CA2159000C - Data processing system for communications network - Google Patents

Abstract

An inter-network call accounting system for use in a communication network such as the public switched telephone network (1) in Britain allows call records to be sorted according to the network operator to be charged in respect of the calls, prior to being priced and charged. A data analyser (7) incorporating an expert system is provided for call records which cannot be validated. The data analyser (7) can apply default or amended date, or can output invalid data to a suspended process awaiting updated reference information. Unfixable data is output to a sump for management purposes. A pricing and charging engine processes data already sorted according to billable entity and incorporates further data analysis means for dealing with data invalid by reason of pricing and charging related information.

Description

WO 94/23530 ~ ~ ~ PCT/GB94100706 nATA PR~'7('RSSTNC. SYSTEM FOR COMMUNICATIONS NETWORK
The present invention relates to a data system for data collection and processing in mufti network communications.
Where communication instances, for instance telephone calls or data transfers, occur within a single network, it is known to log and process data related to those communication instances. Commonly, in a public switched telephone network (PSTN), data will be collected concerning call duration, and processed with respect to at least time of day and call type, so that the network operator can generate an item on a bill destined for the subscriber wno initiated a call.
Over recent years, the data systems for PSTNs have necessarily become increasingly complex as the choice of service and call type available to subscribers has greatly increased. For instance, with the introduction of 0800 numbers, it is no longer the initiating subscriber who will be billed. Many more complicated services are already being trialled, or available, on PSTNs, such as call forwarding where a call initiated by a first subscriber to a selected number is forwarded automatically by the network to a different number, the difference in cost being borne by the receiving subscriber.
Another aspect of communication networks which is in the course of considerable change is the multiplicity of network operators in existence. In the past, PSTNs have been run primarily by government organisations as part of the national infra structure. Nowadays and increasingly, privatisation of the PSTNs and the relaxation of regulatory monopolies means that there are many more network operators available to the subscriber and these network operators must, for practical reasons, provide i.~.ter network connection. This means that a network operator must take into account not only communication instances arising in their own network or in a limited number of i nter-connected networks of independent but similar administrations, but also communicatior_ instances arising in a theoretically very large number of competing eW 1 . ~;~'.: ~iJ.1 '.ll W 'ric:v v 1 :'='?- :3-J~ '': 1.~: ~U : U1 ~ 1 ~-'zi~~'~ 1V ~- +-1.~ E35 '?399-~-~~~:~: t~ 4 ' 21590~E1 networks of different types and prow=di:g a w_de variety o=
services to subscribers.
It is, therefore, o' increasi:lg importance that data be collected and processed in co:necticn wit: commu~.ication instances arising outs=de an cperaLOr's network :u~
trrm-_nating in or simplll crossvng t_~e operator's re=work.
Ir_ t~.e paper "Access Charge and Rsvenue ~rchitectu= e"
~,T&T Techn-_cal Journal, Vol 0'5, No.3, May 198~, New ~'crk US, pages 73-81, by Cbuchovrs~i, a data system for predic=ing 1~ access charges and revenue is disclosed, for use by an inter exchange carrier in a ?STN o~ the US type after tha separation of local a_~d long distance carriers in the '8Cs ~;d-_vesticure) .
When calls pass t'=rough the network c= more t~~an o::e operator, price and charging agreements between operators for the carriage of each other's calls come into play. Such arrangemer_ts can vary Trom the simple Sender deeps All !5~.~?
arrangement to complex pricing formulae.
It has been an established practice between separate network operators or administratior_s, in teecommunicatio:s, that call data would be collected by the administratio~l resvonsible for the network in which a call arises. If that call then terminates in a second revwerk, the administration co::cerned with the second r_etwork relies on she da~a collected by the administration respor_sible for the f_rsc network, for instance for accounting purposes. In British patent application number GB 2254224, in the name cf ruj_tsu Li mited, a system is described for avoiding double accour_ti__~_g of outgoing international calls based on intelligent ne~work technology.
However , the telecommunications a nvir onment is c:~ang== g quickly, politically as well as techn_ca=ly. w'_th the adven=
of greater competition-, .t is increasingly attractive to network admini$trations to monitor not on~y traffic a=icing in the,_r own network but also traffic arising elsewhere but crossing or terminating in their own network. T_. the network in which traffic ar':ses belongs to a competing opera=or c.
AUfEfVDED S!-~EET

. ,. . c.t' ~ ~Il .....~ . ~ -- ~ m L ~ - ml ~ U 1 . 1 _b :, ~ 1 ~! t - r-~;1 b:J _J;J:J-r-l t~J ~ rc 2i530~a - 2a -admiris~rat.ion, it is desirable that it is at 'east possible tc cross check the competing operator's accounts.
=:1 knOWI2 a=rangemerts, data collectio:~ poi:~ts concerning calls n a PS=h' have been at local exchanges of a ~_etwork since the loca= exchange o_cks -gyp ;.raffic as it arises. This a=ra~_gement, however, does nct nrcvide for data callecrion with respect to -_nter-network traffic. Even were 1~ there to be data collection points to collect dGta on calls incoming to a network, t:~e logistics involved .n processing such data to any level of detail are daunting. For insta~:ce, it has been estimated that calls incoming to the PSTN
AMEiJDED ~~-IEET

W..'. v" . _.. , n. .v.~iir.v .m -- ..)w~ ~ ~ a.u m. ~ . n . .-a . mrw~2::
~... -J.=.;-_ _~ _.

operated in Britain by Britis:: Telecommunications plc (BTi from o;.her ne~work administrations including the =sle of Man and the Cellnet cellu'_ar network totalled i5.4 million calls per day in the twelve mor_ths to March '-99Z. This =ig-~re is expected to incre«se to the order o~ 2? mil''_io~_ ells a cav in the year to March 195. Taking all cal? instan:~_es in=o account, i~?cluding those arising w_thin the ~T . S'I'N, 5u million call instances per day :have been p_edicted =or =9~5.
_ In spite of the very :arge quantity o= data _nvol Yed, '_0 hGs been fc~.:nd pcssible ~n mak;_ng t::e present i nven~ion to design a p~ocess for collecting and processing data relating to calls incoming to a major telecommunications network, =he British PSTN, whic:~ can prod~~ce an output i n suff_cient detai 1 to all ow the associated network adm:.ris;.rat~.on 5 generate account infcrmation which riot only can be al' ocate~a to outside network administrations appropriately, buy also supports itemised billing. That is, tha account infor:naticrl can be broken down in suff:.c=ent detail even to identify ndi vidual call s, so far as they =ulfil preselected criteria, 2o in the marner of item:.sed billing currently available in t=~e national billing system for the British PSTN from Br_tieh Telecommunications plc.
According to a =first aspect of the present -:riven do n, there is provided a process for collecting and proc=sling 25 data concerning communication instances in a fiys~
communications network, wherein the netwcrk inc=odes at _east one point of connec~icn, either directly cr ir_d=rectly, to a second communications network, by means of which poi=t o=
cor_necti on a common ication instance arising in said seconc 30 network can be transmitted into, and either cross o=
term=note in, sai3 first netwcrk, the process comps=sing t~_a steps of i) collecting data at a data access pgint at said pcint o connection, sa=d data concur~:ing a communication 35 instance arising in said second network and comp=is-rig route in'ormation and at least one parameter mea3urement A~iF~;rJE~'~ Sfi ~~ET

rC~,v. W wr.r'.1 vl~ wnt:~ vi .:.~_~- :3-ba : ia:~l . Ulil i_t3 r lW- t,.19 fi~J- '_t:~~-1-lEi~:~ i susceptible of billing, -_'or example duration, with respect to s=id cor~rnunicatic~!~ inst«nce;
ii) transmitting said data into a data proc2saing system;
and ii=_) processing said d g n ~' ata to a erate ~~'__ing inrermat_on.
~y collecting the data at a paint of connection between t:e first network and another ~etwork, it becomes availabl=
to the administration assoc-~ated with the first network to obta-in first hand information about communication instances ~0 incoming to the first network, and thus potentially ~o cross check data provided by other na_work operators e:
administrators.
Accordvng to a second aspect of the gresent invent:.on, there is provided a data processing arrar_gement, ;cr 'S processing data collected -_n a PSTN at « point of connection with another network, the arrangement comprising:
i ) a data i nput for inputting data concern ~ ng communicatior_ instances from a communications r_etwork, said data comprising at least one of a plurality of sort 2G characteristics;
ii) verifying means nor checking she integrity ar_d sufficiency of data received at the da'a input;
iii) a data a_~alyzer ier analyzing data rejected by the verify;ng mear_s, and for submitting amended cr de'aul:.
25 data to the verifying means;
i~~) pricing means for pricing verified data which has beer.
output by the verifying means, in accordance w=th updatable reference informatior_; and vi output means for outputting priced, verified da~a from 30 the gricing mear:s in=o memor~r locations, each memory location being dedicated to data reieva~~t to one or more of said sort characteristics.
Preferably, the pricing means can also validate data, and output errcred data to a data a:~alyzer, wh_ch may be t==a 35 above data ar_alyser or a different or_e, so that data w~:~ ch has been corrupted can potentially be reformatted, or AME~DFD ~'~tET

K~.v. v", ._..-, ... ,...~iL.~ vi -- -3-JJ ~ ~l.:u::~l . vm1 Wb .1V!- --~~J
d'~ ~:~;J~-'r-c!~:J~:- b otherwise corrected, and, therefore, re-entered to :.:~e system as a val id record of a cornmun=ration ins tar~ce .

It may al so (or al ternatively) be that th'_s --yr then data analysis step is used to analyse the data with Yespect to a different type ef fault. For it:sta:~ce, data analysis carried out on errored data which has been located by t:~.e verifying means might be errored princ_paliy in respect eL ~orr2at and routing information while the ergo-ed data prom the pricing mear~s rr.ight be errored principally in respect of pricing to information.

'-'he sort characteristics will typical 1 y be such that the memory '_ocations each hold dta releva: t to corr;unicatio n instances which will be billable to a oommon accountinc entity, for instance, arising in a common respect_~:e '!5 communications network.

The sort characteristics :night be applied at any one of several stages of the data processing ar_angement described above. ~okrever, ir_ a PS'T-'nF for. example, the nature of erro_e data usually arising makes it preferable to provide sorting 20 means between. ~_i':? , the data analyser associated with the verifying means, and tiv), the pricing means. The pricing means therefore acts on data already sor~ed. If the sort characteristics relate to the different e__~.tities who will be billed in respect of the commun:.cation instances repres=_nted 25 by the data, then this arrangement can also have the advantage that the pr~c~~ng means can potentially bm simplified in applying constraints relevar.~ to individual entities.

It might be noted that a network such as the BT FSTN

30 comprises both 1 oral and tru::k exchanges a::d provides not only inter-exchange call transmission but also loca_ call delivery to the er_d user. I'h is means that the data collection and processing ~ecessary to, support bi,-1,_r_g or bill verification has to be sufficie~.tly complex tc deal with 35 an extremely wide range ew variables . '?'his i s ~ r_ contrast tc the situation where a network provides only inter-tr1"=J.k exchange transmission, or only the local call delivery.

AMENDED SHEET

fly . , y ~.u .. m .~...-.. W --. .)-~J:u ~ L~W J~ ~ V 1 n 1 y0 W n ~ T1~ LS:J
~~J:~J'T-tCl~~
- 5a -Accordi ng to a third aspect of the present invention, the=a is provided a data collec.t:on and processing ~=rangement nor ::se in a first communication network which is connected to and receives commun-_cation instances =rom m.u'-tip'-a further networks, the aYrangemenc comprising:
a': reg,.'_stering mea:~s for registering a communication instance incoming to the first r_etwork having ar~_sen in one of said further networks, b; means for formatting a record cT said commur_icatior.
instance, the reccrd comprising data identifying said one of the Further netwcrks and a parameter value such as durat=~on associated with the ccmmunicat_o~: instance, cj validating means for validating said record, d) pricing and charging means for asso-ciati::g p=icing ar_d 2o charging data with a validated record and prcvid-_ng a sorted array of priced, charged and validated records, the array being sorted according to the ide__~_tities o=
~he further networks, and e) analyzing mear_s for analyzing records which are rejeCtrd by the validating means, the analyzing means being arranged to deal with the rejected records in one cz ac least three ways according to the cause o. rej ection, said three ways being to set values in a nor.-validated record to a best-fit value, or to set values in a non-validated record to default values, or t0 archsve or dump the non-validated record, records which have been dealt with ir~ either t he f first way or .he second way being transmitted, directly or ir_directly, to the pricing and charging :n2ans as validated records.
Accordv_ng to a fourth aspec~ of the present ir_venticn, there is prcvided a data colt ection and processirlg jystern, .or use in collecting and processing communication records relevant to a plurality cf networks, wherein said syste:~
AME~~~'D ~'~~'~T

.w , . . ,~ , ~. ., "- _..~, , . __ .~ -,~.J ,..~ ._ ~ ~ . ~ . _o ~ . v . ~ r-, :7 U:J '=:3LJ.~-~-~c.;> -: i., 21~~~~~
- 5b -comprises at least one input for receiving commur_ication records generated at a point o= conneCtzon bstween a first of said p?urality of networks and at least one ocher of said Lluraiity of networks, said records providing identification of the network ir, which a:~ associated communication instance arose cr from w~~ch it entered said first network and a parameter measurement susceptib'_e of bill_ng, such as '~ duration, associated with the communication in6tance, the system further comprising validation means for validat-ng orrnat and routing informatior_ aspects of the records, data analysir_g means for analysing errored records rejected by said lra.~..l~3tio:~ means, the analysing means being capabl a of categorising said errored records and applyi=~c 3efault values to a~ least one category or the errored records, ca~a sorting means, for sorting va_idated and defaulted records according to said ::etwor~C identi ficaticn, and p tic=ng means for receiving the sorted records and, based or_ the information contai ned therain, generating billing inforr..aation fvr use i::
billing entities relevant to the identified netwcrKS.
AMENDED SH~~T

WO 94/23530 2 ~ ~ ~ ~ PCT/GB94100706 A system according to an embodiment of the present invention is now described, by way of example only, with reference to the accompanying drawings, in which:
Figure i shows diagrammatically the architecture of a system for collecti.~.g and processing data comprising call information so as to support an accounts system for call instances incoming to a telecommunications network;
Fi gures 2, 3 and 4 s how overvi ew fl ow di agrams f or a system as shown in Figure 1;
Figure 5 shows a hardware and communications diagram for the system of Figure 1;
Figure 6 shows a software architecture for use between a streamer and a data analyser in a system according to Fi gure 1.
Figure 7 shows an architecture for hardware providing a company system for use in the system of Figure 1;
Figure 8 shows a schematic diagram of batch array processing architecture for use in the company system of Fi gure 7;
Figures 9 and 10 show an exchange file and Advanced Protocol Data Unit (APDU) format, in connection with polling of data from exchanges for use in the system of Figure 1;
Figures il to 21 show flow diagrams for use in a streamer and data analyser of a system according to Figure 1;
Figure 22 represents process interactions between elements of the system in Figure 1;
Figures 23 to 30 provide entity life history diagrams, showing the status that a record within each entity might be in, and from that status which other statuses can be reached by which actions;
Figures 31 and 32 represent the state of an agenda, and a pattern net, following data population and firing of a rule, in an expert system for use in the system of Figure 1;

_ 7 _ Figures 33 and 34 show object hierarchies, for a rule base system and case base system respectively, for use in a data analyser of a system according to Figure 1;
Figure 35 shows design principles involved in building an expert systemjORACLE interface for a data analyser in a system according to Figure 1;
Figure 36 shows a data model for a company system for use in a system according to Figure 1;
Figures 37 to 43 show flow diagrams relevant to operation of a data analyser for use in the system of Figure 1; and Figure 44 shows data flow, with emphasis on data relevant to a company system for use in a system according to Fi gure 1.
In some parts of the following description and Figures, the terms "INCA" and "IDA" may have been occasionally used.
These stand for Inter-Network Call Accounting, a description o f the whol a s ys tem, and f or I NCA Data Anal ys er. The 1 atter is a reference to the data analyser 7 comprising an expert system and interfacing with the Streamer 6.

WO 94/23530 ~ ~ ~ ~ ~ ~ PCT/GB94/00706 _ g _ The description below is set out in the following man ner:

1. FI GURE : BLOCK VI EW OF ~ ARCHI TECTURE

FI GURES 2, ~, AND 4; ~;v FLOW DI AGRAMS FOR PROCESS
OVERVI EW

i) Point of Interconnect and DDC

ii) Streamer iii) Company System (or Box) 3. FI GURES 1 AND 5 TO 8: HARDWARE, COMMUNI CATI ONS AND

SOFTWARE ARCHITECTURES

i) POI and DDC

ii) Streamer and Data Analyser iii) Company System iv) Client Boxes 4. FIGURES 9 AND 10: CALL RECORDS AND DATA FORMATS

i) Call Records ii) Mapping Data Structures onto Exchange Data 5. FIGURES 11 TO 19, AND 22 TO 30: MORE DETAILED FLOW

DIAGRAMS FOR STREAMER AND DATA ANALYSER PROCESSES

i) Streamer: DDC Polling ii) Streamer: FILE PROCESS

iii) Streamer: DDC Deletion iv) Data Analyser: Process v) Entity Life Histories o. FIGURES 31 TO 35: EXPERT SYSTEM

i) Overview ii) Rule Base Generic Rules iii) Case Base iv) Oracle Interface 7. FIGURES 20, 21 AND 37 TO 43: USE OF EXPERT SYSTEM BY

DATA ANA LYSER

8. FI GURES 3 6 and 4 4: COMPANY SYSTEM, DATA ANALYSI S
AND

PRI CI AND CHARGI NG
NG

AUDI T
TRAI L

WO 94/23530 a PCT/GB94/00706 _ g _ 1. FT BUR . ~ ~ BLOCR VT EW OF ARCH. TECTURE
Referring to Figure 1, the system is provided so as to collect data in a First network i, for example the BT PSTN, relating to call instances arising in, or incoming from, a second network 2. The data is collected, at a Point of Interconnect (POI) 3 provided by an exchange of said first network 1, and brought to one of about ten district data collectors (DDCs) 5 in the PSTN. These hold data which comprises route information for each incoming call, thus allowing identi~ication of for instance the intended destination of the call, the carrier from which the call is incoming, itemisation data so that each call is treated as an event, and (preferably) calling line identity so that calls which were simply transit calls in the second network 2 can also be accounted accurately with respect to the network in which they arose.
Each district data collector (DDC) 5 is polled by a streamer system 6 which expands and validates the call data at both file and call record level, primarily against the Routing Reference Model. (Although the district data collectors 5 of the BT PSTN pick up the relevant data, their role may equally be provided by other component systems of an accounting arrangement, such as that known as a network mediation processor.) Data which is found invalid by the Streamer 6 is diverted to a data analyzer 7 where a knowledge-based system is used to assess the invalidity and, where possible, reform the data in an attempt to solve the problem. This is an important component of the system since invalid data will generally be lost as an accountable input.
Validated data is meanwhile streamed, according to the operator associated with the second network 2 from which the associated call was received, and passed on to the company system 8.
The streamer 6 provides the following functions:
~ Poll each DDC 5 for files awaiting processing by the data system of the present invention.

WO 94/23530 ~ ~ ~ ~ PCT/GB94/00706 1 Validate the file and its call records against the Routing Reference Model.

1 Expand the call records and Allocate to the correct Telecom Network Operator.

1 Reject the invalid data tdv the IDA 7.

1 Copy the raw file received from the IDA
7 to the Raw Record Backup Interface directory.

1 Delete the file from the DDC 5 once the data has been secured on the interface directory.

1 Provide the user with a user interface to enter the Routing Reference Model data 1 Provide a complete audit trail through the streamer.

1 Provide the user with the ability to monitor the operation and data integrity of the streaming operation.

The data analyser 7 provides the following functions:
1 Poll an interface directory for files containing one or more errors.
1 Hold the incorrect call records in a suspense area if they are valid call records but do not match the Routing Reference Model.
1 Provide a user interface so that users can re stream the data after the Routing Reference Model has been updated.
1 Apply default call record values to fields that are incorrect in accordance with the rules specification.
1 Stream any correct data that has not been streamed already, due to the error thresholds being exceeded.
1 Stream any corrected data.
1 Provide a complete audit trail through the IDA 7 at a call record level.
The company system 8 also nowadays has an important role to play because it is the company system which imports factors derived not only from call parameters but also from the relationship between the operators of the two interconnected networks 1, 2. The impact a call will have in an accounting procedure will be partly determined by such WO 94123530 215 9 0 0 ~ PCT/GB94100706 factors as the "service level agreement" between the relevant operators. It is at the company system 8 that these factors are brought into play, by reference to various information sources which may include look-up tables and/or the National Charging Database (NCDB)9. With particular regard to the latter, account is also taken here of for instance time-varying charge rates.
The output from the company system 8 is thus finally information for use in an accounting system, representing the raw call data collected from the point of connection 3, and processed with reference to the relevant parameters, such as operator-specific and time-varying parameters, which should apply. This output is provided to a client system 10 which gives user access by means of a personal computer.

WO 94/23530 ~ ~ ~ ~ ~~ PCT/GB94/00706 2. FI GURES 2. 3 AIQD 4: PROCESS OVERVI EW
Referring to Figures 2, 3 and 4, flow diagrams can be used to give a process overview of the above, in operation in response to a call instance.
2 (i) POINT OF INTERCONNECT AND DDC
Figure 2 shows process steps carried out by the POI
exchange 3 and by the DDC 5 in response to an incoming call.
Al l thes a s teps are known, the exchange 3 and DDC 5 bei ng unmodified for the purposes of the present invention.
Referring to Figure 2, a call incoming to or outgoing from the relevant network 1, at step 200 generates a call record in the POI exchange 3. At step 210, the exchange 3 gives every call instance a "File Generation Number" in the series 0-9999. At step 220, the exchange 3 groups the call records into Advanced Protocol Data Units (APDUs ), and groups the APDUs into files.
At step 230, the DDC 5 polls the exchange 3 for all call record data in APDU format. At step 235, the DDC 5 adds control data in the form of a header APDU and a trailer APDU.
The DDC 5 also, at step 240, gives each file a file sequence number in the range from 0-999999, and at step 245 gives each APDU an APDU sequence number in the range 0-16353, APDUs bei ng i n bi nary format. At step 2 50, the DDC 5 pl aces the files in a directory structure, from which the Streamer 6 is able to pick them up by polling. At the same time, at step 260, an entry is made for the file, in a catalogue file which is called DI RINDEX. This catalogue file contains a list of all files available to be polled by the Streamer 6.
2 ( i i ) ~$~$
Referring to Figure 3, at step 300, the Streamer 6 polls the DDC directory structure periodically, entering the call records into a random access memory (RAM), each file being loaded into 1 Mbyte. This polling process includes the step of copying the latest DIRINDEX file. At step 310, which can be in practice part of the DDC polling process at step 300, ~:~S~~OJ

the data is converted from binary to ASCII (American Standard Code for Information Interchange) format.
At step 320, the Streamer o carries out validation of the call records. If a call record is incomplete or incorrect so that it cannot be verified, instead of proceeding to subsequent processing steps in the Streamer 6 and ultimately to a billing process, it is diverted to an interface directory (step 330) for further analysis in the incorrect data analyser 7.
Valid data however progresses, at step 340, to an identification process in which data in the call record is used to establish what other network the call originated in, or entered the BT PSTN from, or in some circumstances was destined to terminate in. A code representing the relevant network operator for billing is added to the call record and the files are then broken down and restructured, at step 350, according to that code. Hence the call records at this point can be sorted according to the network operator, or other relevant entity, who is liable at least at first instance for the cost of those calls.
At steps 360 and 370, the Streamer 6 then outputs the newly structured files to the Company System 8 and deletes file data from the FTAM filestore on the DDC 5.
Looking at the data analyser 7, this has an important role to play since data which cannot be validated cannot be billed. The data analyser 7, at step 380, polls the interface directory for errored files entered by the Streamer 6 at step 330. The data analyser 7 then has three different chances to put errored data back into the system.
At step 382, it looks to repair the data. If it can, the repaired data is returned to the interface directory, from which the Streamer 6 can pick it up. At step 384, the data analyser 7 looks to apply default values to unrepairable data. Some data elements cannot be "patched" in this manner, for instance because it would affect an audit trail. Lastly, at step 386, the data analyser 7 checks whether there is simply a mismatch between the data and the Routing Reference WO 94123530 ~ ~ ~ ~ ~ PCT/GB94/00706 Model (RRM). The latter is a database giving routing information and is used at the DDC 5 to identify for instance the destination of a call. Copies of the RRM are held at different places in a communications network and, if one copy is updated out of time or incorrectly, can give rise to a mismatch in data. If this appears to be the case, the data analyser 7 enters those call records into a suspend file (step 388) which allows them to be put back into the Streamer 6 process after the RRM has been checked.
If the data analyser 7 cannot deal with the data in any of the above ways, it outputs it, at step 390, to a "sump".
This means the data is effectively lost and will never be billed. It might however be useful in analysis so that changes and corrections can be made to the system in the long term.
2 (iii) COMPANY SYSTEM
Referring to Figure 4, data, at the file level, which has been validated and processed by the Streamer 6 is input to the Company System 8 where the first step, step 400, is validation of the file sequence number. The Company System 8 processes files in file sequence number order, but the Streamer 6 has processed data in parallel from different exchanges 3. If the file sequence number is wrong, the Company System invalidates the file and stops processing it (step 410).
If the file sequence number is acceptable, the Company System 8 moves on at step 420 to validate the call record, this time not primarily in terms of the RRM, as at the Streamer 6, but with more emphasis on data relevant to the billable entity and the relationship between the billable entity and the operator of the first network 1, for instance BT. The billable entity and BT will have entered into a service level agreement (SLA) and the call record might indicate a call type not available to that billable entity under the current SLA. The Company System 8 will pick that up as an invalidit,- and, at step 430, attempt to fix the call 2~~~000 record in error. I f the call record can be fixed, i t is sent to be bulked, at step 440, and re-entered to the Company System 8. I f it cannot be fixed, it is stored, in step 450, for analysis.
Valid call records meanwhile are forwarded to the Company System pricing engine, step 460, at which individual call records are priced in accordance with the NCDB 9, the SLA between the relevant billable entity and BT, and any other relevant information. The priced call records can then be loaded into a summary database, step 470, for charging to the relevant billable entity, and the call records are output to optical disk (step 480) for storage.
Client Boxes 10 receive downloaded information from the summary database on a weekly basis. Each Client Box 10 is dedicated to a single billable entity and can also be used to access the optical disk storage, to obtain its "own" call records only.

WO 94123530 2 ~ J 9 ~~ ~ t~ PCT/GB94/00706 3, FT(;ttRFS 1, 5 6. 7 AND 8: uARDWARE. COMMUNICATION AND
yOFTWARE ARCHITECTURES
3 ( i ) p~T'~"' OF I NTERCONNECT 3 AND DDC 5 The exchanges 3 and DDCs 5 are of known type and are not described in detail herein. They operate, briefly, as follows.
Referring to Figures 1 and 2, any call coming into or leaving the British PSTN operated by British Telecommunications plc (BT) will nowadays pass through a digital telephone exchar_ge ~s t:e °oint of Interconnect (POI)3. All such exchanges relevant to the data system of the present invention are currently System X telephone exchanges of types Digital Junction Switching Unit (DJSU), Digital Local Exchange (DLE) or Digital Main Switching Unit ( DMSU ) .
Every telephone call going into or leaving the BT
network 1, as shown at step 200 of Figure 2, generates a call record within the POI exchange 3 in the format known as British Telecom Call Record Type 6. The System X POI
exchanges 3 are polled daily by the DDCs 5, at step 230, for all call record data in APDU format. Polling takes place using the File Transfer Access and Management (FTAM) protocol across high-speed BT data links. DDCs 5 act purely as collectors of call record files from POIs: no processing of call records takes place within a DDC. DDCs are not dedicated to call record polling, but perform a variety of other data collection, processing and forwarding tasks.
In order for the streamer system 6 to gain access to the FTAM filestore on the DDC 5, it is necessary to provide identification. This is done by allocating a Network Nodal Identity (NNI) to the streamer 6 as a relevant end system.
The NNI is then used as a username for gaining access to the FTAM filestore, along with a password.

_ 1 7 _ 3 (ii) STREA:~SER 6 AND DATA ANALYSER 7 Referring to Figure 5, the hardware and communications diagram for the streamer 6 and the data analyser 7 may be as follows. (I t should be understood that the communications architecture of Figure 5 represents only one of any number of communications architectures that might be suitable in different enviro.~.:~e~ts. ) The streamer 6 has a "hot-standby"
Streamer Box Backup (SBB) 6a which cuts in as soon as a fault on the main. streamer system b occurs, and both can be provided on Hewlett-Packard HP857S minicomputers running the UNIX operating systam. Th a strGa~~~er a and SBB oa might be connected to local area networks (LANs) 515.
Raw data polled by the streamer 6 (or hot-standby 6a) from the DDCs 5 (not shown in Figure 5) is backed up using an optical disc storage system (not shown). The data is polled using FTAM (File Transfer, Access and Management) over BT
Megastream high-speed data links and a Multi-Protocol Routing Network (MPRN) 510. The MPRN 510 is OSI (Open Systems Interconnection) compliant. There are direct communication links 515 between the streamer 6 and the data analyser 7 and an "Ethernet" bridge 505 gives the streamer 6 and the data analyser 7 access to at least one wide area network (WAN) 510, for instance that used by BT for the PSTN. The WAN 510 in turn gives access to a company system 8 and client boxes 10 situated at the primary BT PSTN network management and data centre. This means that the network management and data centre can input and output data, for instance for analysis and to input init'_al and updated routing reference data.
Referring to Figure 6, the Data Analyser 7 might be provided on a Hewlett-Packard HP9000. Software for the Streamer 6 and the Data Analyser 7 utilises the following technologies:
1 IEF for Batch Processes 1 ART/IM for Expert System Capabilities 1 HP/UX Version 9.0 1 Business Objects as a PC Client for reports 1 Oracle Versie_~. 5 2159~4c~

_ ,3 _ - SQLFORMS
- SQL*Report Writer 1.1 - PL/SQL Version 1. 0 - PRO*~
- SQL*NET TCP/IP Version 1. 2 Ail these are known ~~rrc publicly available. For instance "IEF" is the "I'r.~~ormation Engineering Facility"
Computer Aided Software Engir_eering (CASE) software from James Martin Assoc~ ates, a software engineering tool which generates executable code. The data analyser processes run ph;~sically on the rata araZyse_ 7 F,latiorm and use SQL*NET to connect to an Oracle database 60 on the Streamer 6 platform.
SQL*NET TCP/IP (Transport Control Protocol/Internet Protocol ) can also be used by Streamer/Data Analyser Business Objects Oracle users 65 in order to access the Oracle database 60 located on the Streamer 6 over the MPRN 510, or a suitable TCP/IP bearer network.
The Streamer 6 and the data analyser 7 share database facilities 60 and the users may require access for instance to check or update reference data used in validation by the Streamer 6. The database facilities 60, inter alia, maintain control over which files from the DDCs 5 have been processed and contain a version of the Routing Reference Model.
PRO*C code is generated by the IEF into the IEF code 61 and External Action Blocks (EABs ) 62 as shown in Figure 6.
The Streamer/Data Analyser software library 63 is a set of "C" and PRO*C modules, callable from within EABs 62 or from the ART-IM (Automated Reasoning Tool for Information Management) 64. The ART-IM is proprietary, expert system, application development software. The ART-IM development is conducted within " studio" , a Motif interface to the expert system. Once the expert system has been unit tested within the " s tudi o" , i t i s depl oyed by generati ng " C" modul es f rom within the "studio". Hence, for instance, processes can be created by generating the IEF Code 61 on an OS/2 workstation, and linking the code with EABs 52, the Streamer/Data Analyser WO 94123530 ~ ~ PCT/GB94/00706 _ Z~
software library 6~ and the ART-IM code library 64 on the deployment platform.
( i i i ) coMPA~~ s~sm~'~: a Referring to figures 7 and 8, the Company Sox (or System) 8 compr~.ses a Hewlett-Packard minicomputer 70, " Emerald 890!400" , running the UNIX operating system, the ORACLE relational database management system (RDMS) and a custom application written using the IEF CASE software from ?0 James Martin Associates.
l9i thin the Company Boy: 3, ail :,all records are priced according to complex pricing and charging reference tables, and ORACLE summary tables are incremented. Reference tables provide exchange set-up data, routing reference data, accounting agreements, pricing and charging data and various classes of exception. Pricing and charging reference tables are derived from BT's National Charging Data Base (NCDB) and inter-operator wholesale pricing agreements.
To help the minicomputer with the very high volume of processing tasks involved, Hewlett-Packard UNIX workstations 80, for example " 735s" , are attached as co-processors which bid for processing tasks. A virtually unlimited number of such workstations may be connected to the minicomputer 70 to increase the number of call records that the Company Box can process but a reasonable minimum for the BT PSTN might currently be for instance twelve. As stated earlier, it may be that the data system of the present invention will be required to process 60 million call records per day by 1995.
The arrangement relies on the Hewlett Packard product known as "Task Broker" 81, the data system of the present invention being set up to run on a batch array. In order to do so, custom parameters need to be fed into Task Broker and an appropriate set of these parameters are listed below:
i) Global Parameter Settings (which are optional) - which clients may access server - which machines may remotely administer Task Broker WO 94/23530 ~ ~ ~ PCT/GB94/00706 - ~0 -which network mask to be used - smallest and largest UID (user identity) al l owed - loggi:.g verbosity - maximum number of task submittals to be processed concurrently.
- list ."achines that client should contact for ser~~ices.
ii) Client Parameter Settings which are optional) 0 - list ror each service, the servers the client should contact for service.
iii) Class Parameter Settings - every service must be in a class; set up a class for each type of service each machine will provide.
iv) Service Definitions ( for every service, the following must be specified) class - affinity - arguments Note, affinity is a number between 0-1, 000 which indicates how well a node is able to provide a service.
Task Broker ~s a queuing system which controls which work stations bid for and process files. In order to use Task Broker with the company system 8, there are three programs and a configuration file. The configuration file sets up the parameters Task Broker needs to operate in the company system environment including which work stations it can communicate with, what programs to call to process a file, and how to prioritise. It is the configuration file parameters which are set out above.
The three control programs operate (in summary) as follows. When a fil a comes to the Emerald minicomputer of the company system 8, a master program " run_cp. sh" sends it to be processed via Task Broker and kicks off a monitoring program " cleanup-cp. sh" in the minicomputer. Task Broker allocates the file to a work station, which processes the ~,0 94123530 PCTIGB94100706 fil a according to a third program " cp. sh" . I f things go smoothly, the file returns to the minicomputer where "cleanup ct. sh" allocates i t to the correct directories of a client system 10. "Cleanup_cp. sh" also monitors the work stations.
If there is an overlong delay in processing by a work station, it will shut down Task Broker on that work station since there is clearly then a problem. Lastly, "cleanup cp. sh" also control s recording and event logging.
Finally, as well as an output to the client system 10, 1 0 pri ced call records f rom the Company Box 8 are saved to an array of Optical Disc Drives 71, so that individual priced call records may be retrieved and analyzed in future.
3 (iv) CLIENT SYSTEM (OR BOXES) 10 15 Summary ORACLE database tables of interconnect calls are downloaded weekly from the Company Box 8 to Client Boxes 10.
Each Client Box (CLB) 10 is a Hewlett-Packard UNIX
workstation, and deals only with summary database tables and call records generated under a single interconnection 20 agreement between BT and another operator, for example Manx Telecom. A Client Box 10 runs an ORACLE RDMS, and Business Objects software. Information from each Client Box 10 allows BT not onlZ~ to bill. another network operator in respect of their use of BT~s network, but also to verify incoming bills 25 from another network operator to BT. Each Client Box 10 can also interrogate the Optical discs 41, but only for call records under the interconnection agreement associated with that Client Box 10. It is not possible for a Client Box to interrogate the Company Box 8 directly for its own call 30 records, let alone those relating to other agreements between BT and other operators. Personal Computers are connected to a Client Box 10 to allow analysis of the Summary Tables.

s. FIGURES 9 AND '0: CALL RECORDS AND DATA FORMATS
.~ CALL RECORDS' (i) British Teleco~~ Type 5 call records are generated for the primary purpose of billing customers. Call records should contain sufficient information to price a call accurately, based on date, time, duration, distance to be travelled and other factors.
Each Type call record can include the following:

0 1 length Of bi 1 Brig ra~p_rrj;

record use;

1 record type;

call type & call effectiveness;

clearing cause;

time disconti.~.uity flag (change to/from GMT from/to BST

duri ng cal l ) ;

1 calling line identity (CLI);

1 route group type;

sampling category;

route group;

nodal point code (NPC): unique identifier for POI

exchange producing record;

1 linking field (used when call straddles more than one time-charge band);

calling party category (business, residential, payphone);

charge band;

1 date and time of address complete;

date and time of answer;

date and time of calling party clear;

1 date and time of called party clear;

called number field.

Call records are captured by the Call Accounting Sub-system (CAS) Revenue Apportionment and Accounting (RAA) facility on all System X exchanges. As mentioned above, at step 220 call records are grouped together into APDUs, and APDUsare further grouped into a file, with each file being WO 94/23530 215 9 ~ 0 ~ PCT/GB94/00706 up to 1 Mbyte i.~. size. Nothing in this grouping process within a System X POI exchange destroys any parts of individual call records. all call records are in simple binary format.
Referring to :'figure 9, each exchange file 40 contains a number of APDUs 51, which are of variable length. Each APDU
51 contains a number of billing records which are also of a variable length. The following are, however, fixed ~ Exchange File Maximum Size 1 Megabyte ~ APDU Maximum Size 512 Bytes ~ Billing Recor$ P~ta::imum Size 70 Eytes The DDC Header and 'frailer APDUs are identical apart from the APDU type which is 241 for header APDU, and 245 for trailer APDU.
The following information is available in the header and trailer APDU: -APDU Length .........,.,. Length of header/trailer APDU
APDU type ........... " ,. 241 for header, 245 for trailer Unique File Identifier.. . See below concerning DI RINDEX
Destination NNI ......... NNI of _TNCA Streamer Application Group .....,. Application Group of INCA data =14 Input tapeicartridge Seq. No ................ Sequence Number of tape/cartridge Output File Seq. No .... DDC Sequence Number Timestamp DDC rece;ved data . . . . . . . , . . . , , , , , , . , Date and Time data received by DDC
Partfile Indicator ..... Indicates whether file is a part file Exception Indicators ... Indicates what may be wrong with file Read Count ............. No. of times this file has been read Filesize .............,. Size in bytes of this file Count of unselected APDUs .........,..,. " ,. No. of APDUs of wrong APDU type Selected APDU type ..... APDU type of INCA data type WO 94/23530 ~ ~ ~ PCT/GB94100706 .~
APD'J Count . . . . . . . . . . . . . ::umber of APDUs in t'.:is file First Sea. No .......... Starting APDU Sequence Number Last Seq. No ........... Endi:ng APDU Sequence :lumber The read count represents the number of times this file has been tolled from the DDC by the Streamer 6. The partfile indicator indicates whet:~er the whole file was received by the DDC 5 successfullv or whether harts of the file were mi s s mg.
?0 The exception indicators are two 1 byte bitmask fields which indicate a :y cri3rs that were detected by the DDC 5 relating to this transfer.
The valid values for all of the fields above will be validated within the computer aided software engineering (CASE) application software described below with reference to the " COMPANY SYSTEM ( OR BOX ) " 8.
Referring to Figure 10, a brief description of the APDU
structure 51 would include the APDU header 52, the actual billing records 53 concerned, and the APDU trailer 54.
The format ror the billing records 53 is of standard type and a "C" structure can be designed to map exactly onto that format.
When data has been polled from the exchanges 3 to the DDC 5, some of the data which is at the head of each data APDU is stripped out by the DDC 5. This data is representative of the DDC 5 and of the exchange 3 and is not relevant as a data feed for an end-processing system.
When the file is copied into an appropriate directory by a DDC 5, such that it is made available for the streamer 6 to copy using FTAM, an entry is made for it in a catalogue file, called DI RINDEX. The DI RINDEX file entry carries the following data:
i) activity marker (1 byte) which may show a) inactive entry b) file available for transfer c) file currently being used (eg in FTAM
transfer) d) file successfully transferred (not yet deleted) ii) INCA filename format iii) output routing, which may show a) file available for FTAM
b) magnetic tape only iv) unique file identifier, including details such as the creation time and the relevant exchange NNI.
file size in bytes vi) number of APDUs in file.
Looki ng at i i ) , the I NCA fi 1 ename format, that i ncl udes vii) the streamer NNI
viii) NNI and cluster number of exchange ix) application group of INCA data x) DDC file sequence number of exchange file.
4 (ii) MAPPING DATA STRUCTURES ONTO EXCHANGE DATA
The streamer 6 maps the data into data structures for use in the model for the Company Box 8, using the following principles . -1 It is assumed that the APDU length fields and the billing record length fields will be correct. If they are not then the validation will fail at either the APDU
level or the Billing Record level, and the file will be streamed to the Data Analyser 7.
1 The input data will initially be scanned to find the Header APDU 52. This will be identified by an APDU type of 241 (Hex F1). The selected APDU type field will then be checked along with the Unique File Identifier to establish that this is indeed the header APDU 52.
1 After the header APDU 52 has been found and the header APDU data structure has been mapped, it is assumed that ~15~(~~J
WO 94/23530 PCTlGB94/00706 all of the APDUs in the file will follow the data standard of a one word record length followed by an APDU. eg.
/HEADER APDU/RL/APDU/RL/APDU.../RL/APDU/RL/TRAILER APDU
where RL is tze Record Length.
If the structure of the file deviates from this standard then the file will be streamed to the Data Analyser 7 for further analysis. This error condition will be detected within the validation of the APDU immediately following the deviation.
~ Within each APDU it is assumed that the structure foflows that of Figure 6. Again any deviation from this structure will cause the whole data structure mapping to become mis-aligned, and will lead to the file being rejected and streamed to the Data Analyser 7.
~ It is assumed that there will be no data after the trailer APDU 54. Any data that does appear after the trailer APDU 54 will be lost.

WO 94/23530 ~ PCT/GB94/00706 5. FIGURES 11 TO 19 AND 22 TO 30: STREAMER AND DATA
ANALYSER PROCESSES
(i) STREAMER: DDC POLLING PROCESS
When files are received by the DDCs 5 they are validated 5 (using checksumming) and some extra information is added to the beginning and end of the file, in the APDU Header and Trailer 52, 54, as mentioned above with reference to Figure 2. These files are then made available for polling by the Streamer 6 by placing them in the directory structure to be used by the streamer 6, and updating the DI RINDEX file. This DI RINDEX file contains a list of all files available to be polled by the Streamer 6, and the Streamer 6 uses that list to ensure it has polled all new files.
Referring to Figure 11, the Streamer 6 will prepare to poll multiple DDCs 5 by going into a "Stream all DDCs"
process. At step 700, the streamer 6 "stream all DDCs"
process is triggered, for instance at a specified time. At step 710, it runs a check that the Streamer 6 is available to receive files from the DDCs 5. If the Streamer 6 is available, it goes into a cycle, steps 720, 730 and 740, in which it runs through a list of the DDCs 5 and creates a "DDC
Process" for each DDC 5 to be polled. At the end of the list, this process finishes (step 750).
For each DDC 5, the Streamer 6 will now run the "DDC
process" . Referring to Figure 12, at steps 800, 805, the DDC
process starts with a check as to whether either the DDC 5 concerned for that process, or the Streamer 6, is shut down, and a check at step 810 as to whether DDC polling is due.
There are certain times at which the DDCs 5 cannot be polled and step 815 runs a check as to whether a non-poll window applies. If not, step 820 looks for a free process slot to process files. If all these checks are clear, the streamer o accesses the DDC DI RINDEX, step 825, and initiates the process list, step 830, and file list, step 835, which will ensure the streamer 5 applies all relevant processes to each of the exchange files received from the DDC 5. In step 840, 845 and 850, the st=earner 6 runs through the files from the DDC DIRINDEX, creating its own log of the exchange files to be processed, and provides a file processing capacity, steps 855 and 860, to process the files in the file list. Once all the exchange files from the DDC DIRINDEX list have had processing capacity allocateel;- the "DDC process" updates its record of when the next poll is due, step 865, and goes back to sleep, step 870.
The DDC process will be stopped of course, step 875, if either the DDC 5 or the streamer 6 has shut down, and will remain in sleep mode, step 870, whenever a poll is not due, the DDC is in a non-poll window, or there is no available processing capacity.
~vnical event cycle Assume the following DDC POLLING MPH = 17 ... This is the minutes past the hour to Poll DDC POLLING INT_HRS - 1 ... This is how long to wait for next Poll in hours DDC DELAY IN DELETE = 12 ... How long to wait after the file has been marked for deletion before actual deletion request occurs.
The System has been booted at 23:30 on the previous day.
Time Schedule 00: 17 DDC Process wakes up, copies over DI RINDEX file, and creates processes to stream data to the Streamer 6.
00:30 DDC Process finishes creating processes to stream files because either the Maximum number of processes have been created OR all of the files available have been given to file processes to download.
The wakeuptime is calculated as 00:30 +
DDC POLLING INT HRS and set minutes to DDC POLLI NG MPH.

~~ WO 94/23530 0 ~ PCT/GB94/00706 _> Next Polling Time = 00: 30 + 1: 00 = 1: 30 (SET MPH) - O1: 17 Calculate the number of seconds to sleep -TO SECONDS (01:17 - CURRENT TIME) Sleep (seconds to_sleep) ... File Processes complete streaming of data 01:17 DDC Process Wakes up ...
5 (ii) STREAMER: FILE PROCESS
Referring to Figure 13, the operation of the File Process, created at step 855 during the DDC Process, is as follows. The File Process works from a file list received from the DDC Process, step 1302. Running through the file list, step 1303, for each exchange file listed, File Process reads the exchange file log, step 1305, validates the call records, step 1306, copies the file to a raw record backup, step 1307, for use if for instance the streamer 6 goes down subsequently, diverts the file to the data analyser 7 if there was a validation failure, step 1310, or streams the file to the Company Box 8, step 1312.
File Process stops, step 1313, if the DDC 5 or the Streamer 6 is shut down, step 1304 or if the files are seriously corrupted, step 1311, for instance because communications with the DDC 5 have failed. The exchange file log monitors what stage an exchange file has reached in relation to the Streamer 6 and will carry a status selected from active, processed and deleted for each file, where "active" indicates it is being processed by the Streamer 6, "processed" indicates it has been processed by the Streamer 6, and "deleted" means it has been deleted from the DDC 5 by the Streamer 6.
Referring to Figure 14, the step 1306 in which call records are validated can be expanded as follows. At this point, steps 1401 and 1402, the exchange file is copied from the DDC 5 and the file header and first APDU header 52 validated, steps 1403, 1405. If either fails, a file error log is created, step 1412. If both are acceptable, the call WO 94/23530 ~ ~ ~ ~ PCT/GB94100706 records are each validated, steps 1407, 1408, and a call record error log created, step 1409, if one fails.
Validation is repeated for each APDU 51. Whether validation has shown all is correct, or errors have been logged, the audit trail is updated 1413 and File Process moves on to step 1307 as described above.
Referring to Figure 15,'f.=les which have been validated during the File Process are now ready to go to the Company Box 8. At this stage, the file structures are broken down so i0 that the individual call records 53 can be sorted according to the billabla entity they are relevant to. The call records 53 are now written to physical files, step 1503, for the different billable entities.
5 (iii) DDC FILE DELETION PROCESS
Once a data file has been successfully downloaded from the DDC 5 to the streamer 6, and the data has been expanded and streamed to the appropriate Company Box 8, the data file must be deleted from the FTAM filestore on the DDC. The streamer 6 will delete the file using an FTAM delete request a number of hours after the file has been secured on either the company box 8 (or local storage for the company box 8 if the link to the company box 8 has gone down). The exact time between the data being secured and the files being deleted can be set on a per DDC basis.
5 (iv) DATA ANALYSER: PROCESS
Referring to Figure 16, the step of validating call records in FILE PROCESS, step 1306 in Figure 13, generates a file error log, step 1412, and a call record error log, step 1409. The data analyser 7 runs two processes, the "DA
?ROCESS" and the " SUSPENSE FILE PROCESS" , which are initiated during the boot-up sequence of the HP9000.
DA PROCESS monitors continuously whether data which has been sent by the Streamer 6 is available to be processed by ?5 the Data Analyser ~. This data will always exist initially as the original exchange file, irrespective of whether the WO 94/23530 ~ ~ ~ PCT/GB94/00706 _ ~i _ data contained individual call records which could not be streamed, or the failure was at file level.
As 1 ong as t he Data Anal ys a r 7 i s not f 1 agged as s hut down, step 1602, DA PROCESS will first pick up the earliest file error log to be processed, step 1603, and check whether it was a failure at file /APDU level or at call record level, step 1604.
Referring to Figures 16, 17 and 20, if the failure was at call record level, DA PROCESS will pick up the next call l0 record error log with regard to the file, step 1702, and send the relevant call record to the ART xM rule base for correction, step 2000. If the failure was at file level, the whole exchange file has been rejected by the Streamer 6. In this case, the complete file is loaded to memory, step 1606, and the file header and APDUs 51 sent to the ART IM, step 1607, for correction.
There are several outcomes to analysis done by the Data Analyser 7. Fixable data will be sent to the ART IM to be corrected, and subsequently can be validated and streamed to the Company Box 8. If a routing error is involved, the data may be put into suspense in case there is a problem with a record of routing information somewhere in the system, for instance because it needs updating. It may be possible to validate call data after all, once the routing information has been corrected. If a whole file is unreadable, it might have to be sent, still in binary format, to a Binary File Dump. If data, for instance a file, is determined by the ART
IM to be unfixable, and the error is not concerned with routing so as to justify suspension, it may be archived. The data will never be billed but may be used in analysis to identify long term or significant problems which themselves can be put right and so avoid losing billable items in the future.
Returni ng to Fi gure 16, the mai n DA PROCESS, havi ng us ed the ART IM to run checks at step 1605 and 1607, will next sort out files which have been returned from the ART IM as unfixable. If they cannot even be read, step 1608, they are WO 94/23530 ~ ~ ~ ~ ~ PCT/GB94100706 _ ~~ _ forwarded to the binary file dump. These files can potentially be read, since they may be in hexadecimal, octal or ASCII format, and might be used at a later time for anal_:sis. Alternatively, r_ies might be readable by the Data Analyser, but are still rated ." unfixable" by the ART IM.
These are, at step 1009, loaded to a " SUMP" database where, again, they will never provide billable data but can be queried and analysed.
If a file has been sent to the ART IM and was fixable, :0 the ART I~d will return each call record sequentially for validation, steps 1610 and 1611. DA FR~CESS will then validate these call records first by checking for a routing failure, step 1612, and creating a call record error log, step 1615, in the event that there is call record failure.
These will get picked up and rerun through the ART IM, steps 1603 to 1605 and :701 to 1703. If the call record is acceptable, it will be streamed to the Company Box 8, via steps 1616 to 1618.
Referring to Figure 18, where there has been a call record routing failure, detected at steps 1612, 1704 or 1907 (see below), the call records are categorised and suspended.
That is, the failure is analysed to the extent that it can be matched to an existing routing error pattern, step 1802, and then the call record is added to an existing pattern file which contains all call records showing the same routing error pattern, step 1803. These pattern files are held in suspension, the primary BT PSTN network management and data centre being notified. A separate process, SUSPENSE FILE
PROCESS, then deals with these files.
SUSPENSE FILE PROCESS is an important aspect of the data analyser 7 because it takes a category of errored files, which can potentially be corrected, out of the "mainstream"
of data processing. These files may only have been picked up as errored because routing data somewhere in the system has not been updated. They are potentially billable. By means of SUSPENSE FILE PROCESS, the primary network management and data centre has the opportunity to update routing data in the WO 94123530 ~ ~ ~ PCT/GB94/00706 system and still catch files found errored previously.
Further, by appending call records to an existing pattern file, a "Route Pattern Suspend File", for a particular route pattern, files can be selected for reattempting validation by simply running a selected Route Pattern Suspend File.
Referring to Figure 19, as long as the process is not shut down, step 1902, SUSPENSE FILE PROCESS starts by locating the earliest routing pattern which has been amended, for instance, by the network management and data centre, step 1903. It will then pick up the next suspended file containing that routing pattern, step 1904, and attempt to validate the call records, steps 1905 and 1906. There may of course be more than one routing error in the call record. If that is the case, SUSPENSE FILE PROCESS will revert to step 1801, on Figure 18, and create a routing error entry in a routing error pattern file, thus re-suspending the call record. However, if there is no other routing failure, SUSPENSE FILE PROCESS will attempt to stream the call record to the Company Box 8, by reverting to step 1501 on Figure 15.
The PROCESS runs through all of the call records in the suspended file in this way, step 1910, and all the files which have been suspended with respect to that particular route pattern, step 1911.
Referring to Figure 22, this shows the process interactions between the streamer system 6, the company box 8 and the data analyser 7. The main process area of the streamer 6 is the one called "FILEPROCESS". This does all the validation and intrinsic operations on a file. In the data analyzer area there is the "IDA FILEPROCESS" which enters data to the expert system. Importantly, this process triggers the Route Pattern Suspend File and "SUSPENSE
FILEPROCESS" by appending data to a Route Pattern Suspend File. It is this which avoids a large backlog of data building up because SUSPENSE FILEPROCESS operates outside the main IDA FILEPROCESS. Another area of interest is the "SUMPDATABASE" receiving output from the "SUMPLOADER".
Although data in the SUMPDATABASE cannot be put right, it can WO 94/23530 ~ ~ ~ ~ ~ PCT/GB94/00706 .. _ 34 -be queried and analysed so that, potentially, rules at the IDA FILEPROCESS can be changed so that subsequent data can be re-streamed.
In Figure 22, processes are shown in circles, the Company Box as a block, data files, logs and the like are shown between open-ended parallel lines and archived data is represented by the conventional symbols for databases.

The process, and stored data, interactions referenced (a) to (y) on Figure 22 can be listed as follows, the arrow heads denoting relevant transfer directions:

a) NNI and list of file names to be processed, transferred b) Exchange file log, STATUS - A, created c) DI RINDEX file accessed d) FTAM exchange file copied e) FTAM exchange file deleted f) Exchange files, where STATUS - P, read g) STATUS set to D if exchange files deleted successfully (at (e) above) h) Exchange file log read where STATUS - A

i) Exchange file log data updated. STATUS set to P

j) File is in error so file error log created k) Call record is in error so call record error log created 1) File copied to Data Analyser directory if file is in error m) File error log read n) Call record error log read o) Raw (binary) data file looked up p) Data appended to route pattern suspend file for this route pattern q) Entry made in route error pattern r) ART/IM created closest matches s) ART/IM has identified that this data cannot be fixed.
Data is placed in the SUMP for further analysis or deletion t) User has identified the problem cannot be fixed. File is placed into the sump for further analysis or deletion u) When file structure unintelligible, file thrown into binary file dumps v) Streamed file created.
w) SUSPEND FILE PROCESS is initiated by status on route error pattern being set to ready. If problems persist then Count field updated and status set to SUSPENDED
x) Closest matches are updated if the chosen solution fails to fix the problem y) Streamed file created 5 ( v ) ENTI TY LI FE HI STORI ES
Referring to Figures 23 to 30, entity life history diagrams can show the statuses that a record within that entity can be in and, from that state, which other states can be reached by which actions. In each of these Figures, the statuses are simply identified by the reference numeral 2300 and the definitions of the statuses are given below.
Figure 23: File Error Log;
READY - the file is ready to be streamed by the data analyser 7.
SUSPENSE - either the whole file or a least one call record within the file has been sent to the suspense area.
BIN - the file could not be read by the data analyser 7 and has been sent to the bin area.
SUMP - the whole file has been sent to the sump area.
COMPLETE - the data analyser 7 has streamed the file and any of the files call records in the suspense area have been successfully re-streamed or archived.
Figure 24: Call record error log;
READY - call record is ready to be streamed by the data analyser ?.

SUSPENSE - call record has been sent to the suspense area.
SUMP - call record has been sent to the sump area.
ARCHIVED - call record has been sent to the trash area (ie ARCHI VED ) .
COMPLETE - the data an.alvser 7 has streamed the call record successfully.
VAL FAILURE - there are differences in the ART-IM and IEF
validation procedures.
_igure 25: Route error pattern;
UNSELECTED - created by ART-IM and waiting analysis by data analyser user, or re-streamed after analysis but failed.
PENDING - selected by data analyser user for analysis.
READY - data analyser user completed analysis and is ready to be re-streamed.
CLOSED - successfully re-streamed or ARCHIVED.
Figure 26: Closes matches;
UNSELECTED (OR NULL) - generated by ART-IM.
SELECTED - selected by data analyser user for anal ysis.
Figure 27: Sump file log;
SUMP - a file is ready for the SUMP PROCESS.
PROCESSING - t:~e file is ready to be viewed by the data analyser user.
ARCHIVED - the file has been archived.
Figure 28: File route error link;
SUSPENDED - the file is in the suspense area.

~~~oooo COMPLETE - fhe file ras beer. successfully re-streamed from the suspense area.
F figure 29: F,xchange file iog;
AfCTIVE) - exchange file is being processed by the streamer.
PROCESSED) - exchange file has been processed by the Streamer.
DELETED) - exchange pile has been deleted by the Streamer Figure 30: District data collector;
(All statuses are changed by Streamer 6 users via SQL*Forms. ) P(REBIS) - DDC is prebis.
L(IVE)- - DDC is live.
CEASED) - DDC has been ceased.
Referring to Figure 6, it will be seen that the Streamer 6/Data Analyser 7 software architecture includes IEF
external action blocks (EABs) 62. The EABs 62 are used where it is inappropriate or not possible to implement within the IEF. For instance, the following functions might be carried out by means of the EABs 62:
~ "Add call record to suspense"
This module will create a new entry of a call record, within a linked list containing call records for an exchange file which are to be sent to the suspense file.
~ "Add call record to archive".
Creates a new entrv of a call record, within a linked list containing call records for an exchange file which have been fixed but cannot be re-streamed, to the archive directorv.

WO 94123530 ~ ~ ~ ~ ~ ~ ~ PCT/GB94100706 _ 3g _ ~ "Add network operator record".
Checks whether this is a new Network Operator and if so will create a _new entry in the linked list of "network-operator-structure". If it is an already used network operator name it will add a linked list entry into the linked list of call, records for that network operator.
Where a fix has been applied to a call record it will update the "call-record-error-log" with the "network operator record" identity and streamed call record sequence number.
1 "Call =ecord to IDA rules"
Passes a single call record to the data analyser ART-IM
rule base. The call record is identified by the APDU
sequence number and call record sequence number passed in by IEF. The data structure loaded in memory is then searched for the call record and owning APDU and exchange file header data. This data is then fed into the rule base and validated. Any errors found will each generate a call record rule log row entry. The call record error log record status will also be updated by the rule base.
~ " Commi t"
Commits all current database changes.
~ "Create DDC process"
Creates an occurrence of the DDC process which will be responsible for polling a particular DDC. It will create/open a fifo (file in/file out) to the child process and will write into the fifo the value of the DDC NNI.
1 "Create file process"
Creates the process which will perform the task of streaming the file _names passed in the array of file names.
~ "Delete file from DDC"
Deletes a file using the FTAM protocol from disc on the DDC.

~' WO 94/23530 ~ ~ fl PCT/GB94100706 '9 J
1 "Delete data analyser file"
Deletes a file from the streamer/data analyser directory.
1 " Delete suspense file"
Deletes a file from the suspense file directory.
1 "File to bin"
Passes a file which cannot be read into the ART; IM rule base to the binary file dump.
1 "File to data analyser rules"
Passes a whole file to the data analyser ART-IM rule base and initialises the rule base. The initialisation of the rule base involves clearing old data, selecting default and routing reference data and populating the rule base with that data. The data analyser binary file is then loaded into a data structure in memory. This data is then fed into the rule base and validated. Any errors found will each generate the appropriate rule log row entry. Call record error logs will be created by the rule base where appropriate, together with route error pattern, closest matches and file route error link records. Once validated, the rule base will return a validation status to IEF and retain its internal data for later retrieval.
1 " File to sump"
Passes a file which cannot be fixed by the ART-IM rule base to the sump.
1 " FTAM HLCOPY"
Copies, using the FTAM protocol, the DDC file name from the DDC FTAM addres s us i ng the DDC us er name, DDC pas sword and DDC account to the Streamer 6. The user name, password, account and FTAM address of the Streamer 6 can be defaulted to NULL if required. This routine is not called directly from the IEF and hence does not return an IEF style status.

WO 94/23530 ~ ~ ~ ~ ~ t~ PCT/GB94/00706 1 "Get DDC process parameters"
Creates or opens a fifo in the streamer/TMP directory which will be named "DDC process fifo < PID >". It will read the values of the DDC NNI from the fifo, the data having been inserted into the fifo by the "create-file_process" routine.
~ "Get file process parameters"
Creates or opens a fifo in the streamer/TMP directory which will be named "file process fifo < PID >". It will read the values of the above variables from the fifo, the data having been inserted into the fifo by the "create_file process" routine.
~ "Get exchange file"
Copies a file using the FTAM protocol from disc on the DDC, straight onto disc on the Streamer 6. The file will then be read into memory on the Streamer 6 and then renamed into the raw record backup directory from where it will be archived. This module calls "map data_structure to-file" in order to set up the initial pointers to the first billing record, first APDU, and the header and trailer APDUs.
1 "Get no. invalid data analyser APDUs"
Returns a count of invalid APDUs re-processing of call records which have failed the Streamer validation process.
1 "Map data analyser file"
Reads a file into memory for subsequent processing.
~ " Process active"
Establishes whether a particular PID is active and returns a flag accordingly.
~ "Read exchange file header"

2~~~000 Uses the pointers to the header and trailer APDU to return in a structure all of the fields from the header and the APDU type from the trailer.
1 "Read data analyser exchange file header"
Uses the pointers to the header and trailer APDU to return in a structure all of the fields from the header and the APDU type from the trailer for a file which has been sent to the data analyser.
1 "Read first DI RINDEX record"
Copies, using the FTAM protocol, the DIRINDEX file from the DDC to temporary storage, and opens the file and returns the first record to the caller.
1 " Read next APDU"
Returns the APDU structure pointed to by the current APDU pointer and sets the current APDU pointer to the next APDU. Also sets the current billing record pointer to the first billing record within the returned APDU, and copies and byte reverses the data into the current APDU array.
1 " Read next DI RINDEX records"
Reads the next record from the DI RINDEX file on the DDC.
1 "Read next data analyser record"
Returns the next billing record output from the ART-IM
rule base. Successfully processed records will appear first, followed by those which require sending to the suspense file.
1 "Read next suspense record"
Returns the next billing record output from the suspense file.
1 "Read next record"
Returns the billing record currently pointed to by the current billing record pointer, and sets the pointer to the WO 94/23530 ~ ~ ~ ~ PCT/GB94100706 next billing record if this record is not the last in the current APDU. (This is determined using the APDU length and the minimum length of a billing record.) ~ "Rename network operator files":
Renames anv network opexator files that nave been written to the temporary.~directory in the operational directory ready to be processed by the Company Box 8.
1 " Sleep"
Will sleep for the specified number of seconds.
~ " Stream file"
Dumps the file in memory to the data analyser ready for data analyser processing.
1 "Stream file network operator"
Uses the pointer to the first network operator to get to all of the validated, expanded records for that operator. It then attempts to write the records from the linked list into a nfs temporary directory. If successful, the file is renamed into the nfs directory. If the file cannot be reopened on the nfs temporary directory, the file is opened on the local temporary directory, and upon successful writing the file is renamed into the local directory.
1 " Stream file RRB"
Dumps the file in memory to the raw record backup di rectory.
"write console"
Writes a message to a network management workstation.
~ "Write to suspend file"
Writes the records from the linked list of suspended call records for an exchange file into the suspend directory.

~"~ WO 94/23530 ~ ~ PCT/GB94/00706 ~ "Write to archive file"
Writes the records from the linked list of archive call records for an exchange file into the archive directory.

~1.~9~~~

c. .~t rLlR~'B '~ i m0 ~ :,. ~XprRm gvgTrM. ART-I M
( 1 ) OVERVI E~~
The expert system uses the facilities of the ART-IM
knowledge based expert system tool kit supplied by Inference 5 Corporation. It is a knowledgejrule base programming system which allows for a flexible model of decision making, and therefore modelling of the real ~,aorld, within the knowledge hierarchy, as well as providing a more heuristic method for problem solving. The tool kit contains the ART-IM language as well as an integrated editor, an interactive development environment, tools for tre development of end-user interfaces, a method of deploying run time versions of developed applications and the facility to interpret external data intelligently.
In the data analyser 7, the expert system is split into two subsystems, the rule base and the case base. In general, the case base is used to deal with routing based errors, and the rule base is used for defaulting and calculable errors.
Both make use of the ART-IM functionality.
The rule base uses the ART-IM procedural languages including rule, function and methods. Each error is defined within a schema and instances of these schemas are used on the data structures. All schemas within an in-data object hierarchy are populated via the IEF/ART-IM interface using the "DEF-EXTERNAL FUN" facility of ART-IM.
The mechanism of programme flow control used by ART-IM
is very different from seauential statement-by-statement flow, as usually found in programming languages. Referring to Figures 31 and 32, the expert system holds all its internal data, that is schemata and facts, in a pattern net 3100. This is represented in Figure 31 by a plurality of patterned circles, each representing a piece of internal data (a schema or a fact). This data can be set up by ~ loading an ART-IM test case file (more usually done in a development/unit testing context).
~ by populating from an external source (eg Oracle or IEF;
more usual in a productionisystem test environment).

.~ ~159~~~

1 by generating from ART-IM rules (used as very flexible "working storage" eg generatio~ of error schema after validation test failure).
Once set up, data is compared directly with the conditions specified within the rules. A rule resembles an "IF< conditions ~ THEN < action >" of a more traditional programming language. If conditions of the rule match exactly an instance of data, an activation is created within an associated agenda 3105. All instances are checked against all rules. In terms of performance, the pattern net and rule conditions are managed by an efficient pattern-matching al gori thm wi thi n t:'~e ART-I P~i run ti me s ys tem.
At the end of the evaluation part of the cycle, all rule activations are placed in order on the agenda stack. The first rule activation on the stack will be fired. The order of appearance of activations defaults to random unless salience, that is priority of rules, is set by the developer.
Referring to Figure 32, after firing of the topmost rule activation on the agenda 3105, the action of the rule has actually changed data in the pattern net which will in turn alter what appears on the agenda stack following the next evaluation cycle.
It might be ..~.oted that the data instance causing the initial firing (the circled instance 3110) will not be reevaluated, thereby avoiding continuous looping although if the data within the data instance changes and the new pattern matches a rule condition, then a rule activation will be created.
The~ART-IM run will finish when:
1 no matching conditions and patterns found.
1 all matching conditions and patterns have already fired rul es .

2~59a~3~
,16 _ The above car_ be summarised as follows:
1 ) rule ac tivations are generated by matching data patterns with rule conditions 2) rules can, by default, fire in any order although priorities can be set all data is evaluated ~n parallel re-evaluation occurs each time a rule has fired 5) the same rule can fire many times during a run, depending on the number of matching data instances i0 0) rule conditions are sensitive to changes in the pattern net 7) ART-I?d stops if no matching rule conditions or pattern net data is found or all matched activations have fired al ready.

WO 94/23530 ~ ~ ~ PCT/GB94/00706 _ 97 _ Referring to Figure 33, the rule base system is based on an object hierarchy, as shown. Each of the objects 3300 is defined in ART schemas and the connecting lines between objects 3300 depict inheritance from the object above.
The exchange file, APDU and call record contain slots for each data item ~n their structure. Each slot has a corresponding slot in the appropriate default object to declare whether t::e resultant has a default value, a calculable value or ~s an un-modifiable field. The rule base :0 uses the default system to check what form an error correctfcn gust be, if all~wed.
The above covers the data schemas. With regard to error schemas, every possible data analyser error has its details described within an appropriate schema. Each error '_5 description and its instances contains a slot for each of the following:
The object on which the error relates, that is an exchange file.
An error descr=ption.
20 The affected slot.
The specific data object for an error instance.
The name of the repair value.
The source of the error.
The resultant repair value.
25 The rule position in fire order.
The value of t::e slot prior to any fix being applied.
o (ii) RULE BASE GENERIC RULES
The rule base operational flow is controlled by a number 30 of generic rules, these performing the following functions:
1 for each occurrence of an error trigger, that error' s repair method is fired to generate a repair value and its fire order allocated.
1 for a fixable error where there is only one affected 35 slot, the affected slot is updated with the repair value generated and the t~~~e stamp of the change is stored with the instance of the error.

WO 94123530 ~, ~ a ~ ~ ~ PCTIGB94I00706 _ s8 ~ for each i.~.stance of an error where the repair description declares the error type as suspendable, the data item affected is moved to the suspense file and the time stamp of the move is stored with the instance of the error.
1 for each instance of an error where the repair description declares that the error type is sumpable, the data item affected is moved to the sump and the time stamp of the sumping of the file is stored with the instance of the error.
1 a record is created on the file structure rule log for each fix on an APDU or ex.char~g2 file.
~ an Oracle record is created on the call record error log for each fix on a call record with the appropriate error i of ormati on.
~a xable errors 1 ) Default values can be allocated to the following fields:
APDU type and trailer billed call indicator called party clear PBX suffix record use record type DDC time stamp header APDU type class of data transfer format version number node time stamp part file indicator table size trailer APDU type called party clear application group 2) The following errors are calculable:

- WO 94/23530 ~ ~ ~ ~ PCT/GB94100706 - .~9 -APDU lengt:n; the lenat:~ of the APDU.
APDU count; the length of the APDU sequence.
End APDU sequence number; start sequence number plus the number of valid APDUs.
Start APDU sequence number; obtained from the sequence number of the first APDU in the exchange file.
Dialled digit count; the length of the dialled digit s t ri na.
i0 There are error exceptions with regard to the above, such as where the checksumming for an APDU shows an error.
Errors of this type are immediately sumped within the rule base. Some errors with regard to the APDU sequence result in the complete range of sequence numbers being re-sequenced ?5 from "1", and the related exchange files being updated. It may be that the last digit of a dialled digit string is a character between A and F. The repair value here is the dialled digit string minus the last digit.
20 non-fixable errors On a non-fixable error occurrence, the data item in error, ie a call record, is passed to the sump, as described above, and the appropriate error log updated. Areas which cannot be amended, and which therefore generate non-fixable 25 errors are as follows:
address seizure time stamp address completion time stamp either address or answer time stamp 30 calling party clear time stamp calling line directory number seizure time stamp dialled digit string (except when the last digit is between A and F).

WO 94/23530 ~ ~ J PCTIGB94/00706 o (iii) THE CASE: BASE SYSTEM
The routing =eference case base is a case base of routing patterns (ie TUN, Youte group, route number, NNI, Nodal Point Codei plus other reference data, for example the billable network operator name, and Live and Ceased Node time stamps. The case base reference data is populated from the Routing Reference schemata which in turn are populated from data contained within the Streamer Reference Data subject area 3600 of the data model (See Figure 36).
Referring to Figure 34, it can be seen that the object hierarchy for the case base system is similar to that for the rule base system, shown in Figure 33, with the addition of three obj ect classes 3400; " suggested solutions" , "potential solutions" and "routing reference". It might be noted that "suggested solutions" and "potential solutions" are only created following identification of a routing reference error and contain mainly pointers to other data, that is incoming call record in error and routing reference schema that are most closely matched. The "routing reference" schemata are created from the routing reference data on the Oracle database.
With regard to the routing case base, and initialisation, the routing case base is populated at the start of a run from routing-description schemata. One case is created for every routing description schema. The routing case base will be set up with the following parameters.
~ Maximum of three matches ~ Any matches below threshold value of zero probability will be ignored so as to weed out highly unlikely matches.
~ Only the following slots on the case base are used in pattern matching;
TUN (ie Telephony Unit Number), route group, nodal point code, route number, NNI, and direction ~ The following are ignored for purposes of pattern matching:

.,~._ ~,~,G 94/23530 ~ 0 4.f PCT/GB94/00706 Live node rime stamp Ceased node time stamp Telecom network operator role and name ~ Direction is treated as slightly different for matching purposes. It is the least significant matching slot and is given a fixed weighting ceiling of 50 of the overall weighing. The other slot weights will be split equally between the remaining 950 of the overall weighing.
i0 Pattern matching, together with other case base functions such as setting initialisation parameters, is achieved by sending messages to the case base. The pattern matching is done in two steps, these being to send an incoming call record schema to the case base, which will return the number of matches found, and to send a retrieve-match-score message which will determine the closeness of match for each returned case together with the key of the Routing Reference schema associated with the returned case.
The case base is used for an error code validation relating to call pattern, in the cases of Nodal Point Code or Route Group Not Found, Invalid Route Number, or Direction Invalid, as follows:
~ attempt to ri.~.d an exact match between each incoming call record and a case on the Routing Reference case base.
If there is an exact match the call record has a valid routing pattern and no further validation with regard to the above error will be required.
~ if no exact match if found, an error schema will be generated which triggers the rule base generic rules, as above, which will apply a repair method.
1 the specific ~~epair method will create one suggested-solution schema which will contain (looking at Figure 34):
i) up to three potential-solution schemata, each containing a pointer to the associated Routing Reference schema. The potential-solution schema will also contain the match position ( ie closes t, next closest etc ) and a % measure of the closeness of t~:e match, and i;) a pointer to the incoming call record in error.
I t s hould be noted that the repai r method wi l l di f f er from the usual repair method invoked by generation of an error schema instance because it will consist of a function irouting-mismatch, which wi~.l assert the suggested-solution schema instance a:.3 facts containing keys to the Routing Reference schema) and another rule (generate-closest-matches, which will trigger on generation of the facts created by routing-mismatch and will generate one instance of a cotential-solution schema for each case base match found).
Where node time stamp validation is concerned, the case base will be used as follows:
1 to attempt to find an exact match between each incoming call record schema and a Routing Reference schema. If there is an exact match the rule will then check for time stamp discrepancies (ie seizure time stamp should fall between node live and cease times) on the matching incoming call record schema and the Routing Reference schema. If no discrepancy exists, no further processing associated with this error will take place.
~ if a time stamp discrepancy is found, an error schema will be generated which triggers the rule base generic rules, as above, which will apply a repair method.
~ the specific repair method will create one suggested-solution schema which will contain (see Figure 34):
- one potential-solution schemata each containing a pointer to the associated Routing Reference schema. The potential-solution schema will also contain the match position (ie closest, next closest etc.) and a % measure of the closeness of the match.
- a pointer to the incoming call record schema in error.
It should be noted that the repair method will again differ from the usual repair method invoked by generation of an error schema instance, because it will consist of a WO 94123530 ~ PCT/GB94/00706 - ~3 -func~ion (node-timestamp-discrepancy - which will assert the suggested-solution schema instance and facts containing keys to the Routing Reference schema) and another rule (generate-node-time-discrepancies which will trigger, on generation of the facts created by routing-mismatcn and will generate one instance of a potential-solution sci:ema).

WO 94123530 ~ ~ ~ ~ ~ ~ PCTIGB94100706 o i i v j AR''t'- ~ M AND ORACLE I NTERFACE
Referring to Figure 35, direct access to the ORACLE
database from ART-IM is required to exploit fully the "parallel" validation features of the ART-IM rule base.
There are four main interfaces:
1 the population of Routing Reference data 3500 1 the populatio~ of default data 3505 1 the output of fix data to form an audit trail 3510 1 the output of routing error patterns as a precursor to suspense data handling 3515.
Looking at the population of Routing Reference data, this interface 3500 involves refresh of internal ART-IM
schema and casebase from data in the Routing Reference Model physically held within ORACLE tables:
1 the refresh is triggered during the initialisation phase of an ART-I M run.
1 existing internal ART-IM Routing Reference schema are cleared together with their casebase entries.
1 data is SELECTed from ORACLE tables from a ProC program ( EAB_I NI TI ALI SE I DA-RULEBASE ) whi ch wi 11 be us ed as part of two External Action Blocks (file to ida_rules and call record to ida rules).
1 I nternal ART-I M s chema are popul ated by the ProC program 2 5 1 The Routi ng Re f erence Cas ebas a i s i n turn popul ated f rom the internal Routing Reference schema by a function (inca-ida initialise-casebase) as part of the casebase initialisation process.
Looking at the population of default data:
1 the refresh is triggered during the initialisation phase of an ART-I M run.
1 existing internal ART-IM default (df-call-record, df apdu etc) schemata are cleared together with their casebase entries.
~ data is SELECTed from ORACLE tables from a ProC program ( EAB_I NI TI ALI SE I DA_RULEBASE ) whi c h wi 11 be us ed as part WO 94/23530 0 ~ PCTIGB94/00706 of two External Action Blocks (file to ida rules and call record to i~a rules).
1 I eternal ART-I ?~! s chema are popul aced by the ProC program Looking at the creation of Error and Fix data, if errors are detected during incoming data validation which are associated ~f~ith date that car. be fixed then an audit trail of the fixes applied by the rule base needs to be maintained:
~ for every file structure in error a row entry is created in the FILE-ERROR-LOG table. This is done by the streamer process.
1 for every call record in error a row entry is created in the CALL_RECORD-ERROR-LOG. This can be done by the streamer process or by ART-IM.
1 For every error detected and fix applied at the file structure level a row entry is created in the FILE STRUCTURE RULE LOG on the ORACLE database. This is best done by the rule base using a generic rule which is triggered when all file level error detection and fixing has completed. The rule should fire once for each error detected/fix applied and when fired will invoke a user-defined-procedure call sql_exec limited which does the necessary insertion.
1 for every error detected and fix applied at the file structure level a row entry is created in the CALL RECORD RULE LOG on the ORACLE database. This is best done by the rule base using a generic rule which is triggered when all call record level error detection and fixing has completed. Again, the rule should fire once for each error detected/fix applied and when fired will invoke a user-defined-procedure call sql_exec immed which does the necessary insertion.
1 the ART-IM rules will populate the inserted values from slots on internal schemas.

WO 94/23530 ~ ~ 4j ~ PCT/GB94/00706 Looking at the creation of Routing Error Patterns and Closest Matches data, if errors are detected during incoming data validation which are associated with data that is suspended then a record of the i ncoming call record error pattern (based on TUN, NNI,. route group number, route group, direction) together with~the three closest matches (based on the closest patterns on t he routing reference model to the incoming call record in error) needs to be stored on the ORACLE database for later suspense file processing. Patterns are stored following completion of all validation/fix processing. In more detali:
1 for every error generated that is a suspense file error (and assuming no unfixable errors have been generated on the same call record - these unfixable call records are weeded-out using the move to sump generic rule), a generic rule (move_to suspense_file area) is fired. The rule tries to select the pattern in error from the database, and, if the pattern in error exists:
i ) tests for any entry in FILE ROUTE ERROR-LINK with relevant pattern exchange file and foreign keys.
ii) if the entry exists no further action is required.
iii ) if the entry does not exist then inserts a row entry into the FILE ROUTE ERROR LINK.
If the pattern in error does not exist:
i v ) i ns erts a row entry i nto a ROUTE ERROR PATTERN tabl a populated by error pattern data from incoming call records.
v) inserts a row entry into FILE ROUTE ERROR_LINK.
vi ) inserts up to 3 row entries into the CLOSEST MATCHES
table populated by routing reference patterns found from previous casebase processing to be closest to the route pattern in error.
~ A user defined procedure is used to pass SQL command to 3 5 ORACLE.
1 The ART-IM rules will populate the inserted values from slots on internal schemas.

WO 94/23530 ~ ~ ~ ~~ PCT/GB94/00706 _ 57 _ ;, FTGURES 20 21 37 TO 43: USE OF EXPERT SYSTEM BY DATA

In the flow diagrams referenced below, it might be noted that a slightly different format has been applied from that of earlier flow diagrams in this specification. That is, function calls are denoted by boxes with double vertical lines, simple statements are denoted by boxes with single vertical lines, and yes/no decisions are denoted by a simple di amond.
The use of the ART-Iid expert system by the data analyser 7 pan be expressed in flow diagrams. Referring to Figures 16, 17 and 20, once it has been determined that there is a failure at call record level, step 1605, and the next call record error log has been selected from a file, step 1702, the relevant call records are sent to the expert system, step 2000. The expert system locates the correct APDU, steps 2005, 2010, and then the errored call record, steps 2015, 2020.
The expert system then checks whether the call record is correctly itemised (step 2025), in this example according to System X itemisation, and, if it is not, directs the call record to sump by setting the IEF status to "SUMP", step 2030, while updating the call record error log, step 2035.
If the call record is correct'_y itemised, it is "put through"
the expert system, steps 2040, 2045, 2050, and the results assessed by the data analyser 7, in step 1704 onwards.
Referring to Figure 16 and 21, it may have been decided that there is failure at file or APDU level, step 1604. In that case, the file is loaded to memory and the file header and APDUs sent to the expert system; step 2100. The expert system database is called up, step 2105, and the APDU schemas from the previous run deleted, step 2110. The first test run is to refresh the expert system version of the Routing Reference Model, step 2115, which may immediately result in correcting the apparent error. If not, the default data for the expert system is refreshed, step 2120, in case for instance default data for the error concerned has previously WO 94/23530 ~ ~ ~~ PCT/GB94/00706 _ 5$ _ been missing. If either of these is successful, the data analyser process reasserts itself, Figure 16, and the results from the expert system refresh'steps will allow the file to go to validation of its cal~l'records, step 1611. If neither is successful, the call records themselves must be individually validated. This is described below.
Referring to Figure 37, the function box 2125 of Figure 21, "map header and APDU schema", expands to include loading (steps 3700 to 3725, 3735 ) and running (steps 3730, 3740, 3745, 3750) the expert system, ART-IM, with respect to call records from the errored files which could not be processed successfully after refreshes of the Routing Reference Model and default data. This loading process includes getting data not avai 1 abl a on the ART databas a ( " Forei gn Keys" ) , f or instance data from the Streamer 6, in step 3715, to enable the expert system to access the files. Having analysed each call record, the ART supplies a status (step 3755), which may indicate the call record is fixed or should be suspended or Bumped. The data analyser process (IEF) keeps a count of the call records to be sumped, step 3760, and sets a flag in the ART-IM, step 3765, which triggers clean-up by the ART-IM, step 3770, to clear out each call record and relevant schemas to avoid these simply building up.
Referring to Figures 38 to 43, the application of the expert system file rules can also be expressed in flow diagrams, and the following examples are shown, the flow diagrams being self-explanatory:
i) Figure 38; ART File Rules (exchange file header) This can be applied to trailer APDU
format version number file type node timestamp DDC/NMP timestamp (NMP stands for Network Mediation Processor) class of data transfer WO 94/23530 ~ PCTIGB94/00706 _ 5A _ node cluster identity streamer NNI
application group part file indicator file byte size table size selected aPDU type ii) Figure 39; APDU first sequence number rule iii) Figure 40; APDU last sequence number rule ivl FiQUre 41; APDU secruence number count rule v) Figure 42; ART APDU rules This can be applied to -retransmission indicator linking field vi) Figure 43; ART call record rules This can be applied to -record use billed call indicator clearing cause PBX suffix CLI cluster identity network circuit network band circuit identity circuit number charge band call sampling method sampling mode count reset indicator value of N (where N relates to a count made while running a test set of call records for example) called party clear timestamp t - n0 -8. FIGURES 36 AND 44: COMPANY SYSTEM
Referring to Figure 4, the output from the Streamer 6 to the Company System d comprises call records sorted according to billable entity, and validated as described above using a data analyser incorporating the ART-IM expert system.
The primary role of the Company System 8 is to price the call records and to output t:.e priced records so that they can be billed to clients. However, it also has a validation role, as mentioned above, with emphasis on data relevant to IO the billable entity and the reiati onship between the billable entity and the operator of r_he first network 1. The company system 8 therefore incorporates or accesses a company system data analyser, referred to in the following as "cIDA".
The cIDA application can reside alongside the data analyser 7 which validates data from the Streamer 6, described above. In Figure 4, the steps of fixing errored call records, 430, bulking the fixed call records, 440, and investigating unfixable call records, 450, can all be carried out by means of the cIDA application.
Interestingly, it has been noted that the majority of errors, of the order of 90% of the errors picked up by the company system 8, concern decode anomalies, mainly to do with " time lines" such as " 123" and " emergency services" ( 999 ) calls. The bulk of the remainder of errors can be attributed to discrepancies in reference data. There can therefore be two primary aspects to building a data analyser for use with the company system 8, these being to tackle records providing the majority of the errors, the decode anomalies, and then to provide an infrastructure capable of representing files back to the company system 8 after correction.
Processing Overview A suitable arrangement might be as follows. Error and warning files are sent from the company box 8 to the cIDA
where they are loaded to specific directories, one per operator. A single file can hold zero or many records.
Preferably, the cTDA provides a parallel processing facility WO 94/23530 ~ o ~ PCT/GB94/00706 for all operators, ~unnina concurrently, with the capability of manual override. A log is maintained in order to control the sequence of files into and out of the cIDA.
Once an error file has been selected for processing, the cIDA selects each record in turn, assuming the file is not empty, and evaluates the error into one of two categories:
fixable and unfixabl e. U.~.fi~abl a records are written to a table, reported on, and can later be removed from the database for archiving. Where a record has been deemed to be fixable, it might be fixed automatically by applying rules, or -t might need manual intervention before it can be fixed.
Each record, irrespective of error type, is inserted into an ORACLE database table, with all details passed from the company box 8 and a flag set to indicate the "state".
The state might, in accordance with the above, be selected from suspense unfixable rul es Users, using Business Obj ects run at regular intervals, have the capability to view all records currently held and the state allocation they have been given. An audit log can be held for a relevant period, such as for one month for all " charging number string" corrections.
2 5 I t mi ght be noted that the us a of automati c rul es may well be found unnecessary. By correcting errors caused by decode anomalies, hat is 90% of current errors, the error rate has been found to be reduced to 0.01%. Hence, the simplicity of errors arising means that a system employing automatic rules would be over complicated.
Referring to Figure 44, the dataflow routes about the data collection and processing system of the present invention can be seen. In this Figure, data stores such as files and tables are represented by the horizontally extending rectangles with vertical dotted lines, and processes are represented by the bigger blocks, incorporating ~1~9~~J

rectangles. Entities external to the whole system, such as the NCDB 9, are represented by the "lozenges".
As already described, raw call data is entered to the Streamer, which converts the raw call data, validates and processes the call records, involving a data analyser so far as necessary, and outputs~vaiidated, itemised call records to the company box. The company box firstly performs operator specific validation, and secondly aggregates itemised call records. At this stage, the call records are priced, using charging information for instance from the national charging database (NCDB) 9, and output in summarised form to produce a bill report for the relevant client system 10. Other outputs include the expanded call records, stored on optical disc 71, and summarised call records for a management reporting system 4400.
It can be seen in Figure 44 that there is also an output from the data analyser to an auditing system "CARDVU" 4405.
Although embodiments of the present invention can provide extremely detailed information for audit purposes, the auditing system itself is not part of the invention and is not therefore described herein, beyond the comments below at " 9. AUDI T TRAI L" .
Referring to Figure 36, a data model for the company system 8 shows clearly the data sources for use at charging and pricing by the company system 8. Much the greatest amount of data, the " C&P reference data" , is derived from the NCDB 9. However, there are constraints set by the accounting agreement 4500 between the billable entity and the operator of network 1. Many issues can be dealt with from the network management centre and the data model of Figure 36 provides appropriate visibility thereto by means of the "telecoms network operator role" box 4505.
The following initials, used in Figure 36, can be expanded as follows:
CBM Charge Band Matrix CB Charge Band NN Network Node WO 94/23530 ~ ~ PCT/GB94/00706 KCH Kingston Communications, Dull (an operator in the UK of a network interconnected to the BT
PSTN
TE Telecom Eirann (as above) NCIP National Charging Information Package (an interface to data on the NCDB) Pricing and charging engines, complying with the type of constraints offered by the system of the present invention, are known and specific description of the charging and pricing engine is not therefore offered here. Indeed, although the data model of Figure 36 shows all entities involved, not all the relationships are shown as the representation would become too complicated. Overall, however, it must be borne in mind that the call records handled by the company system 8 are already sorted according to billable entity. This aspect of the data needs to be maintained, clearly, so that relevant reports can be allocated to the correct client systems 10. This can be done, as indicated above, for instance by maintaining allocated directories for the billable entities.

WO 94/23530 ~ ~" ~ ~ PCT/GB94/00706 9. AUDI T TR.AI L
An arrangement as described above can provide a sophisticated audio trail. Data from the exchange at the point of interconnect comes in a file, and is packaged into APDUs. The streamer system 6 polls data off the DDCs 5 using the FTAM protocol, ~he data being in binary, in call records.
The streamer system 6 validates the data against the data base containing reference data, the Routing Reference Model, and assesses which other network operator should be billed.
The streamer system 6 writes a full call record in ASCII with operator and exchange information added.
An audit trail arises as follows. On the exchange, call instances are numbered with a File Generation Number which cycles from 0-9999. The DDC 5 also adds a sequence number which cycles from 0-999999, at the file level. Within the file, APDUs are also sequenced with an APDU sequence number which cycles from 0-16353, being binary.
This means that there is stored a record of the number of records in a file, the APDU start and finish numbers, and the number of APDUs.
Because a sequence number is added to each number at the exchange, it can be ensured that the company box 8 receives the numbers in sequence, although they will not necessarily be processed in order. The streamer system 6 actually processes in parallel from different exchanges at the same time.
In the data analyser, where a "pattern net" is used, by means of which data will "fire" a rule if it does not fit valid content, the analyser can patch data items only where the data item concerned would not affect price or the audit trail. Patch in this context means set to a standard value.
Hence, the data analyser cannot change the call record sequence number because that identifies the call record. If the call record sequence number were to be changed, there would be no audit trail.

,y.., v,. . ~.., ,. .'...~., .. __~- ~-JW ~ 1~:~~uJ ~ ml~y ,_5 y!~~- r-f'J-CS:J _J'J'.~J-~-~C~J~illl T_he system described above is, as stated., or_ly ore specific embodiment of the invention. It relates to a STN
~-~d, as described, deals with call records in a voice commv.~nicat_ons system. Further, the specific form of ca'_1 records ir_vo~'_ved, SV~stem X '"ype 5, relate to only one type of exchange w~icz might be used at a point of interco:lnection POI ) be ~WeE~ rletWCrlC3 .
A simple extension cf the application of t .e irve :t~_o:l is that, as well as using caii record data to gerlerate billing information, traffic analysis inrormatiory can also be picked up a~-~d processed. For instance, calls which are ineffective is reaching a destination, "ireffeccives", car be counted by t~e exchange at the PoI and t'~e "bulited" outcome input to the data processing system.
3owever, more significant c:.anges might incl::da the use cf the system with commun,_catiors other than voice communicatians, even excluding voice comrnur.ications, ar~:d, as already mentioned, it is clearly not esssntza~, that a PST?~1 is involved, although the benefit of embodiments of the invention is c_early significant with a PSTN in the light a=
the sheer volume of records and complexity cf sources vnvclved.
AhIE~!DED E~E~j

Claims (15)

- 66 -

1. A process for collecting and processing communication records in a first communications network, the records concerning communication instances, wherein the network includes at least one point of connection to a second communications network, the process comprising the steps of:
i) collecting data at a data access point at said point of connection, said data concerning a communication instance arising in an originating network other than said first network, and creating a respective data record for each communication instance, each such record comprising route information identifying the originating network and at least one parameter measurement susceptible of billing, such as duration, with respect to said communication instance;
ii) transmitting said records into a data processing system which includes a data analyser;
iii) validating data in said records followed by analysing invalid data;
iv) the analysis of invalid data identifying data which can potentially be set to a default value;
v) setting the invalid data to a default value; and iv) processing each said record to generate biding information;

2. A process according to Claim 1, wherein said first network comprises a public switched telephone network.

3. A process according to either one of claims 1 or 2, wherein said data processing step comprises streaming said data according to the identity of said originating network.

4. A process according to any one of the preceding claims wherein the first network comprises a communications network including both local exchanges and trunk exchanges and the data processing step includes correlating pricing and charging data from a database in accordance with the route information identifying the originating network.

5. A process according to Claim 4 in combination with claim 3 wherein said correlation is carried out subsequent to streaming the data.

6. A data processing arrangement, for processing data collected in a communications network but concerning call instances arising outside the network, the arrangement comprising:
i) a data input for inputting said data, said data comprising at least one of a plurality of sort characteristics;
ii) verifying means for checking the data received at the data input;
iii) a data analyzer for analyzing data rejected by the verifying means, and for substituting amended or default data therefor;

iv) pricing means for pricing data output by the verifying means or by the data analyser, in accordance with updatable reference information; and v) output means for outputting priced data from the pricing means into memory locations, each memory location being dedicated to data relevant to one or more of said sort characteristics.

7. A data processing arrangement according to claim 6 wherein each sort characteristic identifies a further network outside said communications network in which further network an associated communication arose.

8. An arrangement according to either one of claims 6 or 7 wherein said communications network is a public switched telephone network.

9. An arrangement according to any one of claims 6, 7, or 8 wherein said data analyser comprises means far storing data which cannot be amended or defaulted in a suspended data store, for potential subsequent processing.

10. A data collection and processing arrangement for use in a first communication network which is connected to and receives communication instances from multiple further networks, the arrangement comprising:
a) registering means for registering a communication instance incoming to the first network having arisen in one of said further networks, b) means for formatting a record of said communication instance, the record comprising data identifying said one of the further networks and a parameter value such as duration associated with the communication instance, c) validating means for validating said record, said validating means rejecting records which include one or more invalid data fields:
d) pricing and charging means for associating pricing and charging data with a validated record and providing a sorted array of priced, charged and validated records, the array being sorted according to the identities of the further networks: and e) analyzing means for analyzing rejected records, the analyzing means processing the rejected records according to the cause of rejection, said processing comprising:
f) setting values in a non-validated record to best-fit values, or g) setting values in a non-validated record to default values, or h) archiving the non-validated record, or i) dumping the non validated record records which have had values set by the analysing means being transmitted, directly or indirectly, to the pricing and charging means as validated records.

11. An arrangement according to claim 10 wherein an exchange of said first network is arranged to receive a communication instance, a data collector is arranged to register a record of the communication instance as said registering means, routing information incorporated in said record providing the data identifying said one of the further networks, and wherein the validating means has access to a routing reference data model and is arranged to use the degree of correlation between the routing information and the routing reference data model as one of the criteria in validating a record.

12. An arrangement according to either one of claims 10 or 11, wherein the analysing means is further arranged to process rejected records in by appending data concerning a non-validated record to a in a suspended data file for subsequent access and analysis..

13. An arrangement according to claim 14 wherein each suspended data file stores non-validated records having the same error pattern.

14. An arrangement according to any one of claims 10 to 13 wherein the pricing and charging means comprises validating means, or has access to validating means, and is arranged to output non-validated records to the analysing means so as to allow reprocessing of data which has become corrupted since first being validated in the arrangement.

15. A data collection and processing system, for use in collecting and processing communication records relevant to a plurality of networks, wherein said system comprises at least one input for receiving communication records generated at a point of connection between a first of said plurality of networks and at least one other of said plurality of networks, said records providing identification of the network in which an associated communication instance arose or from which it entered said first network and a parameter measurement susceptible of billing, such as duration, associated with the communication instance, the system further comprising validation means for validating format and routing information aspects of the records, data analysing means for analysing errored records rejected by said validation means, the analysing means being capable of categorising said errored records and applying default values to at least one category of the errored records, data sorting means, for sorting validated and defaulted records according to said network identification, and pricing means for receiving the sorted records and, based on the information contained therein, generating billing information for use in billing entities relevant to the identified networks.