WO2009078001A1

WO2009078001A1 - Prediction of mobile subscriber's location

Info

Publication number: WO2009078001A1
Application number: PCT/IE2008/000123
Authority: WO
Inventors: Jan Olsak; Michal Karasek
Original assignee: Markport Limited
Priority date: 2007-12-19
Filing date: 2008-12-19
Publication date: 2009-06-25

Abstract

To reduce the number of SRIs sent to a HLR in GSM, a neural network-based prediction function (neural network, NN) is incorporated in an SMSC. The task of the NN is to predict the location of an MS (e.g. routing information such as Point Codes and Sub System Numbers or Global Titles for recent serving MSCs for a subscriber). The prediction is based on previous SRIs and takes into account the actual date (including distinction between working and non- working day) and time of the day. Thus, instead of an HLR, the NN is interrogated (via a LSRI as shown) to obtain the routing information (serving MSC Point Code or Global Title) needed to deliver a SM to the MS. If delivery of the SM based on the routing information obtained from the NN fails, the HLR is interrogated to obtain up-to-date routing information and the weight matrix of the NN is updated.

Description

"Prediction of Mobile Subscriber's Location"

INTRODUCTION

Field of the Invention

The invention relates to transmission of messages from a mobile network element to a mobile device.

Prior Art Discussion

To send a short message from an SMSC to a mobile station in a GSM network, the routing information must be known. The routing information in a GSM network is obtained by sending an SRI request across the core signalling backbone to the appropriate HLR (as shown in Fig. 1).

Naturally, transmission of the request increases the load on the network - two messages (SRI, SRI response, for example) being exchanged between a transmitting network element and the location register to obtain the routing information. Similarly, if the routing information is not known to the location register, the location register must interrogate the network, which can involve paging the mobile equipment, hi such a case, the network overhead is significantly higher.

It is known to provide an SMSC in which the routing information associated with the last location of a mobile subscriber is retained for a configurable period and used for routing.

JP2004180024 describes a location prediction system.

The aim of the invention is therefore to reduce network overhead necessary to acquire routing information. In particular, the invention is directed towards reducing the number of routing information requests and responses which need to be sent across the core signalling backbone of the mobile network, such as across the SS7 core mobile network infrastructure. Glossary of terms and their definitions

AIM Application Interface Module ANSI-41 American National Standards Institute standard for mobile phone networks

(previously referred to as IS-41 i.e. Interim Standard-41) CAPEX Capital Expenditure (Money spent to acquire or upgrade physical assets such as HW boxes.)

FSM Forward Short Message (GSM operation)

FSM-ACK Positive Response to Forward Short Message (GSM operation) FSM-NACK Negative Response to Forward Short Message (GSM operation) GSM Global System for Mobile Communications standard for mobile phone networks

GT Global Title

HLR Home Location Register

IMSI International Mobile Subscriber Identity LSRI Logical SRI

MDN Mobile Directory Number

MIN Mobile Identification Number

MO Mobile Originated (message)

MO-FSM Mobile Originated Forward Short Message (GSM operation) MO-FSM-ACK Positive Response to Mobile Originated Forward Short Message (GSM operation)

MS Mobile Station

MT Mobile Terminated (message)

NN Neural Network OPEX Operating Expenditure (On-going costs for running the product or system.)

PC Point Code

SRI Send Routing Information

SMDPP Short Message Delivery Point to Point (ANSI-41 operation)

SMDPP(ACK) Positive Response Short Message Delivery (ANSI-41 operation) SMDPP(NACK) Negative Response Short Message Delivery (ANSI-41 operation)

SMSC Short Message Service Centre

SMSREQ Short Message Service Request (ANSI-41 operation for requesting routing information) SMSREQ(NACK) Negative Response to Short Message Service Request (i.e. a response containing errors such as Unavailable)

SSN Sub System Number

Subscriber ID Subscriber identification - this general term is for example realized by MSISDN or IMSI in GSM networks, and by MIN or MDN in ANSI-41

(CDMA or TDMA) networks. In addition to this information, or alternatively, it may include an internally-assigned (i.e. within the prediction function) subscriber identifier.

MSC Mobile Switching Center Routing information The complete information required to route data to a destination including the type of network on which the subscriber may be available. The type of network is relevant for a subscriber who can roam on or be available on multiple network technologies for example GSM, TDMA, CDMA, IMS and/or multi-mode combinations of same.

SUMMARY OF THE INVENTION

According to the invention, there is provided method of transmitting a message to a destination mobile device via one or more mobile networks, the method comprising the steps of: a prediction means, in response to a query, predicting location of the destination mobile device, and a sending element performing a message delivery attempt to the mobile device on the basis that it is at the predicted location.

In one embodiment, the prediction means is incorporated in the sending element.

In one embodiment, the prediction means is in a separate element linked with the sending element by a link outside of signalling infrastructure of a mobile network, such as by a local area network.

hi one embodiment, the sending element is an SMSC or an MMSC. In one embodiment, the method comprises the further step of, if the message delivery attempt fails, interrogating a location register to determine actual location and performing a message delivery attempt to the device at the actual location.

In one embodiment, the method comprises the further steps of updating the prediction means with location data in real time, for learning by the prediction means.

hi one embodiment, said updating is according to actual location of the mobile device determined from a location register.

In one embodiment, the prediction means derives from an MO message location data for an originating mobile device, and subsequently provides said location data for a subsequent message delivery attempt to the originating mobile device.

In one embodiment, the prediction means performs conversion from one location data format such as PC and SSN to another format such as GT.

hi one embodiment, the prediction means accesses a translation table for said conversion.

In another embodiment, updates are also received asynchronously by the prediction means.

In one embodiment, said asynchronous updating is performed in a batch manner.

In one embodiment, the prediction means predicts with time and date as inputs.

In one embodiment, the prediction means provides a probability value for the prediction, and the probability value is used for determining whether the predicted location should be used for message delivery attempt.

In one embodiment, the prediction means comprises a neural network. Preferably, the neural network comprises a plurality of sub-networks, each associated with one or more subscribers. In one embodiment, the prediction means stores only recent location information involving a limited number of locations for at least some subscribers, so that the prediction means is dynamically optimized.

In one embodiment, the prediction means comprises a neural network and said optimization comprises dynamically modifying the number of neural network nodes.

In one embodiment, mobile device location data returned by the prediction means includes identification of the serving MSC and a subscriber identifier.

In one embodiment, the prediction means includes technology protocol specific routing information in a prediction; and wherein the routing information includes type of network on which a roaming subscriber may be available, and a subscriber identifier for said type of network.

In one embodiment, the network technologies include GSM, TDMA, CDMA, UMTS or IMS or a combination of these.

In one embodiment, the prediction means maps a subscriber identifier to a weight matrix, converts time of a routing query into a whole number of time periods such as hours, multiplying the individual digits of this number with the weight matrix and the results are added and finally an algorithmic function is applied to return a result to identify an MSC.

In one embodiment, the prediction means uses a translation table to convert the result into the MSC address.

In one embodiment, the prediction means also provides an internally-generated subscriber identifier; and said internally-generated subscriber identifier is used in a subsequent request made by the sending element.

In one embodiment, the prediction means updates a table upon receipt of new subscriber identifier in order to translate between different subscriber identifiers for the same subscriber, and to subsequently provide an appropriate subscriber identifier when queried. In another embodiment, the method comprises the farther steps of notifying the prediction means of failure of delivery, the prediction means updating itself, and a second routing query being made to the prediction means.

In one embodiment, the prediction means is instructed by the sending element to exclude a set of one or more MSCs from its prediction.

In another aspect, the invention provides a mobile network system for sending messages, the system comprising a network element and a prediction means, and the network element comprises means for requesting from the prediction means, without use of a network signalling infrastructure, predicted location information for a message delivery attempt.

In one embodiment, the element and the prediction means are co-hosted.

hi one embodiment, the prediction means is in a separate element linked with the sending element by a link outside of signalling infrastructure of a mobile network, such as by a local area network.

In one embodiment, the network element comprises means for automatically updating the prediction means according to learned subscriber location information.

In one embodiment, said updating is according to actual location of the mobile device determined from a location register.

In one embodiment, the prediction means comprises a neural network. In one embodiment, the neural network comprises a plurality of sub-networks, each associated with one or more subscribers.

In one embodiment, the element is a message service centre such as an SMSC or an MMSC.

In a further aspect, the invention provides a computer program product comprising software code for performing operations of any method defined above when executing on a digital processor.

DETAILED DESCRIPTION OF THE INVENTION

Brief Description of the Drawings

The invention will be more clearly understood from the following description of some embodiments thereof, given by way of example only with reference to the accompanying drawings in which:-

Fig. 1 is a prior art message sequence diagram referred to above;

Figs. 2 to 4 are message sequence diagrams illustrating message transmission methods according to the invention; Fig. 5 is a diagram illustrating a neural network prediction element, and Fig. 6 illustrates an example of an algorithm used by the neural network element;

Figs. 7 to 12 are further message sequence diagrams illustrating various embodiments of the prediction function.

Description of the Embodiments

The invention reduces network overhead required to determine the routing information for equipment in telecommunications networks. The reduction of routing information updates is achieved by the use of a prediction means (referred to henceforth as a "prediction function" or in some embodiments as a "neural network" or "NN") that predicts the location of the mobile equipment. Prediction of the routing information is based on the previous behaviour of the telecommunication equipment. The prediction function may have a neural network at its core in one embodiment, but may alternatively have a different artificial intelligence technology such as intelligent rule-based learning technology.

The following description is based on the GSM and ANSI-41 types of telecommunication network. Other implementation possibilities serving in other types of telecommunications networks and their combinations are possible as will be appreciated by those skilled in the art.

Referring to Fig. 2, to reduce the number of SRIs sent to a HLR in GSM for mobile terminated messages, a neural network-based prediction function (NN) is incorporated in an SMSC. The task of the NN is to predict the location of an MS (e.g. routing information such as Point Codes and Sub System Numbers or Global Titles for recent serving MSCs for a subscriber). The prediction is based on previous SRIs and takes into account the actual date (including distinction between working and non- working day) and time of the day. Thus, instead of an HLR, the NN is interrogated (via a LSRI as shown) to obtain the routing information (serving MSC Point Code and Sub System Number if needed or Global Title) needed to deliver a SM to the MS. If delivery of the SM based on the routing information obtained from the NN fails (e.g. the response includes an error code such as Absent Subscriber SM in GSM or Destination no longer at this address in ANSI-41), the HLR is interrogated to obtain up-to-date routing information and the weight matrix of the NN is updated, as shown in Fig. 3. The updating of the prediction function for automatic learning is particularly advantageous, as it ensures that the percentage of correct responses is kept high. Also, the updating can be based on information derived from an MO message. A simple example, is where user A sends a message to user B, and B responds to A. The prediction function would have the up-to-date information on location of user A for the response message. Thus it is advantageous for the NN to learn location information (e.g. serving MSC) for the subscriber originating the message for subsequent use for predictive routing when messages are being routed to that subscriber. Of course such an approach is only relevant for mobile originating subscribers and does not serve any purpose for fixed location originators such as a fixed location application originating messages.

Referring to Fig. 4 the SMSC updates the NN with the location information of the originating subscriber such as the PC and SSN or GT (as available) of the originating subscriber's MSC, when it receives a mobile originated SM. Such behaviour can be configurable for example to be always done, intermittently done or indeed not carried out.

The MO message arriving at the SMSC may not always contain both PC and SSN, and GT of the serving MSC. Therefore, the AIM or the NN must be able to perform conversion from PC and SSN to GT or vice versa. If the NN element is to implement this feature, it can adopt various approaches, the simplest of which would be a static translation table. The table includes information indicating to which network each MSC is associated.

Referring to Fig. 5, the neural network consists of many sub-neural networks, each dedicated to a subscriber ,and each sub-neural network having its own weight matrix. Alternatively, there may be multiple sub-neural networks and/or weight matrices per subscriber. Also, there may in another embodiment not be a weight matrix per sub-neural network, and instead a smaller number of super matrices the entries of which relate to subscribers. The routing information produced by the neural network depends on date and time. The probability of successful prediction may also be provided and can be used to determine whether a HLR request is required or whether a delivery attempt based on the routing information obtained from the neural network should be made.

A neural network is represented by an oriented graph the edges of which are ascribed their values. The graph may be implemented as a table in which each entry represents a single node of the neural network. Each node is associated with a function which computes an output value based on input values. The output value then becomes the input value for the function of the next node. Edges (connections between nodes) are implemented by another table. Each connection is assigned a value determining the level of significance of the input/output values of this connection.

We assume that the graph structure is fixed, yet it is possible to change the number of connections and some connections will represent dead ends. At the beginning, all nodes are interconnected, yet only those connections which are active (not dead ends) have non-zero values.

The dynamic part of neural networks (i.e. the part that is changing most often) is the so-called weight matrix - a set of values of individual connections organized in a matrix. In this embodiment, for each subscriber there is a dedicated weight matrix.

Fig. 6 illustrates a very simple algorithm used for predicting location by the NN element. In this example, it is known to the NN from previous message traffic, that a particular subscriber with ID (e.g. 11111111) can be reached via MSC with ID 1 every day between 10 am and 10 pm and via MSC with ID 0 at all other times (i.e. between 10pm and 10am). The above subscriber identifier is of course for illustrative purposes only.

The subscriber ID is mapped to a matrix (a very simple one in this example). Advantageously there can be multiple identifiers stored by the NN for a particular subscriber for a particular technology or multiple technologies, for example MSISDN and IMSI in GSM, or in the case of ANSI-41 MIN and/or MDN, and IMSI thus allowing support for multiple technologies. This enables great flexibility in for example scenarios where there are multiple technologies, (Refer to Fig. 11 and the associated description).

The time of the LSRI query is converted into a whole number of time periods, hours in this embodiment, the individual digits (1 and 0 for hour 10) of this number being matrix multiplied with the NN weight matrix pertaining to this particular subscriber, and the results are added and finally modulo 2 is applied. Thus, either 0 or 1 is returned. This number identifies the MSC (a translation table may convert this ID into a MSC address). If the algorithm yields an incorrect result, an update function must modify the matrix to ensure that subsequent results are provided with higher probability.

If the result produced by the neural network is incorrect, an update function is called. This update function receives the following arguments: the same input originally presented to the neural network, and the correct result. Based on this input, the update function performs changes in weight matrixes (or even in neural network structure) to ensure that when the same input is provided for the neural network again, the result is correct.

The neural network structure is the same for all subscribers but each subscriber has its own weight matrix. The weight matrix specifies the weight of each edge in the neural network structure represented as an oriented graph.

For learning, the neural network needs the following inputs: - subscriber identity,

- additional subscriber data (this is not necessary but could be used for improving the provided prediction)

- routing information (including type of network on which the subscriber may be available which is relevant for a subscriber who can roam on or be available on multiple network technologies, and/or multi-mode combinations of same),

- actual time and date (including working / non-working day distinction), and

- result(s) of the previous delivery attempt(s).

For retrieval of routing information, the neural network obtains: - subscriber identity,

- actual time and date (including working / non-working day distinction).

The neural network will retrieve routing information and provide the following output:

- routing information (including type of network on which the subscriber may be available which is relevant for a subscriber who can roam on or be available on multiple network technologies, for example GSM, TDMA, CDMA, IMS and/or multi-mode combinations of same),

- the probability that the subscriber is at the predicted location (depends on configuration - may not be provided in all implementations, Subscriber ID,

Error Code (conditional)

The following table summarizes the aforementioned input and output data.

The input is processed by the neural network according to weight matrixes assigned to the particular subscriber ID, date, and time. The output depends mainly on the weight matrix dedicated to each subscriber ID, while the structure (algorithm) of the neural network is fixed and common to all subscriber IDs. The weight matrixes reflect the fact for example that younger subscribers are more likely to move than older ones.

The data exchanged between the AIM and the NN element in LSRI and LSRI response can be illustrated by the following protocol pseudo-definitions: 1. LSRI

+ + ₊ +

I Invoke ID | Sub ID | Rejected Node Number | +_{+ + +}

• Invoke ID is used for identification of the corresponding response. This is a mandatory field.

• Subscriber ID (as defined in the glossary section) is a mandatory field.

• Rejected Node Number - this field is an array which may contain Network Node Numbers (e.g. MSC addresses) which the AIM does not want to obtain as a result since it suspects that these would be incorrect. This field is optional.

2. LSRI response

• Invoke ID is used for identification of the corresponding request. This is a mandatory field.

• Subscriber ID - This is typically a different identifier from what is provided in the LSRI request. Thus in GSM, the MSISDN is typically the Subscriber ID provided in the request with the IMSI being the Subscriber ID provided in the LSRI response. Similarly in ANSI-41, MIN, IMSI or MDN is provided in the request, while MIN or EvISI is typically returned in the response. This topic is described further in Fig. 11 and the accompanying text. In addition to this information, or alternatively, the Subscriber ID may include an internally-assigned subscriber identifier.

• Network - Identification of network type (e.g. GSM or ANSI-41). This field is optional (not needed in single network setups). The network type indicates the network technology associated with the information provided in the Subscriber ID of the LSRI response. Advantageously, this can be used by the sending element, for example an SMSC in determining the encoding to be used for the target network technology.

• Node Number - Network Node Number may be reported as Point Code and Subsystem Number, Point Code, Global Title, or any combination of the preceding. This field is conditional (i.e. may not be filled out if the Error Code field is populated).

• Error Code provides means for reporting error conditions instead of routing information. This field is conditional. Referred to as "Error" in the response. An Error Code may have any type of characters, such as numeric or alphabetical.

• Probability is an optional field, depends on configuration. The probability of the subscriber being at the predicted location can be computed in a number of ways. The simplest implementation would be represented by the ratio between the number of previous correct predictions for a particular subscriber ID and the total number of LSRI(s) for this subscriber ID.

The weight matrices are created during the learning phase (e.g. capturing the relevant information from responses to SRI requests sent to the HLR, or capturing information in incoming MO-FSMs). Subsequently, they are updated (modified) based on the result of each delivery attempt. Advantageously, both the learning phase and the updating phase do not pose any strain on the network as they do not generate any additional network traffic.

The prediction element (neural network-based in this embodiment) is highly configurable to enable flexible optimization and adjustment to specific conditions. The typical aspects of the prediction element functionality which could be configured are the following:

• Additional subscriber data processing o The prediction element may be configured to receive and process additional subscriber data (such as age, gender, occupation, etc.) Such data could but would not have to be used for further improving the prediction results. • Probability reporting o The prediction element may be configured to report the probability of the predicted location. o If the probability is reported the receiving party (e.g. SMSC) can decide how to use the predicted location (e.g. if the probability is too low it can perform a standard SRI request).

• Low probability threshold o If the probability is not reported, the prediction element may have a configurable probability threshold level under which it responds with an error code to signify that it cannot provide a prediction with sufficient probability level. • Network interface o The prediction element may be configured to function as a part of another network element (e.g. a plug-in module for SMSC) or it can function independently. In the independent setup, it can either function as a prediction element only or it can also serve as both a prediction and message transfer element.

Note that the above-mentioned configurable options do not represent a complete list. They serve as an example of the most important configurable options only.

The number of NN nodes can be optimized so that instead of storing all possible locations (which is the default approach), only a configurable limit of recent locations associated with the recent routing information related to SM delivery for a subscriber (e.g. routing information such as Point Codes and Sub System Numbers or Global Titles for recent serving MSCs for a subscriber) would be stored, hi practical terms, this would limit the amount of routing information stored per subscriber to a more manageable number, which would be useful especially in large-scale networks.

The prediction function used in Figs. 2 and 3 is incorporated into an SMSC. However, it may alternatively be separate from the element which interrogates it. For example, referring to Figs. 7 and 8, the neural network element may serve as a front end communication point substituting the HLR as well as a message transfer element from the AIM's point of view. The AIM would then always interrogate only the NN and it would in turn (if delivery of the SM based on the routing information obtained from the NN fails, or possibly under other circumstances, which might be based on location prediction probability) interrogate the HLR to obtain up-to-date routing information and update the NN weight matrix. Thus, advantageously in this embodiment the NN would represent a standalone learning, aggregation and interrogation service network element. This embodiment is depicted in Figs. 7 and 8.

In this embodiment, the prediction element (NN element) can actually be seen as a proxy for all the traffic going out of the AIM. The AIM can be a standard GSM network interface which may use a regular SRI request or a special local SRI (IP message, function call, or many other possibilities). If it is a standard SRI, in this embodiment when acting also as a proxy message transfer element, the NN always reports its own address as the MSC address and thus the AIM sends the FSM MAP operation towards the NN which takes care of all the delivery operations. It selects the MSC address (either predicts it or queries the HLR) and then forwards the FSM to the appropriate MSC. Subsequently, it relays the response from the addressed MSC back to the AIM. If the AIM is not a standard GSM AIM but a specifically modified AIM, the LSRI query may be omitted completely and each FSM operation can be forwarded directly to the NN element.

Regardless of the method via which the NN element is connected to the AIM (SS7, SS7oIP, TCP, library function call etc.), NN always brings the advantage of reducing the average load on the HLR and reducing the total amount of messages exchanged with the HLR.

Fig. 8 also depicts message flow in a situation where the NN functions as a proxy, however this diagram shows a situation when the predicted mobile station location is incorrect. In this case, the NN receives a negative acknowledgement in response to its FSM operation (the error code identifies that the recipient cannot be reached via this MSC). This prompts the NN to query HLR and then to use the obtained address for routing the FSM operation as well as for updating its own neural network data. In this particular situation, the amount of traffic and load on HLR is not decreased by the NN element. Yet, it should be noted that after a certain "learning time" this would be a rather rare scenario.

Since the location query and delivery operations involved in the whole message flow in Fig. 8 last a certain amount of time, it would be advisable to reckon with this possible delay when configuring the AIM application (especially network timeout values).

Fig. 9 depicts another embodiment, hi this embodiment, the NN element exists as a stand-alone element and provides the prediction function only (unlike in the embodiments depicted in Figs 7 and 8, in this embodiment the NN is not acting as a proxy message transfer element). In this particular situation, the NN element provides routing information including network identification, which is crucial when multiple networks are available. The AIM sends a local SRI query and the NN responds with routing data which identifies also which network the destination MSC finds itself in. The AIM then sends a SMDPP operation to the identified network and MSC (in this particular example it is SMDPP in an ANSI-41 network). It is expected that the AIM can communicate with different networks and can deal with subscriber identifiers from different networks. Fig. 9 shows GSM and ANSI-41 networks, however, the NN element could be used with any other mobile network types.

Fig. 10 expands on the multiple network embodiment shown in Fig. 9. In this particular situation, the NN element predicts incorrect routing information. The SMDPP operation carried out by the AIM fails. This arises where the MSC cannot deliver the message because the mobile device is not available, the MSC returning errors such as Absent Subscriber SM in GSM or Destination no longer at this address in ANSI-41. In this type of situation, the AIM may be configured to perform another LSRI. Fig. 10 illustrates a case when a second LSRI is performed. In this LSRI, the NN receives update information that its previous prediction was incorrect and more specifically the MSC address(es) it should not provide this time. Based on this data, the NN updates its internal structures and provides another (more precise) prediction. The resultant routing information is used by the AIM for the second delivery attempt.

Fig. 11 shows a situation very similar to that illustrated by Fig. 10. Again, incorrect location data is provided by the first LSRI response. Upon being informed that the prediction was incorrect (by the second LSRI), the NN element provides a second prediction. This time (unlike in Fig. 10), the prediction points to a MSC address from a different network. Thus, in Fig. 11 where there is a multi-network configuration and more than one LSRI query needs to be performed for a successful message delivery, the prediction element internally assigned Subscriber ID (together with routing information) returned in the first LSRI response can be used for a subsequent LSRI query. It is not essential that an internally-generated identifier be used, as the sending element may use any subscriber identifier which is available to it.

To be able to provide routing information for multiple networks (as shown in Fig. 11), the NN element or the connected AEVI must be able to perform conversions between different types of subscriber identifiers used in different networks. The NN element can implement this functionality in a number of ways, the simplest being a table which would contain the following columns: • ID - primary key, arbitrary unique value assigned by the NN element itself

• MSISDN o IMSI

• MIN

• MDN

The NN element updates this table any time it receives new data in order to be able to provide correct translations between different identifications. Thus the NN is capable of providing a relevant subscriber ID in its LSRI response. Similarly as in the previously mentioned figures, Fig. 12 shows a multi-network embodiment of the NN element. This diagram illustrates a situation where the NN provides not only a predicted location but also the probability of this information, hi this particular situation, the provided probability is low (lower than the probability threshold defined in the AIM configuration) and therefore, the AIM decides to query HLR for up-to-date information. It receives this information, updates the NN data and delivers the message.

Alternatively, the NN element may not report the probability value and only compare it with its own probability threshold configuration option. If the computed probability were lower than the threshold, the NN could return an error instead of standard response. The AEVI would then have to perform HLR query (or queries) and the delivery operation.

It should be noted that the NN element might be employed in multi-network setups with more than two networks. In such setups, the NN may return not only a prediction with network and routing data (within that network) but also a list of networks ordered according to probability. This again may lead to less HLR queries since the AIM may use the provided data to query the HLR' s with higher probabilities first.

An advantage of the above embodiments lies not in the reduction of the total number of messages exchanged among network elements, rather that (Logical) SRI messages are exchanged between AIM(s) and the NN element across a network (such as a local network with attendant responsiveness) based on non-SS7 based infrastructure. An example is TCP/IP. Thus, the NN does not replace the HLR in the general sense but serves as a location register access point for those network elements (AIMs), resulting in them not requiring to access the HLR via SS7 infrastructure, with attendant cost reduction benefit.

Moreover, even in all-IP networks where SIGTRAN solutions are used instead of SS7 networks, the invention still brings an advantage. In such networks, the response provided by the NN element may not only be quicker than that of HLR but also the invention leads to decreased load on the HLR system, which is beneficial for all other elements in the network which require information from the HLR.

It will be appreciated that the invention achieves much reduced control message communication in a network such as a core SS7 mobile network. For example, it is envisaged that the invention will be able to correctly predict destination location for approximately 50% of messages. In the prior art scenario it is required to exchange four messages across the SS7 infrastructure (for example in GSM, SRI, SRI resp, FSM, FSM ack ) to send an SMS. Using the invention just two messages will be exchanged across the SS7 infrastructure (FSM, FSM ack) will be exchanged. This saves 25% of traffic on the SMSC SS7 interface and thus significantly reduces the signalling traffic on the SS7 network and HLR load.

The cost saving on the mobile network operator side will be achieved by the reduction of the SS7 signalling traffic both on OPEX and CAPEX side. For SS7 networks the CAPEX cost per message is high. Moreover based on the type of licence regular licence payments are usually required based on the peak number of messages that the infrastructure is able to transfer per second. The same licence model is valid for the HLR capacity and SMSC interface capacity.

In the case of an onboard prediction solution the costs are limited just to the increased processing power and memory of the SMSC. hi the case of the standalone solution CAPEX costs of the TCP/IP network infrastructure are significantly lower than across the SS7 infrastructure, OPEX is limited just to network maintenance. In many cases existing infrastructure may be used.

The invention is not limited to the embodiments described but may be varied in construction and detail. For example, as stated above the invention is applicable to multiple technologies (for example GSM, TDMA, CDMA, IMS, UMTS and/or multi-mode combinations of same) in which case the specific appropriate technology's subscriber information (such as MSISDN, IMSI for GSM, MIN and/or MDN or IMSI for ANSI) is retained and used by the prediction function. It can therefore return the appropriate routing information for the technology for the location at which the subscriber is predicted to be. In the above embodiments the network is a GSM network or an ANSI-41 network in which there is SMSC - HLR interaction. However, the prediction function can be deployed to provide routing information to deliver data to another network element, including end delivery to a mobile station.

Also, while the invention has been described in the case where an SMSC is sending a message, it also applies to other sending network elements, such as MMSCs. The latter employing for example direct SMS delivery for MMS notifications or alternatively using SMS bearer by interacting with an SMSC for delivery of MMS notifications. The invention is generally applicable to services using bearer SMS for example, WAP services employing WAP gateways or PPG sending WAP Push over SMS bearer, or mobile e-mail systems that send notifications or other data over SMS bearer.

Claims

1. A method of transmitting a message to a destination mobile device via one or more mobile networks, the method comprising the steps of: a prediction means, in response to a query, predicting location of the destination mobile device, and a sending element performing a message delivery attempt to the mobile device on the basis that it is at the predicted location.

2. A method as claimed in claim 1, wherein the prediction means is incorporated in the sending element.

3. A method as claimed in claim 1, wherein the prediction means is in a separate element linked with the sending element by a link outside of signalling infrastructure of a mobile network, such as by a local area network.

4. A method as claimed in any preceding claim, wherein the sending element is an SMSC or an MMSC.

5. A method as claimed in any preceding claim, comprising the further step of, if the message delivery attempt fails, interrogating a location register to determine actual location and performing a message delivery attempt to the device at the actual location.

6. A method as claimed in any preceding claim, wherein the method comprises the further steps of updating the prediction means with location data in real time, for learning by the prediction means.

7. A method as claimed in either of claims 5 or 6, wherein said updating is according to actual location of the mobile device determined from a location register.

8. A method as claimed in claim 6, wherein the prediction means derives from an MO message location data for an originating mobile device, and subsequently provides said location data for a subsequent message delivery attempt to the originating mobile device.

9. A method as claimed in claim 8, wherein the prediction means performs conversion from one location data format such as PC and SSN to another format such as GT.

10. A method as claimed in claim 9, wherein the prediction means accesses a translation table for said conversion.

11. A method as claimed in any of claims 6 to 10, wherein updates are also received asynchronously by the prediction means.

12. A method as claimed in claim 11, wherein said asynchronous updating is performed in a batch manner.

13. A method as claimed in any preceding claim, wherein the prediction means predicts with time and date as inputs.

14. A method as claimed in any preceding claim, wherein the prediction means provides a probability value for the prediction, and the probability value is used for determining whether the predicted location should be used for message delivery attempt.

15. A method as claimed in any preceding claim, wherein the prediction means comprises a neural network.

16. A method as claimed in claim 15, wherein the neural network comprises a plurality of sub-networks, each associated with one or more subscribers.

17. A method as claimed in any preceding claim, wherein the prediction means stores only recent location information involving a limited number of locations for at least some subscribers, so that the prediction means is dynamically optimized.

18. A method as claimed in claim 17, wherein the prediction means comprises a neural network and said optimization comprises dynamically modifying the number of neural network nodes.

19. A method as claimed in any preceding claim, wherein mobile device location data returned by the prediction means includes identification of the serving MSC and a subscriber identifier.

20. A method as claimed in any preceding claim, wherein the prediction means includes technology protocol specific routing information in a prediction; and wherein the routing information includes type of network on which a roaming subscriber may be available, and a subscriber identifier for said type of network.

21. A method as claimed in claim 20, wherein the network technologies include GSM, TDMA, CDMA, UMTS or IMS or a combination of these.

22. A method as claimed in any of claims 19 to 21, wherein the prediction means maps a subscriber identifier to a weight matrix, converts time of a routing query into a whole number of time periods such as hours, multiplying the individual digits of this number with the weight matrix and the results are added and finally an algorithmic function is applied to return a result to identify an MSC.

23. A method as claimed in claim 22, wherein the prediction means uses a translation table to convert the result into the MSC address.

24. A method as claimed in any of claims 19 to 23, wherein the prediction means also provides an internally-generated subscriber identifier; and said internally-generated subscriber identifier is used in a subsequent request made by the sending element.

25. A method as claimed in claim 24, wherein the prediction means updates a table upon receipt of new subscriber identifier in order to translate between different subscriber identifiers for the same subscriber, and to subsequently provide an appropriate subscriber identifier when queried.

26. A method as claimed in any of claims 5 to 25, comprising the further steps of notifying the prediction means of failure of delivery, the prediction means updating itself, and a second routing query being made to the prediction means.

27. A method as claimed in claim 26, wherein the prediction means is instructed by the sending element to exclude a set of one or more MSCs from its prediction.

28. A mobile network system for sending messages, the system comprising a network element and a prediction means, and the network element comprises means for requesting from the prediction means, without use of a network signalling infrastructure, predicted location information for a message delivery attempt.

29. A mobile network system as claimed in claim 28, wherein the element and the prediction means are co-hosted.

30. A mobile network system as claimed in claim 28, wherein the prediction means is in a separate element linked with the sending element by a link outside of signalling infrastructure of a mobile network, such as by a local area network.

31. A mobile network system as claimed in any of claims 28 to 30, wherein the network element comprises means for automatically updating the prediction means according to learned subscriber location information.

32. A system as claimed in claim 31 , wherein said updating is according to actual location of the mobile device determined from a location register.

33. A system as claimed in either of claims 31 or 32, wherein the prediction means derives from an MO message location data for an originating mobile device, and subsequently provides said location data for a subsequent message delivery attempt to the originating mobile device.

34. A system as claimed in any of claims 28 to 33, wherein the prediction means provides a probability value for the prediction, and the probability value is used for determining whether the predicted location should be used for message delivery attempt.

35. A system as claimed in any of claims 28 to 34, wherein the prediction means comprises a neural network.

36. A system as claimed in claim 35, wherein the neural network comprises a plurality of sub-networks, each associated with one or more subscribers.

37. A system as claimed in any of claims 28 to 36, wherein mobile device location data returned by the prediction means includes identification of the serving MSC and a subscriber identifier.

38. A system as claimed in any of claims 28 to 37, wherein the prediction means includes technology protocol specific routing information in a prediction; and wherein the routing information includes type of network on which a roaming subscriber may be available, and a subscriber identifier for said type of network.

39. A system as claimed in any of claims 28 to 38, wherein the prediction means maps a subscriber identifier to a weight matrix, converts time of a routing query into a whole number of time periods such as hours, multiplying the individual digits of this number with the weight matrix and the results are added and finally an algorithmic function is applied to return a result to identify an MSC.

40. A mobile network system as claimed in any of claims 28 to 39, wherein the element is a message service centre such as an SMSC or an MMSC.

41. A computer program product comprising software code for performing operations of a method as claimed in any of claims 1 to 27 when executing on a digital processor.