAR) senso s, can be combined wi h odome y in o ma ion o co ec o d i and o p o ide e icien localiza ion and acking a he same ime [ 10 – 12 ]. How- e e , such ision-based auxilia y ea u es a e no always a ailable. Fo ins ance, isual ea- u es ail in da k en i onmen s and LIDAR ea u es ail in hea ily e lec i e en i onmen s. In his pape , we e isi he undamen al ques ion: I only sma phone da a (including he 5G NR ne wo k) and an en i onmen map is p o ided, is i possible o localize a pedes ian o a long pe iod o ime? To achie e his, map suppo is no enough. Di e en possibili ies a e discussed, and i is shown ha he special pe o mance o he u u e mobile communica ion s anda d 5G NR plays a cen al ole as an enabling echnology. The e o e, 5G posi ioning, which is an in as uc u e-based solu ion ha p o ides absolu e posi ion in o ma ion, is conside ed. The 5G echnology is used in ou inal app oaches, hus showing ha 5G can po en ially se e as a base equi emen . Dense NR ne wo ks a e en isioned o ange om a ew me e s up o ens o me e s; o example, assuming se e al access nodes pe oom in indoo en i onmen s [ 13 ]. Thus, 5G NR, as an in as uc u e- based posi ioning solu ion, can be made a ailable bo h indoo s and ou doo s and is an ideal way o na iga e in ci ies (e.g., be ween buildings, subways, and ain s a ions). The emainde o he pape is s uc u ed as ollows: he ela ed wo k and challenges a e desc ibed in he nex sec ion, sec ion h ee p esen s he s udy da ase , in sec ion ou , wo main posi ioning echniques which can wo k independen ly— ha is, only sma phone senso s and a simula ion o In as uc u e-based implemen a ion (e.g., 5G)—a e discussed; sec ion i e p esen s he au onomous posi ion es ima ion me hod, whe e pa icle il e ing map-ma ching is discussed; in sec ion six, we p esen an ini ial usion concep o he au onomous app oach and 5G—which can be ealized wi h and wi hou aking map- ma ching in o conside a ion; and inally, sec ion se en p o ides ou conclusions and di ec ions o u u e wo ks. 2. Rela ed Wo k and Challenges O e ime, indoo posi ioning me hods ha e e ol ed o he cu en s age, which now show an accu acy pe o mance up o he le el o one me e . Howe e , he posi ioning echniques a e s ill mos ly limi ed o senso e o accumula ion, de ice placemen s o ini e labeled da a. A wide ange o di e en senso echnologies ha e been used in o de o o e come hese limi s du ing he las ew yea s, especially h ough sma phones, which ha e almos all he in o ma ion one needs in one place. Wi hin he a ailable senso s in sma phones, ine ial measu emen uni s (IMUs) a e in demand. This is la gely because IMUs con ain senso s (such as accele ome e s and gy oscopes) and hey p o ide in o ma- ion abou he o ien a ion and posi ion o any objec ha hey a e a ached o. Howe e , di e en applica ions and en i onmen al condi ions in luence he senso selec ion and usion ype. Figu e 1shows a his o ical imeline o he sma phone-based posi ioning echniques, beginning wi h accele ome e s. The posi ioning mus be accu a e and eliable, e en when he use o he de ice is a ied in e ms o placemen (e.g., handheld o in a packe ) o o ien a ion (e.g., po ai o landscape mode). Au ho s in [ 14 ] ha e ied o combine di e en physics-based me hods o o e come placemen challenges, such as he Ine ial Na iga ion Sys em (INS), Pedes ian Dead Reckoning (PDR), o di e en il e ing app oaches such as he Kalman Fil e (KF) o o ien a ion acking and he Pa icle Fil e (PF) o map-ma ching. Howe e , hey could no a gue ha all can be done in eal ime applica ions. The e o e, open challenges and Elec onics 2021,10, 397 3 o 17 ques ions can be de ined clea ly based on he imeline e iew and by aiming o a obus and eal ime posi ioning solu ion. Elec onics 2021, 10, x FOR PEER REVIEW 3 o 17 (KF) o o ien a ion acking and he Pa icle Fil e (PF) o map-ma ching. Howe e , hey could no a gue ha all can be done in eal ime applica ions. The e o e, open challenges and ques ions can be de ined clea ly based on he imeline e iew and by aiming o a obus and eal ime posi ioning solu ion. Figu e 1. The imeline o sma phone-based posi ioning echniques ob ained om ex ensi e li e a- u e e iew [15–29]. A esea ch challenge ega ding pose acking is how o de e mine he ini ial pose o he de ice. This is a e y impo an ques ion—i he ini ial posi ion calcula ion is possible, hen i should be also possible o es ima e he new posi ion o he nex poin s. The pedes- ian na iga ion should be compa able in use o GNSS-based na iga ion, in a way ha he app oach wo ks wi h e e y sma phone in any ime. Howe e , in o de o p o ide high- quali y posi ioning, absolu e me hods using Wi eless Local A ea Ne wo k (WLAN), Blue- oo h, o Ul a-Wide Band (UWB) a e among hose indi idual solu ions which o en e- qui e big in as uc u al changes, o example he addi ional moun o access poin s in inge p in app oaches [18,20,30]. GPS ini ializa ion [19] has been accep ed as a s anda d ini ializa ion me hod; howe e , i ob iously imposes a s ong limi on he s a ing poin . Manual use inpu is he only emaining way o s a a na iga ion jou ney, pe haps by using a QR code as ano he specialized solu ion [23]. The ini ial o ien a ion has ewe p oblems, as pi ch and oll a e calculable based on he g a i y es ima ion om accele om- e e measu es [30]. Calib a ed geomagne ic senso measu emen ollowing a local mag- ne ic map can also help in his p ocess, bu his is highly elian on a labeled da ase [22]. The nex challenge ega ds pedes ian posi ioning. Many s udies on o ien a ion acking in he con ex o indoo posi ioning ha e a emp ed o es ima e he heading o he de ice, a he han he heading o he pedes ian. They, he e o e, gene ally assume ha he posi ion whe e he mobile de ice is a ached is ixed and known, o hey neglec he misalignmen be ween sma phone heading and he use ’s walking di ec ion, lea ing i as an open challenge [31]. Howe e , he e ha e been some a emp s o calcula e he heading independen o o ien a ion, ei he ollowing he e y i s a emp s based on ac- cele a ion alues and PCA calcula ion [32], o simila app oaches such as equency do- main analysis; howe e , such me hods mus be imp o ed o each he desi ed accu acy [33]. The las wo challenges become e en mo e p oblema ic when he placemen o he de ice a ies o e ime as a use pe o ms di e en asks, such as making a call, ca ying he de ice in a bag, o walking on an escala o . The e a e also limi less human beha io s, such as side walking, back walking, s ange o jump s epping, aking s ai s wo a a ime, and many o he possibili ies. Resea che s ha e ecen ly ied o acqui ed senso da a ac oss a limi ed numbe o human subjec s and sma phone placemen s. In his manne , he i s supe ised aining da ase o ine ial na iga ion was in oduced in 2017, using one Suppo Vec o Machine (SVM) and eigh Suppo Vec o Reg ession (SVR) models [27]. Realis ic and open benchma king da ase s o pedes ian indoo posi ioning using Figu e 1. The imeline o sma phone-based posi ioning echniques ob ained om ex ensi e li e a u e e iew [15–29]. A esea ch challenge ega ding pose acking is how o de e mine he ini ial pose o he de ice. This is a e y impo an ques ion—i he ini ial posi ion calcula ion is possible, hen i should be also possible o es ima e he new posi ion o he nex poin s. The pedes ian na iga ion should be compa able in use o GNSS-based na iga ion, in a way ha he app oach wo ks wi h e e y sma phone in any ime. Howe e , in o de o p o ide high-quali y posi ioning, absolu e me hods using Wi eless Local A ea Ne wo k (WLAN), Blue oo h, o Ul a-Wide Band (UWB) a e among hose indi idual solu ions which o en equi e big in as uc u al changes, o example he addi ional moun o access poin s in inge p in app oaches [ 18 , 20 , 30 ]. GPS ini ializa ion [ 19 ] has been accep ed as a s anda d ini ializa ion me hod; howe e , i ob iously imposes a s ong limi on he s a ing poin . Manual use inpu is he only emaining way o s a a na iga ion jou ney, pe haps by using a QR code as ano he specialized solu ion [ 23 ]. The ini ial o ien a ion has ewe p oblems, as pi ch and oll a e calculable based on he g a i y es ima ion om accele ome e measu es [ 30 ]. Calib a ed geomagne ic senso measu emen ollowing a local magne ic map can also help in his p ocess, bu his is highly elian on a labeled da ase [22]. The nex challenge ega ds pedes ian posi ioning. Many s udies on o ien a ion acking in he con ex o indoo posi ioning ha e a emp ed o es ima e he heading o he de ice, a he han he heading o he pedes ian. They, he e o e, gene ally assume ha he posi ion whe e he mobile de ice is a ached is ixed and known, o hey neglec he misalignmen be ween sma phone heading and he use ’s walking di ec ion, lea ing i as an open challenge [ 31 ]. Howe e , he e ha e been some a emp s o calcula e he heading independen o o ien a ion, ei he ollowing he e y i s a emp s based on accele a ion alues and PCA calcula ion [ 32 ], o simila app oaches such as equency domain analysis; howe e , such me hods mus be imp o ed o each he desi ed accu acy [33]. The las wo challenges become e en mo e p oblema ic when he placemen o he de ice a ies o e ime as a use pe o ms di e en asks, such as making a call, ca ying he de ice in a bag, o walking on an escala o . The e a e also limi less human beha io s, such as side walking, back walking, s ange o jump s epping, aking s ai s wo a a ime, and many o he possibili ies. Resea che s ha e ecen ly ied o acqui ed senso da a ac oss a limi ed numbe o human subjec s and sma phone placemen s. In his manne , he i s supe ised aining da ase o ine ial na iga ion was in oduced in 2017, using one Suppo Vec o Machine (SVM) and eigh Suppo Vec o Reg ession (SVR) models [ 27 ]. Realis ic and open benchma king da ase s o pedes ian indoo posi ioning using mode n- day sma phones oge he wi h g ound- u h has become a ecen esea ch ac i i y [ 28 , 29 ], employing deep lea ning-based app oaches [ 2 , 26 ]. I has been ound ha elying on limi ed labeled da a sou ces can lead o an end esul . These au ho s did no discuss how ealis ic he da ase could be, due o he expe imen al si ua ion; howe e , he ques ion is: How can Elec onics 2021,10, 397 4 o 17 a da ase be c ea ed, which well- ep esen s a use ac ing eely, no mally, and away om he expe imen al si ua ion? In o de o es ablish an unlimi ed posi ioning app oach capable o ope a ions e e y- whe e and a any ime, as well as o ind a solu ion o he abo emen ioned challenges, he po en ial o 5G NR echnology can be conside ed as a key equi emen . Owing o he p e ious popula i y o sma phones o posi ioning es ima ion, 5G-based me hods could be a pe ec ma ch o hem. Some esea che s ha e epo ed eaching an accu acy o one me e o be e o 70% o he usage ime di e en posi ioning measu emen s, such as Time o A i al (ToA) and Di ec ion o A i al (DoA), in combina ion wi h Kalman Fil e ing [ 13 ]. 3. The S udy Da ase This pape con ibu es a da ase o sma phone senso measu emen s and he esul s o a 5G-based simula ion. I also includes he equi ed map in o ma ion and in e p e ed Py hon sc ip s. We used a handheld Samsung S10 5G sma phone (Samsung, Suwon, Gyeonggi-do, Ko ea) [ 34 ], o eco d accele a ion, magne ic ield, ba ome e and o ien a ion senso alues, as well as angula eloci ies by he gy oscope ia an applica ion named “senso log” [ 35 ]. The simula ion esul , discussed in Sec ion 4.2.2, is p esen ed along wi h he building plan (in WKT o ma ). The 5G simula o —in combina ion wi h manual ideo con ols—was used o es ima e a good g ound u h ajec o y. The eade sc ip ead he senso log ile and ex ac ed he da a wi h desi ed equencies. We solely used IMU da a a 100 Hz o all o he ollowing app oaches. Fo he au onomous app oaches, he ini ial poin was manually se o be 566578.7 and 5932830.1 me e s o he ini ial posi ion and 200 deg ees o ini ial heading. The da ase is a ailable online a he ollowing link (h ps://gi hub.com/Hossein-Shoush a i/Elec onicsDa a.gi ) (accessed on 15 Decembe 2020). I should be no ed ha he heigh componen was omi ed in he implemen a ions. 4. Selec ion o Posi ioning Techniques In he ollowing sec ion, wo gene al echniques which p o ide posi ion in o ma ion sou ces a e discussed. The i s one only elies on senso in o ma ion, while he o he jus uses in as uc u e measu emen s. 4.1. Pedes ian Dead Reckoning The accele ome e –gy oscope combina ion is o pa icula in e es o pedes ian dead eckoning. These MEMS ine ial senso s a e o en combined as h ee-axis senso s in sma phones, such as he MPU-9250 (TDK In enSense, Tokyo, Japan) [ 36 ]. When he sma phone is in es ing posi ion, he accele a ion senso egis e s accele a ion due o g a i y on i s h ee axes (i.e., 9.81 m/s 2 ). This allows o il calcula ion o o a ion in he sma phone–senso coo dina e ames. S ep de ec ion is also based on he egis e ed accele a ion. The h ee-axis gy oscope egis e s angula eloci ies. Wi h app op ia e ime measu emen s, he ela i e angles o o a ion in space can be de e mined by in eg a ion. The ine ial senso s a e subjec o colo ed noise, which is e en mo e appa en in MEMS. This can be seen in he angle o o a ion, due o d i . The empe a u e dependence also has a g ea in luence. These e ec s p oduce an inc ease o posi ioning e o along wi h ime o use in PDR. The e o e, he usion o gy oscope da a (d i in angle) and accele a ion da a (high noise in angle) helps o educe unce ain ies o wo o h ee o ien a ion pa ame e s. Fo ins ance, Gy oscope and Accele a ion senso usion has been made possible by means o he well-known Madgwick algo i hm [37]. The basis o PDR is he combina ion o a pedome e , known o es ima ed s ide leng h, and o ien a ion. The e a e nume ous implemen a ion examples o he ealiza ion o he PDR, in which he posi ioning—bo h in e ms o s ep de ec ion and heading es ima ion— mus be accu a e and eliable. Once a s ep is de ec ed, he sys em needs o es ima e he s ide leng h. The s ide leng h o an indi idual can a y signi ican ly o e ime, due o speed, e ain, and o he en i onmen al cons ain s [ 25 ]. The same p oblem applies o Elec onics 2021,10, 397 5 o 17 he exac heading di ec ion o each s ep, despi e he ac ha he phone can be loca ed anywhe e on he body. Using he benchma king me hods om [ 21 ] and [ 25 ], wo di e en PDR ealiza ions we e ep oduced (see Figu es 1and 2). These igu es show he esul s o a eal ajec o y da ase in a building o Ha enCi y Uni e si y. Elec onics 2021, 10, x FOR PEER REVIEW 5 o 17 The basis o PDR is he combina ion o a pedome e , known o es ima ed s ide leng h, and o ien a ion. The e a e nume ous implemen a ion examples o he ealiza ion o he PDR, in which he posi ioning—bo h in e ms o s ep de ec ion and heading es i- ma ion—mus be accu a e and eliable. Once a s ep is de ec ed, he sys em needs o es i- ma e he s ide leng h. The s ide leng h o an indi idual can a y signi ican ly o e ime, due o speed, e ain, and o he en i onmen al cons ain s [25]. The same p oblem applies o he exac heading di ec ion o each s ep, despi e he ac ha he phone can be loca ed anywhe e on he body. Using he benchma king me hods om [21] and [25], wo di e en PDR ealiza ions we e ep oduced (see Figu es 1 and 2). These igu es show he esul s o a eal ajec o y da ase in a building o Ha enCi y Uni e si y. (a) (b) Figu e 2. Realiza ion o PDR elemen s ep oducing a ailable benchma ks, lis ed as: (a) S ep head- ing calcula ion using Madgwick usion o IMU and le eled Gy oscope conside ing ZUPT; and (b) K-based s ide leng h es ima ion. Figu e 2 shows he use o Madgwick usion o IMU da a o azimu h, implemen ed using a Eule ee qua e nion-based me hod and he in eg a ion o gy oscope da a h ough Ze oVeloci yUPdaTe (ZUPT: O se o o a ion a e o Ze o a s ands ill) a he beginning. Based on bo h o hese examples, a di e ence o mo e han 20° is isible; how- e e , hese esul s a e based on he same da a. They make i clea ha he selec ion o o ien a ion es ima ion me hod has a signi ican in luence on posi ion es ima ion. How- e e , one disad an age o Eule based sys ems is he singula i y limi a ion. S ide leng h es ima ion based on accele ome e da a wo ks dynamically and he e is no need o manual inpu , in compa ison wi h he use o a ixed s ide leng h. Based on he expe imen s, he k-based me hod gene ally wo ks be e o di e en use s wi hou changing he se o pa ame e s. Howe e , a a ia ion o 20 cm in s ide leng h be ween wo indi idual’s s ides in a building is un ealis ic. Mo eo e , he co ec ion o a ixed s ide leng h can be done mo e easily and quickly. The eliable de e mina ion o s ide leng h equi es a ealis ic da ase and lea ning app oaches. The e o e, i can be s ill be seen as an unsol ed challenge. In [21], he au ho s concluded ha a ixed p e-es ima ed s ide leng h causes he same de ia ions o an indi idual as he k-based me hods. The nex ques ion ega ds he pedome e . The au ho s in [38] used a que y o wo successi e condi ions o he pedome e . Fo his,  wi h 9.81 m/s 2 had o be sub ac ed om he le elled accele ome e da a and wo h esholds o he accele a ion alue and one o he ime we e used. O he simila h eshold-based pedome e s ha e been p e- sen ed in he li e a u e (see, e.g., [25]). Howe e , he obus ness o he s ep de ec o me h- ods emains ques ionable. The a o emen ioned e e ence assumed ha people gene a e a pe iodic accele a ion signal only when walking and could no dis inguish be ween a eal s ep and ac ions ha gene a e e y simila accele a ion signal pa e ns, such as shaking he sma phone [30]. This may pose conside able p oblems, conside ing human daily be- ha io s such as walking on s ai s, s anding, si ing, sending ex s, making calls, and yp- ing. We seek o esol e his challenge by conside ing bo h ho izon al and e ical displace- men in he usion concep in oduced in Sec ion 5. Figu e 2. Realiza ion o PDR elemen s ep oducing a ailable benchma ks, lis ed as: ( a ) S ep heading calcula ion using Madgwick usion o IMU and le eled Gy oscope conside ing ZUPT; and ( b ) K-based s ide leng h es ima ion. Figu e 2shows he use o Madgwick usion o IMU da a o azimu h, implemen ed using a Eule ee qua e nion-based me hod and he in eg a ion o gy oscope da a h ough Ze oVeloci yUPdaTe (ZUPT: O se o o a ion a e o Ze o a s ands ill) a he beginning. Based on bo h o hese examples, a di e ence o mo e han 20 ◦ is isible; howe e , hese esul s a e based on he same da a. They make i clea ha he selec ion o o ien a ion es ima ion me hod has a signi ican in luence on posi ion es ima ion. Howe e , one disad an age o Eule based sys ems is he singula i y limi a ion. S ide leng h es ima ion based on accele ome e da a wo ks dynamically and he e is no need o manual inpu , in compa ison wi h he use o a ixed s ide leng h. Based on he expe imen s, he k-based me hod gene ally wo ks be e o di e en use s wi hou changing he se o pa ame e s. Howe e , a a ia ion o 20 cm in s ide leng h be ween wo indi idual’s s ides in a building is un ealis ic. Mo eo e , he co ec ion o a ixed s ide leng h can be done mo e easily and quickly. The eliable de e mina ion o s ide leng h equi es a ealis ic da ase and lea ning app oaches. The e o e, i can be s ill be seen as an unsol ed challenge. In [ 21 ], he au ho s concluded ha a ixed p e-es ima ed s ide leng h causes he same de ia ions o an indi idual as he k-based me hods. The nex ques ion ega ds he pedome e . The au ho s in [ 38 ] used a que y o wo successi e condi ions o he pedome e . Fo his, g wi h 9.81 m/s 2 had o be sub ac ed om he le elled accele ome e da a and wo h esholds o he accele a ion alue and one o he ime we e used. O he simila h eshold-based pedome e s ha e been p esen ed in he li e a u e (see, e.g., [ 25 ]). Howe e , he obus ness o he s ep de ec o me hods emains ques ionable. The a o emen ioned e e ence assumed ha people gene a e a pe iodic accele a ion signal only when walking and could no dis inguish be ween a eal s ep and ac ions ha gene a e e y simila accele a ion signal pa e ns, such as shaking he sma phone [ 30 ]. This may pose conside able p oblems, conside ing human daily beha io s such as walking on s ai s, s anding, si ing, sending ex s, making calls, and yping. We seek o esol e his challenge by conside ing bo h ho izon al and e ical displacemen in he usion concep in oduced in Sec ion 5. Finally, PDR is hen pe o med o each de ec ed s ep. In Equa ion (1), he coo dina es o he cu en posi ion Xi a e calcula ed, depending on he p e ious posi ion Xi−1 , summing wi h a mul iplica ion o he scala ansla ion o T (s ide leng h), and heading ma ix R. Xi=Xi−1+T×R (1) The wo PDR ealiza ions led o di e en ajec o ies (see Figu e 3). The ajec o ies we e bo h subjec o signi ican o ien a ion d i and scale e o s caused by inco ec s ide Elec onics 2021,10, 397 6 o 17 leng h es ima ion. In o de o e alua e and ack he app oaches quali y on-boa d, a cumula i e dis ibu ion unc ion (CDF) plo was placed nex o each o he es ima ed ajec o ies. The CDFs we e calcula ed by i ing a no mal dis ibu ion o dis ances be ween he g ound u h s ep coo dina es and he es ima ed one, i.e., e o s. In o he wo ds, he e o is de ined as he nominal alue minus he es ima ed alue. I he e was a con as be ween he pedome e and he ac ual numbe o he s eps, a simple in e pola ion educed o added a ew s ep posi ions. The e alua ion o he PDR me hods shows ha addi ional in o ma ion is needed o map conside a ion and mo e accu a e o ien a ion and s ide leng h es ima ion. The CDF p esen s he de ia ion quali a i ely and quan i i ely, o ins ance, i can be seen ha 90% o he calcula ed poin s in PDR 2 had an accu acy which was be e han 15 m. Elec onics 2021, 10, x FOR PEER REVIEW 6 o 17 Finally, PDR is hen pe o med o each de ec ed s ep. In Equa ion (1), he coo di- na es o he cu en posi ion X  a e calcula ed, depending on he p e ious posi ion X  , summing wi h a mul iplica ion o he scala ansla ion o T (s ide leng h), and heading ma ix R. X  =X  + T × R (1 ) The wo PDR ealiza ions led o di e en ajec o ies (see Figu e 3). The ajec o ies we e bo h subjec o signi ican o ien a ion d i and scale e o s caused by inco ec s ide leng h es ima ion. In o de o e alua e and ack he app oaches quali y on-boa d, a cu- mula i e dis ibu ion unc ion (CDF) plo was placed nex o each o he es ima ed ajec- o ies. The CDFs we e calcula ed by i ing a no mal dis ibu ion o dis ances be ween he g ound u h s ep coo dina es and he es ima ed one, i.e., e o s. In o he wo ds, he e o is de ined as he nominal alue minus he es ima ed alue. I he e was a con as be ween he pedome e and he ac ual numbe o he s eps, a simple in e pola ion educed o added a ew s ep posi ions. The e alua ion o he PDR me hods shows ha addi ional in o ma ion is needed o map conside a ion and mo e accu a e o ien a ion and s ide leng h es ima ion. The CDF p esen s he de ia ion quali a i ely and quan i i ely, o in- s ance, i can be seen ha 90% o he calcula ed poin s in PDR 2 had an accu acy which was be e han 15 m. (a) (b) Figu e 3. Th ee di e en PDR ealiza ions, named PDR 1 ( he k-based s ide leng h and Madg- wick-based heading), PDR 2 ( ixed s ide leng h and le eled gy oscope heading), and PDR mix ( he k-based s ide leng h and le eled gy oscope heading): (a) T ajec o ies shown on he map; and (b) E o s. CDF p obabili y o he me hods. 4.2. 5G-Based Posi ioning The deploymen o cellula sys ems, om he Fi s Gene a ion (1G) o he cu en (Fi h Gene a ion, 5G, o New Radio, NR) s age, has a long his o y. The Thi d Gene a ion Pa ne ship P ojec (3GPP) eleases ha e a emp ed o answe he disco e ed equi e- men s. The 3GPP uni es se en s anda d elecommunica ions de elopmen o ganiza ions. Thei membe s p oduce epo s and speci ica ions ha de ine 3GPP echnologies, includ- ing adio access, co e ne wo k, and se ice capabili ies [39]. A e a while, he demand on cellula ne wo ks came no only om mobile phone p o ide s, bu also wide indus y. Fu he mo e, 4G sys ems ha e inco po a ed signi ican upda es add essing ma ke s di - e en han adi ional mobile phone businesses [40]; o ins ance, he de ice- o-de ice (D2D) pa adigm allows de ices o di ec ly communica e wi h each o he using a local wi eless channel. D2D echnology laid he ounda ions o he ehicle- o-e e y hing (V2X) echnology in oduced in 3GPP Release 14 [41]. Howe e , he 5G ne wo k is he i s cel- lula ne wo k ha was no designed wi h a sole ocus on mobile phones. Towa d he la e pa o he 2010s, he las -gene a ion ne wo ks began o each hei limi s. The easons o his we e mani old, including a emendous g ow h in sma phone pene a ion and an inc ease in bandwid h-hung y applica ions, as well as he la ge num- Figu e 3. Th ee di e en PDR ealiza ions, named PDR 1 ( he k-based s ide leng h and Madgwick- based heading), PDR 2 ( ixed s ide leng h and le eled gy oscope heading), and PDR mix ( he k-based s ide leng h and le eled gy oscope heading): ( a ) T ajec o ies shown on he map; and ( b ) E o s. CDF p obabili y o he me hods. 4.2. 5G-Based Posi ioning The deploymen o cellula sys ems, om he Fi s Gene a ion (1G) o he cu en (Fi h Gene a ion, 5G, o New Radio, NR) s age, has a long his o y. The Thi d Gene a ion Pa ne ship P ojec (3GPP) eleases ha e a emp ed o answe he disco e ed equi e- men s. The 3GPP uni es se en s anda d elecommunica ions de elopmen o ganiza ions. Thei membe s p oduce epo s and speci ica ions ha de ine 3GPP echnologies, including adio access, co e ne wo k, and se ice capabili ies [ 39 ]. A e a while, he demand on cellu- la ne wo ks came no only om mobile phone p o ide s, bu also wide indus y. Fu he - mo e, 4G sys ems ha e inco po a ed signi ican upda es add essing ma ke s di e en han adi ional mobile phone businesses [ 40 ]; o ins ance, he de ice- o-de ice (D2D) pa adigm allows de ices o di ec ly communica e wi h each o he using a local wi eless channel. D2D echnology laid he ounda ions o he ehicle- o-e e y hing (V2X) echnology in o- duced in 3GPP Release 14 [ 41 ]. Howe e , he 5G ne wo k is he i s cellula ne wo k ha was no designed wi h a sole ocus on mobile phones. Towa d he la e pa o he 2010s, he las -gene a ion ne wo ks began o each hei limi s. The easons o his we e mani old, including a emendous g ow h in sma phone pene a ion and an inc ease in bandwid h-hung y applica ions, as well as he la ge numbe o connec ed de ices. All o hese new equi emen s we e ga he ed and he 3GPP ini ia ed he de ini ion o he 5G echnology. The 5G NR equi emen s ha e been s uc u ed unde h ee main ca ego ies [40]: • Enhanced mobile b oadband (eMBB), he aim o which is o p o ide wi eless connec- i i y wi h e y high bandwid h. • Massi e machine ype communica ions (mMTC), p o iding connec i i y o a la ge numbe o IoT de ices, such as sma me e s, wa ches, o wea ables. mMTC equi es e y la ge cell and ne wo k capaci ies. • Ul a- eliable low la ency communica ions (URLLC), a ge ed a p o iding low la ency, obus communica ion links o V2X, emo e su ge y, and o he sa e y-c i ical applica ions. Elec onics 2021,10, 397 7 o 17 The 5G ne wo k also includes o he de ices and use-cases, e e ed o as indus y e icals. Ano he signi ican di e ence, wi h espec o p e ious cellula ne wo ks, is he use o highe equencies in he millime e -wa e spec um, s a ing a 24 GHz. Impo an ea u es ha e also come om 5G, such as massi e mul iple-inpu and mul iple-ou pu (MIMO), beam o ming, cloud compu ing, and ne wo k i ualiza ion. All o hese ea u es, in oduced in Releases 16–17 [ 42 , 43 ], help o inc ease he scalabili y and modula i y o he ne wo k and aid in eaching peak da a a es o 20 Gbps wi h e y high use densi y [40]. 4.2.1. Algo i hms and Technologies We ha e classi ied he 5G-based posi ioning app oaches based on hei geome ic bases, analyzing p oximi y-, dis ance-, angle-, and ime di e ence-based posi ioning ech- nologies. In his classi ica ion, he e is no conside a ion o he da a d i en app oaches such as he well-known Finge p in ing me hod. P oximi y is he simples way o de e mine he loca ion o an objec . I is a me hod which has been long used in cellula ne wo ks; o example, in cell ID (CID), Wi-Fi, and Blue oo h posi ioning sys ems. P oximi y is based on he knowledge ha he objec o be loca ed is nea a e e ence objec whose posi ion is known. Hence, he accu acy o p oximi y as a posi ioning me hod basically depends on he ange o he signal used. Fo ins ance, dense 5G ne wo ks, which a e en isioned o ange om a ew me e s up o ens o me e s, ha e shown accep able accu acy pe o mance when using p oximi y [ 13 ]. The poin ep esen ing he posi ion o he mobile e minal can be ob ained by geome ic calcula ions, o which he cen oid me hod is he mos common [ 44 ]. The weigh o each An enna Node (AN) is p opo ional o i s Recei ed Signal S eng h (RSS) alue; as a esul , he AN wi h he highes RSS pulls he cen oid mos s ongly o i s own posi ion [ 45 ]. The same weigh ing analysis can be pe o med using ime measu emen s. Howe e , powe measu emen s a e no as accu a e as iming measu emen s in he posi ioning echnologies used in legacy cellula ne wo ks (up o LTE). This p oblem has been sol ed in 5G NR [ 40 ]. T iangula ion is a echnique which de e mines he posi ion o an objec by measu ing he angles o he objec om known poin s. I elies on Angle o A i al (AoA) mea- su emen s, in which he incoming angle o he ecei ed signal is known, such ha he ecei e can es ima e he di ec ion o he ansmi e . A leas wo angle measu emen s om wo di e en known poin s a e su icien o 2D localiza ion, by applying igonome ic iden i ica ion. Al hough angle-based posi ioning echnologies a e no new, due o he beam o ming ea u e, i is wi h 5G NR ha hei ull po en ial can be eached. DL-AoD and UL-AoA a e wo 5G angula posi ioning echnologies ha a e always linked o a base s a ion beam by 3GPP speci ica ion (38.305) [46]. T ila e a ion is he name gi en o posi ioning algo i hms e e ing o a posi ion de e - mined om dis ance measu emen s. These a e also called ange measu emen echniques. In T ila e a ion, he “ i” s ands o he (a leas ) h ee ixed poin s ha a e necessa y o de e mine a 2D posi ion [ 47 ]. In [ 44 ], he use o he leas squa e me hod o minimize he localiza ion e o was discussed, whe e he absolu e dis ance e o be ween he ansmi e and ecei e should ideally be ze o. Posi ioning based on ila e a ion can be pe o med using se e al di e en posi ioning measu emen s, such as Time o A i al (ToA), RSS, and so on. Enhanced Cell ID is a me hod based on a p oximi y algo i hm o LTE as an enhancemen o a iming ad ance p ocedu e. The aim is o educe he a ea o unce ain y o he posi ioning [ 48 – 50 ]. Al e na i ely, i he ansmi ed powe is known, he RSS can be used o es ima e he dis ance he signal has a elled, by using pa icula p opaga ion models ha es ima e he a enua ion o he signal in ela ion o he a elled dis ance. This is due o he ac ha he powe o he ansmi ed signal dec eases wi h he a elled dis ance [ 44 ]. In 5G NR, he Round-T ip Time (RTT) o a neighbo cell has been used o de ine a mul i-RTT me hod. The NR E-CID is also a e-de ini ion o ECID; his ime, o 5G ne wo ks [40]. Mul ila e a ion e e s o posi ioning algo i hms based on he di e ence in dis ances om he objec o wo e e ence poin s o known loca ion. In pa icula , hey ely on Time Elec onics 2021,10, 397 8 o 17 Di e ence o A i al (TDoA) measu emen s. In gene al, a leas wo measu emen s ( om h ee ansmi e s) a e needed o ob ain a 2D posi ion, while a leas h ee measu emen s a e equi ed o calcula e a 3D posi ion. In 5G NR, DL-TDOA and UL-TDOA a e based on he same p inciple, whe e he ime di e ence measu emen is ca ied ou by he ne wo k o base s a ions [40]. A simila me hod is he 4G OTDOA. 4.2.2. 5G Simula ion Rega ding 5G, posi ioning me hods can gene a e coo dina es wi h a p e-de ined esolu ion, accu acy, and la ency. Pe o mance a ies in 5G-based echnologies and 3GPP eleases. Fo ins ance, in Release 17, he aim was o gene a e a ho izon al posi ioning p ecision o a leas 1 me e o highe o indus ial use-cases; o example, an accu acy o 0.2 me e is desi able in some indoo use-cases. A posi ioning la ency o less han 100 milliseconds is also desi ed [ 43 ]. Al hough a eal 5G Campus Ne wo k is going o be buil in he amewo k o he Le el 5 Indoo Na iga ion (L5IN) p ojec unning a he Ha enCi y Uni e si y, an indoo cellula -based posi ioning simula ion was designed o e alua e di e en an enna placemen s, ne wo k esolu ions, quali ies, e c. We ca ied ou a simula ion based on he poin s discussed abo e ela ing o 5G posi ioning me hods and conside ing he 5G eleases. Howe e , simula ion poin s we e de i ed o simplici y in he ollowing expe imen s, mainly by using he loo plans and e e ence poin s wi h andom e o s in a ange o 5 m. 5. Au onomous Posi ion Es ima ion In he ollowing sec ion, he mos p omising app oaches o au onomous pedes ian na iga ion a e p esen ed and discussed. Fo be e unde s anding o he esul s, he echnical de ails a e also explained. 5.1. PDR Pa icle Fil e The au onomous posi ion es ima ion was ini ialized by he PDR 1. PDR-based ap- p oaches equi e co ec ion o pedes ian na iga ion, in he case o longe -las ing posi ion es ima ions. The e o e, he a ailable componen s o na iga ion we e conside ed; he loo plan and he ou ing g aph. These we e in elligen ly combined wi h he PDR using PF algo i hms, which has been used p e iously in [38]. In his a ian , he PDR ep esen s he P opaga ion p ocess, and he pedome e con ols he compu a ion a e. Du ing he es ima ion s ep, he loo plan and he ou ing g aph we e used as co ec ion pa ame e s. Figu e 4g aphically ep esen s he p ocessing o he a ailable da a o de e mine co ec ions. In Figu e 4a, he loo plan is used wi h a Boolean logic p inciple: I a pa icle is behind a wall, change i s weigh o ze o ( ed). The o he pa icles change hei weigh s, based on hei o ien a ion wi h espec o he wall posi ion (g ey o black pa icles). As Figu e 4b shows, he weigh ing p inciple was only based on he ou ing edges. The o hogonal o se has been used as he weigh de ined on he no mal dis ibu ion in Equa ion (4). an ai=yi−ym xi−xm. (2) wi=exph−0.5·(αi− wall)·R−1 ·(αi− wall)i(3) wi=exph−0.5·do hogonal i·R−1 d·do hogonal ii (4) Equa ion (2) shows he calcula ion o he cou se be ween ac ual pa icle i and weigh ed mean alue o las es ima ion s ep. This angle was used o compa e he o ien a ion o walls a ound his pa icle. The g ey-scaled colo o pa icles in Figu e 4a shows he weigh s based on hese compa isons o nea es wall included in a p obabili y densi y unc ion which ep esen s a no mal dis ibu ion Equa ion (3). Equa ion (4) shows he e he p obabili y densi y unc ion, which ep esen s a no mal dis ibu ion. A longe dis ance o ou ing Elec onics 2021,10, 397 9 o 17 edge p oduced a lowe weigh o his pa icle. The weigh ed mean alue ou o all pa icle mo ed mo e o ou ing edge, based on he idea, ha his was he ypical a ea o walking. A he end, hese co ec ions gene a e a ajec o y which is posi ioned mo e nea ou ing edges, which a e ypically o ien ed in he mean axis o co ido s. Elec onics 2021, 10, x FOR PEER REVIEW 9 o 17 based on he ou ing edges. The o hogonal o se has been used as he weigh de ined on he no mal dis ibu ion in Equa ion (4). (a) (b) Figu e 4. P inciples o suppo : (a) loo plan— ed pa icles a e behind wall and a e weigh ed ze o; and (b) ou ing g aph— he o hogonal o se o a pa icle o he nea es ou ing edge is used o calcula e i s weigh [21]. ana  =y  −y  x  −x  . (2) w  =exp [−0.5∙(α  −  )∙ R   ∙(α  −  )] (3) w  =exp [−0.5 ∙ d   ∙ R   ∙d  ] (4) Equa ion (2) shows he calcula ion o he cou se be ween ac ual pa icle i and weigh ed mean alue o las es ima ion s ep. This angle was used o compa e he o ien a- ion o walls a ound his pa icle. The g ey-scaled colo o pa icles in Figu e 4a shows he weigh s based on hese compa isons o nea es wall included in a p obabili y densi y unc- ion which ep esen s a no mal dis ibu ion Equa ion (3). Equa ion (4) shows he e he p obabili y densi y unc ion, which ep esen s a no mal dis ibu ion. A longe dis ance o ou ing edge p oduced a lowe weigh o his pa icle. The weigh ed mean alue ou o all pa icle mo ed mo e o ou ing edge, based on he idea, ha his was he ypical a ea o walking. A he end, hese co ec ions gene a e a ajec o y which is posi ioned mo e nea ou ing edges, which a e ypically o ien ed in he mean axis o co ido s. Using he app oach o map suppo on he es da a yielded a signi ican ly di e en ajec o y, compa ed o he PDR ajec o y. Howe e , he app oach is al eady clea ly less e ec i e. Wi hou u he co ec ion using he ou ing g aph, he il e can change he oom a e a ce ain ime, depending on he noise pa ame e s. 5.2. Topological App oach In he opological app oach i s p esen ed by [38], he PF is eplaced by a s a e de- ec ion app oach. A posi ion es ima ion wo ks only based on he ou ing g aph. The idea he e is ha he change o p io i y o o ien a ion es ima ion makes he PDR mo e obus o longe du a ions o na iga ion. An addi ional ad an age compa ed o il e ap- p oaches is he clea s a e de ec ion wi hou he in luence o andomly dis ibu ed pa am- e e epea abili y. A u he ad an age is he equi emen o less compu ing powe , com- pa ed o a PF. The opological app oach wo ks wi h ou s a e que ies. These di e en s a es a e necessa y o handle he di e en e o si ua ions based on PDR in luences by s ep leng h e o and d i e o o calcula ed o a ion angle. The basis is a compa ison o he cu en ly calcula ed di ec ion using gy oscope da a and he o ien a ion o su ounding ou ing edges. A selec ion o he app op ia e ou ing edge is applied. This p ocess is epea ed o each s ep de ec ion. Thus, he PDR uns along he mos p obable ou ing edges. One dis- ad an age is ha i canno assume posi ions eely in space. Figu e 4. P inciples o suppo : ( a ) loo plan— ed pa icles a e behind wall and a e weigh ed ze o; and ( b ) ou ing g aph— he o hogonal o se o a pa icle o he nea es ou ing edge is used o calcula e i s weigh [21]. Using he app oach o map suppo on he es da a yielded a signi ican ly di e en ajec o y, compa ed o he PDR ajec o y. Howe e , he app oach is al eady clea ly less e ec i e. Wi hou u he co ec ion using he ou ing g aph, he il e can change he oom a e a ce ain ime, depending on he noise pa ame e s. 5.2. Topological App oach In he opological app oach i s p esen ed by [ 38 ], he PF is eplaced by a s a e de ec- ion app oach. A posi ion es ima ion wo ks only based on he ou ing g aph. The idea he e is ha he change o p io i y o o ien a ion es ima ion makes he PDR mo e obus o longe du a ions o na iga ion. An addi ional ad an age compa ed o il e app oaches is he clea s a e de ec ion wi hou he in luence o andomly dis ibu ed pa ame e epea abili y. A u he ad an age is he equi emen o less compu ing powe , compa ed o a PF. The opological app oach wo ks wi h ou s a e que ies. These di e en s a es a e necessa y o handle he di e en e o si ua ions based on PDR in luences by s ep leng h e o and d i e o o calcula ed o a ion angle. The basis is a compa ison o he cu en ly calcula ed di ec ion using gy oscope da a and he o ien a ion o su ounding ou ing edges. A selec ion o he app op ia e ou ing edge is applied. This p ocess is epea ed o each s ep de ec ion. Thus, he PDR uns along he mos p obable ou ing edges. One disad an age is ha i canno assume posi ions eely in space. Fo he es da ase was used o his app oach, see Figu e 5. This app oach wo ks well i he ajec o y ollows a clea and simple ou ing g aph. This was no he case he e and he algo i hm ailed a e abou 60% o he un (Figu e 5). The ajec o y could no ma ch he ou ing g aph and he calcula e i s pa h. Unce ain ies in posi ioning, he e o e, inc ease s ongly i he a e sed pa h is no well- ep esen ed by he ou ing edges, o i i de ia es s ongly. Elec onics 2021,10, 397 16 o 17 8. Yang, J.; Cao, Z.; Zhang, Q. A as and obus local desc ip o o 3D poin cloud egis a ion. In . Sci. 2016 ,346–347, 163–179. [C ossRe ] 9. Dube, R.; Dugas, D.; S umm, E.; Nie o, J.; Siegwa , R.; Cadena, C. SegMa ch: Segmen based place ecogni ion in 3D poin clouds. In P oceedings o he 2017 IEEE In e na ional Con e ence on Robo ics and Au oma ion (ICRA), Singapo e, 29 May–3 June 2017; pp. 5266–5272. 10. Hess, W.; Kohle , D.; Rapp, H.; Ando , D. Real- ime loop closu e in 2D LIDAR SLAM. In P oceedings o he 2016 IEEE In e na ional Con e ence on Robo ics and Au oma ion (ICRA), S ockholm, Sweden, 16–20 May 2016; Volume 2016, pp. 1271–1278. 11. G ise i, G.; Tipaldi, G.D.; S achniss, C.; Bu ga d, W.; Na di, D. Fas and accu a e SLAM wi h Rao–Blackwellized pa icle il e s. Robo . Au on. Sys . 2007,55, 30–38. [C ossRe ] 12. Mon eme lo, M.; Th un, S.; Kolle , D.; Wegb ei , B. Fas SLAM: A ac o ed solu ion o he simul aneous localiza ion and mapping p oblem. In P oceedings o he AAAI Na ional Con e ence on A i icial In elligence/IAAI, Edmon on, AB, Canada, 28 July–2 Augus 2002; p. 593598. 13. Koi is o, M.; Hakka ainen, A.; Cos a, M.; Kela, P.; Leppänen, K.; Valkama, M. High-E iciency De ice Posi ioning and Loca- ion- Awa e Communica ions in Dense 5G Ne wo ks. IEEE Commun. Mag. 2017,55, 188–195. [C ossRe ] 14. Xiao, Z.; Wen, H.; Ma kham, A.; T igoni, N. Robus Indoo Posi ioning wi h Li elong Lea ning. IEEE J. Sel. A eas Commun. 2015 , 33, 2287–2301. [C ossRe ] 15. Mizell, D.W. Using g a i y o es ima e accele ome e o ien a ion. In P oceedings o he Se en h IEEE In e na ional Symposium on Wea able Compu e s, Whi e Plains, NY, USA, 21–23 Oc obe 2003; Ins i u e o Elec ical and Elec onics Enginee s (IEEE): Los Alami os, CA, USA, 2004; pp. 252–253. 16. Kunze, K.; Lukowicz, P.; Pa idge, K.; Begole, B. Which way am I acing: In e ing ho izon al de ice o ien a ion om an ac-cele ome e signal. In P oceedings o he 2009 In e na ional Symposium on Wea able Compu e s, Linz, Aus ia, 4–7 Sep embe 2009; pp. 149–150. 17. All Abou he Senso s ha Make he iPhone so Cool. A ailable online: h ps://www.li ewi e.com/senso s- ha -make-iphone-so- cool-2000370 (accessed on 5 No embe 2020). 18. Woodman, O.; Ha le, R. Pedes ian Localisa ion o Indoo En i onmen s. Ph.D. Thesis, Uni e si y o Camb idge, Camb idge, MA, USA, 2010. 19. Cons andache, I.; Choudhu y, R.R.; Rhee, I. Towa ds Mobile Phone Localiza ion wi hou Wa -D i ing. In P oceedings o he 2010 IEEE INFOCOM, San Diego, CA, USA, 15–19 Decembe 2010; pp. 1–9. 20. Google Maps’ Bigges Momen s o e he Pas 15 Yea s. A ailable online: h ps://blog.google/p oduc s/maps/look-back-15 -yea s-mapping-wo ld/ (accessed on 5 No embe 2020). 21. Willemsen, T. Fusionsalgo i hmus zu Au onomen Posi ionsschä zung im Gebäude, Basie end au MEMS-Ine ialsenso en im Sma phone. Ph.D. Thesis, Ha enCi y Uni e si y, Hambu g, Ge many, 2016. 22. F assl, M. Ha nessing Geomagne ic Field Dis u bances o Ubiqui ous Na iga ion. Ph.D. Thesis, Ulm Uni e si y, Ulm, Ge many, 2018. 23. Lukian o, C.; S e nbe g, H. STEPPING—Sma phone-Based Po able Pedes ian Indoo Na iga ion. A ch. Fo og am. Ka og . Telede ekcji 2011,22, 311–323. 24. B ajdic, A.; Ha le, R.K. Walk de ec ion and s ep coun ing on uncons ained sma phones. In P oceedings o he 2013 ACM In e na ional Join Con e ence on Pe asi e and Ubiqui ous Compu ing, Zu ich, Swi ze land, 8–12 Sep embe 2013; Associa ion o Compu ing Machine y (ACM): New Yo k, NY, USA, 2013; pp. 225–234. 25. Li, F.; Zhao, C.; Ding, G.; Gong, J.; Liu, C.; Zhao, F. A Reliable and Accu a e Indoo Localiza ion Me hod Using Phone Ine ial Senso s. In P oceedings o he 2012 ACM Con e ence on Ubiqui ous Compu ing, Pi sbu gh, PA, USA, 5–8 Sep embe 2012; pp. 421–430. 26. Chen, C.; Lu, X.; Ma kham, A.; T igoni, N. IoNe : Lea ning o cu e he cu se o d i in ine ial odome y. In P oceedings o he AAAI Con e ence on A i icial In elligence, New O leans, LA, USA, 2–7 Feb ua y 2018; Volume 32, pp. 6468–6476. 27. Yan, H.; Shan, Q.; Fu ukawa, Y. RIDI: Robus IMU Double In eg a ion. In P oceedings o he Eu opean Con e ence on Compu e Vision (ECCV), Munich, Ge many, 8–14 Sep embe 2018; pp. 621–636. 28. Chen, C.; Zhao, P.; Lu, C.X.; Wang, W.; Ma kham, A.; T igoni, N. Deep-Lea ning-Based Pedes ian Ine ial Na iga ion: Me hods, Da a Se , and On-De ice In e ence. IEEE In e ne Things J. 2020,7, 4431–4441. [C ossRe ] 29. Co és, S.; Solin, A.; Rah u, E.; Kannala, J. ADVIO: An Au hen ic Da ase o Visual-Ine ial Odome y. Min. Da a Financ. Appl. 2018,11214 LNCS, 425–440. [C ossRe ] 30. Xiao, Z. Robus Indoo Posi ioning wi h Li elong Lea ning. Ph.D. Thesis, Uni e si y o Ox o d, Ox o d, UK, 2014. 31. Kuang, J.; Niu, X.; Chen, X. Robus Pedes ian Dead Reckoning Based on MEMS-IMU o Sma phones. Senso s 2018 ,18, 1391. [C ossRe ] [PubMed] 32. Hoseini aba abaei, S.A.; Gluhak, A.; Ta azolli, R.; Headley, W. Design, ealiza ion, and e alua ion o uDi ec —An app oach o pe asi e obse a ion o use acing di ec ion on mobile phones. IEEE T ans. Mob. Compu . 2014,13, 1981–1994. [C ossRe ] 33. Roy, N.; Wang, H.; Choudhu y, R.R. I am a Sma phone and I can Tell my Use ’ s Walking Di ec ion. In P oceedings o he 12 h Annual In e na ional Con e ence on Mobile Sys ems, Applica ions, and Se ices, B e on Woods, NH, USA, 16–19 June 2014; pp. 329–342. Elec onics 2021,10, 397 17 o 17 34. Samsung Galaxy S10 5G. A ailable online: h ps://www.samsung.com/de/sma phones/galaxy-s10/galaxy-s10-5g/ (accessed on 14 Decembe 2020). 35. Senso Log-Apps on Google Play. A ailable online: h ps://play.google.com/s o e/apps/de ails?id=com.h alan.ac i i ylog&hl= en (accessed on 14 Decembe 2020). 36. In ensense MPU-9250 | TDK. A ailable online: h ps://in ensense. dk.com/p oduc s/mo ion- acking/9-axis/mpu-9250/ (accessed on 11 Decembe 2020). 37. Madgwick, S.O.H. An E icien O ien a ion Fil e o Ine ial and Ine ial/Magne ic Senso A ays. 2010. A ailable online: h ps://www.x-io.co.uk/ es/doc/madgwick_in e nal_ epo .pd (accessed on 3 Ap il 2010). 38. Willemsen, T.; Kelle , F.; S e nbe g, H. A opological app oach wi h MEMS in sma phones based on ou ing-g aph. In P oceedings o he 2015 In e na ional Con e ence on Indoo Posi ioning and Indoo Na iga ion (IPIN), Ban , AB, Canada, 13–16 Oc obe 2015; pp. 1–6. [C ossRe ] 39. Abou 3GPP. A ailable online: h ps://www.3gpp.o g/abou -3gpp/abou -3gpp (accessed on 5 No embe 2020). 40. Ga cía, A.; Maie , S.; Philips, A. Loca ion-Based Se ices in Cellula Ne wo ks: F om GSM o 5G NR; A ech House: London, UK, 2020. 41. Release 14. A ailable online: h ps://www.3gpp.o g/ elease-14 (accessed on 5 No embe 2020). 42. Release 16. A ailable online: h ps://www.3gpp.o g/ elease-16 (accessed on 5 No embe 2020). 43. Release 17. A ailable online: h ps://www.3gpp.o g/ elease-17 (accessed on 9 Decembe 2020). 44. Alya awi, I. Real-Time Localiza ion using So wa e De ined Radio. Ph.D. Thesis, Uni e si ä Be n, Be n, Swi ze land, 2015. 45. Celik, G.; Celebi, H.; Tuna, G. A no el RSRP-based E-CID posi ioning o LTE ne wo ks. In P oceedings o he 13 h In e na ional Wi eless Communica ions and Mobile Compu ing Con e ence (IWCMC), Valencia, Spain, 26–30 June 2017; pp. 1689–1692. [C ossRe ] 46. 3GPP Speci ica ion Se ies: 38se ies. A ailable online: h ps://www.3gpp.o g/DynaRepo /38-se ies.h m (accessed on 10 Decembe 2020). 47. Co ea, A.; Ba celo, M.; Mo ell, A.; Vica io, J.L. A Re iew o Pedes ian Indoo Posi ioning Sys ems o Mass Ma ke Applica ions. Senso s 2017,17, 1927. [C ossRe ] [PubMed] 48. Mike, T.; Ewald, Z. LTE Loca ion Based Technology In oduc ion. A ailable online: h ps://www. ohde-schwa z.com/k / applica ions/whi e-pape _230854-122561.h ml (accessed on 29 May 2015). 49. Campos, R.S. E olu ion o Posi ioning Techniques in Cellula Ne wo ks, om 2G o 4G. Wi el. Commun. Mob. Compu . 2017 ,2017, 1–17. [C ossRe ] 50. Bo kowski, J.; Niemelä, J.; Lempiäinen, J. Enhanced pe o mance o Cell ID+RTT by implemen ing o ced so hando e algo- i hm. In P oceedings o he IEEE 60 h Vehicula Technology Con e ence, Los Angeles, CA, USA, 26–29 Sep embe 2004; Volume 60, pp. 3545–3549.