hash
stringlengths 32
32
| doc_id
stringlengths 5
12
| section
stringlengths 4
1.47k
| content
stringlengths 0
6.67M
|
|---|---|---|---|
14ce710049a3c611ebb26f3972649957
|
21.978
|
7.2.3 Functional Architecture
|
The proposed functional architecture is discussed in [6] and is shown in Figure 6. The main idea behind this functional architecture is to be able to reuse existing (circuit switched) CAMEL services for VoIP.
Figure 6: Proposed functional architecture
|
14ce710049a3c611ebb26f3972649957
|
21.978
|
7.3 Message Flows
| |
14ce710049a3c611ebb26f3972649957
|
21.978
|
7.3.1 Registration
|
It is not within the scope of this feasibility study to investigate or select the registration and location update process that must exist within a multimedia service network. However this section outlines a possible registration process that allows CAMEL Subscriber Information (CSI) to be stored in the Gatekeeper/IPSSF. Two distinct levels of registration exist; one at the PDP transport capability (Attach/Detach and PDP context establishment) and another level at the VoIP/multimedia level. Within H.323, this latter registration takes places with a Gatekeeper and roaming users register with a gatekeeper in the visited network. Gatekeeper discovery can be achieved via H.323 gatekeeper discovery procedure or via the resolution of ras://gk via DNS.
Figure 7 outlines a proposed registration process. MS is a mobile station in a visited network.
{1} The MS attaches to the network via existing GPRS procedures. This involves an attach request to the SGSN and a location update sequence from the SGSN to the HLR.
{2} The MS activates a PDP context to establish an IP session with the gatekeeper.
{3} The MS checks that it can register with that particular gatekeeper using a RAS GatekeeperRQ message. Assuming that the request is accepted, the gatekeeper confirms that registration can take place.
{4} Upon confirmation that the MS is allowed to register with the gatekeeper, a RAS Registration Request is sent.
{5} At this point, the gatekeeper sends a location update to the HLR in the user’s home network to register the alias address. The HLR responds with an InsertSubscriberData message that contains the CAMEL subscription Information (CSI). At this stage it is assumed that the Originating CSI (O-CSI) and the Terminating CSI (T-CSI) is sent to the gatekeeper. Supplementary Service CSI (SS-CSI) applicability is for further study. The HLR keeps a record of the address of the gatekeeper that the MS is registered with.
{6} Once the registration process is complete, the PDP session with the gatekeeper may be terminated.
Figure 7: MS registration with gatekeeper
|
14ce710049a3c611ebb26f3972649957
|
21.978
|
7.3.2 MO call requiring CAMEL interaction
|
The call flows for a mobile originated (MO) calls are shown in Figure 8 and are further explained below:
{1} The MS wishes to place a VoIP call. A PDP context must be established to allow an IP session to be set up.
{2} The MS checks that it is allowed to place a call via RAS Admission Request message.
{3} If the MS is allowed to place the call (gatekeeper routed) an H.225 Setup message is sent to the gatekeeper.
{4} When the Setup message is received at the gatekeeper, if the O-CSI for the calling user is present, it would be possible to analyse its contents and if necessary invoke the IPSSF. The IPSSF is now assumed capable of implementing the CAMEL O-BCSM. Later sections in this report cover the mapping of the O-BCSM points in call and H.323 call state model. The IPSSF starts a dialogue with the gsmSCF. The gsmSCF address and service key to invoke are obtained from the triggering criteria in the O-CSI.
{5} The gsmSCF sends the instructions to the IPSSF according to the service logic invoked
{6} The gatekeeper forwards the call according to the instructions received from the gsmSCF. H.225 Setup message is sent to the destination address.
{7} Dialogue between the gsmSCF and the IPSSF may continue according to the service logic. The remainder of the information flows will vary according to the service logic and are not shown.
Figure 8: MO call with CAMEL interaction
|
14ce710049a3c611ebb26f3972649957
|
21.978
|
7.3.3 MT call requiring CAMEL interaction
|
The call flows for mobile terminated (MT) calls are shown in Figure 9 and are further explained below:
{1} An incoming H225 Setup message is received by the mediation gatekeeper in the home network of the called subscriber. (The incoming call could be from an other H.323 gatekeeper or H.323 signalling gateway representing a call originating from external networks such as the PSTN).
{2} The mediation gatekeeper queries the HLR to discover the IP address of the gatekeeper that the called user is registered with. SendRoutingInformation/Ack messages can be used for this purpose.
{3} The mediation gatekeeper forwards the call to the required gatekeeper by sending a Setup message.
{4} The gatekeeper receives the H225 Setup message and checks the T-CSI of the called user to determine whether any CAMEL services should be invoked.
{5} If analysis of the T-CSI shows that the triggering criteria are met, the IPSSF is invoked to create a T-BCSM and the IPSSF initiates a dialogue with the gsmSCF in the home network of the called user. Instructions are received from the gsmSCF on how the call is to proceed.
{6} The gatekeeper will route the call according to the instructions received from the gsmSCF and will send an H225 Setup message to the destination party
{7} Dialogue between the gsmSCF and the IPSSF may continue according to the service logic. The remainder of the information flows will vary according tot he service logic and are not shown.
Figure 9: MT Call with CAMEL interaction
|
14ce710049a3c611ebb26f3972649957
|
21.978
|
7.4 State Models
| |
14ce710049a3c611ebb26f3972649957
|
21.978
|
7.4.1 H.323 and CAMEL O-BCSM
|
Figures 10, 11 and 12 show the relationship between the H.323 protocol state model and the points in call (PICs) defined in the CAMEL phase 3 Originating Basic Call State Model (O-BCSM) for successful and unsuccessful call attempts. The O-BCSM represented is based on CAMEL phase 3 state models as described in reference [9]. For the sake of simplicity, the information flows between the GPRS network nodes (GGSN and SGSN), the gsmSCF and Gatekeeper/SSF are not shown in any of the Figures. The sequence of messages flows relating to the SGSN, GGSN and the gsmSCF are as covered in section 7.3.2 and Figure 8. An O-BCSM object can be created upon receipt of an H.225 Setup message and the analysis of the O-CSI.
The figures assume that the MS has previously registered with a gatekeeper in the network. The destination entity shown can be an other MS, gatekeeper, mediation gatekeeper or H.323 signalling gateway. As the call control protocol, H225 used in an H.323 environment is based on Q931, the mapping between the messages and the PICs are similar to existing GSM mechanisms.
|
14ce710049a3c611ebb26f3972649957
|
21.978
|
7.4.2 Successful call establishment
|
Figure 10 shows the PICs and relationship to H.323 messages for a successful call establishment of MO call with CAMEL interaction.
{1} The H.225 Setup message indicates that a dialled number is received from the MS and detects that an O-CSI exists for this subscriber.
{2} Analysis of the O-CSI takes place and gsmSCF is invoked.
{3} Instructions are received from the gsmSCF to route the call to the destination address.
{4} Upon receipt of the Connect message from the called party, a transition to the PIC O_Active takes place. O_Answer DP can be reported.
{5} Either called or calling party can release the call with a ReleaseComplete message, upon which a transition to the PIC O_Null&Authorise_Origination_Attempt_Collect_Info takes place and the DP O_Disconnect may be reported.
Figure 10: H.323 message flows and corresponding CAMEL O-BCSM for successful call establishment
|
14ce710049a3c611ebb26f3972649957
|
21.978
|
7.4.3 Unsuccessful call establishment due to gatekeeper admission reject
|
Figure 11 shows the PICs and relationship to H.323 messages for the situation where the destination is refused permission to accept the incoming call by the gatekeeper. An AdmissionReject message may be sent by the gatekeeper to the destination is registered with for reasons specified in RAS procedures. In order to keep the Figures relatively simple, the PICs O_Null&Authorise.. and Analyse_Info and corresponding message flows are not shown.
{1} Instructions received from the gsmSCF on routing of the call and an H.225 Setup message is sent to the destination.
{2} The destination is not given permission by the gatekeeper to accept the incoming call. A number of reasons could exist, such as insufficient bandwidth, permission expired etc. This information is carried in the rejection reason of the AdmissionReject and transported into the ReleaseComplete message back to the gatekeeper.
{3} ReleaseComplete message arrives at gatekeeper indicating that the call could not be completed. DPs Route_Select_Failure or O_Routing_and_Alerting_Failure may be reported according to the rejection reason provided in the RelaseComplete message.
Figure 11: H.323 messages flows and corresponding CAMEL O-BCSM for the case where the destination is refused permission to accept the call by the gatekeeper
|
14ce710049a3c611ebb26f3972649957
|
21.978
|
7.4.4 Unsuccessful call establishment due to refusal by destination party
|
Figure 12 shows the PICs and relationship to H.323 messages for the situation where the destination does not accept the incoming call request. Once again, in order to keep the Figures relatively simple, the PICs O_Null&Authorise.. and Analyse_Info and corresponding message flows are not shown.
{1} Instructions received from the gsmSCF on routing of call and an H225 Setup message is sent to the destination.
{2} The gatekeeper receives a ReleaseComplete message with a release reason. Potentially, depending on the release reason, the DPs O_No_Answer, O_Busy, O_Abandon, O_Routing_and_Alerting_Failure or Route_Select_Failure may be reported.
Figure 12: H.323 messages flows and corresponding CAMEL O-BCSM for the unsuccessful establishment (other than admission refusal)
|
14ce710049a3c611ebb26f3972649957
|
21.978
|
7.4.5 H.323 and CAMEL T-BCSM
|
Figures 13 and 14 show the relationship between the H.323 protocol state model and the points in call (PICs) defined in the CAMEL phase 3 Terminating Basic Call State Model (T-BCSM) for successful and unsuccessful MT call attempts. The T-BCSM represented is based on CAMEL phase 3 state models as described in reference [9]. Once again, for the sake of simplicity, the information flows between the GPRS network nodes (GGSN and SGSN), the gsmSCF and Gatekeeper/SSF are not shown in any of the Figures. The sequence of messages flows relating to the SGSN, GGSN and the gsmSCF are as covered in section 7.3.3 and Figure 9. A T-BCSM object can be created upon receipt of an H.225 Setup message and the analysis of the T-CSI.
|
14ce710049a3c611ebb26f3972649957
|
21.978
|
7.4.6 Successful MT call delivery
|
Figure 13 shows the PICS and relationship to the H.323 protocol state model for a successfully established MT call requiring CAMEL interaction.
{1} The H.225 Setup message arrives at the gatekeeper the destination MS is register with. If a T-CSI exists for the subscriber, a T-BCSM instance is created.
{2} Analysis of the T-CSI takes place and gsmSCF is invoked
{3} Instructions from the gsmSCF are received and upon receipt of the Connect message from the called MS a transition to T_Answer takes place. The DP T_Answer may be reported.
{4} Either calling or called party can release the call with a ReleaseComplete message, upon which a transition to PIC T_Null takes place with potentially the DP T_Disconnect being reported.
Figure 13: H.323 message flows and corresponding CAMEL T-BCSM for successful call establishment
|
14ce710049a3c611ebb26f3972649957
|
21.978
|
7.4.7 Unsuccessful MT call delivery
|
Figure 14 shows the PICs and relationship to H.323 message for the situation where the terminating MS does not accept the incoming call.
{1} The H.225 Setup message arrives at the gatekeeper the destination MS is register with. If a T-CSI exists for the subscriber, a T-BCSM instance is created.
{2} Analysis of the T-CSI takes place and gsmSCF is invoked
{3} The called party refuses to accept the call and sends a ReleaseComplete message to the gatekeeper. The ReleaseComplete message may contain release reason. Potentially, depending on the release reason, the DPs T_No_Answer, T_Busy, T_Abandon or T_call_handling_Failure may be reported.
Figure 14: H.323 message flows and corresponding CAMEL T-BCSM for unsuccessful call establishment
|
14ce710049a3c611ebb26f3972649957
|
21.978
|
7.5 CAMEL Integration
|
Integration of CAMEL functions with H.323 Gatekeeper functions may require enhancements to H.323 standards or CAMEL standards. Some initial work has been done to identify the standards that may need to be enhanced and to assess the extent of the changes required.
|
14ce710049a3c611ebb26f3972649957
|
21.978
|
7.5.1 Impact on H.323 Standards
|
Section 7.4 describes and illustrates how H.225 messages may be mapped to O_BCSM and T_BCSM Detection Points (DPs) at the Gatekeeper/SSF. When studying the Figures it can be seen that some CAMEL Phase 3 DPs (such as O_Active) have good correspondence with H.225 messages at the Gatekeeper, while other DPs require more detailed analysis. At this level of analysis it seems likely that changes to H.323 may not be needed.
As the H.323 standards are already well established it seems likely that changes would be unwelcome in and it seems more likely that, if necessary, some reduction in CAMEL functionality may be a more acceptable compromise.
|
14ce710049a3c611ebb26f3972649957
|
21.978
|
7.5.2 Impact on CAMEL and other UMTS Standards
|
It is likely the some functional modelling additions and some CAP message information element additions will be needed. It is also likely that some HLR/VLR data and MAP protocol additions may be needed. It is of course very difficult to assess and describe exactly how much effort and meeting time is required to effect these changes. In comparison to previous CAMEL work on Phase 1, Phase 2 and Phase 3, in the author's opinion the effort required is probably more than was necessary to complete GPRS inter-working in CAMEL Phase 3, but probably less than the work necessary to complete CAMEL Phase 1.
The major impact of these changes concerns documents 22.078, 23.018, 23.078, 29.078 and 29.002.
|
14ce710049a3c611ebb26f3972649957
|
21.978
|
7.6 Service Impacts
|
The combination of CAMEL and H.323 (as described in this report) may support some GSM supplementary services, such as unconditional call forwarding (subject to some re-engineering) and perhaps most operator specific services (depends on the extent of any reduced functionality necessary for integration). Every service needs to be studied in detail to determine exactly what can be supported. Interworking with H.323 based supplementary services is another aspect. Again every service needs to be studied in detail, however, it seems likely that a comparable level of inter-working may be possible to that achieved with a combination of an MSC and ISDN, i.e. low-level services like CLI probably are feasible but more complex services may not be feasible.
|
14ce710049a3c611ebb26f3972649957
|
21.978
|
7.7 Multimedia Evolution
|
The proposal in this report is focused on voice not multimedia. Control of multimedia services requires further study. However, the proposal does not in any way preclude CAMEL evolution to control multimedia services supported by the H.323 family of recommendations. It is likely that further enhancements to the protocols and functions listed in section 7.5.2 (e.g.. MAP, CAP, HLR, VLR, CSE) may be necessary, depending on the nature of the multimedia service control required.
|
14ce710049a3c611ebb26f3972649957
|
21.978
|
7.8 Advantages
|
The following advantages have been identified:
Maximises the re-use of existing functional entities, protocols and services. Such reuse decreases the development and ownership costs allowing existing familiar 2G services to be provided to 3G subscribers at an early stage.
Minimum changes to the CSE for the support of legacy services. There are several IN/CAMEL services already deployed such as PrePaid, VPN, Mobile Number Portability etc., which may be required in a voice over IP network.
This approach is in line with the work currently underway in ETSI SPAN 3 (Services and Protocols), in particular a work item addressing IN support for voice over IP in the H.323 architecture and associated protocols in association with the TIPHON project. The study will investigate how an H.323 gatekeeper can act as a virtual Service Switching Point (SSP). It is worth noting that ETSI plan to harmonise the fixed line IN protocol (ETSI Core INAP CS3.1) and mobile equivalent (CAP Phase 3) into a common protocol targeted for ETSI Core INAP CS4.
|
14ce710049a3c611ebb26f3972649957
|
21.978
|
7.9 Disadvantages
|
The following disadvantages have been identified:
Introduces new functional entity ‘IPSSF’, which provides the necessary mapping between the Gatekeeper and the CSE. However, this functional entity is based on the functions already provided by a VMSC/GMSC, where already standardised process such as the gsmSSF can be reused.
The interface between the SCS/CSCF and the IPSSF requires further study.
|
14ce710049a3c611ebb26f3972649957
|
21.978
|
8 SIP Solution
| |
14ce710049a3c611ebb26f3972649957
|
21.978
|
8.1 Description
|
The IETF Multiparty Multimedia Session Control (MMUSIC) working group specifies an IP telephony architecture (Figure 15). The architecture is seen as an alternative to ITU-T H.323 family of recommendations that is simpler to implement. It allows for the possibility of interworking with H.323 terminals. However, H.323 currently has the greatest industry backing.
The Session Description Protocol (SDP) can be likened to H.245 (channel open/close and terminal capability handling). Session descriptions have a list format containing information about the session (see RFC 2327). The Session Initiation Protocol (SIP), reference [4], can be likened to H.225 (Registration, Admission, Status (RAS), and Q.931 messages). Message headers are in plain text and look similar to Email headers.
SIP uses a client server model similar to the Hypertext Transfer Protocol (HTTP) and many others (Figure 16 (a)). It is used in conjunction with other protocols such as SDP, RTP and RSVP. SIP can establish connections via TCP or UDP (Figure 16 (b)).
Figure 15: IETF IP Telephony Architecture
Figure 16: (a) Client/Server Model using TCP, (b) Protocol Stack using UDP/TCP
SIP Messages: To initiate a session a client sends an INVITE message to a server. An INVITE message typically contains a session description in SDP sufficient to establish communication. SIP Request and Response messages are listed below.
SIP Request Messages
SIP Response Messages
INVITE
ACK
BYE
CANCEL
OPTIONS
REGISTER
1xx Informational
2xx Success
3xx Redirection
4xx Client Error
5xx Server Error
6xx Global Failure
INVITE - Invites client or server to establish a session.
ACK - Confirmation reception of a final response to an INVITE message.
BYE - The sender wishes to close the session.
CANCEL - Cancels pending requests.
OPTIONS - Asks for information about capabilities before establishing a session.
REGISTER - Informs a Location Server of the client's IP address.
SIP response messages use a 3 digit number, e.g. 1xx. The first digit defines the category. The next two digits allow up to 100 variations, e.g. 200 OK (successful invitation).
Session Establishment: Sessions may be established using direct point-to-point communication or by using a SIP Server for personal mobility. A SIP server may be a Redirection Server or a Proxy Server. To establish a session using a SIP server the originator sends an INVITE message to the server. The server communicates with a Location Server to retrieve the IP address of the terminating party. When a Redirection Server is employed (Figure 17) the IP address is passed to the originator. The originator sends a new INVITE message and a session is established with the terminating party. When a Proxy Server is employed (Figure 18) the address of the terminating party is not passed to the originator. A Session is established between the originator and terminating party via the Proxy Server. One or more intermediate Proxy Servers may take part in the session. It is envisaged that a Proxy Server may be a network entity where CAMEL service control could be applied. This possibility is investigated further in this report.
Figure 17: Session Establishment Using a Redirection Server
Figure 18: Session Establishment Using a Proxy Server
Interworking with H.323: RFC 2543 states (reference [4]) that SIP could be used to determine that a party can be reached via H.323, obtain the H.245 gateway and user address and then use H.225.0 to establish the call.
|
14ce710049a3c611ebb26f3972649957
|
21.978
|
8.2 Architecture
| |
14ce710049a3c611ebb26f3972649957
|
21.978
|
8.2.1 Introduction
|
This section of the report provides information flows that illustrate the possible interaction of CAMEL and SIP. In particular it provides a proposal for the triggering of CAMEL services as well as a mapping between the CAMEL call states and the call states of the Session Initiation Protocol (SIP).
An overall objective for the feasibility study is to demonstrate that CAMEL control of VoIP services in 3G networks can be readily specified and implemented by adapting standards and software used in 2G networks. This approach leads to services that function the same when a user roams between 2G and 3G networks, simplifies service evolution from 2G to 3G, and leads to more rapid implementation. This section of the report investigates the possibility of CAMEL service control based on the SIP proxy Server approach, as described in companion contributions (or sections that will form part of the technical report). This means that a locally configured proxy server is required for outgoing calls that require legacy service support based on existing CAMEL services.
The section of the report is organised as follows: Section 8.2.2 outlines the proposed functional architecture for the support of CAMEL/SIP interaction. Section 8.2.3 briefly describes the concepts for IN service triggering based on CAMEL Subscription Information. Section 8.3.1 describes a registration process, section 8.3.2 deals with the detail of triggering services for Mobile Originated Calls, section 8.3.3 deals with the details for triggering Mobile Terminated calls. Section 8.4 describes the mapping of the SIP protocol state to the CAMEL basic call state model for the Originating and the Terminating sides.
|
14ce710049a3c611ebb26f3972649957
|
21.978
|
8.2.2 Functional Architecture
|
The proposed functional architecture is derived from various companion contributions relating to the technical report. It is provided here for completeness. The concept of the ‘IPSSF’ is introduced which acts as an overlay between the IP telephony call control and the Intelligent Network layer provided by the CAMEL Service Environment (CSE) or the GSM Service Control Function (gsmSCF). This ‘IPSSF’ provides the necessary mapping between the SIP protocol state machine and the CAMEL Basic Call State Model (BCSM). Figure 19 outlines the proposed functional architecture at the network level.
Figure 19: Proposed functional Architecture to support CAMEL control of VoIP based on SIP call control
|
14ce710049a3c611ebb26f3972649957
|
21.978
|
8.2.3 Basic concept of the proposal
|
Subscribers may register in the visited SIP network allowing the subscriber to receive incoming calls. A subscriber may use the MSISDN as an additional identifier in the registration process. Upon registration with the server, the CAMEL subscription information for the subscriber is sent to the IPSSF by the HLR in the subscriber’s home network. As incoming calls made to the subscriber terminate at the server the subscriber is registered with, the Terminating CAMEL Subscription Information (T-CSI) may be examined and if necessary the gsmSCF may be invoked on a per incoming call basis. Similarly, calls made by a subscriber already registered with a proxy server allow the Originating CAMEL (O-CSI) subscription information to be examined and potentially allow the gsmSCF to be invoked. Callers not registered will not have any O-CSI information in the proxy server they are using to place the call. The proposal here is as follows: when the initial call request message (or the INVITE method) is received by the SIP proxy server, the IPSSF/VLR establishes a dialogue with the HLR of the home subscribers network to allow the CAMEL subscription information to sent. The O-CSI may then be examined and if necessary the gsmSCF may be invoked.
|
14ce710049a3c611ebb26f3972649957
|
21.978
|
8.2.4 Assumptions
|
a. All the call flows show that the SIP Proxy Server and the IPSSF have been co-located in order to avoid showing an information flows between the two entities. Standardisation of the messages for this interface is for further study.
b. The attach/detach and establishment of the Packet Data Protocol (PDP) session are based on existing GPRS procedures found in UMTS 24.080 and UMTS 29.060.
c. When a subscriber registers or places a call via a SIP Proxy server, the home network of the subscriber can be identified.
d. A subscriber may use the MSISDN as an additional identifier in the registration process.
e. Originating and terminating SIP Proxy servers must operate in a call-state aware mode.
f. As registration with a SIP Proxy server is not mandatory, it shall be possible to determine whether a registration exists for that particular subscriber when an incoming call is placed by a subscriber. This allows the CSI information to be fetched from the HLR is the subscriber is not registered. (Note: Absence of the O-CSI does not necessarily mean that the user is not registered, merely that the O-CSI may not exist for that subscriber).
g. The information flows make no consideration for interworking with other networks (e.g. PSTN via gateways as this is considered to be outside to scope of the technical report.
|
14ce710049a3c611ebb26f3972649957
|
21.978
|
8.3 Message Flows
| |
14ce710049a3c611ebb26f3972649957
|
21.978
|
8.3.1 Proposed Registration process
|
This section outlines a possible registration process based on the SIP REGISTER method, which allows CAMEL subscription information to be stored in the SIP Proxy Server/IPSSF. IETF RFC 2543 [4] defines the term Registrar for registration purposes and it is the SIP registrar that accepts the REGISTER method. In this section it has been assumed that the SIP Proxy Server and the SIP registrar are co-located. This registration process is in addition to the PDP transport layer registration (GPRS Attach/Detach and PDP context establishment as found in UMTS 24.080 and UMTS 29.060). With the SIP REGISTER method, it is assumed that registration with a location server takes place. As this is outside the scope of SIP to specify, the information flows for this procedure are not shown, but are assumed to take place.
Unlike H.323, registration with a server is not mandatory. Only users that wish to receive incoming calls need to register with a SIP Proxy server and a location server. Callers placing calls are not required to register. The information flows for the registration procedure are shown in Figure 21 and elaborated in the following text:
{1} The MS attaches to the network using existing GPRS procedures. This involves an attach request to the SGSN and a location update sequence between the SGSN and the HLR.
{2} The MS activates a PDP context to establish an IP session with the local proxy SIP server in the visited network. Mechanisms for the discovery of the SIP Proxy server are for further study.
{3} The MS sends a REGISTER method to the SIP Proxy server.
{4} The SIP Proxy server notifies the IPSSF/VLR of the registration attempt. The IPSSF in turn notifies the HLR. The HLR responds with an InsertSubscriberData message which contains the CAMEL Subscription Information (CSI) which includes the Originating CSI (O-CSI) and Terminating CSI (T-CSI). The applicability of the Supplementary Service CSI (SS-CSI) is for further study.
{5} The SIP Proxy server acknowledges that the registration process has been completed by a 200 OK response message.
{6} Once the registration process is complete, the PDP session with the SIP Proxy Server may be terminated.
Figure 21: Proposed registration procedure
|
14ce710049a3c611ebb26f3972649957
|
21.978
|
8.3.2 Mobile Originating Call with CAMEL interaction
|
This section deals with the mobile originating calls that require interaction with CAMEL.
The call flows are shown in Figure 22 and are further explained below:
{1} The MS wishes to place a VoIP call. A PDP context is established to allow an IP session to be established over the GPRS network.
{2} The User Agent Server in the MS initiates a SIP request by issuing an INVITE method to the SIP Proxy server.
{3} The VLR functionality in the IPSSF is checked to determine if the calling party has previously registered. If no registration found, then step {4} is followed. If the IPSSF determines that the calling user has a valid registration then step {5} is followed.
{4} The IPSSF establishes a dialogue with the HLR of the subscriber’s network. An UpdateLocation message is sent to the HLR. The HLR responds by sending an InsertSubscribersData message, which may contain the CAMEL Subscription Information, including O-CSI, T-CSI and the SS-CSI.
{5} The O-CSI data is analysed and if the necessary triggering criteria are met, the gsmSCF is invoked via an InitialDP message.
{6} The SIP Proxy server will route the call based on the instructions received by the service logic in the gsmSCF. The remainder of the information flows will vary according to the service logic and are not shown.
Figure 22: MO Call with CAMEL interaction
|
14ce710049a3c611ebb26f3972649957
|
21.978
|
8.3.3 Mobile Terminating Call with CAMEL interaction
|
This section deals with the CAMEL interaction for mobile terminated calls. A CAMEL service is triggered if the triggering criteria held in the called subscriber’s T-CSI matches the characteristics of the incoming call. The information flows are shown in Figure 23 and further explained below:
{1} The terminating SIP Proxy server receives an INVITE method.
{2} The T-CSI is analysed and the triggering criteria are check against the particulars of the incoming call. A terminal must register with a server to be able to accept incoming call and has been assumed that since this registration has taken place; the T-CSI is available at the server.
{3} If the necessary triggering criteria are met, the gsmSCF is invoked and a CAP dialogue established between the IPSSF and the gsmSCF.
{4} Instructions are received from the gsmSCF on how the call is to be routed.
{5} The SIP Proxy server will route the call based on the instructions received by the service logic in the gsmSCF. A network initiated PDP context is established in order to deliver the INVITE method to the User agent server. As the rest of the information flows will vary according to the service logic, the remained of the information flows are not shown.
Figure 23: MT Call with CAMEL interaction
|
14ce710049a3c611ebb26f3972649957
|
21.978
|
8.4 State Models
| |
14ce710049a3c611ebb26f3972649957
|
21.978
|
8.4.1 Mapping SIP message to CAMEL Basic Call State Models
|
This section deals with how the Originating Basic Call State Model (O-BCSM) Points in Call (PICs) and Detection Points (DPs) or ‘triggers’ are mapped to the appropriate SIP messages. Although a mapping is possible, there is not always the same analogy between the circuit switched environment that the CAMEL BCSM were designed for and the packet environment, and as a result a direct mapping is not always possible. The state models for the CAMEL O-BCSM and the T-BCSM are based on the emerging CAMEL Phase 3 draft specifications as identified in reference[11].
For simplicity, the information flows in the Figures of the following subsections do not show the flows between the GPRS network nodes (SGSN and GGSN) and between the gsmSCF and IPSSF.
|
14ce710049a3c611ebb26f3972649957
|
21.978
|
8.4.2 Mapping to Originating BCSM for MO calls
|
The mapping between the SIP methods and responses for O-BCSM relating to Mobile Originating Calls are shown in Figure 24. Only the successful case is described. Further work is necessary to describe all other scenarios. The information flows are further described below :
{1} INVITE method arrives at the proxy server, indicating that the MS has requested to set up a call. SIP Proxy server determines if O-CSI exists for this user.
{2} Analysis of the O-CSI takes places and if necessary triggering criteria are met, gsmSCF is invoked. DP CollectedInfo may be reported.
{3} Instructions received from the gsmSCF on how the call is to be routed, together with which EDPs are armed. State Routing_And_Alerting entered. INVITE method forwarded to destination.
{4} A response ‘200 OK’ indicates that the destination has accepted in session invitation, indicating that a session has been established. State O_Active is entered, DP O_Active may be reported to the gsmSCF and an ACK is sent to the originating party.
{5} Either party may release the call with a BYE method. On receipt of the BYE, transition to the PIC O_Null&Authorise_Oriigination_Attempt_Collect takes place and the DP O_Disconnect may be reported to the gsmSCF.
Figure 24: Message flows and corresponding CAMEL O-BCSM for successful call establishment
|
14ce710049a3c611ebb26f3972649957
|
21.978
|
8.4.3 Terminating BCSM for MT Calls
|
The mapping between the SIP methods and responses for T-BCSM and Mobile Terminating Call is shown in Figure 25. The information flows further described below :
{1} INVITE method arrives at the destination SIP Proxy server. Server/IPSSF determines if a T-CSI exists for the called user
{2} T-CSI is analysed and if necessary triggering criteria are met, gsmSCF is invoked. Transition to state TerminatingCallHandling and DP Terminating_Atttempt_Authorised may be reported.
{3} Call accepted by the terminating party. DP T_Answer may be reported to the gsmSCF, state T_Active entered.
{4} Either party may terminate the call by sending a BYE and transition to PIC T_Null takes place and the DP T_Disconnect may be reported.
Figure 25: Message flows and corresponding PIC in the CAMEL T-BCSM for successful call delivery
|
14ce710049a3c611ebb26f3972649957
|
21.978
|
8.4.4 Unsuccessful MT Call Delivery
|
This subsection explores the mapping and the reporting of the DPs that may be encountered when the call is not successfully established. The information flows are shown in Figure 26 and further explained below:
{1} INVITE method arrives at the destination SIP proxy server. Server/IPSSF determines is a T-CSI exists for the called user
{2} T-CSI is analysed and if necessary triggering criteria are met, gsmSCF is invoked. Transition to state TerminatingCallHandling and DP Terminating_Atttempt_Authorised may be reported.
{3} The destination does not accept the incoming call – reason response may be any value in 4xx response range. The mapping of client error codes (4xx) to the possible Detection Points in PIC Terminating_Call_Handling is not all that straight forward. For example, the DP T_CallHandlingFailure can capture most of the 4xx error codes, and T_Busy can be mapped on to 486 Busy Here, it is not clear how T_Abandon and T_No_Answer can be mapped on to the error codes. Although further work is required, the problem is not unique to GSM and CAMEL. Work in other standards organisations needs to be investigated to determine similar issues.
Figure 26: Message flows and corresponding PIC in the CAMEL T-BCSM for unsuccessful call establishment.
|
14ce710049a3c611ebb26f3972649957
|
21.978
|
8.5 CAMEL Integration
|
Integration of CAMEL functions with SIP Proxy Server functions may require enhancements to SIP standards or CAMEL standards. Some initial work has been done to identify the standards that may need to be enhanced and to assess the extent of the changes required.
|
14ce710049a3c611ebb26f3972649957
|
21.978
|
8.5.1 Impact on SIP Standards
|
Section 8.4 describes and illustrates how SIP messages may be mapped to O_BCSM and T_BCSM Detection Points (DPs) at the SIP Proxy Server/SSF. When studying the Figures it can be seen that some CAMEL Phase 3 DPs (such as O_Active) have good correspondence with SIP messages at the SIP Proxy Server, while other DPs require more detailed analysis. At this level of analysis it seems likely that changes to SIP may not be needed.
|
14ce710049a3c611ebb26f3972649957
|
21.978
|
8.5.2 Impact on CAMEL and other UMTS Standards
|
It is likely that some functional modelling additions and some CAP message information element additions will be needed. It is also likely that some HLR/VLR data and MAP protocol additions may be needed. It is of course very difficult to assess and describe exactly how much effort and meeting time is required to effect these changes. In comparison to previous CAMEL work on Phase 1, Phase 2 and Phase 3, in the author's opinion the effort required is probably more than was necessary to complete GPRS inter-working in CAMEL Phase 3, but probably less than the work necessary to complete CAMEL Phase 1.
The major impact of these changes concerns documents 22.078, 23.018, 23.078, 29.078 and 29.002
|
14ce710049a3c611ebb26f3972649957
|
21.978
|
8.6 Service Impacts
|
The combination of CAMEL and SIP (as described in this report) may support some GSM supplementary services, such as unconditional call forwarding (subject to some re-engineering) and perhaps most operator specific services (depends on the extent of any reduced functionality necessary for integration). Every service needs to be studied in detail to determine exactly what can be supported. Interworking with SIP extensions for call control services is another aspect. Again every service needs to be studied in detail.
|
14ce710049a3c611ebb26f3972649957
|
21.978
|
8.7 Multimedia Evolution
|
The proposal in this report is focused on voice not multimedia. Control of multimedia services requires further study. However, the proposal does not in any way preclude CAMEL evolution to control multimedia services supported by SIP. It is likely that further enhancements to the protocols and functions listed in section 8.5.2 (e.g.. MAP, CAP, HLR, VLR, CSE) may be necessary, depending on the nature of the multimedia service support required.
|
14ce710049a3c611ebb26f3972649957
|
21.978
|
8.8 Advantages
|
The following advantages have been identified:
Maximises the re-use of existing functional entities, protocols and services. Such reuse decreases the development and ownership costs allowing existing familiar 2G services to be provided to 3G subscribers at an early stage.
Minimum changes to the CSE for the support of legacy services. There are several IN/CAMEL services already deployed such as PrePaid, VPN, Mobile Number Portability, etc which may be used in a voice over IP network.
|
14ce710049a3c611ebb26f3972649957
|
21.978
|
8.9 Disadvantages
|
The following disadvantages have been identified:
This approach is not in line with the work currently underway in ETSI SPAN 3 (Services and Protocols), in particular a work item addressing IN support for voice over IP on the H.323 architecture and associated protocols in association with the TIPHON project.
Introduces new functional entity ‘IPSSF’, which provides the necessary mapping between the SIP Proxy Server and the CSE. However, this functional entity is based on the functions already provided by a VMSC/GMSC, where already standardised process such as the gsmSSF can be reused.
The interface between the SCS/CSCF and the IPSSF requires further study.
|
14ce710049a3c611ebb26f3972649957
|
21.978
|
9 Work in other Standards Groups
| |
14ce710049a3c611ebb26f3972649957
|
21.978
|
9.1 3GPP
| |
14ce710049a3c611ebb26f3972649957
|
21.978
|
9.1.1 System Architecture Working Group 2 (S2)
|
The 3GPP S2 group has overall system architecture responsibility for the GSM/GPRS evolved system within the 3GPP organisation and also has close linkage with the ETSI SMG12 (GSM/GPRS) system architecture group.
The S2 group is investigating the architectural impacts and linkages involved with providing IN (CAMEL) based features into the mobile environment including the impacts of aspects such as roaming, Mobile Originated and Terminating calls, interactions with GSM standardised supplementary services and distributed service control/SSF issues across network boundaries. Items currently under study include the addition of CAMEL features for GPRS (beyond CAMEL Phase 3) and the opportunities to include further location based capabilities. CAMEL is also continuing into the UMTS area to include the Open Service Architecture (OSA) aspects.
3GPP document 23.121 specifies architectural requirements.
|
14ce710049a3c611ebb26f3972649957
|
21.978
|
9.2 ETSI
| |
14ce710049a3c611ebb26f3972649957
|
21.978
|
9.2.1 TIPHON
|
TIPHON Working Group 7 is studying the implementation of TIPHON networks in a mobile environment, and has been in existence since the beginning of 1999. TIPHON mobility is user and service mobility in the context of the VoIP application. The group has an understanding with 3GPP S2 to liase over IP activities. It is currently planning on producing 3 main documents:
07.001 Analysis of existing roaming techniques applicable to TIPHON mobility services
07.002 Investigation of synergies and common requirements between TIPHON networks and wireless systems as they are currently being developed by other bodies.
07.003 Mobility and Access to Wireless Systems: Extensions to Requirements, Architectures and Protocols
However, at the present moment only 07.001 has any significant text associated with it. The document is a review of existing mobile networks and specifically lists supplementary services as an open issue. In section 5.1 the GSM network is described, and CAMEL identified as the Intelligent Network service within GSM used to provide service mobility where the end user should not see any difference in the services provided by the IN nodes irrespective of the user’s location and terminal used. Similarly section 5.7 describes the GPRS network, section 5.9.4 introduces Session Initiation Protocol, and section 5.10 details the mobile extension to H.323.
At the July 1999 TIPHON 14 meeting, working group 7 discussed a document (14TD054) received from ITU-T SG16 for information concerning new work recently approved on mobility for H.323. The contribution stirred discussion, as the solution was seen as conflicting with the general attempt within TIPHON to work towards a layered architecture.
|
14ce710049a3c611ebb26f3972649957
|
21.978
|
9.2.2 SPAN3
|
ETSI SPAN (Services and Protocol for Advanced Networks) is ETSI’s core competence centre for fixed networks standardisation including IP based networks. SPAN3 (formerly SPS3) is the competence centre for IN Activities, including generic operations for INAP mobility. SPS3 work on Core INAP CS-3 has been split into two streams. CS-3.1 comprises CS2 and CAMEL Phase 3, while CS-3.2 looks at fixed network requirements and will be based on the ITU CS3 output. Both are due for completion by the end of 1999 and will be merged in Core INAP CS4, due for completion at the end of 2000.
CS-3.2 contains a work item addressing IN support for voice over IP based on the H.323 architecture and associated protocols in conjunction with the TIPHON project. A liaison statement from SPS3 to TIPHON entitled “Using Intelligent Networks in a Tiphon Architecture” was written in June 1999, giving the following proposals
- The IN infrastructure shall be independent of the IP telephony signalling protocol (SIP, H.323,…).
- The call control and consequently IN control shall be at least at the edge of the network i.e. nearest to the user in a Local Exchange and between two operators in an Inter-operator Gateway.
|
14ce710049a3c611ebb26f3972649957
|
21.978
|
9.3 IETF
|
The Routing area contains the IP Routing for Wireless/Mobile Hosts (mobileIP) Working Group. It has developed routing support to permit IP nodes (hosts and routers) using either IPv4 or IPv6 to seamlessly roam among IP sub-networks and media types. The group has produced an Internet draft entitled “Requirements on Mobile IP from a Cellular Perspective” which considers MobileIP as a macro-mobility solution for cellular networks. This document does not address the issues surrounding the provision of supplementary services.
The Transport area contains the IP Telephony (iptel) working group and the Media Gateway Control (MEGACO) working group.
The Internet draft “SIP Call Control Services” describes a set of extensions to SIP which allow for various call control services. Example services include blind transfer, transfer with consultation, multi-party calls, bridged conferences, and ad-hoc conferencing. The Internet draft “Accessing IN services from SIP networks” proposes a mapping from the states of the IN call model to the states of SIP, an Internet call signalling protocol.
|
14ce710049a3c611ebb26f3972649957
|
21.978
|
9.4 ITU-T
| |
14ce710049a3c611ebb26f3972649957
|
21.978
|
9.4.1 SG11
|
Study Group SG11 is responsible for studies relating to signalling requirements and protocols for telephone, N-ISDN, B-ISGN, UPT, mobile and multimedia communications. SG11 is responsible for most of the Q-Series standards, including Q.931. Questions under study by the group include:
5/11 Intelligent network capability sets
6/11 New signalling capabilities and requirements for advanced broadband multimedia services
7/11 Signalling, call handling and management requirements for universal personal telecommunications and for user mobility in future public land mobile systems
13/11 Network signalling for the support of broadband services and third generation land mobile networks (FPLMTS)
22/11 Intelligent Network Application Protocol (INAP)
24/11 Signalling requirements for emerging land mobile and satellite mobile - Network and Inter-network Signalling Requirements
|
14ce710049a3c611ebb26f3972649957
|
21.978
|
9.4.2 SG16
|
Study Group16 is responsible for studies relating to multimedia service definition and multimedia systems, including the associated terminals, modems, protocols and signal processing. SG16 is responsible for the H-Series standards, including H.323, as well as parts of the T-, G-, and F-Series. Questions under study by the group include:
13/16 Packet switched multimedia systems and terminals
This question covers the H.323 standard, of which Annex H studies H.323 Mobility. The annex has the following objectives:
Define functional requirements for H.323 mobility at the application level in a transport independent way.
Examine whether any new messages or message elements need to be created in H.323 for supporting mobility
Facilitate interoperability for implementation of H.323 mobility over any specific networking environment such as wireless/PSTN
|
14ce710049a3c611ebb26f3972649957
|
21.978
|
10 Questions & Answers
|
Early drafts of this FTR were presented to 3GPP CN WG2a and a small number of questions were raised. The questions and answers are recorded here.
Q. In our work, are we considering IN control of RAS signalling or for call control or both?
A. Control of RAS signalling is out of scope.
Q. How does the architecture support the playing of tones and announcements?
A. This is a problem not unique to CAMEL and is one of many issues that we have not addressed. We felt that it is appropriate to address this aspect as part of a more detailed study.
Q. Are we only considering CAP over SS7?
A. We have not excluded the possibility of CAP over IP, this possibility depends on the outcome of SIGTRAN work and the MAP over IP work in CN2B. CAMEL control of VoIP services is independent of whether SS7 or IP is used for the lower layer signalling transport.
Q. What considerations are there for 'fast start' and what impact does this have?
A. Requires further study.
Q. How are we incorporating the work output from MEGACO?
A. As far as we are aware MEGACO work does not directly impact CAMEL control of VoIP services but further study is required.
|
14ce710049a3c611ebb26f3972649957
|
21.978
|
11 Conclusions & Recommendations
|
This report has demonstrated how CAMEL can be used to control VoIP services in an all IP network using H.323 or SIP. The analysis has been carried out at a high level sufficient to identify that further detailed design and specification work should lead to a viable solution provided the architectural assumptions used in this report are correct. At this level of analysis no preference for H.323 or SIP was evident. Based on this evidence, either H.323, SIP or both could be successfully employed.
Use of CAMEL is one service creation approach for H.323/SIP-based services. There are also alternative H.323/SIP and IN interworking options identified in the report. Finally use of “next generation” architectures may enable other types of service creation which are not based on IN. The overall options need to be reviewed to select which of them will be most appropriate for R00 considering service and technical aspects. Based on this analysis the use of CAMEL is technically feasible and would provide a service evolution path from R99, but detailed comparison with other options has not been included.
Finalisation of the architecture and service creation principles for R00 is urgently needed so that work on the selected option(s) can start. It is also necessary to decide what standardised service(s) are required for H.323/SIP and which service(s) will be implemented using tool-kits.
Annex A:
Change history
Change history
Date
TSG #
TSG Doc.
CR
Rev
Subject/Comment
Old
New
Jun 2000
CN#08
NP-000252
Approved by TSG CN
2.1.1
3.0.0
|
4131cd5fefdadf4075e3ec0993ed8fe4
|
34.901
|
1 Scope
|
The present document describes the statistical theory and concepts applied in the conformance test of the user equipment (UE) so as to improve test speed.
|
4131cd5fefdadf4075e3ec0993ed8fe4
|
34.901
|
2 References
|
The following documents contain provisions which, through reference in this text, constitute provisions of the present document.
• References are either specific (identified by date of publication, edition number, version number, etc.) or non‑specific.
• For a specific reference, subsequent revisions do not apply.
• For a non-specific reference, the latest version applies. In the case of a reference to a 3GPP document (including a GSM document), a non-specific reference implicitly refers to the latest version of that document in the same Release as the present document.
[1] 3GPP TS 34.121: "Terminal Conformance Specification; Radio transmission and reception (FDD)".
[2] 3GPP TS 34.122: " Terminal Conformance Specification; Radio transmission and reception (TDD)".
3. Definitions, symbols and abbreviations
Definitions, symbols, abbreviations and equations used in the present document are listed in TR 21.905 [5] and TR 25.990 [6].
|
4131cd5fefdadf4075e3ec0993ed8fe4
|
34.901
|
3.1 Definitions
|
For the purposes of the present document, the following additional terms and definitions apply.
|
4131cd5fefdadf4075e3ec0993ed8fe4
|
34.901
|
3.2 Symbols
|
For the purposes of the present document, the following symbols apply:
[…] Values included in square bracket must be considered for further studies, because it means that a decision about that value was not taken
|
4131cd5fefdadf4075e3ec0993ed8fe4
|
34.901
|
3.3 Abbreviations
|
For the purposes of the present document, the following abbreviations apply:
BER Bit Error Ratio
BLER Block Error Ratio
DUT Device under Test
|
4131cd5fefdadf4075e3ec0993ed8fe4
|
34.901
|
3.4 Equations
|
Void.
|
4131cd5fefdadf4075e3ec0993ed8fe4
|
34.901
|
4 Introduction
|
This technical report includes two distinct approaches made to determine total test time optimisation. For ease of understanding they are just referred to here as the first and second approach. Furthermore the two approaches differ a little in that they use slightly different terminology.
|
4131cd5fefdadf4075e3ec0993ed8fe4
|
34.901
|
4.1 First approach
|
The first approach is found in clauses 5 and 6. It reflects TS 34.121 in that the symbols, abbreviations and equations are consistent with TS 34.121.
|
4131cd5fefdadf4075e3ec0993ed8fe4
|
34.901
|
4.2 Second approach
|
The second approach is found in clauses 7 to 11 and does not directly reflect TS 34.121 although it does use the existing theory from TS 34.121. The difference is that it refines the theory and derives further approaches for test time reduction. Some of the symbols, abbreviations and equations have local meaning and these are identified in clause 7.
5. Definitions of distribution functions and parameters to be used
Summary
|
4131cd5fefdadf4075e3ec0993ed8fe4
|
34.901
|
5.1 Properties of the Poisson Distribution
|
Description of a statistical experiment by a distribution function and basic characteristics of the distribution.
|
4131cd5fefdadf4075e3ec0993ed8fe4
|
34.901
|
5.2 Equivalence between Poisson Distribution and Chi Square Distribution
|
Here it is shown, that both distributions are equal. Just the form is different. On the other hand there are two inverse cumulative operations. One of them is useful for our purpose.
|
4131cd5fefdadf4075e3ec0993ed8fe4
|
34.901
|
5.3 Confidence interval
|
In a single test we apply ns samples and measure ne errors. The result can be member of different distributions each characterized by another parameter NE. We ask for two of them:
1) The worst possible distribution NEhigh , containing our measured ne with [D= 0.0085%] probability in the sense
0.000085= (2)
ni is the integration variable
ne is the measured value
NE is the variable to tune in order to make the integral consistent.
The result of the inverse cumulative operation is NEhigh
2) The best possible distributions NElow , containing our measured ne with [D=0.0085%] probability in the sense
0.000085= (3)
The result of the inverse cumulative operation is NElow
To illustrate the meaning of the range between NElow and NEhigh:
In the case our measured value ne is a rather untypical result (just [0.0085%] probability) nevertheless the final result NE can still be found in this range, called confidence interval.
The probabilities D in (1) and (2) can be independent like in GSM, but we want to have them dependent and equal.
The inverse cumulative Chi Squared distribution gives the wanted results:
Inputs: number of errors ne, measured in this test.
Probabilities D and the complementary probability 1- D
Output: NE, the parameter describing the average of the distribution.
E.G.:
(4)
(5)
Figure 5‑6: Confidence Interval
Same as the width of the distributions the confidence interval increases proportional to SQR(ne), that means, it increases absolutely, but decreases relatively to the measured number of errors.
|
4131cd5fefdadf4075e3ec0993ed8fe4
|
34.901
|
5.4 Application of the confidence interval to decide the outcome of the test
|
If we find the entire confidence range, calculated from a single result ne, on the good side of the specified limit we can state: With high probability 1-D, the final NE is better than the limit.
If we find the entire confidence interval, calculated from a single result ne, on the bad side of the specified limit we can state: With high probability 1-D, the final NE is worse than the limit.
With each new test we update our preliminary data for ns, ne and ber. For each new sample we calculate the confidence interval and check it against the test limit.
Once we find the entire confidence interval on the good side of the specified limit we allow an early pass.
Once we find the entire confidence interval on the bad side of the specified limit we allow an early fail.
If we find the confidence interval on both sides of the specified limit, it is evident neither to pass nor to fail the DUT early.
Transcription of the above text into formulas:
The current number of samples ns is calculated from the preliminary ber and the preliminary ne
ber = ne/ns (6)
BERlim = NElimit / ns (7)
for abbreviation in the formula: bernorm = ber/BERlimit = ne/ NElimit (normalised ber)
Early pass stipulates:
NEhigh < NElimit (8)
Early fail stipulates:
NElow > NElimit (9)
The early fail and the early pass limit are displayed in Figure 5‑7:
early pass limit
(10)
early fail limit
(11)
Figure 5‑7: Early pass and early fail curves
|
4131cd5fefdadf4075e3ec0993ed8fe4
|
34.901
|
5.6 Test time reduction
|
Using 5.4 the outcome of the test is connected with two qualities, a good one and a worse and variable one. Introducing the bad DUT factor M, the quality of the test is now uniform and test time is further reduced.
|
4131cd5fefdadf4075e3ec0993ed8fe4
|
34.901
|
5.7 Calculation of the intersection coordinates (maximum number of sample and the normalized test limit)
|
Calculus for intersection co-ordinates of the early pass and early fail limit.
|
4131cd5fefdadf4075e3ec0993ed8fe4
|
34.901
|
5.8 Wrong decision risk F
|
5.1 to 5.6 applies the wrong decision risk for a single test step D. However it is desirable to have a predefined wrong decision risk for the entire test F. The approach to derive F from D this is explained here.
|
4131cd5fefdadf4075e3ec0993ed8fe4
|
34.901
|
5.1 Properties of the Poisson distribution
|
NOTE: The following text is expressed in terms of BER=bit error ratio. However it can be used for BLER (Block error ratio) as well. Even for 1- Success Ratio, used in RRM delay tests, the theory can be used.
With a finite number of samples (ns), the final bit error ratio BER cannot be determined exactly.
Applying a finite ns, we measure a number of errors (ne).
ne/ns =ber is the preliminary bit error ratio.
In a single test we apply a predefined number of samples ns and we measure a number of errors (ne). ne is connected with a certain differential probability in the Poisson distribution. We don't know the probability and the position in the distribution conducting just one single test.
Repeating this test infinite times, applying repeatedly the same ns, we get the complete Poisson distribution. The average number of errors is NE. NE/ns is the final BER.
Poisson Distribution:
dpois(ne,NE)=(NEne/ne!)e-NE (1)
e.g. :
Figure 5‑1: Example of Poisson distribution curve
The Poisson distribution has the variable ne and is characterised by the parameter NE.
Real probabilities to find ne between two limits are calculated by integrating between such limits.
Note: The Poisson distribution is an approximation: Independent error occurrence is described by the binomial distribution. If the BER approaches 0 the Poisson distribution approximates the binomial distribution.
|
4131cd5fefdadf4075e3ec0993ed8fe4
|
34.901
|
5.2 Equivalence between Poisson distribution and Chi Square distribution
|
The experiment, the Poisson distribution is based on, is having observed a certain number of samples (ns), the number of events (ne) is counted to calculate the ratio ne/ns.
The experiment, the Chi Square distribution is based on, is having observed a certain number of events (ne), the number of samples (ns) is counted to calculate the ratio ne/ns.
Poisson and Chi Square are valid only if ne<<ns
Figure 5‑2: Comparison of Chi-Squared and Poisson distribution
The dotted blue function is the Chi-squared distribution, using the parameters of the Poisson distribution. We scaled, offset and changed the interpretation of variable and parameter such that both distributions match. The Poisson distribution is a discrete distribution. Such scaled the Chi Squared distribution interpolates the Poisson distribution exactly for all NE (degenerated for NE=ne=0).
The experiment of the Chi Square distribution is always terminated by an event,
In contrast the experiment of the Poisson distribution almost never is terminated by an event because of ne/ns-->0. This explains that the Poisson distribution needs one event more, to equal in its form the Chi Square distribution
2*dchisq(2*NE,2*ne) = dpois(ne-1,NE) describes the experiment, terminated by an error.
2*dchisq(2*NE,2*(ne+1)) = dpois(ne,NE) describes the experiment, terminated by any sample.
The terminating error may be the artificial error at the beginning of the test, or the last error, causing the fail.
In the next comparison shows dpois versus dchisq.
The first 3D plot shows the Poisson distribution: (Figure 5‑1)
Variable: ne Range 0 to 10 Column in the table 0 to 10- axis in the plot
Parameter: NE Range 0 to 10 Row in the table 0 to 100 axis in the plot
The second 3D plot shows the Chi Square distribution: (Figure 5‑4)
Variable: NE Range 0 to 10 Column in the table 0 to 100- axis in the plot
Parameter: ne Range 0 to 10 Row in the table 0 to 10 axis in the plot
Columne 0 is degenerated
Table 5‑1: Poisson distribution calculation
Figure 5‑3: 3D plot for Poisson distribution
ne
Table 5‑2: Chi-squared distribution calculation
Figure 5‑4: 3D plot for Chi-squared distribution
Observation:
1) The rows in Poisson distribution correspond the columns in the scaled Chi-squared distribution and vice versa.
2) Poisson distribution at ne=0 versus NE is the exponential distribution
Chi-squared distribution at ne=0 (degree of freedom=1) versus NE is also the exponential distribution
see the next plot:
Figure 5‑5: Comparison between Poisson and Chi-squared distribution
Inverse Cumulative Operation:
We have seen: Chi Square and Poisson both describe the same array: ne versus NE.
The figures above show, that NE and ne in both functions are not commutative.
Hence there are two inverse operations (a) and (b):
D= = 2* (a)
with D=wrong decision probability or confidence level (input).
ni is the integration variable
ne is the measured value.(input, discrete) It is the integration limit
NE (real) is tuned such that the integral is consistent.
It returns an NE as a function of the two parameters D and ne. qchisq(D,ne)
D= = 2* (b)
NI is the integration variable
NE (real) is the integration limit
ne (discrete) is tuned such that the integral is consistent.
It returns ne as a function of the two parameters D and NE: qpois(D,NE)
Our target requires a). This is usually called the Inverse Cumulative Chi Square function.
(b) is the solution for another target. This is usually called the Inverse Cumulative Poisson function.
(a) returns a greater NE than (b) returns with respect to ne. (easily visible in the figures)
The difference (a)-(b) is small. This is also visible from the figures: ne and NE are close to commutative.
(a) returns a continuous NE, (b) returns a discrete ne.
|
4131cd5fefdadf4075e3ec0993ed8fe4
|
34.901
|
5.5 Test time reduction
|
Whichever ne we propose as a final stop condition e.g. ne =200, the test can leave the area between the early pass and the early fail limit through the open end of the right side of Figure 5-7. This situation needs an arbitrary pass or fail decision. E.g. pass, if the test hits the vertical 200-error line. This situation has the following drawback: The test has two different qualities. A good one, when the test hits an early pass or early fail limit, and a worse and variable one, when the test hits the vertical 200 error line; variable, depending on the height, it crosses the line. The quality of the test in terms of wrong decision risk is variable in the range D up to as bad as 50%. We can replace the situation against a better trade-off:
Instead a test with different qualities against one limit,
we design a test with a fixed uniform quality against two limits,
(gaining further test time reduction).
We maintain the definition of the early fail limit:
(a) We fail a DUT and accept the probability of D= 0.0085% that it is actually better than the limit.
We propose a meaningful redefinition of the early pass limit:
(b) We pass a DUT and accept the probability of D=0.0085% that it is actually worse than
M times the limit (M>1).(M = Bad DUT factor)
This produces the following consequences:
(1) The early pass limit is shifted upwards by the factor of M
(2) The early fail and the early pass limit intersect.
(3) The intersection coordinates are:
the normalized test limit
and the maximum number of events
Transcription of the above text into formulas:
berlimbadpass: early pass limit against the bad DUT limit (12)
berlimfail: early fail limit against the specified limit (13)
Figure 5‑8: Early pass and early fail curves with multiplication factor M
|
4131cd5fefdadf4075e3ec0993ed8fe4
|
34.901
|
5.6 Calculation of the intersection coordinates
|
initial guess of target number of events
root finds the zero of the function
target number of events
normalized test limit
|
4131cd5fefdadf4075e3ec0993ed8fe4
|
34.901
|
5.7 Wrong decision risk F
|
Provided a single BER trajectory with final BER on the limit hits the early fail limit. If a fail is decided at this instant of the test, the wrong decision risk is as small as D. For each member of a large population of DUTs a wrong decision can happen, with probability D, accumulating to an amount F > D for the entire population.
D is the wrong decision risk based on the statistical totality of samples with BER on the limit.
F is the wrong decision risk based on the statistical totality of DUTs with BER on the limit.
(The same holds for a bad DUT, hitting the early pass limit.)
We call D the wrong decision risk at a single test step and F the wrong decision risk for the entire test. For a real test it is desirable to define in advance the wrong decision risk F of the entire test. An exact theory is not available for this problem. It is proposed to derive D from F by the following simulation:
A large population of DUTs with BER on the limit (limit-DUT) is simulated and decided against the early pass and early fail bound, with a free D-parameter in the early pass and fail limit. The simulation will show, that a certain fraction F (D<F<1) falsely fails.
The complementary simulation is:
A large population of DUTs with M*BER (bad DUT) is simulated and decided against the early pass and early fail bound, with a free D-parameter in the early pass and fail limit. The simulation will show, that a certain fraction F (D<F<1) falsely passes.
Both false decision fractions are approximately equal and represent the wrong decision probability F for the entire test. D is tuned such that F corresponds to the predefined wrong decision probability.
|
4131cd5fefdadf4075e3ec0993ed8fe4
|
34.901
|
6 F to D conversion in BER BLER tests
| |
4131cd5fefdadf4075e3ec0993ed8fe4
|
34.901
|
6.1 Conversion F to D
|
Annex F.6.1 in TS 34.121 gives a statistical approach for BER BLER tests. It gives early pass and early fail conditions. The formulas for this condition contain the parameter D, the wrong decision probability for a single test step. However it is desirable to have a wrong decision probability for the entire test F. This contribution explains the way, how to derive F from D and gives results for a set of parameters.
|
4131cd5fefdadf4075e3ec0993ed8fe4
|
34.901
|
6.2 Introduction
|
Provided a single BER trajectory with final BER on the limit hits the early fail limit. If a fail is decided at this instant of the test, the wrong decision risk is as small as D. For each member of a large population of DUTs a wrong decision can happen, with probability D, accumulating to an amount F > D for the entire population.
D is the wrong decision risk based on the statistical totality of samples with BER on the limit.
F is the wrong decision risk based on the statistical totality of DUTs with BER on the limit.
(The same holds for a bad DUT, hitting the early pass limit.)
We call D the wrong decision risk at a single test step and F the wrong decision risk for the entire test. For a real test it is desirable to define in advance the wrong decision risk F of the entire test. An exact theory is not available for this problem. It is proposed to derive D from F by the following simulation:
A large population of DUTs with BER on the limit (limit-DUT) is simulated and decided against the early pass and early fail bound, with a free D-parameter in the early pass and fail limit. The simulation will show, that a certain fraction F (D<F<1) falsely fails.
The complementary simulation is:
A large population of DUTs with M*BER (bad DUT) is simulated and decided against the early pass and early fail bound, with a free D-parameter in the early pass and fail limit. The simulation will show, that a certain fraction F (D<F<1) falsely passes.
Both false decision fractions are approximately equal and represent the wrong decision probability F for the entire test. D is tuned such that F corresponds to the predefined wrong decision probability.
|
4131cd5fefdadf4075e3ec0993ed8fe4
|
34.901
|
6.3 The simulation procedure
| |
4131cd5fefdadf4075e3ec0993ed8fe4
|
34.901
|
6.3.1 Equal pass and fail probability
|
A population of DUTs on the limit is established. Quantity 10 000
Preselected BER 1%
An early fail and an early pass limit is established. With D (wrong decision risk) and M (bad DUT factor)
With target ne and test limit as a side result.
D is tuned in the inner loop
M and Fpredefined are varied in the outer loop
During the simulation
A member of the population leaves the statistical totality if
An error happens and the early fail limit is hit (false fail)
An error happens and the early pass limit is hit or crossed (correct pass)
The fraction false fails / 10 000 = F is recorded.
Inner loop: In repeated trials D is tuned, such that F ≤ Fpredefined (conservative approach).
Having decided for a specific D the simulation is repeated again 10 times and Fmin, Fmax, and Fmean are recorded.
The complementary simulation is done with a population of bad DUTs
(same quantity, same M, same Fpredefined , same D)
Observation 1: the false pass fraction is slightly lower than the false fail fraction.
Hence the result is even more conservative for the false pass.
Outer Loop: M is varied from 1.1 to 1.5 in steps of 0.1
Fpredefined is varied from 0.2 %, 0.5 %, 1%, 2% to 5%.
Observation 2: For lower wrong decision risks F the false decisions in 10 000 DUT are less.
Hence the variance of F in the 10 repetitions relatively increases.
For lower wrong decision risks F the simulation time increases.
Hence the compensation of the increasing variance of F by more repetitions is limited by the simulation time, or vice versa: the simulation results for F converge to a final value, investing infinite effort for simulations.
For practical and security reasons the Ds for lower Fs are decided more conservative than the equivalent ones for higher Fs.
|
4131cd5fefdadf4075e3ec0993ed8fe4
|
34.901
|
6.3.2 Unequal pass and fail probability
|
For statistical test, frequently repeated, a single false fail can fail the composite test. This can be combated by a decreased false fail risk. This costs test time.
A single false pass does not have this effect on the composite test. Hence it is not necessary to consume increased test time due to decreased false pass risk for the pass probability. Hence unequal pass and fail probabilities are treated for very low false fail risk.
A population of DUTs on the limit is established. Quantity 10 000
Preselected BER 1%
In the complementary simulation
a population of bad DUTs is established (same parameters)
Common for both simulations:
An early fail limit is established with Dfail and an early pass limit is established with Dpass (D wrong decision risk Dfail < Dpass) and M (bad DUT factor)
With target ne as a side result.
Dfail and Dpass are tuned independently in the inner loop
M is varied in the outer loop
During the simulation
A member of the population leaves the statistical totality if
An error happens and the early fail limit is hit (false fail)
For the complementary simulation: (correct fail)
An error happens and the early pass limit is hit or crossed (correct pass)
For the complementary simulation: (false pass)
The fraction false fails / 10 000 = Ffail and the fraction false pass/10 000 = Fpass are recorded.
Inner loop: In repeated trials Dfail and Dpass are tuned independently , such that F fail and Fpass ≤ Fpredefined (conservative approach).
Having decided for a specific Dfail and Dpass the simulation is repeated again 10 times and Fmin fail, Fmax fail, and Fmean fail and Fmin pass, Fmax pass, and Fmean pass are recorded.
Observation 3: The Dpass must be slightly lower than the equivalent D in the case for equal probabilities (a).
Due to lower Dfail the target number of errors increases (e.g. 345 403), accumulating more single step wrong decisions. This is compensated by a lower Dpass.
|
4131cd5fefdadf4075e3ec0993ed8fe4
|
34.901
|
6.4 Result of the simulation: M-F array
|
Table 6‑1: M-F table
7. Definitions, symbols and abbreviations
Definitions, symbols, abbreviations and equations used in the present document are listed in TR 21.905 [5] and TR 25.990 [6].
|
4131cd5fefdadf4075e3ec0993ed8fe4
|
34.901
|
7.1 Definitions
|
For the purposes of the present document, the following additional terms and definitions apply.
Wrong Decision Probability: Chance of incorrect judgement based on the given test results.
Significance Level: Chance of incorrect judgement based on the given test results. This term is exchangeable with Wrong Decision Probability.
Individual Significance Level: The significance level of a simple test or a single decision. Denoted by "D".
Total Significance Level: The significance level of a test system as a whole (a set of simple tests.) Denoted by "F".
Confidence Coefficient (Level): In this report, this is defined by 1 – Significance Level.
Specified Error Ratio: General term that is referred to BER/BLER or other error ratios that are specified in the test specifications (TS34.121 and TS34.122.)
Average Error Ratio: The error ratio specific to a DUT whose value can be determined by averaging the infinite number of measurement data of error ratio.
Early Pass/Fail Criteria: a set of simple tests, each of which consists of a pair of thresholds of the measured error ratio.
|
4131cd5fefdadf4075e3ec0993ed8fe4
|
34.901
|
7.2 Symbols
|
For the purposes of the present document, the following symbols apply:
[…] Values included in square bracket must be considered for further studies, because it means that a decision about that value was not taken
C Confidence Coefficient (Confidence Level)
D Individual Significance Level
Dp Individual Significance Level of an early pass criterion
Df Individual Significance Level of an early fail criterion
F Total Significance Level
Fp Total Significance Level of a set of early pass criteria
Ff Total Significance Level of a set of early pass criteria
M Bad DUT factor
R0 Specified error ratio
R Measured error ratio (rate) (calculated from the measured error count and time duration or sample number)
Rth Threshold level against measured error ratio
Rth-p Threshold of an eraly pass test
Rth-f Threshold of an eraly fail test
r Average error ratio (rate) of DUT
|
4131cd5fefdadf4075e3ec0993ed8fe4
|
34.901
|
7.3 Abbreviations
|
For the purposes of the present document, the following abbreviations apply:
BER Bit Error Ratio
BLER Block Error Ratio
DUT Device under Test
pdf Probability Distribution Function
CDF Cumulative Distribution Function
|
4131cd5fefdadf4075e3ec0993ed8fe4
|
34.901
|
7.4 Equations
|
Chi-square distribution: of a degree of freedom n
where is gamma function.
Exponential distribution:
|
4131cd5fefdadf4075e3ec0993ed8fe4
|
34.901
|
8 Introduction
|
BER/BLER measurements are inherently statistical processes, and a statistical approach should be introduced in order to make a BER/BLER test method objective and accurate.
In the GSM standard specifications, this has been achieved by introducing "wrong decision probability (significance level)" and "bad DUT factor." That is, the standard requires the test method to give a well-defined (predictable) "wrong decision probabilities", Ff / Fp or confidence coefficients Cf (= 1 - Ff)/Cp (= 1 - Fp). Here, it should be noted that the significance levels are defined against the population of UEs with two error rates (0.01 and 0.015).
These parameters, the significance level and bad DUT factor, should be determined a priori, since there seems to be no good reason to pick a set of values rather than others. So, it would be a reasonable decision to employ the same value with GSM standard: F = 0.002 (0.2 %) and M = 1.5 for BER and BLER measurements since we now have a long enough experience to prove the validity of these parameters.
The GSM standard recommends that BER/BLER should be calculated after 200 errors were observed and the resultant BER/BLER should be compared against a threshold (around 1.24 R0). This test method gives the significance level presented above.
However, this method is not optimal from the viewpoint of test time. Apparently, if the DUT population consists of very good UEs (UE with a very low BER), it would take a long time to observe 200 errors. In practice, such a situation is unacceptable, so some supplementary criterion should be used. For example, if a UE doesn't report any bit/block errors for the certain time period, it should pass the test. This supplementary criterion may save many of the potentially time-consuming cases, but still not optimised for the test time.
In this report, a more sophisticated method will be developed to optimise the test time by introducing the early pass/fail criteria, whose basic idea was inspired by I-95 standard.
|
4131cd5fefdadf4075e3ec0993ed8fe4
|
34.901
|
9 Statistical characteristics of testing processes
| |
4131cd5fefdadf4075e3ec0993ed8fe4
|
34.901
|
9.1 Exact probability distribution functions
|
BER/BLER measurements treat phenomena that are characterized as:
1) The experiment consists of a sequence of independent trials.
2) Each trial has two possible outcomes, E (Error) or N (Non Error).
3) The probability of E (r = f(E)) is constant from one trial to another.
That is, the BER/BLER measurement can be reduced to observation of Bernoulli sequences. Many other test objects, such as RRM delay measurements results, can also be reduced to a process that is characterized by these requirements. In these case, the outcomes are either "Pass or Successful (the measured delay was within the limit value, for example)" or "Fail (the delay exceeded the limit)."
Our objective is to find a way to tell whether r is larger than R0*M or smaller than R0 with a certain significance level from the result of a certain number of trials.
This report proposes to use a set of many simple tests, instead of a single simple test. Here, a test is a comparison of resultant error rate against a predefined criterion (a threshold value). Practically speaking, there would be two ways to do this comparison:
a) To do the comparison for every trial, or
b) To do the comparison each time an error is observed.
For method a) above, the probability of giving an error number, m, is given by a binomial distribution,
.
Where r: average error rate, m: the number of errors, s: the number of samples. The resultant error ratio, R, is m/s, and its probability can be derived from the equation above.
On the other hand, for method b), the probability that m-th error is observed at the s-th sample follows a negative binomial distribution,
.
The error rate R is again given by m/s.
|
4131cd5fefdadf4075e3ec0993ed8fe4
|
34.901
|
9.2 Approximation with continuous distributions
|
In this section, the notion r for average error ratio is used for average error rate too.
The distributions introduced above are exact, but sometimes they are very time-consuming to simulate the process on a computer. So, it is convenient to use appropriate continuous distribution functions that approximate the discrete functions.
If r is very low (that is, m/s << 1), the phenomenon described in the previous section can be reduced to the one that satisfies following requirements:
1) Every error can be described by the time at which it occurred.
2) Each error occurs independently
3) The average error rate r (= m/T) is constant throughout the testing.
It is known that the time interval between the consecutive errors, t, should follow an exponential distribution.
Again, there are two possibilities to determine the measured error rate R and test it against the criteria:
a) To calculate R after the predetermined time period, or
b) To calculate R when m-th error is observed.
In case a), the probability that m errors are observed within the certain time period, T follows the Poisson distribution.
Where : average error count in the time duration T, so = rT. So, the function can be represented as:
The measured error rate, R, is given by m/T.
For the test method b), the time to m-th error, T, follows an m-Erlang distribution.
Where (x, n) is a chi-square distribution function of a degree of freedom n. m-Erlang distribution’s CDF is given by
This function can be transformed to a function of R(=m/T), mERCDF().
Apparently, the CDF is a function of (R/r) and this means the distribution does not depend on the absolute value of average or measured error rate. This would make a handling very easy.
Almost all the RF and RRM measurements that require the statistical approach are of the discrete nature, and then the continuous pdf is accurate only for smaller r values. It’s hard to tell the exact condition under which the approximation should be good, but it can be said that we cannot use the continuous pdf when r = 0.1. This subject will be discussed in the chapters below.
|
4131cd5fefdadf4075e3ec0993ed8fe4
|
34.901
|
9.3 Simple test criteria and significance level
|
For BER/BLER testing, our test can be reduced to a comparison of the measured error rate, R against a threshold. It would follow the procedure like:
if R < Rth-p then the DUT passes
else if R > Rth-f then the DUT fails
else no decision should be made
Where Rth-p and Rth-f are the threshold for pass and fail criteria, and in general Rth-p Rth-f.
The significance level for the first test, Fp, is defined as the probability that a bad DUT (r > R0 M) should pass. Similarly, Ff is defined as the probability that a good DUT (r < R0) should fail in this test.
The severest condition from the viewpoint of the significant level is that r = R0M and r = R0 for bad DUT and good DUT population respectively. So, it would be reasonable to define the significance levels for such populations.
Now that it is presumed that r of the DUT population is unique (R0 or M*R0), we can calculate the probability distribution of R according to the functions introduced in the previous chapter. For a pdf, f(R: r, m), the threshold can be determined so that:
(1)
(2)
Where, it should be noted that x takes only discrete values that are given by m/s.
For the continuous pdf’s, the summation should be replaced with integration, or using their CDF (F(R: r, m))
(1’)
(2’)
When we choose an appropriate combination of M, m, and D(= Dp = Df), we can set the Rth so that the outcome of the test is either of Pass or Fail. For GSM standard, M = 1.5, D = 0.2 %, and then m = 200 and Rth = 1.24 R0. This can be derived from both of m-Erlang distribution (mERCDF(R:m,r) above) and the negative binomial distribution.
|
4131cd5fefdadf4075e3ec0993ed8fe4
|
34.901
|
10 Early Pass/Fail termination of testing
| |
4131cd5fefdadf4075e3ec0993ed8fe4
|
34.901
|
10.1 Framework of test method
|
If a UE reported 50 block errors for the first 100 frames, we would intuitively know that the UE should be bad (its average BLER should be higher than 0.01), even though the number of the observed errors is much smaller than 200. We are quite sure about that, since it should be very rare for a UE with an average BLER of 0.01 to report 50 block errors out of 100 frames. This can be shown using the binomial distribution, which gives its probability as
The probability that 50th error should occur at 100th sample is given by negative binomial distribution:
We can safely say that both cases should be very unlikely. That is, r cannot be 0.01 and should be much larger.
This suggests that, for very good or very poor UEs, the test sequence can be terminated at earlier stages, and as a result, the test time can be cut short. This is the origin of the idea of "early pass/fail criteria."
Such the test method will consist of many simple tests and its pseudo script would look like:
m = 0
for each s: # repeat for sample forever
if an error is reported:
m = m + 1
R = m/s
if R > Rth_f(m): # fail test
UE failed
break # stop procedure
else if R < Rth_p(m):
UE passed
break
else
continue
This method is straightforward. Each time an error reported, the error rate should be calculated and then compared with thresholds that depend on m. However, there is till a room for improvement from viewpoint of test time reduction. The problem is apparent if we imagine the case in which no errors occur at all. The test would not be completed within a finite time period.
Assume that the no error samples last long enough after the m-th error. If the duration is long enough and the error rate calculated from imaginary (m +1)th error is lower than Rth_p(m + 1), the UE can be considered to have passed the test. This idea would be implemented like:
m = 0
for each s: # repeat for sample forever
if an error is reported:
m = m + 1
R = m/s
if R > Rth_f(m): #fail test
UE failed
break
else: # if not error
if s > Tdp(m): #pass test
UE passed
break
else:
continue
Where Tdp(m) is a predefined time duration (in sample number) which gives R < Rth_p for (m +1)th error.
To make the test method complete, a table like the example below should be prepared:
m
Rth_p
Tdp
Rth_f
0
---
417
---
1
0.0024
540
---
2
0.0034
750
0.031
3
0.0040
800
0.025
…
…
…
…
Where, "---" means "any decision should not be made at that error count." For example, while no error is reported, any decisions should not be made based on the error rate, and only when the time duration in terms of the number of samples exceeds 417, DUT is decided to pass.
|
4131cd5fefdadf4075e3ec0993ed8fe4
|
34.901
|
10.2 Early Pass/Fail criteria
|
As described in the previous section, once the framework of the test procedure is established, a test can be defined by presenting the table of Rth_p, Tdp, and Rth_f. These values will be referred to as "Early Criteria."
The requirements for such criteria should be:
1) F should be 0.2 % for BER/BLER measurement, and 5% for RRM delay measurement and others.
2) The test shall be terminated in a finite time period (in terms of the number of samples.)
Presuming these requirements, we will try to minimize the test time.
There seems to be a broad freedom in choosing a set of early criteria. However, it would be a good start point to choose these criteria so that they provide a constant D at each error count m. When the error-count based and discrete distribution approach is presumed, the pdf to be used is a negative binomial distribution, nBi(s: r, m) = nBi(m/R: r, m). Then, equation (1) and (2) will be:
(3)
(4)
Using these equations, Rth_f and Rth_pcan be determined and from the latter, Tdp can also be derived. The problem is how to determine D. The target is clear; we have to get an F of 0.2%, but the population keeps changing in the procedure (some DUT hit either of the criteria and will be removed from the population), so it seems almost impossible to determine D to give a predefined F by an analytical method.
So an experimental method should be used to determine D values, which goes like:
1) Pick a D value
2) Calculate Rth_p and Rth_f
3) With a simulation with random generator of Bernoulli distribution of r = R0 and r = R0M, and the criteria above, determine F value.
4) If the resultant F value is not close enough to the target F value (total significance level), start over with step 1) with a slightly different D value.
In the experiments that give results shown in this report bisection method was sued to perform this iteration efficiently, anda binomial distribution generator was used as Bernoulli generator. The simulations have been done for a population of 100,000 devices.
The resultant criteria for typical parameters are given in the following tables.
Table 10‑1: Early Pass/Fail criteria for RRM delay measurements (R0=0.1, F=0.05)
m
Rth-p
Tdp
Rth-f
0
`---
32
---
1
0.03125000
46
---
2
0.04347826
57
---
3
0.05263158
68
0.75000000
4
0.05882353
79
0.50000000
5
0.06329114
89
0.41666667
6
0.06741573
98
0.35294118
7
0.07142857
108
0.30434783
8
0.07407407
117
0.28571429
9
0.07692308
126
0.26470588
10
0.07936508
136
0.25000000
11
0.08088235
144
0.23404255
12
0.08333333
153
0.22641509
13
0.08496732
162
0.21666667
14
0.08641975
171
0.21212121
15
0.08771930
179
0.20547945
16
0.08938547
188
0.20000000
17
0.09042553
197
0.19540230
18
0.09137056
205
0.18947368
19
0.09268293
213
0.18627451
20
0.09389671
222
0.18348624
21
0.09459459
230
0.18103448
22
0.09565217
238
0.17741935
23
0.09663866
247
0.17557252
24
0.09716599
255
0.17266187
25
0.09803922
263
0.17123288
26
0.09885932
271
0.16883117
27
0.09963100
280
0.16666667
28
0.10000000
288
0.16568047
29
0.10069444
296
0.16384181
30
0.10135135
304
0.16216216
31
0.10197368
312
0.16062176
32
0.10256410
320
0.15920398
33
0.10312500
328
0.15789474
34
0.10365854
336
0.15668203
35
0.10416667
344
0.15555556
36
0.10465116
352
0.15450644
37
0.10511364
360
0.15352697
38
0.10555556
368
0.15261044
39
0.10597826
376
0.15175097
40
0.10638298
383
0.15094340
41
0.10704961
391
0.15018315
42
0.10741688
399
0.14946619
43
0.10776942
407
0.14878893
44
0.10810811
415
0.14814815
45
0.10843373
423
0.14705882
46
0.10874704
430
0.14649682
47
0.10930233
438
0.14596273
48
0.10958904
446
0.14545455
49
0.10986547
454
0.14454277
50
0.11013216
462
0.14409222
51
0.11038961
469
0.14366197
52
0.11087420
477
0.14285714
53
0.11111111
484
0.14247312
54
0.11134021
492
0.14210526
55
0.11178862
500
0.14138817
56
0.11200000
508
0.14105793
57
0.11220472
516
0.14039409
58
0.11240310
523
0.14009662
59
0.11281071
531
0.13981043
60
0.11299435
539
0.13921114
61
0.11317254
546
0.13895216
62
0.11355311
554
0.13839286
63
0.11371841
561
0.13815789
64
0.11408200
569
0.13763441
65
0.11423550
577
0.13742072
66
0.11438475
584
0.13692946
67
0.11472603
592
0.13673469
68
0.11486486
600
0.13627255
69
0.11500000
607
0.13609467
70
0.11532125
615
0.13565891
71
0.11544715
622
0.13523810
72
0.11575563
630
0.13508443
73
0.11587302
637
0.13468635
74
0.11616954
645
0.13454545
75
0.11627907
653
0.13416816
76
0.11638591
660
0.13380282
77
0.11666667
668
0.13368056
78
0.11676647
675
0.13333333
79
0.11703704
683
0.13299663
80
0.11713031
690
0.13289037
81
0.11739130
698
0.13256956
82
0.11747851
705
0.13225806
83
0.11773050
713
0.13216561
84
0.11781206
720
0.13186813
85
0.11805556
728
0.13157895
86
0.11813187
735
0.13149847
87
0.11836735
743
0.13122172
88
0.11843876
750
0.13095238
89
0.11866667
758
0.13088235
90
0.11873351
765
0.13062409
91
0.11895425
773
0.13037249
92
0.11901682
780
0.13012730
93
0.11923077
788
0.13006993
94
0.11928934
795
0.12983425
95
0.11949686
803
0.12960437
96
0.11955168
810
0.12938005
97
0.11975309
818
0.12916112
98
0.11980440
825
0.12911726
99
0.12000000
832
0.12890625
100
0.12019231
840
0.12870013
101
0.12023810
847
0.12849873
102
0.12042503
855
0.12830189
103
0.12046784
862
0.12826899
104
0.12064965
870
0.12807882
105
0.12068966
877
0.12789281
106
0.12086659
884
0.12771084
107
0.12104072
892
0.12753278
108
0.12107623
899
0.12735849
109
0.12124583
907
0.12733645
110
0.12127894
914
0.12716763
111
0.12144420
922
0.12700229
112
0.12147505
929
0.12684032
113
0.12163617
936
0.12668161
114
0.12179487
944
0.12652608
115
0.12182203
951
0.12637363
116
0.12197687
958
0.12622416
117
0.12212944
966
0.12621359
118
0.12215321
973
0.12606838
119
0.12230216
981
0.12592593
120
0.12232416
988
0.12578616
121
0.12246964
995
0.12564901
122
0.12261307
1003
0.12551440
123
0.12263210
1010
0.12538226
124
0.12277228
1017
0.12525253
125
0.12291052
1025
0.12512513
126
0.12292683
1032
0.12500000
127
0.12306202
1039
0.12487709
128
0.12319538
1047
0.12475634
129
0.12320917
1054
0.12475822
130
0.12333966
1062
0.12464046
131
0.12335217
1069
0.12452471
132
0.12347989
1076
0.12441093
133
0.12360595
1084
0.12429907
134
0.12361624
1091
0.12418906
135
0.12373969
1098
0.12408088
136
0.12386157
1106
0.12397448
137
0.12386980
1113
0.12386980
The resultant thresholds (criteria) for BER, BLER and RRM Delay measurements are shown in Figure 10‑1,Figure 10‑2 and Figure 10‑3 respectively.
Figure 10‑1: Early Pass/Fail Criteria for BER Measurement (R0= 0.001, F = 0.002)
Figure 10‑2: Early Pass/Fail Criteria for BLER Measurement ( R0= 0.01, F = 0.002)
Figure 10‑3: Early Pass/Fail for RRM Delay Measurement (R0 = 0.1, F = 0.05)
|
4131cd5fefdadf4075e3ec0993ed8fe4
|
34.901
|
10.3 Estimating the test time
|
The final goal of this report is to reduce the total test time, so each set of criteria should be rated by estimating its test time. This task is fairly simple, and can be done using almost the same simulation program used in the previous section. The sample number at which the measurement is terminated is recorded and averaged to give the test time of the criteria. The estimation is repeated for various average error rates, r.
Some results are shown in Figure 10‑4 through Figure 10‑6.
<Figure will be inserted here>
Figure 10‑4: Test Time for BLER Measurement (R0 = 0.001, F = 0.002)
Figure 10‑5: Test Time for BLER Measurement (R0 = 0.01, F = 0.002)
Figure 10‑6: Test Time for RRM Delay Measurement (R0 = 0.1, F = 0.05)
In Figure 10‑7, the BLER test time (Figure 10‑5) is compared with a conventional test method, in which no early criteria are not incorporated. Apparently, the required test time is the time to see 200 errors, and the test time is simply given by 200/r. It should be noted that the averaged test time values almost coincide at r = 0.0124, and the early criteria greatly reduce the test time in the areas of r << 0.0124 and of r >> 0.0124.
Figure 10‑7: Reduction of the Test Time by Employing Early Criteria
|
4131cd5fefdadf4075e3ec0993ed8fe4
|
34.901
|
10.4 Minimizing the test time
|
The criteria set in Section 10.2 ("constant D criteria") were chosen in an arbitrary way, and there is no guarantee that that should give the shortest test time. So, to seek the optimized method, some other criteria that are basically a variation of the "constant D criteria" will be proposed and their test time will be estimated in the following subsections.
|
4131cd5fefdadf4075e3ec0993ed8fe4
|
34.901
|
10.4.1 Truncation
|
As a matter of fact, while the simple test is terminated at m = 200 (F = 0.2%), the cross point of the criteria is located at m = 374 for the "constant D criteria." Moreover, only few DUTs survive till later stages. So, it seems to be a good idea truncating the procedure at a little earlier stage.
This means that Rth_p and Rth_f should be to 1.24 R0 at an m value between 200 and 374 in the criteria table. The change is simple, but has an impact on F, so we have to re-evaluate D.
The results are shown in Figure 10‑8, which shows no significant improvement in the test time, while the (rare) maximum test time of 1000 samples were reduced to 700 as shown in Figure 10‑9. In the figure, the red solid line shows the average test time as a reference.
Figure 10‑8: Test Time by Truncated Early Criteria (R0 = 0.1, F = 0.05)
Figure 10‑9: Maxim Test Time by Truncated Early Criteria (R0 = 0.1, F = 0.05)
|
4131cd5fefdadf4075e3ec0993ed8fe4
|
34.901
|
10.4.2 Decimation
|
If the simple tests (comparisons) were done at a smaller number of m, that would result in a smaller F value. In other words, a larger D value can be employed for a certain F value, this can mean a reduced test time since the criteria curve with a higher D value will be located at more left hand side. The real expected advantages are that such method can be presented by a much shorter table and that it may reduce the requirement for calculation.
The resultant test time values for the criteria with every m, and one with every 10 m points are shown in Figure 10‑10. Apparently, the decimation didn’t improve the test time.
Figure 10‑10: Test time of 1:10 Decimated Criteria (R0 = 0.1, F = 0.05)
|
4131cd5fefdadf4075e3ec0993ed8fe4
|
34.901
|
10.4.3 Composite criteria
|
The results of the simulation stated above show that almost all the devices are picked out at relatively early stages, and only few reach later stages. This observation would lead to the idea that the test time could be reduced if stricter criteria at the later stages. Suppose that each threshold in a criterion, Rth-fix(m) / Rth_var(m), have individual significance levels, Dfix / Dvar, respectively. A new criteria, Rth_com(m), can be built by merging these two criteria like,
That is, the composite threshold, Rth-com, is a weighted average of Rth-fix and Rth-var. Where mmax corresponds to the cross point of pass/fail criteria of Rth-fix , and Dfix > Dvar.
The experimental method to find D value witch meets the requirement for F was introduced in Section 10.2, and it can be easily modified to find Dvar for a fixed Dfix. The resultant D values are listed in Table 10‑2.
Table 10‑2: Combinations of D values to give F = 0.05 (R0 = 0.1)
Dfix
Dvar
mmax
Comment
---
0.005125
143
Reference
0.01
0.00425
118
0.02
0.00250
92
0.03
0.001125
77
Using these criteria, the test time was evaluated, and the results are shown in Figure 10‑11. In comparison with the reference ("Constant D" criteria), the composite criteria give a slightly shorter test time in the worst case (r = 1.24 R0), but these give a longer test time for r < R0.
Figure 10‑11: Test time for composite test criteria
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.