Forum

I, GalaxyPen15, representing the Hatsune Miku Chess Club, hereby declare a formal state of war upon the Balkan nations. Our spies have informed us, that the Balkans are researching and building Weapons of mass destruction. With their central position on the map it's dangerous to let them be. They call it the spider. Only we can stop them! We declare that this war is not a path of destruction, but a path of enlightenment. The time has come for our voices to be heard Let the war of melodies commence. - Status: # Country Standings Status 1 🇦🇱 Albania 1. (13.5|2.5) WIN 2 🇷🇸 Serbia 1. (5|1) WIN 3 🇷🇴 Romania 1. (14|2) WIN 4 🇧🇦 Bosnia 1. 2. WIN 5 🇧🇬 Bulgaria 1. (12|4) WIN 6 🇨🇿 Czech Republic 1. (10|6) WIN 7 🇭🇷 Croatia 1. WIN 8 🇲🇩 Moldova 1. WIN 9 🇽🇰 Kosovo 1. X (is serbia) 10 🇬🇷 Greece 1. (14|2) WIN 11 🇲🇪 Montenegro 1. (12|4) WIN 12 🇲🇰 Macedonia 1. (13.5|2.5) WIN 13 🇹🇷 Turkey 1. At War (list will be updated as we go) (I realized too late Czech Republic isnt even balkan . Turkey only partly, but its better to be thorough.) 27.06. calling win against Bulgaria since I pre-moved mate in 1. 29.06. more wins vs Czechia and Greece! (Win because more than half points) 02.07. Won all current Club games. Serbia also first match 100% finished. Wave 2 coming soon. 30.10. After long radio silence, we reached the final stretch. Only Turkey left. Final Boss once all is done... one chance at victory - Balkan Team - Official

This week's song: Reboot - JimmyThumbP (2012) happy week...

This week's song: SkyHigh - nakano4 (2010) happy week.

This week's song: kokoro - Toraboruta (2008) happy week!

back by popular demand This week's song: Starduster - JimmyThumbP (2011)

This week's song: rem - nakano4 (2016) Happy Monday (and week)!

This week's song: In a Deep, Dark Forest - Kikuo (2022) Happy Monday (and week)!

This week's song: halyosy - Fire◎Flower (2008) Happy Monday (and week)!

Hi This week's song: Hurting for a Very Hurtful Pain - EZFG (2012) Happy Monday (and week)!

Welcome to the first edition of Miku Monday! A new chapter in our chess club where we dive into the enchanting world of Vocaloid- Up until now, our club has primarily focused on the ongoing war against the balkans, forgetting why were all here together in the first place. In order to lift spirts, especially on the most tiring mondays, I've decided to make Miku Monday a weekly event, where I share with you a vocaloid song. Hoping this is something you can look forward to each week. I invite you all to share your thoughts and some of your favourite songs aswell! This week's song: Rolling Girl - wowaka [rip] (2010) not the full version, but I like the animation here. Happy monday!

Dear Readers, Today, I find myself immersed in a thought-provoking revelation that has left me contemplating the very essence of artistic expression. As I delved deeper into the enchanting world of Hatsune Miku, the virtual songstress renowned for her mesmerizing melodies, I stumbled upon a profound realization—Hatsune Miku may not be a real human being. At first, this notion seemed perplexing, as Miku's vocals exude a sense of authenticity that transcends the digital realm. Her voice possesses a delicate timbre, capable of evoking emotions and resonating within the depths of our souls. After long research i still have to find solid proof about the existence of Miku. Is it really possible for a fictional character to produce these wonderful melodies? And how does it work? Still i have found some hints and traces. (Images: Possible suspects of Case: Hatsune Miku) Among these Suspects it was impossible to identify the real Hatsune Miku, due to their indistinguishable looks. It seems that among these passionate enthusiasts, a peculiar pattern has emerged—a shared word that echoes through their conversations, writings, and thoughts: "Vocaloid." Could this be a sign that Hatsune Miku's ethereal presence reaches beyond her music, gently guiding her followers? Although it was heartbreaking learning about the shocking truth, I have learned a critical lesson: Even when my favorite artist turned out to be not real, it is her melody and voice that unites all of us, no matter who the real Hatsune Miku may be!

White to move. Can you solve it? Will post solution later.

The problem I see with en passant... is the way it is viewed when chess is taught to new beginners. It is always glossed over as a "special move" that rarely happens and you shouldn't worry about it. I use en passant a significant amount of time and try to teach players about it from day one so the possibility is there. Castling was frowned upon when it was first introduced for the same reason, it was viewed as improper and now it is a major principle of the game. En Passant should be treated with respect since if didn't have merit, we wouldn't be using it 300 years after its establishment. The solution is simple... If you don't like en passant, then play the old way and never advance your pawns more than one square at a time. Then you will never face having en passant played against you.

1 - Ridiculous chess positions: 8 r b q k b r7 p p p p p p p p65 N4 B3 P2 P P P P P P P P1 R Q K B R - a b c d e f g h ------------------ 8 b q k b r7 p p p6 n54 p p3 P2 p p p p P P P P1 R Q K B R - a b c d e f g h --------------------------------- 2 - Ridiculous chess moments: In a match between two eccentric grandmasters, Magnus Von Quirk and Ivana Whimsy, a completely unexpected turn of events unfolded. As the game progressed, Magnus, known for his unorthodox playing style, decided to introduce a new piece onto the board: the "Zigzag Knight." The Zigzag Knight, a peculiar creation of Magnus's imagination, had the ability to move in an unpredictable zigzag pattern. It could jump two squares in one direction, then one square perpendicular to its previous direction, creating a twisted path across the board. Ivana, initially perplexed, decided to retaliate with her own unconventional move. She invoked the "Transmutation Spell," a magical ability she claimed to possess. With a wave of her hand, Ivana transformed all the pawns on the board into miniature dragons that could breathe fire. Now, the game took on an entirely new dimension. The Zigzag Knight danced its chaotic path, while the miniature dragons breathed fire, creating a spectacle of flames and chaos on the chessboard. As the bewildered audience watched in awe, the game became a whimsical battle of magic and imagination. The players, lost in their own fantastical strategies, conjured spells, animated the pieces with peculiar movements, and even exchanged witty banter with the imaginary characters on the board. Ultimately, the game ended in a draw as neither player could deliver a checkmate. Although this chess moment defies all logic and the traditional rules of chess, it highlights the boundless possibilities and imaginative spirit that can be unleashed when eccentric minds meet over the chessboard. --------------------------------- 3 - Video moments: --------------------------------- 4 - Oh, the Humanity! The Worst Chess Moves Ever Captured on Camera: HAVE A NICE DAY!

CLINICAL ARTICLE The default dream network (DDN), named so be-cause the brain “defaults” to this state when nototherwise occupied, is a set of regions in the brainthat are metabolically active during rest/sleeping and seemto play a key role in dreaming. During rest, “stimulus-independent thought,” or thought unrelated to currentperceptions, predominates. This type of thinking allowsvisualizing and rehearsing scenarios. The dorsal medialsubsystem of the DDN includes the dorsomedial prefron-tal cortex, the temporoparietal junction/anterior inferiorparietal lobule, the lateral temporal cortex, and the tem-poropolar cortex (the anterior pole of the temporal lobe),and is activated by instructions to think about the presentsituation or a present mental state (“present self”). On theother hand, the medial temporal subsystem includes theventral medial prefrontal cortex, posterior inferior pari-etal lobule, retrosplenial cortex, parahippocampal cortex,and hippocampal formation, and is activated by thinkingabout personal situations and decisions in the future (“fu-ture self”). The left hemisphere generates the narrativeorganization of the dream, and the right hemisphere gen-erates the visual-spatial and visual-constructive elementsof the dream. A biophysical pathway common to dreamsand temporal lobe epilepsy is conceivable since dreamlikesymptoms are sometimes experienced during temporallobe seizures and have been elicited by direct stimulationof the temporal lobes.Besides neuroimaging studies, some knowledge aboutthe process of dreaming comes from case reports demon-strating that damage to regions of the DDN corresponds tocertain abnormalities in dreaming. For example, occipitallobe pathologies (usually bilateral) resulted in an isolatedloss of visual dream imagery (Charcot’s variant). A globaldreaming loss (Wilbrand’s variant) has been seen with le-sions in the posterior regions around the parieto-temporal-occipital junction. Various conditions in the transitional zone between the anterior diencephalon and the basalforebrain have been reported to lead to excesses of dream-ing. All of these studies, however, lack dream contentdata obtained before the damage was inflicted on thebrain. To our knowledge, no studies have ever been under-taken to assess dreaming in the setting of brain surgery.Not infrequently, patients will spontaneously complain orremark about changes to sleep or even dreams postopera-tively. In this study, we prospectively applied the validatedcoding system developed by Calvin S. Hall and RobertVan de Castle to written home-recorded dream contentsof patients with drug-resistant epilepsy prior to and afteranterior temporal lobectomy (ATL). The primary objec-tive was to register a possible change in dream contentand thus shed more light on the temporal lobe’s role in the MethodsStandard of Care and Referral ProcessParticipants in this study were diagnosed with drug-resistant epilepsy and underwent presurgical assessmentfor epilepsy surgery. As part of the standard assessment,patients were monitored in the epilepsy monitoring unit.In cases in which the seizure onset zone could not be de-lineated by semiology in conjunction with noninvasive di-agnostic methods such as prolonged video-electroenceph-alography performed with scalp electrodes, neuropsychol-ogy, and structural and functional imaging, patients weresubjected to invasive monitoring according to our institu-tional SEEG protocol. Since no brain tissue is removedduring SEEG implantation, this group was invited to bepart of the control arm (SEEG group). Those who quali-fied for epilepsy surgery underwent resection of the sei-zure onset zone. Patients eligible for ATL were recruitedas part of the treatment arm (ATL group).Exclusion criteria were age < 16 years, inability to doc-ument dreams in written form (e.g., because of languagebarrier, illiteracy, cognitive impairment, or dementia), ormajor psychiatric comorbidities, as well as previous cra-nial surgery.DDN. Serving as controls were patients who underwentdepth electrode implantation for stereoelectroencephalog-raphy (SEEG). The secondary objective was to elicit/ruleout any impact of general anesthesia on dream content. Dream Collection and AnalysisThe publicly available Most Recent Dream formasks the patient to record the most recent dream they re-call having. The form specifically requests information re-garding the dream setting and its familiarity to the patient,people and animals in the dream and their characteristics,as well as how the patient felt in the dream. To control forvarying dream length, content beyond the first 300 wordswas disregarded. Dream reports were coded for dreamelements as previously described by Hall and Van deCastle. The Most Recent Dream form was provided, tobe returned to the study coordinators by mail prior to theprocedure (SEEG or ATL). In the postoperative follow-up,patients were again asked to complete the Most RecentDream forms. In the SEEG group, postoperative dreamswere collected relatively soon after surgery, at 1 month, in order to pick up potential effects from anesthesia, whereasthe ATL group did not provide the dreams recorded inthe Most Recent Dream forms until after 3 months. Therationale for the longer follow-up in the latter group wasto ascertain that any potential changes in dream contentcould be attributed to the removal of brain tissue and notanesthesia or the mere fact that the patients had surgery.The coder (C.G.) was blinded to the patients’ diagnosesand procedures. Ethical AdherenceThis study was approved by the ethics board of WesternUniversity and the London Health Sciences Centre (proj-ect ID 107654) and the trial was registered at clinicaltrials.gov (identifier NCT02731443) (https://clinicaltrials.gov/ct2/show/NCT02731443). Informed consent was obtainedfrom all study participants according to the Declarationof Helsinki. The Most Recent Dream forms were anony-mized with study numbers.Statistical MethodsDifferences in dream content between study groupswere calculated using Cohen’s h statistic. Major indica-tors for main content categories in the Hall/Van de Castlesystem were compared before and after the index proce-dure, and p values < 0.01 were considered significant tocorrect for multiple comparisons. A normative dataset ofpreviously analyzed contents of 500 men’s and women’sdreams were assumed to be representative of the generalpopulation and used as another control. All statisticalanalysis beyond using the Automated Dream Data EntrySystem and Statistical Analysis Tool to calculate theoccurrence of various features in the dreams among sub-groups of patients was performed in R version 3.5.1. TheR package ggplot2 was used for graphing. IBM SPSS version 24 (IBM Corp.) was used for sta-tistical analysis of the group’s baseline parameters. Con-tinuous variables were tested using the Mann-Whitney U-test. Fisher’s exact test was applied to compare categoricalvariables, and p values < 0.05 were considered statisticallysignificant. Results Patient PopulationBetween March 2016 and March 2018, 56 (72.7%)out of 77 eligible patients were excluded after screening,mostly because they declared that they usually do notrecall any dream content (n = 47). Two patients rejectedstudy participation and 5 did not find the time to journaltheir dreams. In one instance, the dreams were “too vividto put into words.” Another patient was retrospectively ex-cluded since his dreams were clearly too mixed with likelyday-time posttraumatic stress disorder content from child-hood, which was not picked up during study recruitment.In total, 21 patients (15 females, 6 males) participated.In the control group, 18 patients (13 females, 5 males)provided 55 pre- and 60 post-SEEG dreams. Sufficientintraoperative monitoring information was obtained inall SEEG cases after an average of 10.5 ± 7.1 monitoringdays with 9.8 ± 3.6 electrodes. Out of 18 patients, 15 were considered surgical candidates. Three patients from thisgroup later also became part of the ATL group, and theirpre-SEEG dreams were used as the baseline for the as-sessment of changes in dream content. With the additionof another 3 ATL patients without prior SEEG, a total of6 patients (4 females and 2 males) provided 30 pre- and21 post-ATL dreams. The cohort’s baseline characteristicsare given in Table 1. There was no visible mesial temporalsclerosis in the ATL group on preoperative imaging. An-esthesia time (intubation to extubation) was significantlyshorter for the SEEG patients (199.8 ± 68.6 minutes) thanthe ATL patients (333.2 ± 71.1 minutes; p < 0.01); this wasalso the case for operation time (skin to skin), which was105.8 ± 42.4 minutes in the SEEG patients versus 262.7 ±69 minutes in the ATL patients (p < 0.01). Anesthesia wasinduced with propofol and lidocaine as well as midazolamin some cases. Maintenance was achieved using desfluraneand sevoflurane. In the ATL group, laterality was right in4 and left in 2 patients. One left ATL was performed in apatient under awake condition for intraoperative languagecortical mapping. The neocortex was removed at 4 to 5 cmbehind the temporal pole and the mesial structures resect-ed to 1 cm behind a coronal plane of the posterior surfaceof the midbrain, according to Girvin. No complicationsoccurred in either group.Surgical seizure outcome at 3 months was Engel classI in 3 and Engel class II in 2 patients. One patient hadunfavorable seizure outcome (Engel class IV), but showedworthwhile improvement (Engel class III) in the furthercourse. Pathology of the resected temporal lobes revealedgliosis in 2 patients, gliosis in combination with a focalcortical dysplasia type 1 in 3 patients, and no abnormali-ties in 1 patient. Patients With Drug-Resistant Epilepsy Compared to Historical NormsCompared to publicly available normative dreams, the ATL patients’ preoperative dreams were not statisti-cally different (Fig. 1). SEEG patients were preoperativelymore likely to have dreams with indoor settings comparedto the norms (p < 0.01; Fig. 2). Preoperative Dreams Compared to Postoperative DreamsPatients who underwent ATL were significantly lesslikely to have dreams involving physical aggression fol-lowing surgery (p < 0.01; Fig. 3). There are three indi-vidual categories related to this: “aggressor,” “physicalaggression,” and “aggression.” All of them are separateentities. It should be noted that the categories “aggressor”and “aggression” were not significantly different and pos-sibly even more frequent in post-ATL patients by trend.There were no significant differences in the dream contentof patients prior to and after SEEG procedures across themajor categories analyzed (Fig. 4). DiscussionSupporting the role of the temporal lobe in dream pro-duction, Bentes et al. showed that the dream content ofpatients with temporal lobe epilepsy differs from that ofthe general population. For 5 consecutive days, 52 patients recorded their dreams while being monitored withelectroencephalography in-hospital for epilepsy surgicalassessment. These patients had decreased dream recall,dreams with a higher percentage of familiar settings,and fewer dreams featuring dreamer success. This is incontrast to our analysis, which detected no difference in setting familiarity or success compared to normativedata. Instead, we found that pre-SEEG patients had moredreams featuring indoor settings than the historical norm.As social isolation often prevails in these patients’ lives, incorporation of daily experiences into their dreams as so-called “day’s residues” could explain this observation.However, direct comparison is difficult for two reasons.First, many of our SEEG patients had extratemporal lobeepilepsy, and second, the SEEG patients were compared toexternally obtained norms rather than a healthy matchedcontrol group who recorded dreams for 5 days at home. ATL and Dream ContentAs a primary finding of this prospective study on a co-hort of patients with drug-resistant epilepsy, significantly less (in fact no) physical aggressive features were seen indream content of patients following ATL. On the contrary,the “aggressor” and “aggression” remained unchanged.The reason behind this is that the latter categories repre-sent the total aggression, which also includes nonphysi-cal aggression, such as yelling or arguing, for example.Within the limitations of the study design in mind, twocareful explanations of this observation can be attempted.ATL entails amygdalectomy, and the amygdala, whichis time locked to be activated during rapid eye movement(REM) sleep, has previously been attributed to fear con-ditioning in dreams. The amygdala even takes a cen-tral role in Revonsuo’s threat-simulation theory of the an-cestral human. By means of deep brain amygdala stimu-lation, Lai et al. provoked dreamy states enriched with vivid and bizarre emotional elements in 2 patients withposttraumatic stress disorder. Further evidence comesfrom De Gennaro et al., who established a correlation ofvolumetric and ultrastructural measures of the amygdalawith dream emotional load and bizarreness in 34 healthysubjects. In 8 patients with Urbach-Wiethe disease (a rarecondition with bilateral calcification of the basolateralamygdala), Blake et al. showed that threatening dreamcontent still occurred, but was less negatively charged andthus perceived as more pleasant.As several lines of evidence converge on the amygdalaas a node for affective salience, we hypothesize that theunilateral resection of a (potentially dysfunctional) amyg-dala may have exerted a “taming effect” on dream con-tent in our patients. None of the ATL patients preoperatively stood outwith peri- or interictal aggressive behaviors. Absence or amarkable reduction of seizures, which was achieved in 5out of the 6 ATL patients at 3 months, is therefore unlikelyto directly account for a decrease in aggression states inthese patients’ lives and—presumably—dream content.However, it is conceivable that the indirect positive ef-fect of a favorable seizure outcome by means of ATL onthe patients’ well-being created a more at-ease mindset.This, in turn, might be subconsciously reflected by morepeaceful dream contents. Notwithstanding, this argumentis not further substantiated and it should be acknowledgedthat various psychological factors may have influenced theinner life of our patients. For example, it is possible thatanxiety in anticipation of their upcoming operation might have unfolded aggression in dreams during the preopera-tive waiting period. Anxiety peaks have previously beendescribed in a hospitalized neurosurgical patient’s dreamcontent. However, this situation would have applied tothe SEEG group as well, who showed no pre- and post-operative dream content differences. Antiepileptic drugscan also facilitate aggression in certain individuals withepilepsy. Among those with the strongest evidence for ag-gression are levetiracetam, perampanel, and possibly topi-ramate. However, these antiepileptic drugs in questionwere distributed in equal parts in both groups, and left un-changed up to at least 9 months after surgery at our insti-tution. Another explanation for the observed decrease inphysical aggression in postoperative ATL patients’ dreamsmight be a change in sleep patterns. McNamara et al. discovered that aggressive social interactions were morecharacteristic of REM than non-REM sleep phases orwake reports. Although no reports on sleep patterns afterbrain surgery exist, it is known that patients with traumaticbrain injury spend less time in REM sleep and perhapsthis holds true for ATL patients as well. SEEG and Dream ContentAs a secondary finding, the effect of general anesthe-sia on dream content in patients undergoing depth elec-trode insertion for SEEG (if present at all) was statisticallynegligible. The inclusion of this control group undergoing“sham surgery” is a unique strength of the study concept.Some patients were later re-recruited into the surgical group, thus raising intergroup homogeneity. The fact thatpre- and post-SEEG dream content with a shorter follow-up of 1 month remained stable also strengthens the robust-ness of the changes seen in the ATL group, who had alonger follow-up of 3 months. For anesthesiologists andpatients receiving general anesthesia, it is interesting toknow that dream content will remain unaffected. Howev-er, confirmatory studies for patients undergoing extraneu -rological surgery and anesthesia with different anestheticsare needed. Study Strengths and LimitationsTo the best of our knowledge, we are the first to in-vestigate the change of dream content in patients elec-tively undergoing brain surgery. This prospective study has a higher scientific value over the previous case reportssummarized by Solms that did not capture prelesionaldream content. In comparison to tumor-invaded brain,for example, the resected temporal lobes in our temporallobe epilepsy cohort can be regarded as relatively healthy.However, epileptic brain tissue is always dysfunctional tosome degree. Since resective brain surgery is not under-taken for healthy brains, the current cohort is closest tothe normal population in a neurosurgical setting. The lowstudy participant inclusion rate—especially in the ATLgroup—in a high-volume epilepsy surgery center over 2years is a major shortcoming mainly attributable to lackof dream recall. This is unsurprising in a population withepilepsy and confirms the findings of previous studies.Bentes et al. showed that patients with drug-resistant temporal lobe epilepsy have lower dream recall (71.2%)than healthy controls (87.8%). More than half of the pa-tients who consulted our clinic for presurgical investiga-tions could not participate because they never or veryrarely remembered their dreams. The much lower dreamrecall rate in our patients compared to that in the temporallobe epilepsy patients reported by Bentes et al. is likelyexplained by the fact that many of the patients in the pres-ent study had extratemporal epilepsy and also generalizedseizures. Bonanni et al. showed that dream recall was halfas likely in patients with generalized seizures as in pa-tients with complex partial seizures. Perhaps writing upthe most recent dream posed an inconvenience to some ofour patients. Whereas some patients might therefore nothave tried to remember their dreams in the morning, oth-ers perceived this as interesting to do, which leads to se-lection bias. The gold standard remains laboratory dreamreports; however, home-recorded dream reports providevalid data and have therefore been used by others.Interestingly, home-recorded dreams show a higher pre-ponderance of physical aggression than dreams recalled inthe laboratory. More qualitative dream assessment toolsto provide further insight into the emotional component,such as bizarreness or the occurrence of parasomnias suchas nightmares, for example, would have been desirable.Ideally, a second later follow-up had been included. Likely,however, more drop-outs would have weakened the long-term results. Last, our results may not be generalizable asaggression in particular differs by culture and gender.Taking into account the limitations, this study can be re-garded as a pilot paving the way for future high-volumedream studies in neurosurgery. ConclusionsThe study results suggest that perhaps the temporallobe with the amygdala as an emotional integrator fordreaming plays a role in the generation of aggressivedream content within the DDN. As a result of ATL, ag-gression is less predominant in patients’ dreams. All pa-tients receiving general anesthesia can be counseled thattheir dream content will remain unaffected. With the re-sults of the current study and findings of future studiesassessing different brain regions that belong to the DDN,the generation of dream content will be further unraveled.Last, neurosurgeons and physicians involved in the careof patients with drug-refractory epilepsy or other neu-rological conditions are encouraged to become more at-tuned to their patients’ inner life, which may be accessiblethrough their dreams. ------------------------------------------------------------------------------------ We have made strides in the field of dream research, utilizing brain imaging techniques and studying neural activity during sleep. However, these methods only scratch the surface, providing a mere glimpse into the complex tapestry of our dreams. We are left with abstract patterns of brain activity, but no true visual representation of what occurs within the dreamer's mind. Dreams are deeply personal experiences, and being unable to share or visually communicate them poses a significant limitation. How marvelous it would be if we could express the surreal landscapes, vibrant colors, and fantastical characters that dance within our dreams! Such visuals could foster greater understanding and connection among individuals, as we would be able to explore the depths of our collective subconscious. Artists and writers have long attempted to capture the essence of dreams through their work, but these interpretations often fall short of the true experience. The frustration intensifies when we realize that no matter how skilled the artist, their depiction remains a mere approximation, an interpretation of an experience they did not personally witness. I express my disappointment and frustration with the current limitations in visualizing dreams. Let us strive for a future where dreams are not only experienced but also visually shared, understood, and celebrated.

Disclaimer:This document was prepared by the Congressional Research Service (CRS). CRS serves as nonpartisan shared staff tocongressional committees and Members of Congress. It operates solely at the behest of and under the direction of Congress.Information in a CRS Report should not be relied upon for purposes other than public understanding of information that hasbeen provided by CRS to Members of Congress in connection with CRS’s institutional role. CRS Reports, as a work of theUnited States Government, are not subject to copyright protection in the United States. Any CRS Report may bereproduced and distributed in its entirety without permission from CRS. However, as a CRS Report may includecopyrighted images or material from a third party, you may need to obtain the permission of the copyright holder if youwish to copy or otherwise use copyrighted material. AbstractThis paper tests two hypotheses regarding howwell two distinct Long-Term Evolution (LTE) networkproblems can be detected through supervised techniqueswith near-real-time performance. The tested networkproblems are physical-cell-identity (PCI) conflicts androot-sequence-index (RSI) collisions. These were labeledthrough confi gured cell relations that verified these twoconfl icts. Furthermore, a real LTE network was used. Theresults obtained showed that both problems were bestdetected by using each key performance indicator (KPI)measurement as an individual feature. The highest averageprecisions obtained for PCI conflict detection were 31%and 26% for the 800 MHz and 1800 MHz frequency bands,respectively. The highest average precisions obtained forRSI collision detection were 61% and 60% for the 800 MHzand 1800 MHz frequency bands, respectively.1. IntroductionTwo of the major concerns of mobile network operators(MNO) are to optimize and to maintain networkperformance. However, maintaining performance hasproven to be a challenge mainly for large and complexnetworks. In the long term, changes made in the networksmay increase the number of conflicts and inconsistenciesthat occur in them. These changes include changing thetilting of antennas, changing the cell’s power, or evenchanges that cannot be controlled by the mobile networkoperators, such as user mobility and radio-channel fading.In order to assess the network’s performance,quantifiable performance metrics, known as key performanceindicators (KPI), are typically used. Key performanceindicators can report network performance such as thehandover success rate and the channel interference averagesof each cell, and are periodically calculated, resulting in timeseries. A time series can be either univariate or multivariate.As this study uses data samples that represent LTE cellswith several measured key performance indicators, thenthe data consist of multivariate time series.This paper focuses on applying supervised techniquesfor detecting two known LTE network conflicts, namelyphysical-cell identity (PCI) conflicts and root-sequenceindex (RSI) collisions. The labeling used was only possibledue to a CELFINET product that allows obtaining cellrelations that label the two mentioned network conflicts;also, real data obtained from a LTE network was used. Theaim of this paper is to test two hypotheses regarding how welltwo distinct LTE network problems can be detected throughsupervised techniques with near-real-time performance.The resulting conflict-detection solution would then runin an entity external from the LTE architecture during theearly morning. The solution would then alert the networkengineers of any existing conflicts in order to have promptresponses.As this paper aims to create models for near-real-time detection of PCI conflicts and RSI collisions, thepopular k-nearest neighbors with dynamic-time-warpingclassifi cation approach was not tested [1]. The reason forthis decision was based on the fact that it is computationallyintensive and very slow for large data sets, as was the casefor this paper.In order to automatically detect the network faultcauses, some work has been done by using key performanceindicator measurements with unsupervised techniques, asin[2]. 12 The Radio Science Bulletin No 364 (March 2018)The paper is organized as follows. Section 2 introducesthe analyzed network problems, namely PCI conflictsand RSI collisions. Section 3 presents the chosen keyperformance indicators and machine-learning (ML) models,the two proposed hypotheses, and describe how the modelsobtained were evaluated. Section 4 presents the resultsobtained. Finally, conclusions are drawn in Section 5.2. Network Problems Analyzed2.1 Physical Cell Identity ConflictEach LTE cell has two identifiers with diff erentpurposes: the Global Cell Identity (ID) and the PCI. TheGlobal Cell ID is used to identify the cell from an operation,administration, and management perspective. The PCI isused to scramble the data in order to aid mobile phonesin separating information from different transmitters [3].Since an LTE network may contain a much larger numberof cells than the 504 available values of PCIs, the samePCI must be reused by several cells. However, the userequipment (UE) cannot distinguish between two cells ifthey both have the same PCI and frequency, a situationcalled as PCI conflict.PCI confl icts can be divided into two cases: PCIconfusions and PCI collisions. PCI confusions occurwhenever an LTE cell has two different neighbor LTEcells with equal PCIs, in the same frequency band [4]. PCIcollisions happen whenever an LTE cell has a neighborLTE cell with identical PCI in the same frequency band [4].A good PCI plan can be applied to avoid PCI conflicts.However, it can be difficult to do such a plan without gettingany PCI conflicts in a dense network. Moreover, networkchanges – namely increased cell power and variable radioconditions – can lead to PCI conflicts. PCI confl icts canlead to an increase in dropped-call rate due to failedhandovers, as well as an increase of blocked calls andchannel interference [4].2.2.2 Root Sequence IndexCollisionThe user equipment has to perform the LTE random-access procedure to connect to an LTE network, establishor reestablish a service connection, perform intra-systemhandovers, and synchronize for uplink and downlinkdata transfers. The LTE random-access procedure can beperformed using two different solutions: allowing non-contention-based and contention-based solutions. An LTEcell uses 64 physical random-access channel (PRACH)preambles. Twenty-four of those preambles are reservedby the evolved-NodeB for non-contention-based access.The remaining 40 preambles are randomly selected by theuser equipment for contention-based access [3].The 40 physical random-access-channel preamblesthat the user equipment can use are calculated by the userequipment through the RSI parameters that the LTE celltransmits in the system information block 2 through thephysical random-access channel [5]. Whenever two or moreneighbor cells operate in the same frequency band and havethe same RSI parameter, this results in the connected userequipment calculating the same 40 physical random-accesschannel preambles, increasing the occurrence of preamblecollisions. The aforementioned problem is known as RSIcollision, and can lead to an increase of failed serviceestablishments and re-establishments, as well as an increaseof failed handovers.3. MethodologyThis study was performed using real data from anLTE network of a mobile network operator with a PCIreuse factor of three. Furthermore, data were collected forthe same weekday of three consecutive weeks, for everyperiod of 15 minutes, the minimum temporal granularityused by network operators, resulting in a daily total of 96measurements.Using a CELFINET tool, it was possible to labelcells that had PCI conflicts and/or RSI collisions. Sourcecells that had configured neighbor cells with equal PCIin the same frequency band were labeled as having a PCIcollision. Source cells that had two or more neighborcells with equal PCI in the same frequency band betweenthemselves were labeled as having a PCI confusion. Sourcecells that had neighbor cells with equal RSI in the samefrequency band were labeled as having an RSI collision.Cells that did not present any of these conflicts were labeledas non-confl icting.3.1 Proposed Key PerformanceIndicatorsThe fi rst step involved in collecting a list of keyperformance indicators for LTE equipment was to choosethe most-relevant key performance indicators for detectingPCI confl icts and RSI collisions. The key performanceindicators were chosen by taking into account the theorybehind LTE and how PCI and RSI are used. Accordingly,the following key performance indicators were chosen forPCI confl ict detection:• Average CQI: the average channel quality indicatormeasured by the user equipment• UL PUCCH Interference Avg and UL PUSCHInterference Avg: the average measured interferencein the physical uplink control and shared Channel• Service Establish: the amount of established serviceconnections The Radio Science Bulletin No 364 (March 2018) 13• Service Drop Rate: the ratio of the dropped serviceoccurrences• DL Avg Cell Throughput Mbps: the average measuredcell downlink throughput in Mbit/s• DLAvg User Equipment Throughput Mbps: the averagemeasured user equipment downlink throughput in Mbit/s• DL Latency ms: the average duration an Internet protocolpacket takes since being sent by the user equipmentuntil reaching back to it• RandomAcc Succ Rate: the success rate of establishedservices made through the random access channel• IntraFreq Prep HO Succ Rate and IntraFreq Exec HOSucc Rate: the success rate of handover preparation andexecution between cells operating in the same frequencyband.To detect RSI collisions, a subsection of theaforementioned key performance indicators were selected,namely: UL PUCCH Interference Avg, UL PUSCHInterference Avg, Service Establish, IntraFreq Exec HOSucc Rate, IntraFreq Prep HO Succ Rate, and RandomAccSucc Rate.After discarding cells with high null key performanceindicator measurements and interpolating those of theremaining cells, it was decided to separate the data intodifferent frequency bands, namely the 800 MHz and1800 MHz bands. The 2100 MHz and 2600 MHz frequencybands were not considered, as they represented only 9% ofthe data, and had few occurrences of PCI conflicts and RSIcollisions. This decision to separate the data into differentfrequency bands was taken in order to create frequency-dependent models, since different frequency bands havediff erent purposes.The cleaned data for PCI conflict detection in the800 MHz frequency band consisted of 8666 non-conflictingcells, 1551 PCI confusions, and six PCI collisions. The1800 MHz frequency-band data had 16675 non-conflictingcells, 1294 PCI confusions, and no PCI collisions. Thedata concerning each frequency band was split into 80%for the training set and 20% for the test set. Additionally,as PCI collisions are very rare, it was decided to do a 50%split for collisions, yielding three collisions in both thetraining and test sets.The cleaned data for RSI collision detection inthe 800 MHz frequency band consisted of 10128 non-confl icting cells and 6774 RSI collisions. The 1800 MHzfrequency-band data consisted of 17634 non-conflictingcells and 10916 RSI collisions. The data relative to eachfrequency band was split into 80% for the training set and20% for the test set.3.2 Considered Classifi cationAlgorithmsIn order to reduce the bias from this study, fivediff erent classification algorithms were set. The aim of theclassifi ers was to classify cells as either non-conflictingor confl icting, depending on the detection use case. Theconsidered classification algorithm implementations weretaken from the Python Scikit-Learn library [6], and werethe following:3.2.1 Adaptive Boosting (AB)Adaptive Boosting is an ensemble method, whichis a class of a machine-learning approaches based on theconcept of creating a highly accurate classifier by combiningseveral weak and inaccurate classifiers. Adaptive Boostinguses subsets of the original data to produce weak performingmodels (high bias, low variance) and then boosts theirperformance by combining them together based on a chosencost function. Adaptive Boosting was the first practicalboosting algorithm, and remains one of the most used andstudied classifiers [7]. Its implementation uses decision-treeclassifi ers as weak learners.3.2.2 Gradient Boost (GB)Gradient Boost is another popular boosting algorithmfor creating collections of classifiers. It differs from AdaptiveBoosting because it calculates a negative gradient of a costfunction (direction of quickest improvement), and picksa weak learner that is closest to the obtained gradient toadd to the model [8]. The Gradient Boost implementationconsidered uses Decision Trees (DT) as weak learners.3.2.3 Extremely RandomizedTrees (ERT)This belongs to the family of tree ensemblemethods, and uses a technique different from boosting,known as bagging. Bagging-based algorithms aim tocontrol generalization error by perturbing and averagingthe generated weak learners, such as decision trees. TheExtremely Randomized Trees algorithm stands out fromother tree-based ensemble classifiers because it stronglyrandomizes both feature and cut-point choice while splittinga tree node [9]. Extremely Randomized Trees aims tostrongly reduce variance through a full randomization of thecut-point and feature combined with ensemble averagingwhen compared to other algorithms. By training each weaklearner with the full training set instead of data subsets,Extremely Randomized Trees also minimizes bias. 14 The Radio Science Bulletin No 364 (March 2018)3.2.4 Random Forest (RF)Random Forest is another bagging-based algorithm inthe family of tree ensemble methods. Similarly to ExtremelyRandomized Trees, several small and weak trees can begrown in parallel, and these set of weak learners result ina strong classifi cation algorithm either by averaging or bymajority vote [10]. Random Forest is similar to ExtremelyRandomized Trees, but differs in two aspects. RandomForest uses data subsets for growing its trees, whileExtremely Randomized Trees uses the whole training set.Random Forest chooses a small subset of features to bechosen on splitting a node, while Extremely RandomizedTrees chooses a random feature from all features.3.2.5 Support Vector Machines(SVM)Support Vector Machines aim to separate data samplesof different classes through hyperplanes that define decisionboundaries. Similarly to Decision-Trees-based classifiers,Support Vector Machines are capable of handling linear andnonlinear classification tasks. The main idea behind SupportVector Machines is to map the original data samples fromthe input space into a high-dimensional feature space suchthat the classification task becomes simpler [11].3.3 Proposed HypothesesIn order to reduce bias even further, two hypotheseswere proposed to find the one that led to the best-performingmodels for PCI confl ict and RSI collision detection.3.3.1 Statistical Data ExtractionClassificationPCI confl icts and RSI collisions are better detectedby extracting statistical calculations from the daily timeseries of each key performance indicator and using themas features for classification. The Python tsfresh toolwas used to extract statistical data from the time series[12]. tsfresh applies several statistical calculations tothe data, followed by feature elimination through statisticalsignifi cance testing. As it resulted in hundreds of features,Principal Component Analysis (PCA) was applied fordimensionality reduction before applying the data intothe Support Vector Machine classifier. This decision wastaken because Support Vector Machine takes longer toconverge as the dimensionality increases, while it does notsignificantly increase the training and testing times of thetree-based classifiers. It was decided to use a number ofprincipal components (PC) that led to 98% of the cumulativeproportion of variance explained, maintaining most of theoriginal variance.3.3.2 Raw Cell Data ClassificationPCI confl icts and RSI collisions are better detectedby using each cell’s daily key performance indicatormeasurements as an individual feature. This hypothesis wasproposed to compare a more computationally intensive butsimpler approach with the previous hypothesis. Moreover,as there were 96 daily measurements per key performanceindicator in each cell, by using, for instance, 10 keyperformance indicators, this would have yielded 96 × 10= 960 features. Due to the high dimensionality of the datato test this hypothesis, Principal Component Analysis wasapplied (once again) to reduce its dimensionality beforeusing the Support Vector Machine classifier. It was decidedto use a number of principal components that led to 98% ofthe cumulative proportion of variance explained.3.4 Model EvaluationIn a binary decision problem, a classification algorithmlabels predictions as either positive or negative. A predictionfor confl ict detection could fit into one of these fourcategories: True Positive (TP), conflicting cells correctlylabeled as conflicting; False Positive (FP), non-conflictingcells incorrectly labeled as conflicting; True Negative (TN),non-conflicting cells correctly labeled as non-conflicting;False Negative (FN), conflicting cells incorrectly labeledas non-confl icting.As there was a high interest in knowing how wellthe models obtained could classify PCI conflicts and RSIcollisions, the classic accuracy metric by itself was notenough. Classifi cations where a non-conflicting cell waserroneously classified as a conflict were to be avoided; itwas thus chosen to additionally evaluate the models obtainedthrough the precision and recall metrics. The metrics usedcould then be defined as follows:TPRecall TP FN  , (1)TPPrecision TP FP  , (2)TP TNAccuracy TP TN FP FN    , (3)where Recall measures the fraction of conflicting cellsthat are correctly labeled, Precision measures the fractionof cells classified as confl icting that are truly conflicting,and Accuracy measures the fraction of correctly classifiedcells [13]. Precision can be thought of as a measure of aclassifi er’s exactness – a low precision can indicate a large The Radio Science Bulletin No 364 (March 2018) 15number of False Positives – while Recall can be seen as ameasure of a classifier’s completeness: a low recall indicatesmany False Negatives.Since a classification algorithm can output theprobabilities of a sample belonging to a specific class, theprobability decision threshold can be tuned to alter themodel’s classification outputs. For instance, increasingthe probability decision threshold to classify a specificclass may lead to an increase in Precision at the cost ofa lower Recall. Precision-Recall (PR) curves are built bychanging the decision probability threshold for a class. Itthus was decided to also evaluate models through theirPrecision-Recall curves in order to perform a thoroughmodel evaluation. Precision-Recall curves, often used ininformation retrieval [14], have been cited as an alternativeto Receiver Operator Characteristic curves for tasks witha large skew in the class distribution, as in PCI conflictdetection [15]. Additionally, the average Precision is alsorepresented by the Precision-Recall curves through the areasunder the curves. It should be noted that there is a tradeoffbetween the number of samples for model training, trainingduration, and model performance. With more data samplesand more training time, the resulting model generalizesbetter and has more time to learn the data structure.4. Results4.1 Physical Cell IdentityConfl ict Detection4.1.1 Statistical Data ExtractionClassificationThe fi rst hypothesis presented in Section 3.3 wastested using the data presented in Section 3.1. RegardingPCI confusion detection, tsfresh yielded 798 and909 signifi cant features for the 800 MHz and 1800 MHzfrequency bands, respectively. Concerning PCI collisiondetection, a total of 2200 features were extracted for the800 MHz case that were not selected through hypothesistesting, due to the dataset only containing a marginallylow number of six PCI collisions. Principal ComponentAnalysis was applied for dimensionality reduction fora faster Support Vector Machine convergence. For PCIconfusion detection, this resulted in 273 and 284 principalcomponents for the 800 MHz and 1800 MHz frequencybands, respectively.The optimal hyperparameters to create each modelwere obtained through a grid search on the training setwith 10-fold cross validation, maximizing the Precisionmetric. After training the models, they were tested on thetest set, based on a decision probability threshold of 50%.The results are presented in Table 1.It should be added that when a classifier did not classifyany True Positives or False Positives, the Precision wasrepresented as a Not a Number (NaN), since it resulted ina division by zero. The Adaptive Boosting model had thebest performance, with a 50% Precision for the 800 MHzfrequency band. However, no model classified a sample asconfl icting in the 1800 MHz frequency band data.In order to obtain more insights about the models’performance, the Precision-Recall curves were obtained,and are represented in Figure 1. The highest averagePrecision was 27%, by using the Gradient Boost classifier.The Gradient Boost presented the highest Precision mostlythroughout the plot. The Support Vector Machine was clearlythe worst-performing model, especially in the 1800 MHzfrequency band.The training and testing running times to obtain thePrecision-Recall curves were also collected. Gradient Boost,which resulted in the two best models, had a testing timebelow one second and a training time below 30 seconds forboth frequency bands. The learning curves were obtained,and they showed that the average Precision would onlymarginally increase with more data. Gradient Boost thusresulted in the overall best-performing models for bothfrequency bands by using statistical calculations as features.Regarding PCI collision detection, PrincipalComponent Analysis resulted in 619 principal componentsto be used by the Support Vector Machine classifier for the800 MHz frequency band. The optimal hyperparameterswere obtained, and the test results were collected aftertraining the models. A table with the results is not shown, asno tested model was able to classify a sample as conflicting.The Precision-Recall curves were obtained and plotted,showing a maximum Precision of 23% with 100% Recallby Random Forest, while this was approximately zero forthe remaining classifiers (the plot is not illustrated in thispaper as it would not add much information).800 MHz Band 1800 MHz BandModel Accuracy Precision Recall Accuracy Precision RecallERT 85.24% NaN 00.00% 93.27% NaN 00.00%RF 85.24% NaN 00.00% 93.27% NaN 00.00%SVM 85.24% NaN 00.00% 93.27% NaN 00.00%AB 85.24% 50.00% 02.83% 93.27% NaN 00.00%GB 85.18% 46.00% 02.43% 93.27% NaN 00.00%Table 1. Statistical-data-based PCI confusion classifi cation results. 16 The Radio Science Bulletin No 364 (March 2018)4.1.2 Raw Cell Data ClassificationThe second hypothesis presented in Section 3.3 wastested using the data described in Section 3.1. Using eachindividual key performance indicator measure as a feature,an average filter with a window of size 20 was applied toreduce the noise interference. Principal Component Analysiswas applied, which resulted in 634 principal components tobe used by the Support Vector Machine classifier for boththe 800 MHz and 1800 MHz frequency bands.Once again, the optimal hyperparameters wereobtained through grid search, and the test results werecollected after model training. The classification results fora 50% decision probability threshold are shown in Table 2.Overall, Gradient Boost was the classifier that led to thebest performance, having the highest Accuracy and Recallfor both frequency bands, but not the best Precision forthe 1800 MHz frequency band. Both models created by theExtremely Randomized Trees and Random Forest classifiershad a 100% Precision for the 1800 MHz frequency band,which meant that Random Forest could result in the bestmodel, as it had higher Recall.In order to see if Gradient Boost led to the bestperforming model, the Precision-Recall curves wereobtained, and they are presented in Figure 2. Regardingthe 800 MHz frequency band, Gradient Boost showed thehighest average Precision, with a peak of 60% Precisionfor 4% Recall. Concerning the 1800 MHz frequency band,Extremely Randomized Trees presented the best averagePrecision, while Gradient Boost achieved higher Precisionfor a Recall lower than 5%. Additionally, Random Forestwas not the best performing model, as was seen in Table 2.The training and testing running times for eachmodel were obtained. In the 800 MHz frequency band,Gradient Boost, which led to the best-performing model,had a testing time below one second and a training timebelow 14 seconds. Regarding the 1800 MHz frequencyband, Extremely Randomized Trees, which led to thebest-performing model, was one of the quickest to train(i.e., 40.3 seconds), but it was one of the slowest to test(i.e., 1.4 seconds). Nevertheless, its overall performancewas near real time.Regarding PCI collision detection, PrincipalComponent Analysis resulted in 634 principal componentsfor both frequency bands. The test results were collectedwith the optimal hyperparameters. The best performingmodel was the model obtained from Adaptive Boosting,as it detected one out of three PCI collisions with 100%Precision. However, due to the marginally low number ofPCI collisions in the dataset, the results were not sufficientlysignifi cant to draw any conclusions.Figure 1. The smoothed Precision-Recall curves for statistical-data-based PCI confusion detection.800 MHz Band 1800 MHz BandModel Accuracy Precision Recall Accuracy Precision RecallERT 85.37% 22.22% 00.71% 93.57% 100% 00.45%RF 85.63% NaN 00.00% 93.60% 100% 00.90%SVM 85.63% NaN 00.00% 93.54% NaN 00.00%AB 85.63% NaN 00.00% 93.54% NaN 00.00%GB 85.73% 75.00% 01.07% 93.63% 80.00% 01.80%Table 2. Raw-cell-data PCI confusion classifi cation results. The Radio Science Bulletin No 364 (March 2018) 174.2 Root Sequence IndicatorCollision Detection4.2.1 Statistical Data ExtractionClassificationThe fi rst hypothesis presented in Section 3.3 wastested using the data described in Section 3.1. RegardingRSI collision detection, tsfresh yielded 732 and851 signifi cant extracted features for the 800 MHz and1800 MHz frequency bands, respectively. In order toreduce the data dimensionality for applying to the SupportVector Machine model, Principal Component Analysis wasapplied, resulting in 273 and 284 principal components forthe 800 MHz and 1800 MHz frequency bands, respectively.The optimal hyperparameters were obtained throughgrid search, and the test results are presented in Table 3.The Extremely Randomized Trees model delivered thehighest Precision for both frequency bands, but GradientBoost had the highest overall Accuracy and Recall.In order to gain more insights regarding theperformance of the models, Precision-Recall curves wereobtained and are presented in Figure 3. The Gradient Boostmodel was the best for both frequency bands, having aPrecision peak of 85% and an average Precision of 61%.The abnormal curve behavior of the Adaptive Boostingmodel was due to the assignment of several cells with thesame probability values.The training and testing running times for each modelwere obtained. The Gradient Boost model showed testingtimes lower than one second; however, it had one of thehighest training times. More specifically, it required 28.4and 246 seconds of training time for the 800 MHz and1800 MHz frequency bands, respectively. Nonetheless,the Gradient Boost model presented higher performancerelative to other obtained models with near-real-timeperformance, thus overall being the best model. The learningcurves obtained showed that the performance would notsignificantly increase if more data were added to the dataset.4.2.2 Raw Cell Data ClassificationThe second hypothesis presented in Section 3.3 wastested using the data described in Section 3.1. Using eachindividual key performance indicator’s measure as a feature,an average filter with a window of size 20 was applied.Principal Component Analysis was applied, which yielded in332 principal components to be used by the Support VectorMachine classifier for both the 800 MHz and 1800 MHzfrequency bands for RSI collision detection.The optimal hyperparameters were obtained throughgrid search, and the results are presented in Table 4. OnceFigure 2. The smoothed Precision-Recall curves for raw-cell-data-based PCI confusion detection.800 MHz Band 1800 MHz BandModel Accuracy Precision Recall Accuracy Precision RecallERT 60.32% 100% 00.48% 62.27% 72.97% 02.00%RF 64.93% 61.30% 32.62% 64.13% 66.94% 12.12%SVM 60.94% 54.80% 11.55% 61.79% NaN 00.00%AB 64.02% 56.79% 40.83% 66.37% 59.88% 36.29%GB 66.87% 61.60% 44.88% 69.39% 63.97% 45.53%Table 3. Statistical-data-based RSI collision classifi cation results. 18 The Radio Science Bulletin No 364 (March 2018)more, the Gradient Boost model revealed more Accuracy forboth frequency bands. The Random Forest and ExtremelyRandomized Trees models had the highest Precision forthe 800 MHz and 1800 MHz frequency bands.The Precision-Recall curves were obtained and arepresented in Figure 4. The Gradient Boost model had thehighest average Precision, while the Random Forest andExtremely Randomized Trees models showed slightlyworse average Precision.The training and testing running time for each modelwere obtained. The Gradient Boost model showed testingtimes lower than one second, and the third highest trainingtimes for both frequency bands. More precisely, it took 12.8and 24.4 seconds to train in the 800 MHz and 1800 MHzfrequency bands, respectively. However, the GradientBoost model’s performance was in near real time, and itwas thus overall the best-performing model. The learningcurves obtained showed that the results would improve ifmore data were added to the training set, especially for theGradient Boost model.5. ConclusionsThis paper tested two hypotheses regarding howwell two distinct LTE network problems could be detectedthrough supervised techniques with near-real-timeperformance.The PCI confusions were better detected by usingthe measurement of each cell’s daily key performanceindicators as an individual feature. This was concluded dueto the result that the average Precision was higher whiletesting this hypothesis. Specifically, the average Precisionsreached 31% and 26% for the 800 MHz and 1800 MHzfrequency bands, respectively. No conclusions could bereached regarding PCI collision detection due to the lownumber of PCI collisions in the data set.The RSI collisions were detected with similarperformance by two proposed hypotheses. However, onecould say that the best detection was obtained by using themeasurement of each cell’s daily key performance indicatorsas an individual feature because the learning curves showedthat the results would further improve if more data was addedfor the second hypothesis. The best-performing model wasthe model that used the Gradient Boost classifier, reachingaverage Precisions of 61% and 60% for the 800 MHz and1800 MHz frequency bands, respectively.The results showed that supervised techniques forPCI and RSI confl ict detection are not well suited. This isbecause while a cell may have one of these two conflicts,the confl ict’s impact on the key performance indicatorsmight be negligible. This fact can be due to several factors,such as the distance between cells, their azimuth, and theenvironment. For future work, an unsupervised approachfor network confl ict detection followed by manual labelingto be used by a classifier could be investigated. This would800 MHz Band 1800 MHz BandModel Accuracy Precision Recall Accuracy Precision RecallERT 59.49% 50.00% 00.83% 59.83% 75.00% 00.22%RF 61.70% 62.64% 13.52% 65.55% 63.86% 33.07%SVM 60.07% 52.24% 16.61% 59.25% 46.67% 09.14%AB 64.73% 60.38% 37.60% 64.99% 59.59% 40.32%GB 66.41% 60.84% 47.92% 66.22% 62.72% 39.52%Table 4. Raw-cell-data RSI collision classifi cation results.Figure 3. The smoothed Precision-Recall curves for statistical-data-based RSI collision detection. The Radio Science Bulletin No 364 (March 2018) 19result in the labeling of cells with significant differencesbetween them, which could lead to better classificationresults.

#This script will return CUI information for a single search term.#Optional query parameters are commented out below. import requestsimport argparseimport json def results_list(input): parser = argparse.ArgumentParser(description='process user given parameters') parser.add_argument("-k", "--apikey", required = True, dest = "apikey", help = "enter api key from your UTS Profile") parser.add_argument("-v", "--version", required = False, dest="version", default = "current", help = "enter version example-2021AA") parser.add_argument("-s", "--string", required = True, dest="string", help = "enter a search term, using hyphens between words, like diabetic-foot") #args = parser.parse_args() apikey = "xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx" #version = args.version string = input uri = "https://uts-ws.nlm.nih.gov" content_endpoint = "/rest/search/current" full_url = uri+content_endpoint page = 1 try: search = True while search ==True and page == 1: query = {'string'tring,'apiKey':apikey, 'pageNumber'age} #query['includeObsolete'] = 'true' #query['includeSuppressible'] = 'true' #query['returnIdType'] = "sourceConcept" #query['sabs'] = "SNOMEDCT_US" r = requests.get(full_url,params=query) r.raise_for_status() print(r.url) r.encoding = 'utf-8' outputs = r.json() items = (([outputs['result']])[0])['results'] if len(items) == 0: if page == 1: print('No results found.'+'\n') break else: break print("Results for page " + str(page)+"\n") cui = [] for result in items: if str(result['rootSource']) == "" or "" or "" or "" or "" and search == True: if page == 1: print('UI: ' + result['ui']) cui.append(result['ui']) print('URI: ' + result['uri']) print('Name: ' + result['name']) print('Source Vocabulary: ' + result['rootSource']) print('\n') search = False print('*********') return cui except Exception as except_error: print(except_error) print("Eingabe: ")eingabe = input().capitalize()CUI = results_list(eingabe)#url4 = "https://uts-ws.nlm.nih.gov/rest/content/current/CUI/"+ str(CUI) +"/atoms?language=ENG&apiKey=xxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx" namensliste = [] for item in CUI: url4 = "https://uts-ws.nlm.nih.gov/rest/content/current/CUI/"+ str(item) +"/atoms?language=ENG&apiKey=xxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx" response = requests.get(url4) # Print the responsess response_js = response.json() response_json = json.dumps(response_js) # parse the JSON data parsed_data = json.loads(response_json) #print (parsed_data) try: #for lang["language"] in parsed_data: name = [result["name"] for result in parsed_data["result"]] #print (name) for item in name: if not item in namensliste: namensliste.append(item.capitalize()) except: print("skipped") #for item in namensliste: # item.capitalize() print(namensliste)

We take advantage of a variety of devices in our day-to-day life, and we might treat them as just pieces of hardware, elements fulfilling a certain purpose — forgotten about until it’s time to use them. [Jasmine Lu] and [Pedro Lopes] believe that these relationships could work differently, and their recent paper describes a wearable device that depends on you as much as you depend on it. Specifically, they built wrist-worn heart rate sensors and designed a living organism into these, in a way that it became vital to the sensor’s functioning. The organism in question is Physarum polycephalum, a slime mold that needs water to stay alive and remain conductive — if you don’t add water on a regular basis, it eventually dries out and hibernates, and adding water then will revive it. The heart rate sensor’s power rail is controlled by the mold, meaning the sensor functions only as long as you keep the mold alive and healthy. In their study, participants were asked to wear this device for one-two weeks, and the results go way beyond what we would expect from, say, a Tamagotchi — with the later pages describing participant reactions and observations being especially impressive. For one, the researchers found that the study participants developed a unique sense of connection towards the slime mold-powered device, feeling senses of responsibility and reciprocity, and a range of other feelings you wouldn’t associate with a wearable. Page 9 of the paper tells us how one participant got sick, but still continued caring for the organism out of worry for its well-being, another participant brought her “little pet mold friend” on a long drive; most participants called the slime a “friend” or a “pet”. A participant put it this way: […] it’s always good to be accompanied by some living creature, I really like different, animals or plants. […] carrying this little friend also made me feel happy and peaceful.There’s way more in the paper, but we wouldn’t want to recite it in full — you should absolutely check it out for vivid examples of experiences that you’d never have when interacting with, say, your smartphone, as well as researchers’ analysis and insights. With such day-to-day use devices, developing a nurturing relationship could bring pleasant unexpected consequences – perhaps, countering the “kept on a shelf since purchase” factor, or encouraging repairability, both things to be cherished. If you’ve ever overheard someone talking about their car or laptop as if it were alive, you too might have a feeling such ideas are worth exploring. Of course, not every device could use a novel aspect like this, but if you wanted to go above and beyond, you could even build a lamp that needs to be fed to function. _____________________ AbstractHolographic methods from optics can be adapted to acoustics for enabling novel applications in particle manipulation or patterning by generating dynamic custom-tailored acoustic fields. Here, we present three contributions towards making the field of acoustic holography more widespread. Firstly, we introduce an iterative algorithm that accurately calculates the amplitudes and phases of an array of ultrasound emitters in order to create a target amplitude field in mid-air. Secondly, we use the algorithm to analyse the impact of spatial, amplitude and phase emission resolution on the resulting acoustic field, thus providing engineering insights towards array design. For example, we show an onset of diminishing returns for smaller than a quarter-wavelength sized emitters and a phase and amplitude resolution of eight and four divisions per period, respectively. Lastly, we present a hardware platform for the generation of acoustic holograms. The array is integrated in a single board composed of 256 emitters operating at 40 kHz. We hope that the results and procedures described within this paper enable researchers to build their own ultrasonic arrays and explore novel applications of ultrasonic holograms.Keywords:acoustic hologram algorithm; open ultrasonic array; acoustic tweezers 1. IntroductionThe ability to produce dynamic ultrasonic fields with target shapes is of fundamental importance in ultrasonic imaging [1], nondestructive testing [2,3], and high-intensity focused ultrasound HIFU therapy [4]. When operating in air, there are numerous emerging applications that require the generation of acoustic fields with certain shapes, such as noncontact tactile feedback [5,6,7], volumetric displays [8,9], parametric audio generation [10,11], and the contactless manipulation of objects [12,13,14,15,16].In recent years, optical holographic methods have been adapted to acoustics [13,16,17,18], opening the possibility of generating arbitrary acoustic fields that can be controlled in real time. Acoustic holography is normally achieved using either passive metamaterial structures [17,19,20] or an array of ultrasonic transducers [13,16,21]. Metamaterial structures have the main advantage of allowing for the generation of acoustic fields with a higher spatial resolution, but they cannot dynamically change the field. In contrast, phased arrays do not have this limitation, since the emission phase and amplitude of each transducer can be controlled by a computer, allowing to change the acoustic field in real time. This capability of phased arrays is encapsulated in commercially available platforms, e.g., Ultraleap, Bristol, UK; Pixie Dust Tech., Tokyo, Japan; SonicEnergy, California, USA, each of which provides technology development and commercialisation towards specific target market solutions. Despite the numerous scientific advancements made in both industry and academia, there is currently no unifying hardware platform that can flexibly support exploratory research in acoustic holography applications.In this paper, we present SonicSurface, a low-cost open hardware array for generating arbitrary acoustic fields in mid-air. We also present an algorithm for calculating the emission amplitude and phase for each transducer in order to create a target amplitude field at a certain distance from the array. Additionally, we offer a comparison of the accuracy of the generated fields depending on the size of the ultrasonic emitters as well as their phase and amplitude resolution. This paper is accompanied by video instructions, available at www.upnalab.com (accessed on 20 February 2021), Do-it-Yourself. Given the low-cost and the use of off-the-shelf components, we hope that researchers can build and use these ultrasonic arrays for their own experiments. We also note the companies commercializing ultrasonic phased arrays offer proprietary solutions that are certified for their use in various commercial applications.2. Related WorkOur review of related work gives an overview of projects that designed and built ultrasonic arrays that typically operate at 40 kHz. Additionally, we provide a review of algorithms for creating an arbitrary pressure field.An ultrasonic phased array consists of a collection of elements that can transmit or receive ultrasonic waves with specific time delays (phases offsets) and amplitudes. This technology enables the generation of arbitrary pressure fields by controlling the phases and amplitudes of each emitter. Moreover, it provides an interesting setup for a wide spectrum of novel applications, such as mid-air displays [22], wireless power transfer [23], acoustic imaging [24], or delivering food through acoustic levitation [25], to mention a few.Iwamoto et al. first demonstrated ultrasonic mid-air haptic feedback [26], who developed a prototype consisting of 12 annular channels with a total of 91 ultrasound transducers in a hexagonal arrangement, a single focal point could be refocused along the central axis perpendicular to the array. Shinoda’s group [27,28,29,30] developed a more sophisticated system that was capable of controlling individually 249 transducers, being able to focus at different 3D positions in space, their boards have the capability to be chained to operate as a larger array system. Carter et al. [6] developed a phased array that can produce multi-point haptic feedback. Ultraleap (Bristol, UK) is a company that commercializes ultrasonic phased arrays for haptic applications related to automotive [31], digital signage [32], and AR/VR [33] applications. The company has also been exploring the effects on humans of high intensity ultrasound exposure [34] and has been releasing multiple prototypes that explore optimized array designs [35,36]. For example, transducer array in a Fibonacci spiral arrangement can suppress unwanted secondary focal points [37]. Pixie Dust Technologies (Tokyo, Japan) provides a parametric speaker [10] and an acoustic levitator [38] based on ultrasound phased arrays. The parametric prototype array has 269 transducers populating a circular array, π/32phase resolution, and can be refreshed at 1 kHz. Their levitator prototype has four orthogonally placed phased arrays with 285 transducers with a phase resolution of π/8and it is updated at 1 kHz. These ultrasound phased arrays have a fast update rate, high-power output, and sufficient phase and amplitude resolution; however, they are comparatively expensive, the software is closed, and the hardware cannot be easily modified.Some researchers have developed open platforms of acoustic phased arrays operating at 40 kHz in air. These platforms allow developers to create their own low-cost array [39,40,41]. For example, TinyLev [42] is a single-axis acoustic levitator that uses two ultrasonic arrays facing each other, reducing the number of independent channels by arranging transducers within the same distance to the trapping positions. Hirayama et al. [9] presented an acoustic levitator display with two opposed arrays that was capable of creating and modulating a large number of focal points at high speeds (20 kHz update rate) for delivering tactile feedback and parametric audio at the same time. While some part of the code is public, the hardware was not provided. Other projects have released both the hardware and software. For example, Ultraino [41] is a multi-purpose phased array that is accompanied by a platform that helps designers to build small phased arrays. The hardware is based on an Arduino MEGA microcontroller and provides 64 channels with π/5phase resolution. Furthermore, multiple boards can be chained together, expanding the number of individual controlled channels. The software is capable of customising phased array arrangements and visualising the pressure field in real-time. Despite the advantage of being a low-cost platform, the operating voltage is limited, reaching a large number of channels is cumbersome, and the transducers need to be wired to the boards. This last part gives some flexibility, but it makes the setups complicated to build, even when just flat geometries are required.A more detailed review of the available ultrasonic phased arrays can be found in [40]. We reckon that the presented hardware, SonicSurface, provides the most affordable and simple flat phased-array. More importantly, within this paper, we provide an algorithm that is capable of generating arbitrary acoustic fields using SonicSurface or other arrays that provide phase control.Acoustic holography [43] involves obtaining the near field of a radiating surface by taking measurements on the far field. It is a fundamental technique in health structure monitoring or mechanical vibration analysis. During the last years, a new trend in acoustic holograms has emerged [13,16,17,18], which is defined as the application of techniques, previously used in optics, to obtain target acoustic fields of different shapes by engineering the amplitude and phase of an array of emitters or an emission modulating surface.From an algorithmic point of view, researchers first implemented single-point algorithms [5,26] or single traps with different shapes [13]. Later, multi focal-point algorithms [16,44,45] enabled creating high-amplitude points at independent positions. For example, Plasencia et al. [46] proposed a method for optimizing the phases and amplitudes of the acoustic field, obtaining higher-quality points than previous phase-optimization approaches.Other strategies used a phase modulation plate on top of a flat radiating piston. Melde et al., used an iterative algorithm [17] in order to calculate the required phase modulation to create a target field at a given distance; they employed a static 3D printed modulator that encoded the phases for reconstructing the target hologram. Brown et al. [47] introduced a second holographic plate to modulate both phase and amplitude surface.These algorithms assume a high-resolution modulation plate with almost pixel-like shape for each point that modulates the field. Differently, here we introduce a modification on the previous algorithms to obtain target amplitude fields using discrete ultrasonic arrays that are made of circular emitters.3. Hardware DesignSonicSurface is a phased array consisting of 256 transducers emitting at 40 kHz. The transducers are arranged in a 16 × 16 grid and built on a single integrated printed circuit board (PCB). On one side of the PCB, ultrasonic emitters are soldered, whereas, on the other side, the field-programmable gate array (FPGA) (EP4CE6E22C8N—ALTERA IV Core Board, Waveshare), shift registers (74HC595, TI), drivers (MIC4127 from MT), and decoupling capacitors (ceramic 50V 0.1 μF) are mounted. The signals for each emitter are generated by the FPGA. The shift registers demultiplex each digital line coming from the FPGA into eight channels, and the drivers boost the voltage of the channels from logic voltage to the supplied power voltage (up to 20 peak-to-peak voltage (Vp-p)). A block diagram can be seen in Figure 1. Figure 1. Schematic of the SonicSurface ultrasonic array. A field-programmable gate array (FPGA) receives the phases to be emitted from a computer, they are stored on a double buffer and constantly output. The FPGA multiplexes 8 channels into one line so that only 32 output pins are needed. There are 32 blocks of shift registers, being able to drive a total of 256 emitters.The calculation of the phases and amplitudes to be emitted is performed on an external computer and then sent to the FPGA via Serial Universal asynchronous receiver-transmitter (UART) protocol at 203,400 bauds. A double buffer has been implemented in the FPGA to generate the signals uninterruptedly [48]; one of the buffers stores emission patterns coming from the computer, whereas the second buffer is the one that is used by the FPGA to continuously generate the emission signals, a command from the computer swaps the buffers at once. This method avoids latency and waiting issues. Different versions of the firmware are available for the FPGA to support phase and amplitude control, or amplitude modulation of the 40 kHz main signal.The protocol used to communicate with the FPGA is presented in Table 1; 1 byte specifies commands or emission patterns. If the byte value is larger than 127, it is a command; otherwise, it represents an emission phase offset or amplitude, depending on the mode. By default, the FPGA has a resolution of 32 divisions per period, so numbers from 0 to 31 represent phases from 0 to 2π, 32 represents no emission. Receiving a value of 254 indicates that new phases are going to be sent, the read pointer of the buffer is set to channel 0; afterwards, each phase sent will be assigned into the current read pointer and the pointer increased by one. The command 253 indicates swapping of the buffers. Other commands are: 252, to toggle amplitude modulation at 200 Hz for haptic feedback applications; 252 indicates that instead of phases, amplitudes are going to be sent. From 192 to 196, indicates the board number to activate (being 192 board number 1), in the case that multiple boards were chained together.Table 1. Communication protocol commands. The FPGA can generate 256 square-wave signals at 40 kHz. Each of the signals supports a phase delay control of 32 divisions per period or π/16radians, the amplitude can be modulated with up to 16 divisions. A multiplexing scheme strategy was employed for reducing the number of needed output pins and, thus, reduce the price of the FPGA. Packs of eight channels are multiplexed into one digital line. Later, this line is demultiplexed back into eight channels while using the shift registers. Figure 2 illustrates the channel multiplexation scheme from the FPGA and circuit implementation. Figure 2. At the left, the FPGA blocks in charge of generating the signals are presented, 8 phaseLine blocks (signal generators) are multiplexed into one digital line to reduce the required number of output pins. At the right, the circuit schematic represents a shift register that demultiplexes the signal into 8 channels that get amplified by four dual Metal–oxide–semiconductor field-effect transistor (MOSFET) drivers and fed into the ultrasonic emitters.The shift and the latch clock are generated by the FPGA. The shift clock controls when the shift registers shift data in, the latch clock determines when the data that were shifted should be output. The shift clock operates at 10.24 MHz (8 multiplexed channels × 40 kHz × 32 divisions per period), whereas the latch clock operates at 1.28 MHz (40 kHz × 32 divisions per period). The number of divisions per period (i.e., the resolution in phase or amplitude) could be doubled to 64, but the shift clock would operate slightly above 20 MHz, which would require better filtering and traces on the PCB.Once the digital signal for each channel has been demultiplexed, it is amplified from 5 V up to 20 V using a dual Metal–oxide–semiconductor field-effect transistor (MOSFET) driver (e.g., TC4427a or MIC4127 from MT). After testing different electronic components for amplifying the signals (e.g., L293D or BJT transistors), MOSFET drivers were found to efficiently drive the ultrasonic transducers. Dual Mosfet Drivers can amplify two channels and have a small footprint; larger components would not fit on the integrated board. Subsequently, the output of the drivers is fed into the ultrasonic emitters (a comparison of suitable transducers can be found in the supplementary information of TinyLev [42]). Given the narrowband nature of the emitters, it is possible to use a half-square wave to drive them without generating a significant amount of harmonics [41]. This technique is widely employed for airborne ultrasonic phased arrays, because generating a digital square signal is less complex than creating an analog sinusoidal signal, they are also easier to amplify.We present two models of the ultrasonic array. In the first one, the electronic components (i.e., shift registers, drivers, and decoupling capacitors) are surface mounted device (SMD) and the ultrasonic emitters have a diameter of 10 mm (Figure 3). The second model uses emitters of 16 mm diameter and through-hole (TH) components (Figure 4). The first model is more compact and faster to assemble if SMD equipment is available (e.g., stencils, solder paste, and a reflow oven). The TH model is larger and it takes more time to assemble, but it can be done with entry level electronics equipment (i.e., a soldering iron). Throughout the paper, we focus our experiments on the SMD board, since we think that it will be employed more often in the scientific community. Figure 3. Board with surface mounted devices and emitters of 10 mm diameter. (left) Top view of the Sonic surface where 16 × 16 ultrasonic emitters can be seen. At the sides there are connectors for power, Universal asynchronous receiver-transmitter (UART) in, grounds, sync out and sync in. (center) bottom view where the shift register blocks can be seen with the FPGA on top. (right) closer view on a shift register block where a shift register demultiplexes a digital line into eight signals that are fed to four dual-drivers and then into eight ultrasonic emitters. Figure 4. Ultrasonic array built with Through-hole components and emitters of 16 mm diameter. (Left) transducers of 16 mm diameter soldered on the printed circuit board (PCB). (Center) back of the board with the shift registers, drivers and decoupling capacitors. The FPGA board is connected through an expander board. (Right) detailed view of a shift register block.The program synthesized for the FPGA delegates the phase calculations on an external computer, thereby the cost of the board itself can be kept low. A UART reader block gets the bytes coming from the external computer [49]. A distributor block stores the current channel and sets the phases on the 256 signal generator blocks, each generator block outputs a digital signal of 40 kHz. Each generator block stores two phases, the one to be emitted and the previously read phase. The generator blocks have an internal amplitude counter that represents the number of divisions that the output should be HIGH, there is a global counter (from 0 to 31) that reaches all of the blocks, when the phase of a generator block coincides with the global counter, the internal amplitude counter is set to the target amplitude. The generator blocks have a dataline of five bits to read phases or amplitudes when the line setPhase or setAmp goes high. It also has a swap line, which swaps the phases/amplitudes when it goes high; this is to implement the double buffer. Eight generator blocks are grouped into a multiplexer, giving a total of 32 multiplexed lines that are output from the FPGA, as well as the shift and latch clocks. There are six auxiliary general-purpose inputs/outputs (GPIOs) (we have denominated them from A to F) that can be operated, defined, and implemented by the user. For example, B is used as the UART input, D is used as sync out (internal 40 kHz reference), E can be used as sync in (40 kHz signal to synchronize the global counter), and A can be used as a UART out; C and F are free for custom applications. Figure 5 shows the block diagram of the FPGA firmware. Figure 5. Block diagram of the code that is synthesized in the FPGA. On the top-left, the MasterClock is a phase-locked loop (PLL) to transform the internal 50 MHz clock into a 10.24 MHz clock named CLK_8. At the top-right, there is a global counter that acts as a frequency divider of CLK_8: COUNT[7] sets at 40 kHz and is output as the reference signal on MISC_D; COUNT[2] is the latch clock. If the board acts as a slave, the counter is synchronized with a 40 kHz external signal filtered by a RSS filter. On the bottom left, the UART input is filtered, read, and sent to the distributor. The distributor updates the emission phases of the generator blocks. AllChannels contain 256 generator blocks that connect to 32 Multiplexers of eight channels each. The generator blocks and multiplexers are timed by the outputs of the global counter. At the bottom-right, the multiplexed data channels as well as the latch and shift clocks are output.The UART Reader and Distributor blocks operate with the internal clock, the generator blocks and multiplexers operate with a clock that is synchronized with the sync in signal. Thereby, when multiple boards operate together, the emission waves have exactly the same frequency. If the emission clocks were not synchronized, traveling waves would be created [41], making the generation of static fields impossible. A master board has its sync out connected to its sync in, slave boards take the sync signal from the master board.The presented hardware has been optimized for an operating frequency of 40 kHz. This is the most common frequency for airborne ultrasonic phased arrays [9,38,41,42], operating at higher frequencies is not straightforward. On the one hand, the multiplexation of signals is used to reduce the required traces on the PCB and pins on the FPGA, our current system requires just a two-layer PCB and 40 GPIOs of the FPGA. However, this multiplexation leads to a 10.24 MHz shift clock. Increasing the frequency or phase resolution would require a higher clock frequency, which is beyond what is recommended for a simple PCB or the specs of the shift registers. On the other hand, commercially available transducers that operate at higher frequencies (e.g., 100 kHz or 400 kHz from MultiComp) are 10 mm in diameter and, thus, emit a very narrow beam. The emission from an array of these emitters would not interfere between each other and, thus, would not be suitable for the techniques presented here or phased-array techniques in general.4. AlgorithmThe algorithm that was developed by Melde et al. [17] is a modification of the Gerchberg–Saxton algorithm [50]. It calculates the phase modulation necessary at each point in the emitter plane in order to obtain a target amplitude field at the desired distance. The issue is that the algorithm is designed to produce modulation profiles that are almost continuous with more than 100 × 100 elements that are smaller than half-wavelength. However, available airborne ultrasonic arrays have a resolution of 16 × 16 or 24 × 24 at most, with element sizes that are larger than the wavelength and a circular shape instead of a square. We introduced a modification on this algorithm to consider the discrete nature of ultrasonic arrays and their lower number of elements when compared to passive modulators.The proposed algorithm is an iterative approach with four steps per iteration, as described in Figure 6. The FOCUS library was employed for the forward propagation and the backpropagation of the emission and target field slices [51]. Figure 6. Iterative algorithm to determine the emission phases and amplitudes for an array of emitters. Step (1) fix the amplitude into the target slice, the phase is not modified. Step (2) Backproject the target into the emission. Step (3) Apply on the emission slice a discretization on phase, amplitude, and spatial resolution, as well as the mask with the shape of the emitters. Step (4) Project the emission into the target. After 50 iterations of steps 1 to 4, the target amplitude is shown at the left.5. Results5.1. Comparison between Simulations and ExperimentsThe experimental setup of Figure 7 was used to measure the acoustic pressure distribution generated by the array in order to compare the emitted experimental amplitude slices with the simulated ones. In this setup, an ultrasonic receiver (MA40S4S, Murata) is attached to the head of a delta stage (Anycubic Kossel) and the emitter array sits on its bed. A Matlab script communicates with the delta stage and it moves the receiver to different positions on a grid of 16 × 16 cm with 2.5 mm spacing. At each measuring point, the computer reads the peak-to-peak voltage that was captured by the oscilloscope (Hantek 6074BE). The voltage is linearly proportional to the amplitude and, thus, can be directly translated to amplitude in arbitrary units (a.u.). The computer sends the emission phases to the array through the UART protocol and it controls the stage using the G-Code protocol. Figure 8 shows the obtained experimental amplitude slices, which are in reasonable qualitative agreement with the simulation slices, except for the Brazilian flag pattern. Figure 7. Experimental Setup used to scan the emitted amplitude slice. Figure 8. Amplitude slices obtained for different patterns, plotted using the function imagesc of Matlab. The first row is the target, the second one is the simulated slice, and the third row is the experimental measurement.5.2. Effect of Phase, Amplitude, and Spatial ResolutionWe carried out multiple simulations using the algorithm that is described in Section 4 with different parameters for emitter size, phase emission resolution and amplitude emission resolution. All of the target amplitude fields were generated 16 cm above the array, since we tested that, at that distance, the best results were obtained. The default simulation parameters are those from the SMD board, i.e., emitterSize = 10 mm, phaseResolution = 32, and no amplitude modulation. One parameter was varied at a time and the mean square error (MSE) of the obtained imaged was obtained. The results can be seen in Figure 9. Figure 9. Simulated amplitude fields at 16 cm from the array for different target patterns and array parameters. (First column) target amplitude field. (Second column) obtained amplitude field when the emitter array is the surface mounted device (SMD) board presented in the paper, i.e., emitterSize = 10 mm (transducer diameter), phaseResolution = 32 and no amplitude modulation. (Third column) obtained amplitude fields with an array with emitterSize = 2 mm, phaseResolution = 32 and amplitudeResolution = 16. (Fourth column) mean square error (MSE) as the emitter size decreases. (Fifth column) MSE as phase resolution increases, emitterSize = 10 mm. (Sixth column) MSE as the amplitude resolution increases, emitterSize = 10 mm.The patterns employed were: the flag of Brazil (non-binary image), the letter A, a Dove, and a smiley face. In general, it can be seen that as the emitter size decreases (i.e., more spatial resolution), the quality of the images improves. It is important to note that significant reductions of MSE are obtained, even when emitters get smaller than half-wavelength (4.3 mm), and that no further improvement is obtained below 2 mm (1/4 of the wavelength); this is different from the generation of regular focal points that do not increase its amplitude once the emitters are reduced below half-wavelength size [13]. The phase resolution significantly improves the pattern quality, but quickly plateaus when the phase resolution reaches eight divisions per period; this is in accordance with the simulations performed for simple focal points [41]. For amplitude resolution, it is clear that having amplitude modulation reduces the MSE by half even when only four different amplitudes can be emitted. In summary, the sweet-spot is obtained with a phase resolution of eight divisions per period and amplitude resolution of four divisions; the MSE improves as the emitters get smaller (i.e., more spatial resolution), but no improvement is found once the emitter size reaches quarter-wavelength.These findings could be specific to the patterns that were selected in the study and to our setup characteristics (e.g., wavelength, number of emitters or distance to the target slice); however, the code was made public, so that other researchers could run simulations for their specific setups (e.g., operating in water or with static metamaterials).6. ConclusionsIn recent years, Acoustic holography has found numerous applications and has advanced rapidly due to the adaptation of methods found in the optics community. In this paper, we have attempted to advance, test, and unify algorithms and hardware used for acoustic mid-air holography. Namely, we have described a novel iterative algorithm that calculates the emission phases and amplitudes for an array of emitters that can be used to generate a desired target amplitude field. To our knowledge, this is the first algorithm capable of determining the amplitude and emission phases for discrete arrays comprised of finite sized emitters. We have then used this algorithm to investigate the effects of increased phase, amplitude and spatial resolution in the obtained amplitude field. Our analysis demonstrates that diminishing returns are observed at some point on-wards. Meaning that depending on the application requirements there is no need to use expensive hardware or that the computations can be accelerated by further discretizing the solution domain. Finally, to support the growth of the acoustic holography research community, we have described an open hardware platform named SonicSurface which is an affordable FPGA-based ultrasound phased array. Two different models for the array of emitters have been provided (SMD and TH), so that researchers from different fields and backgrounds can customise these further for their own experimental requirements. We hope that the algorithm and hardware presented in this paper facilitates further research on the field of ultrasonic arrays and enables novel applications of crafted amplitude fields. 4. MODIFICATIONS TO THE INTERNATIONAL RELEASE4.1 Subject to clause 2.1.4, the Licensee may not modify any part of the SNOMED CT Core distributed as part of theInternational Release or as part of a Member's National Release.4.2 Subject to any express and specific statement to the contrary in the documentation distributed as part of the InternationalRelease, the Licensee may not modify any of the documentation (including Specifications) or software (unless provided insource code form) distributed as part of the International Release.4.3 The Licensee may, by written notice, request the Licensor to modify the SNOMED CT Core. Upon receipt of such writtennotice, the Licensor shall consult with the Licensee and shall give due consideration as to whether the proposed modificationshould be made based on the Licensor's editorial guidelines and policies. Following due consideration of the matter, includingconsideration of any information presented by the Licensee, the Licensor shall inform the Licensee whether the proposedmodification shall be made and if the Licensor agrees that the proposed modification should be made, the Licensor shall give anon-binding indication of when, reasonably and in good faith, it anticipates that the proposed modification will be made. If theLicensee would like the content of the proposed modification to be developed more quickly than the Licensor has indicated, theLicensee may itself undertake or procure the undertaking of the development of the content of the proposed modification(outside of any existing Licensor's support services contract). On receipt of the developed content of the proposed modification,the Licensor will then give due consideration as to whether the developed content meets the Licensor's quality assurance, othergovernance processes, Standards and Regulations. If the developed content meets the Licensor's quality assurance, othergovernance processes, Standards and Regulations then the Licensor shall incorporate the modification into the SNOMED CTCore according to its schedule which will give due consideration as to when the proposed modification shall be incorporatedinto the SNOMED CT Core, taking into account other proposals for the modification of the SNOMED CT Core and the workrequired to include the proposed modification in the SNOMED CT Core.5. TERM AND TERMINATION5.1 This License Agreement shall commence on the date on which it comes into effect in accordance with the notice at thebeginning of this License Agreement, and shall continue until terminated in accordance with this clause 5.5.2 Either party may terminate this License Agreement if the other party commits a material breach of any of its obligationsunder this License Agreement (which, in the case of the Licensee, shall include, without limitation, any failure to pay LicenseFees when due under clause 7) in accordance with the following procedure:5.2.1 the party seeking to terminate the License Agreement (the "Terminating Party") shall serve an escalationnotice (the "Escalation Notice") on the other party (the "Defaulting Party") requiring the Defaulting Party tonominate a member of its senior management team to meet with a member of the Terminating Party's seniormanagement team to seek to resolve in good faith the matter giving rise to the service of the escalation notice.5.2.2 The representatives of the parties identified in accordance with clause 5.2.1 shall meet in good faith to seek toresolve the matter. If they are unable to resolve the matter within 45 days of the date of the Escalation Notice theTerminating Party may serve a formal breach notice (the "Breach Notice") on the Defaulting Party requiring it toremedy the breach within 90 days.5.2.3 If the Defaulting Party does not remedy the breach within 90 days of the date of the Breach Notice theTerminating Party may terminate the License Agreement by giving 180 days' written notice to the Defaulting Party(the "Termination Notice").5.3 Neither party may terminate this License Agreement except in accordance with this clause 5.5.4 The Licensee may terminate this License Agreement by giving up to twelve (12) months' prior written notice to the Licensor.5.5 Upon termination of this License Agreement in accordance with this clause 5, all licenses granted under this LicenseAgreement shall automatically and immediately be revoked.5.6 The Licensee shall, by no later than forty five (45) days after termination of this License Agreement for any reason, removeall copies of the International Release from its computer systems and destroy all copies of electronic, paper copy and othermedia containing or representing any part of the International Release. The Licensee shall, if requested by the Licensor, certifyin writing to the Licensor that the Licensee has complied with its obligations under this clause 5.6. 5.7 The Licensee shall, as soon as reasonably practicable following either party giving a Termination Notice for any reason, andin any event by no later than ninety (90) days after such Termination Notice is given, give written notice of such termination toeach End User that the Licensee reasonably believes to be a current user of a Licensee Product and to each Member in eachMember Territory in which the Licensee has distributed or licensed any Licensee Product.5.8 The Licensee may not grant any new sub-license under clause 2.1.5 after either party has given notice under clauses 5.2 or5.4.5.9 The Licensor shall be entitled to publicize the termination of this License Agreement to such persons (including Members,other Affiliates of the Licensor and End Users) and in such manner as it sees fit.5.10 Clauses 5.6, 5.7, 5.8, 5.9, 5.11, 5.12, 7, 8 and 10 to 14 inclusive shall survive termination of this License Agreement.5.11 The Licensee shall, by no later than thirty (30) days after termination of this License Agreement for any reason, submit astatement of account in accordance with clause 7.3 in respect of all periods that have not previously been covered by a statementof account under that clause.5.12 Any termination of this License Agreement, for any reason, is without prejudice to the accrued liabilities of each party as atthe date of termination (including, without limitation, any liability of the Licensee to pay License Fees that has accrued as at thedate of termination), or to the Licensee's obligation to pay License Fees arising from the statement of account submitted underclause 5.11.6. NEW VERSIONS AND CHANGES TO LICENSE TERMS6.1 The Licensor shall notify the Licensee when each new version of the International Release is made available and there shallbe a mechanism for the Licensee to access or obtain copies of the new version of the International Release. The Licensee shallbe liable for any reasonable distribution charge, if applicable, established by the Licensor for each copy of the new version of theInternational Release.6.2 Within one-hundred and eighty (180) days after the Licensor has notified the Licensee of the release of a new version of theInternational Release, the Licensee must upgrade the version of the International Release in its own systems and in the LicenseeProducts to that new version (or alternatively, if a subsequent version of the International Release is or has been released duringthe 180-day period, to that subsequent version at the Licensee's option).6.3 The Licensor may vary the terms of this License Agreement by giving written notice to the Licensee. Any such variationshall take effect not less than ninety (90) days after the notice is given, as specified in the notice. If the Licensee does not wishthis License Agreement to continue subject to the variation, the Licensee may terminate this License Agreement in accordancewith clause 5.4, and if the Licensee gives notice of such termination before the variation takes effect then the variation shall notapply as between the Licensor and the Licensee.6.4 The College of American Pathologists, as originator of Intellectual Property Rights in the International Release, shall as alicensee have a specific [exception] to the Licensor's rights in clause 6.3 in specific circumstances and for a specific fixed termperiod to be agreed with the Licensor, and the terms of such special [ exemption] shall be deemed part of such licensee'sAffiliate License Terms. The Licensor will publish the terms of the special exemption with the Articles.7. LICENSE FEES7.1The Licensee shall pay the License Fees to the Licensor in respect of the Licensee's activities in Non-Member Territories.The License Fees shall be payable annually in arrear.7.2 All License Fees and other amounts payable to the Licensor under this License Agreement are exclusive of value added taxand any other tax of a similar nature, which shall be payable by the Licensee at the prevailing rate in addition to those amounts.7.3 The Licensee shall, at least once in each calendar year, submit a statement of account to the Licensor in such manner andform as the Licensor may prescribe from time to time, setting out the Licensee's activities in Non-Member Territories since theend of the period covered by the previous statement of account submitted under this clause 7.3 (or, in the case of the firststatement of account under this clause 7.3, since the date on which this License Agreement became effective), and theLicensee's calculation of the License Fees and other amounts payable to the Licensor in respect of that period. Each such statement of account shall include, without limitation, a list of all license agreements in respect of Licensee Products that were inforce during the period covered by the statement of account and, in relation to each such license agreement, the dates on which:(a) that license agreement was entered into or otherwise became effective; (b) the Licensee Product was first provided or madeavailable to the licensee under that license agreement; and (c) the International Release (or any part of it) was first madeavailable to the licensee under that license agreement.7.4 The Licensee shall provide the Licensor with such information as the Licensor may reasonably request for the purpose ofverifying any statement of account submitted to the Licensor under clause 7.3.7.5 The Licensor shall, following receipt of a statement of account from the Licensee under clause 7.3, submit an invoice to theLicensee setting out the License Fees and other amounts payable by the Licensee in respect of the period to which the statementof account relates. The Licensee shall pay to the Licensor all amounts set out on each invoice submitted under this clause 7.5within thirty (30) days of receipt of that invoice. The Licensee shall make payment under this clause 7.5 by wire transfer or bysuch other means as the Licensor may make available to the Licensee from time to time.7.6 Interest shall accrue on any outstanding License Fees and other amounts at the rate of the lesser of (a) 500 basis points abovethe European Inter-Bank Offer Rate (EURIBOR), calculated daily from the date on which payment was due and compounding atthe end of each calendar month or (b) the maximum amount allowed under applicable law.8. PROTECTION OF THE LICENSOR'S INTELLECTUAL PROPERTY8.1 Nothing in this License Agreement transfers to the Licensee any right, title or interest in or to the Intellectual PropertyRights in the International Release or any part of it, or grants the Licensee any license in respect of the International Release orany part of it except as expressly set out in clause 2.8.2 The Licensee shall not:8.2.1 use any trademark or service mark (or any registrations thereof) other than the Licensor's trademarks, in anyname that includes the word "SNOMED" or that is confusingly similar to SNOMED CT or any other similartrademark;8.2.2 apply for any trade mark or service mark (or any registrations thereof) in any name that includes the word"SNOMED", or that is confusingly similar to SNOMED, SNOMED CT or any other similar trade mark;8.2.3 abbreviate the marks SNOMED or SNOMED CT; or8.2.4 do anything with respect to the foregoing trade marks that damages or could reasonably be deemed to reflectadversely on the Licensor or such trade marks.8.3 The Licensee shall:8.3.1 include the following notice on all media on which the Licensee Products are distributed and on thedocumentary form of each sub-license granted by the Licensee under clause 2.1.5:"This material includes SNOMED Clinical Terms® (SNOMED CT®) which is used by permission of theInternational Health Terminology Standards Development Organisation (IHTSDO). All rights reserved. SNOMEDCT®, was originally created by The College of American Pathologists. "SNOMED" and "SNOMED CT" areregistered trademarks of the IHTSDO."8.3.2 specify in all media on which any Licensee Product is distributed the version and date of the InternationalRelease contained in the Licensee Product.8.4 The Licensee shall be entitled to use the "SNOMED" and "SNOMED CT" trade marks only on the Licensee Productsdistributed and modified in accordance with this License Agreement and any services relating thereto but not otherwise andsubject to the trade mark utilization Regulation developed by the Licensor and published by the Licensor from time to time. Alluse by the Licensee of the "SNOMED" and "SNOMED CT" trade marks, and all goodwill resulting from that use, shall inure tothe Licensor's benefit.8.5 The Licensee shall maintain quality standards with respect to modifying, supplementing, marketing and distributing the Licensee Products, and any services relating thereto, that are in accordance with applicable law and are at least as stringent asthe Regulations developed by the Licensor and published by the Licensor from time to time.8.6 Upon reasonable written notice from the Licensor, the Licensee shall provide the Licensor with representative samples ofmaterials, software products, advertising, agreements for use of the Licensee Products (other than the terms of those agreementsthat are unrelated to the Licensor's rights and obligations under this License Agreement) and/or other written materials relatingto the Licensee's use of the International Release and the Licensor's trade marks to enable the Licensor reasonably to ascertainthe Licensee's compliance with its obligations under this License Agreement. In the absence of circumstances giving theLicensor reasonable grounds to suspect a breach of this License Agreement, the Licensor may not give notice under this clause8.6 more frequently than once per year.8.7 If any use of the International Release (including without limitation use through a Licensee Product) is reasonablydetermined by the Licensor to be below the standards of quality required under this License Agreement, the Licensor shall notifythe Licensee of such deficiency in writing. Upon receipt of such notice, the Licensee shall take all necessary steps to correctsuch deficiency (including such steps as the Licensor may reasonably specify).8.8 The Licensee shall maintain a complete, accurate and up-to-date register of all sub- licenses granted by the Licensee underclause 2.1.5, and shall make that register available for inspection during normal business hours by the Licensor and itsrepresentatives upon the Licensor giving not less than fourteen (14) days' prior written notice. The register maintained by theLicensee under this clause 8.8 shall at a minimum contain the following information in respect of each sub-license: the nameand registered office of the sub-licensee; the Licensee Product subject to the sub-license; and the version of the InternationalRelease included in that Licensee Product. In the absence of circumstances giving the Licensor reasonable grounds to suspect abreach of this License Agreement, the Licensor may not give notice under this clause 8.8 more frequently than once per year.9. USE IN MEMBER TERRITORIES AND NON-MEMBER TERRITORIES9.1 The Licensee may only exercise its rights under this License Agreement in a Member Territory in accordance with suchconditions as the Member for that Member Territory may prescribe from time to time.9.2 Conditions prescribed by a Member under clause 9.1 may:9.2.1 include, without limitation, a requirement that the Licensee notify the Member before exercising its rightsunder this License Agreement in that Member's territory and a requirement that the Licensee enter into a licenseagreement with the Member in respect of that Member's National Release; and9.2.2 relate to the International Release, the Member's National Release or any part of either of them.9.3 The Licensee shall notify the Licensor (and, if the Licensee's registered office or principal place of business is situated in aMember Territory, shall also notify the Member for that Member Territory) in writing before exercising its rights under thisLicense Agreement in any Non-Member Territory in respect of which the Licensee has not previously given notice under thisclause 9.3. The notice shall be in such form and manner as the Licensor may prescribe from time to time, and shall include suchinformation about the Licensee's current and proposed activities in that Non- Member Territory as the Licensor may require (butthe Licensor may require only the same kinds of information as it requires to be provided by new Affiliates proposing to use,license or deploy the International Release or Licensee Products in Non-Member Territories).9.4 In any case where the Licensee gives notice to a Member in accordance with clause 9.3, the Licensee consents to thatMember providing the content of that notice to the Licensor.9.5 For purposes of this clause 9, the Licensee exercises its rights under this License Agreement in any Member Territory orNon-Member Territory if, without limitation, it:9.5.1 performs any act permitted by this License Agreement in that Member Territory or Non-Member Territory (asthe case may be);9.5.2 deploys the International Release (or any part of it) or any Licensee Product in that Member Territory or Non-Member Territory (as the case may be); or9.5.3 distributes or licenses a Licensee Product for use in, or to any person who is situated in, that Member Territoryor Non-Member Territory (as the case may be). 10 AFFILIATE STATUS10.1 During the term of this License Agreement the Licensee shall be an Affiliate.10.2 As an Affiliate, the Licensee shall be entitled to participate in the Licensor's Vendor Liaison Forum, which is a forum inwhich the Licensee and other Affiliates may communicate with the Licensor and with each other. The Licensor may makeRegulations from time to time governing the Licensee's participation in the Vendor Liaison Forum. New Regulations that theLicensor shall make from time to time governing participation in the Vendor Liaison Forum shall not remove the Licensee'sright to participate in that forum.11. REPRESENTATIONS AND WARRANTIES11.1 To the extent permitted by law, the Licensor excludes all representations, warranties and conditions that would otherwisebe implied by law in this License Agreement (including, without limitation, all implied warranties of quality or fitness for aparticular purpose).11.2 Without limiting clause 11.1, the Licensor does not represent or warrant that the International Release or any part of it willsatisfy any of the Licensee's requirements, operate in combinations selected by the Licensee or be free from defects or errors.12. LIMITATION OF LIABILITY12.1 The Licensor shall not be liable to the Licensee or to any other person, whether in contract, tort (including negligence),misrepresentation, breach of statutory duty or otherwise, for any of the following arising under or in connection with thisLicense Agreement (including, without limitation, in respect of the Licensee's use of or inability to use the International Releaseor any part of it):12.1.1 indirect or consequential loss;12.1.2 special or punitive damages;12.1.3 loss of profits, loss of savings and loss of revenue;12.1.4 loss of business, loss of reputation and loss of goodwill; and12.1.5 loss of data.12.2 Neither the Licensor nor any Member shall be liable to the Licensee or any other person for any failure by the Licensor orthe Member (as the case may be) to maintain or distribute any Extension (or part thereof) or Derivative transferred to theLicensor or the Member (as the case may be) in accordance with clauses 3.4 or 3.5.12.3 The liability of the Licensor arising in any year under or in connection with this License Agreement, whether in contract,tort (including negligence), misrepresentation, breach of statutory duty or otherwise, shall not in any event exceed the LicenseFees paid by the Licensee in respect of that year.12.4 Nothing in this License Agreement excludes or limits the liability of either party for:12.4.1 fraud (including fraudulent misrepresentation);12.4.2 death or personal injury caused by the negligence of that party;12.4.3 any breach of its obligations implied by section 12 of the Sale of Goods Act 1979; or12.4.4 any other liability that by law cannot validly be excluded or limited (but only to the extent that the liabilitycannot validly be excluded or limited).13. ASSIGNMENT13.1 The Licensee may not assign, novate or otherwise transfer any of its rights or obligations under this License Agreement toany person without the prior written consent of the Licensor, not to be unreasonably withheld. 13.2 The Licensor may transfer all of its rights and obligations under this License Agreement to any person to whom theLicensor transfers the Intellectual Property Rights in respect of which the licenses under this License Agreement are granted.14. GENERAL PROVISIONS14.1 This License Agreement contains the entire agreement between the parties relating to the subject matter of this LicenseAgreement, supersedes all previous agreements between the Parties relating to that subject matter and sets out the entirety of theLicensee's rights in respect of the International Release.14.2 Each party acknowledges that, in entering into this License Agreement, it has not relied on any representation, warranty,collateral contract or other assurance made by or on behalf of the other party before the date of this License Agreement.14.3 Except as provided in clause 6.3, this License Agreement may not be varied except in writing signed by both parties andexpressed to vary this License Agreement.14.4 Nothing in this License Agreement shall give either party the ability to act or incur obligations or liability on behalf of theother party or constitutes a joint venture, agency, partnership or employment relationship between the parties.14.5 If any term of this License Agreement is or becomes illegal, invalid or unenforceable in any jurisdiction, that shall notaffect the legality, validity or enforceability in that jurisdiction of any other term of this License Agreement, or the legality,validity or enforceability in any other jurisdiction of that or any other term of this License Agreement.14.6 The Licensee agrees that the Licensor may appoint third parties to process personal data provided by the Licensee to theLicensor under or in connection with this License Agreement (including without limitation payment details provided inconnection with the payment of License Fees). In connection with any such appointment, personal data provided by the Licenseemay be transferred to, and processed in, a country outside the European Economic Area (EEA). The laws governing theprocessing of personal data may be less stringent in such a country than in the member countries of the EEA.15. GOVERNING LAW AND JURISDICTION15.1 This License Agreement shall be governed by, and construed in accordance with, English law.15.2 The English courts shall have exclusive jurisdiction to settle any dispute arising out of or in connection with this LicenseAgreement (including a dispute regarding its existence, validity or termination).15.3 Clause 15.2 is for the benefit of the Licensor only. As a result, the Licensor shall not be prevented from taking proceedingsrelating to any dispute in any other courts with jurisdiction. To the extent permitted by law, the Licensor may take concurrentproceedings in any number of jurisdictions.Appendix ADefined TermsIn this License Agreement, the following defined terms have the following meanings:Term MeaningAffiliate an affiliate of the Licensor in accordance with the Licensor's Articles of Association;Cross-Map a work consisting of (i) SNOMED CT Content and (ii) content of another nomenclature, classification orknowledge structure, together with a set of relationships between (i) and (ii);DataAnalysisSystema computer system that is used to analyze records or other data that is encoded using SNOMED CT, but not ifthat system is also a Data Creation System;DataCreation a computer system that is used to create records or other data that is encoded using SNOMED CT; SystemDerivativea work consisting of (a) SNOMED CT Content, from the SNOMED CT CORE or an Extension; together with(b) either (i) additional properties and/or information about such SNOMED CT content; and/or (ii) any set ofrelationships between that SNOMED CT Content and content of other nomenclature, classification orknowledge structure, and includes a Cross-Map and a Sub-Set;End User a third party user of a Licensee Product;Extension a work consisting of SNOMED CT Content alone that is supplementary to the SNOMED CT Core and thatdepends on the SNOMED CT Core;Hospital a health care body or organisation providing secondary and/or tertiary care;IntellectualPropertyRightspatents, trade marks, service marks, copyright(including rights in computer software), moral rights, databaserights, rights in designs, trade secrets, know-how and other intellectual property rights, in each case whetherregistered or unregistered and including applications for registration, and all rights or forms of protectionhaving equivalent or similar effect in any jurisdiction;InternationalReleasethe release produced and distributed by or on behalf of the Licensor, consisting of the SNOMED CT Core, theSpecifications and the Licensor's Derivatives and other documents and software;License Fees the license fees set out in Appendix B (License Fees in Non-Member Territories);LicenseeProductsproducts distributed or licensed by the Licensee that(1) include or interoperate with the International Release(or any part of it) and/or any Extensions or Derivatives created by the Licensee under this License Agreement,or (2) read or write records or other data that is encoded using SNOMED CT;Member a member of the Licensor;MemberTerritory a territory that is represented by a Member (as published by the Licensor from time to time);NamespaceIdentifiera code or that part of a code that identifies theorganization responsible for creating and maintaining aStandards-Based Extension or a Standards-Based Derivative and is used as an element of SNOMED CTIdentifiers;NationalReleasein respect of each Member, the release produced and distributed by the Member, consisting of the InternationalRelease, the Member's Extensions, the Member's Derivatives and other documents and software;Non-MemberTerritorya territory that is not a Member Territory;Practice(a) a single department of a Hospital (subject to paragraph 2.2 of Appendix B); or(b) any health care body or organisation that provides principally primary care, including without limitation apharmacy, an optician's facility, a physiotherapy centre, a general medical practice or a family medicalpractice;QualifyingResearchProjecta discrete research project that meets all of the following criteria:(a) it is supported by a formal proposal that has been peer reviewed;(b) it has been ethically approved in accordance with the prevailing legislation, regulations and guidelines ineffect in the relevant territory;(c) it is conducted within a definite timeframe;(d) the results of the research are offered for publication in peer-reviewed public journals and are provided tothe Licensor free of charge prior to publication;Regulations regulations made by the Licensor; Relationship a relationship, of a kind defined by the Licensor in Specifications, between concepts (which may be, withoutlimitation, a hierarchical or an associative relationship) or between a concept and a description;SNOMEDCTthe concept-based work of clinical nomenclature and classification with multiple hierarchies and semanticdefinitions known as SNOMED Clinical Terms (SNOMED CT);SNOMEDCT Contentterminological content, consisting of concepts,descriptions and Relationships, each of which is identified usinga SNOMED CT Identifier;SNOMEDCT Core the SNOMED CT Content that is controlled, maintained and distributed by the Licensor from time to time;SNOMEDCTIdentifiera code, of a kind defined by the Licensor in Specifications, for identifying concepts, descriptions andRelationships;Specification specifications promulgated by the Licensor for products and processing relating to SNOMED CT, includingspecifications of the internal logic of SNOMED CT, editorial policies, guidelines and characteristics;SponsoredTerritorya Non-Member Territory that has been recognized and designated by the Licensor as a sponsored territory (aspublished on the Licensor's web site);Standard a Specification that is formally adopted by the Licensor;Standards-Basedin respect of an Extension or a Derivative, an Extension or Derivative the creation of which is the subject ofone or more Standards; andSub-Set a sub-set of SNOMED CT Content that is grouped together for one or more purposes.Appendix BLicense Fees in Non-Member Territories1. Introduction1.1 This Appendix B sets out the license fees payable by the Licensee in respect of its activities in Non-Member Territories.1.2 The license fees set out in this Appendix B do not apply in respect of the Licensee's activities in any Non-Member Territoryif that Non-Member Territory is a Sponsored Territory or was a Sponsored Territory at the time when the Licensee's activities inthat Non-Member Territory were carried out.1.3 The Licensor may, in its sole discretion, waive the Licensee's obligation to pay any or all of the license fees set out in thisAppendix B if the Licensor considers that the Licensee's activities in any Non-Member Territory are in support of charitable orhumanitarian causes in that Non-Member Territory. Any waiver by the Licensor under this paragraph 1.3 may be revoked by theLicensor at any time, shall be without prejudice to any of the Licensor's other rights and remedies under this License Agreementand shall not relieve the Licensee of any of its other obligations under this License Agreement.1.4 Beginning in 2015, license fees payable by the Licensee in respect of its activities in Non-Member Territories for eachfinancial year shall be adjusted by the same percentage as the General Assembly of the Licensor agrees to adjust the AggregateAnnual Fee (as defined in the Licensor's Articles of Association) relative to the Aggregate Annual Fee in the previous year.1.5 The license fees in respect of Hospitals that are set out in this Appendix B apply only to Hospitals that are located on a singlecontiguous physical site. Any Hospital that is located on multiple physical sites shall be treated as falling within paragraph 4 ofthis Appendix B (and not within paragraphs 2 or 3).1.6 For purposes of this Appendix B, if a Practice is located on multiple physical sites then each such site is treated as a separatePractice.1.7 Notwithstanding anything else in this Appendix B, the deployment, distribution or licensing of any software that operates ona mobile device of any kind (including without limitation a mobile phone or tablet device), or any software or service that isaccessed via the internet and enables users to extract or download any substantial portion of SNOMED CT, shall be treated as falling within paragraph 4 of this Appendix B (and not within paragraphs 2 or 3).1.8 The Licensee's obligation to pay license fees in respect of any deployment of the International Release or any LicenseeProduct is not dependent on that deployment of the International Release or Licensee Product being used in a live or productionenvironment.1.9 In any case where the Licensee is exempt from the requirement to pay license fees by reason of a Licensee Product, a DataAnalysis System or a Data Creation System being used exclusively in connection with a Qualifying Research Project, theLicensee shall report to the Licensor on the progress of that Qualifying Research Project in such manner as the Licensor mayreasonably require. The Licensor may revoke the Licensee's exemption for Qualifying Research Projects provided in thisAppendix B if the Licensee fails to comply with this paragraph 1.9.2. Data Creation Systems2.1 The Licensee shall pay the following fees in respect of each Hospital or Practice in a Non-Member Territory in or to whichthe Licensee:(a) deploys the International Release (or any part of it) or any Licensee Product that contains the InternationalRelease (or any part of it) in a Data Creation System, unless that Data Creation System is used exclusively inconnection with a Qualifying Research Project; or(b) deploys, distributes or licenses a Licensee Product that is or includes a Data Creation System, unless thatLicensee Product is used exclusively in connection with a Qualifying Research Project.Fee Band FeeHospital in Band A Territory US$ 1,954 per annum baseline fee adjusted as per paragraph 1.4Hospital in Band B Territory US$ 1,303 per annum baseline fee adjusted as per paragraph 1.4Hospital in Band C Territory US$ 652 per annum baseline fee adjusted as per paragraph 1.4Practice in Band A, B or C Territory US$ 652 per annum baseline fee adjusted as per paragraph 1.4Hospital or Practice in Low Income Band US $0 per annum baseline fee, adjusted as per paragraph 1.4Hospital or Practice in other territory As per paragraph 5.2.2.2 The total fees payable by the Licensee in respect of a number of Practices that are departments of a single Hospital shall notexceed the fee applicable to the Hospital itself. For purposes of this Appendix B, a Practice is treated as a department of aHospital only if: (a) it is located on the premises of that Hospital; and (b) it is funded solely by that Hospital. In any case whereeither or both of the conditions in the preceding sentence are not met in respect of any Practice, fees shall be payable in respectof that Practice in addition to any fees that are payable in respect of any Hospital.3. Data Analysis Systems3.1 The Licensee shall pay the fees set out in paragraph 3.4 if the Licensee:(a) deploys the International Release (or any part of it) or any Licensee Product that contains the InternationalRelease (or any part of it) in a Data Analysis System in a Non-Member Territory, unless that Data Analysis Systemis used exclusively in connection with a Qualifying Research Project; or(b) deploys, distributes or licenses a Licensee Product that is or includes a Data Analysis System in a Non-MemberTerritory, unless that Licensee Product is used exclusively in connection with a Qualifying Research Project.3.2 The fees set out in paragraph 3.4 apply in respect of each deployment, distribution or license of the International Release (orany part of it), a Licensee Product or a Data Analysis System, and vary according to the Non-Member Territory in which thedeployment, distribution or licensing takes place. 3.3 If any Data Analysis System to which the fees in paragraph 3.4 apply consists of more than one database, the fees applicableto that Data Analysis System shall be multiplied by the number of databases in that Data Analysis System.3.4 The fees under this paragraph 3 are as follows:Fee Band FeeBand A Territory US$ 1,954 per annum baseline fee adjusted as per paragraph 1.4Band B Territory US$ 1,303 per annum baseline fee adjusted as per paragraph 1.4Band C Territory US$ 652 per annum baseline fee adjusted as per paragraph 1.4Low Income Band US $0 per annum baseline fee, adjusted as per paragraph 1.4Other territory As per paragraph 5.2.4. Other Activities4.1 The Licensee shall notify the Licensor in writing before deploying the International Release (or any part of it) or deploying,distributing or licensing any Licensee Product (in each case, other than exclusively in connection with Qualifying ResearchProjects) in, for use in, or to any person situated in, any Non-Member Territory in a manner that does not fall within paragraphs2 to 3 of this Appendix B, explaining the Licensee's proposed activities.4.2 Upon receiving notice from the Licensee under this paragraph 4, the Licensor may request, and the Licensee shall provide,such additional information in relation to the Licensee's proposed activities as the Licensor considers reasonably necessary todetermine an appropriate license and reasonable fee in respect of the Licensee's proposed activities.4.3 The Licensee shall be liable to pay such license fees as the Licensor may determine in accordance with this paragraph 4.5. Non-Member Territory Bandings5.1 The allocation of a Non-Member Territory into Band A, Band B, Band C, or Low Income Band shall be as determined bythe Licensor (based on the Non-Member Territory's relative Gross National Income (GNI) or other measure adopted by theLicensor) and published by the Licensor on its web site.5.2 The Licensee shall notify the Licensor in writing before carrying out any activity of a kind described in paragraphs 2 or 3 ofthis Appendix B in a Non-Member Territory that has not been allocated by the Licensor under paragraph 5.1. Upon receivingnotice from the Licensee under this paragraph 5.2, the Licensor shall allocate the Non-Member Territory as described inparagraph 5.1