Tech-invite3GPPspaceIETF RFCsSIP
in Index   Prev   Next

RFC 4730

A Session Initiation Protocol (SIP) Event Package for Key Press Stimulus (KPML)

Pages: 56
Proposed Standard
Part 1 of 3 – Pages 1 to 21
None   None   Next

Top   ToC   RFC4730 - Page 1
Network Working Group                                          E. Burger
Request for Comments: 4730                      Cantata Technology, Inc.
Category: Standards Track                                       M. Dolly
                                                               AT&T Labs
                                                           November 2006

           A Session Initiation Protocol (SIP) Event Package
                     for Key Press Stimulus (KPML)

Status of This Memo

   This document specifies an Internet standards track protocol for the
   Internet community, and requests discussion and suggestions for
   improvements.  Please refer to the current edition of the "Internet
   Official Protocol Standards" (STD 1) for the standardization state
   and status of this protocol.  Distribution of this memo is unlimited.

Copyright Notice

   Copyright (C) The IETF Trust (2006).


This document describes a SIP Event Package "kpml" that enables monitoring of Dual Tone Multi-Frequency (DTMF) signals and uses Extensible Markup Language (XML) documents referred to as Key Press Markup Language (KPML). The kpml Event Package may be used to support applications consistent with the principles defined in the document titled "A Framework for Application Interaction in the Session Initiation Protocol (SIP)". The event package uses SUBSCRIBE messages and allows for XML documents that define and describe filter specifications for capturing key presses (DTMF Tones) entered at a presentation-free User Interface SIP User Agent (UA). The event package uses NOTIFY messages and allows for XML documents to report the captured key presses (DTMF tones), consistent with the filter specifications, to an Application Server. The scope of this package is for collecting supplemental key presses or mid-call key presses (triggers).
Top   ToC   RFC4730 - Page 2

Table of Contents

1. Introduction ....................................................4 1.1. Conventions Used in This Document ..........................5 2. Protocol Overview ...............................................5 3. Key Concepts ....................................................6 3.1. Subscription Duration ......................................6 3.2. Timers .....................................................7 3.3. Pattern Matches ............................................8 3.4. Digit Suppression .........................................12 3.5. User Input Buffer Behavior ................................14 3.6. DRegex ....................................................16 3.6.1. Overview ...........................................16 3.6.2. Operation ..........................................18 3.7. Monitoring Direction ......................................20 3.8. Multiple Simultaneous Subscriptions .......................20 4. Event Package Formal Definition ................................21 4.1. Event Package Name ........................................21 4.2. Event Package Parameters ..................................21 4.3. SUBSCRIBE Bodies ..........................................22 4.4. Subscription Duration .....................................22 4.5. NOTIFY Bodies .............................................22 4.6. Subscriber Generation of SUBSCRIBE Requests ...............22 4.7. Notifier Processing of SUBSCRIBE Requests .................23 4.8. Notifier Generation of NOTIFY Requests ....................25 4.9. Subscriber Processing of NOTIFY Requests ..................27 4.10. Handling of Forked Requests ..............................28 4.11. Rate of Notifications ....................................28 4.12. State Agents and Lists ...................................28 4.13. Behavior of a Proxy Server ...............................29 5. Formal Syntax ..................................................29 5.1. DRegex ....................................................29 5.2. KPML Request ..............................................30 5.3. KPML Response .............................................33 6. Enumeration of KPML Status Codes ...............................34 7. IANA Considerations ............................................34 7.1. SIP Event Package Registration ............................34 7.2. MIME Media Type application/kpml-request+xml ..............35 7.3. MIME Media Type application/kpml-response+xml .............35 7.4. URN Sub-Namespace Registration for urn:ietf:xml:ns:kpml-request ..............................35 7.5. URN Sub-Namespace Registration for urn:ietf:xml:ns:kpml-response .............................36 7.6. KPML Request Schema Registration ..........................37 7.7. KPML Response Schema Registration .........................37 8. Security Considerations ........................................37 9. Examples .......................................................38 9.1. Monitoring for Octothorpe .................................38
Top   ToC   RFC4730 - Page 3
      9.2. Dial String Collection ....................................39
   10. Call Flow Examples ............................................40
      10.1. Supplemental Digits ......................................40
      10.2. Multiple Applications ....................................45
   11. References ....................................................52
      11.1. Normative References .....................................52
      11.2. Informative References ...................................53
   Appendix A.  Contributors .........................................54
   Appendix B.  Acknowledgements .....................................54
Top   ToC   RFC4730 - Page 4

1. Introduction

This document describes a SIP Event Package "kpml" that enables monitoring of key presses and utilizes XML documents referred to as Key Press Markup Language (KPML). KPML is a markup [14] that enables presentation-free User Interfaces as described in the Application Interaction Framework [15]. The Key Press Stimulus Package is a SIP Event Notification Package [5] that uses the SUBSCRIBE and NOTIFY methods of SIP. The subscription filter and notification report bodies use the Keypad Markup Language, KPML. The "kpml" event package requires the definition of two new MIME types, two new URN sub-namespaces, and two schemas for the KPML Request and the KPML Response. The scope of this package is for collecting supplemental key presses or mid-call key presses (triggers). This capability allows an Application Server service provider to monitor (filter) for a set of DTMF patterns at a SIP User Agent located in either an end-user device or a gateway. In particular, the "kpml" event package enables "dumb phones" and "gateways" that receive signals from dumb phones to report user key- press events. Colloquially, this mechanism provides for "digit reporting" or "Dual Tone Multi-Frequency (DTMF) reporting." The capability eliminates the need for "hair-pinning" (routing media into and then out of the same device) through a Media Server or duplicating all the DTMF events, when an Application Server needs to trigger mid-call service processing on DTMF digit patterns. A goal of KPML is to fit in an extremely small memory and processing footprint. The name of the XML document, KPML, reflects its legacy support role. The public switched telephony network (PSTN) accomplished signaling by transporting DTMF tones in the bearer channel (in-band signaling) from the user terminal to the local exchange. Voice-over-IP networks transport in-band signals with actual DTMF waveforms or RFC 2833 [10] packets. In RFC 2833, the signaling application inserts RFC 2833 named signal packets as well as, or instead of, generating tones in the media path. The receiving application receives the signal information in the media stream. RFC 2833 tones are ideal for conveying telephone-events point-to- point in a Real-time Transport Protocol (RTP) stream, as in the context of straightforward sessions like a 2-party call or a simple, centrally mixed conference. However, there are other environments where additional or alternative requirements are needed. These other environments include protocol translation and complex call control.
Top   ToC   RFC4730 - Page 5
   An interested application could request notifications of every key
   press.  However, many of the use cases for such signaling show that
   most applications are interested in only one or a few keystrokes.
   Thus a mechanism is needed for specifying to the user's interface
   what stimuli the application requires.

1.1. Conventions Used in This Document

RFC 2119 [1] provides the interpretations for the key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" found in this document. The Application Interaction Framework document [15] provides the interpretations for the terms "User Device", "SIP Application", and "User Input". This document uses the term "Application" and "Requesting Application" interchangeably with "SIP Application". Additionally, the Application Interaction Framework document discusses User Device Proxies. A common instantiation of a User Device Proxy is a Public Switched Telephone Network (PSTN) gateway. Because the normative behavior of a presentation-free User Interface is identical for a presentation-free SIP User Agent and a presentation-free User Device Proxy, this document uses "User Device" for both cases.

2. Protocol Overview

The "kpml" event package uses explicit subscription notification requests using the SIP SUBSCRIBE and NOTIFY methods. An Application that wants to collect digits creates an application/kpml-request+xml document with the digit patterns of interest to the Application and places this document in its SUBSCRIBE request. SIP SUBSCRIBE messages are routed to the User Interface using standard SIP request routing. KPML Subscriptions do not fork. The KPML request contained in the SUBSCRIBE message identifies the target media stream by referencing the dialog identifiers corresponding to the session responsible for the media stream. Once a subscription is established, the User Interface sends application/kpml-response+xml documents in NOTIFY requests when digits are collected or when timeouts or errors occur. A KPML subscription can be persistent or one-shot. Persistent requests are active until the subscription terminates, the Application replaces the request, the Application deletes the request by sending a null document on the dialog, or the Application explicitly deletes the subscription by sending a SUBSCRIBE with an expires value of zero (0).
Top   ToC   RFC4730 - Page 6
   One-shot requests terminate the subscription upon the receipt of DTMF
   values that provide a match.  The "persist" KPML element specifies
   whether the subscription remains active for the duration specified in
   the SUBSCRIBE message or if it automatically terminates upon a
   pattern match.

   NOTIFY messages can contain XML documents.  If the User Interface
   matches a digitmap, the NOTIFY message (response) contains an XML
   document that indicates the User Input detected and whether the User
   Interface suppressed the representation of User Input, such as tones,
   or RFC 2833, from the media streams.  If the User Interface
   encountered an error condition, such as a timeout, this will also be

3. Key Concepts

3.1. Subscription Duration

KPML recognizes two types of subscriptions: one-shot and persistent. Persistent subscriptions have two sub-types: continuous notify and single-notify. One-shot subscriptions terminate after a pattern match occurs and a report is issued in a NOTIFY message. If the User Interface detects a key press stimulus that triggers a one-shot KPML event, then the User Interface (notifier) MUST set the "Subscription-State" in the NOTIFY message to "terminated". At this point, the User Interface MUST consider the subscription expired. Persistent subscriptions remain active at the User Interface, even after a match. For continuous-notify persistent subscriptions, the User Interface will emit a NOTIFY message whenever the User Input matches a subscribed pattern. For single-notify persistent subscriptions, the user device will emit a NOTIFY message at the first match, but will not emit further NOTIFY messages until the Application issues a new subscription request on the subscription dialog. NOTE: The single-notify persistent subscription enables lock-step (race-free) quarantining of User Input between different digit maps. The "persist" attribute to the <pattern> tag in the KPML subscription body affects the lifetime of the subscription. If the "persist" attribute is "one-shot", then once there is a match (or no match is possible), the subscription ends after the User Interface notifies the Application.
Top   ToC   RFC4730 - Page 7
   If the "persist" attribute is "persist" or "single-notify", then the
   subscription ends when the Application explicitly ends it or the User
   Interface terminates the subscription.

   If the User Interface does not support persistent subscriptions, it
   returns a NOTIFY message with the KPML status code set to 531.  If
   there are digits in the buffer and the digits match an expression in
   the SUBSCRIBE filter, the User Interface prepares the appropriate
   NOTIFY response message.

   The values of the "persist" attribute are case sensitive.

3.2. Timers

To address the various key press collection scenarios, three timers are defined. They are the extra, critical, and inter-digit timers. o The inter-digit timer is the maximum time to wait between digits. Note: unlike Media Gateway Control Protocol (MGCP) [11] or H.248 [12], there is no start timer, as that concept does not apply in the KPML context. o The critical timer is the time to wait for another digit if the collected digits can match more than one potential pattern. o The extra timer is the time to wait for another digit if the collected digits can only match one potential pattern, but a longer match for this pattern is possible. The User Interface MAY support an inter-digit timeout value. This is the amount of time the User Interface will wait for User Input before returning a timeout error result on a partially matched pattern. The application can specify the inter-digit timeout as an integer number of milliseconds by using the "interdigittimer" attribute to the <pattern> tag. The default is 4000 milliseconds. If the User Interface does not support the specification of an inter-digit timeout, the User Interface MUST silently ignore the specification. If the User Interface supports the specification of an inter-digit timeout, but not to the granularity specified by the value presented, the User Interface MUST round up the requested value to the closest value it can support. The purpose of the inter-digit timeout is to protect applications from starting to match a pattern, yet never returning a result. This can occur, for example, if the user accidentally enters a key that begins to match a pattern. However, since the user accidentally entered the key, the rest of the pattern never comes. Moreover, when the user does enter a pattern, since they have already entered a key,
Top   ToC   RFC4730 - Page 8
   the pattern may not match or may not match as expected.  Likewise,
   consider the case where the user thinks they entered a key press, but
   the User Interface does not detect the key.  This could occur when
   collecting ten digits, but the device actually only receives 9.  In
   this case, the User Interface will wait forever for the tenth key
   press, while the user becomes frustrated wondering why the
   application is not responding.

   The User Interface MAY support a critical-digit timeout value.  This
   is the amount of time the User Interface will wait for another key
   press when it already has a matched <regex> but there is another,
   longer <regex> that may also match the pattern.  The application can
   specify the critical-digit timeout as an integer number of
   milliseconds by using the "criticaldigittimer" attribute to the
   <pattern> tag.  The default is 1000 milliseconds.

   The purpose of the critical-digit timeout is to allow the application
   to collect longer matches than the shortest presented.  This is
   unlike MGCP [11], where the shortest match gets returned.  For
   example, if the application registers for the patterns "0011", "011",
   "00", and "0", the critical-digit timeout enables the User Interface
   to distinguish between "0", "00", "011", and "0011".  Without this
   feature, the only value that the User Interface can detect is "0".

   The User Interface MAY support an extra-digit timeout value.  This is
   the amount of time the User Interface will wait for another key press
   when it already has matched the longest <regex>.  The application can
   specify the extra-digit timeout as an integer number of milliseconds
   by using the "extradigittimer" attribute to the <pattern> tag.  The
   default is 500 milliseconds.  If there is no enterkey specified, then
   the User Interface MAY default the exteradigittimer to zero.

   The purpose of the extra-digit timeout is to allow the User Interface
   to collect the enterkey.  Without this feature, the User Interface
   would match the pattern, and the enterkey would be buffered and
   returned as the next pattern.

3.3. Pattern Matches

During the subscription lifetime, the User Interface may detect a key press stimulus that triggers a KPML event. In this case, the User Interface (notifier) MUST return the appropriate KPML document. The pattern matching logic works as follows. KPML User Interfaces MUST follow the logic presented in this section so that different implementations will perform deterministically on the same KPML document given the same User Input.
Top   ToC   RFC4730 - Page 9
   A kpml request document contains a <pattern> element with a series of
   <regex> tags.  Each <regex> element specifies a potential pattern for
   the User Interface to match.  The Section 5.1 describes the DRegex,
   or digit regular expression, language.

   The pattern match algorithm matches the longest regular expression.
   This is the same mode as H.248.1 [12] and not the mode presented by
   MGCP [11].  The pattern match algorithm choice has an impact on
   determining when a pattern matches.  Consider the following KPML

   <?xml version="1.0" encoding="UTF-8"?>
   <kpml-request xmlns="urn:ietf:params:xml:ns:kpml-request"
           "urn:ietf:params:xml:ns:kpml-request kpml-request.xsd"

                         Figure 1: Greedy Matching

   In Figure 1, if we were to match on the first found pattern, the
   string "011" would never match.  This happens because the "0" rule
   would match first.

   While this behavior is what most applications desire, it does come at
   a cost.  Consider the following KPML document snippet.


                        Figure 2: Timeout Matching

   Figure 2 shows a typical North American dial plan.  From an
   application perspective, users expect a seven-digit number to respond
   quickly, not waiting the typical inter-digit critical timer (usually
   four seconds).  Conversely, the user does not want the system to cut
   off their ten-digit number at seven digits because they did not enter
   the number fast enough.

   One approach to this problem is to have an explicit dial string
   terminator.  Often, it is the pound key (#).  Now, consider the
   following snippet.
Top   ToC   RFC4730 - Page 10

                   Figure 3: Timeout Matching with Enter

   The problem with the approach in Figure 3 is that the "#" will appear
   in the returned dial string.  Moreover, one often wants to allow the
   user to enter the string without the dial string termination key.  In
   addition, using explicit matching on the key means one has to double
   the number of patterns, e.g., "x{7}", "x{7}#", "x{10}", and "x{10}#".

   The approach used in KPML is to have an explicit "Enter Key", as
   shown in the following snippet.

   <pattern enterkey="#">

                 Figure 4: Timeout Matching with Enter Key

   In Figure 4, the enterkey attribute to the <pattern> tag specifies a
   string that terminates a pattern.  In this situation, if the user
   enters seven digits followed by the "#" key, the pattern matches (or
   fails) immediately.  KPML indicates a terminated nomatch with a KPML
   status code 402.

      NOTE: The enterkey is a string.  The enterkey can be a sequence of
      key presses, such as "**".

   Some patterns look for long-duration key presses.  For example, some
   applications look for long "#" or long "*".

   KPML uses the "L" modifier to <regex> characters to indicate long key
   presses.  The following KPML document looks for a long pound of at
   least 3 seconds.
Top   ToC   RFC4730 - Page 11
   <?xml version="1.0" encoding="UTF-8"?>
   <kpml-request xmlns="urn:ietf:params:xml:ns:kpml-request"
           "urn:ietf:params:xml:ns:kpml-request kpml-request.xsd"
     <pattern long="3000">

                                Long Pound

   The request can specify what constitutes "long" by setting the long
   attribute to the <pattern>.  This attribute is an integer
   representing the number of milliseconds.  If the user presses a key
   for longer than "long" milliseconds, the Long modifier is true.  The
   default length of the long attribute is 2500 milliseconds.

   User Interfaces MUST distinguish between long and short input when
   the KPML document specifies both in a document.  However, if there is
   not a corresponding long key press pattern in a document, the User
   Interface MUST match the key press pattern irrespective of the length
   of time the user presses the key.

   As an example, in the following snippet in Figure 6, the User
   Interface discriminates between a long "*" and a normal "*", but any
   length "#" will match the pattern.

     <regex tag="short_star">*</regex>
     <regex tag="long_star">L*</regex>

                     Figure 6: Long and Short Matching

   Some User Interfaces are unable to present long key presses.  An
   example is an old private branch exchange (PBX) phone set that emits
   fixed-length tones when the user presses a key.  To address this
   issue, the User Interface MAY interpret a succession of presses of a
   single key to be equivalent to a long key press of the same key.  The
   Application indicates it wants this behavior by setting the
   "longrepeat" attribute to the <pattern> to "true".

   The KPML document specifies if the patterns are to be persistent by
   setting the "persist" attribute to the <pattern> tag to "persist" or
   "single-notify".  Any other value, including "one-shot", indicates
Top   ToC   RFC4730 - Page 12
   the request is a one-shot subscription.  If the User Interface does
   not support persistent subscriptions, it returns a KPML document with
   the KPML status code set to 531.  If there are digits in the buffer
   and the digits match an expression in the KPML document, the User
   Interface emits the appropriate kpml notification.

   Note the values of the "persist" attribute are case sensitive.

   Some User Interfaces may support multiple regular expressions in a
   given pattern request.  In this situation, the application may wish
   to know which pattern triggered the event.

   KPML provides a "tag" attribute to the <regex> tag.  The "tag" is an
   opaque string that the User Interface sends back in the notification
   report upon a match in the digit map.  In the case of multiple
   matches, the User Interface MUST choose the longest match in the KPML
   document.  If multiple matches match the same length, the User
   Interface MUST choose the first expression listed in the subscription
   KPML document based on KPML document order.

   If the User Interface cannot support multiple regular expressions in
   a pattern request, the User Interface MUST return a KPML document
   with the KPML status code set to 532.  If the User Interface cannot
   support the number of regular expressions in the pattern request, the
   User Interface MUST return a KPML document with the KPML status code
   set to 534.

      NOTE: We could mandate a minimum number of regular expressions
      that a User Interface must support per subscription request and
      globally.  However, such minimums tend to become designed-in,
      hard-coded limits.  For guidance, one should be able to easily
      handle tens of expressions per subscription and thousands
      globally.  A good implementation should have effectively no
      limits.  That said, to counter possible denial-of-service attacks,
      implementers of User Interfaces should be aware of the 534 and 501
      status codes and feel free to use them.

3.4. Digit Suppression

Under basic operation, a KPML User Interface will transmit in-band tones (RFC 2833 [10] or actual tone) in parallel with User Input reporting. NOTE: If KPML did not have this behavior, then a User Interface executing KPML could easily break called applications. For example, take a personal assistant that uses "*9" for attention. If the user presses the "*" key, KPML will hold the digit, looking for the "9". What if the user just enters a "*" key, possibly
Top   ToC   RFC4730 - Page 13
      because they accessed an interactive voice response (IVR) system
      that looks for "*"?  In this case, the "*" would get held by the
      User Interface, because it is looking for the "*9" pattern.  The
      user would probably press the "*" key again, hoping that the
      called IVR system just did not hear the key press.  At that point,
      the User Interface would send both "*" entries, as "**" does not
      match "*9".  However, that would not have the effect the user
      intended when they pressed "*".

   On the other hand, there are situations where passing through tones
   in-band is not desirable.  Such situations include call centers that
   use in-band tone spills to initiate a transfer.

   For those situations, KPML adds a suppression tag, "pre", to the
   <regex> tag.  There MUST NOT be more than one <pre> tag in any given
   <regex> tag.

   If there is only a single <pattern> and a single <regex>, suppression
   processing is straightforward.  The end-point passes User Input until
   the stream matches the regular expression <pre>.  At that point, the
   User Interface will continue collecting User Input, but will suppress
   the generation or pass-through of any in-band User Input.

   If the User Interface suppressed stimulus, it MUST indicate this by
   including the attribute "suppressed" with a value of "true" in the

   Clearly, if the User Interface is processing the KPML document
   against buffered User Input, it is too late to suppress the
   transmission of the User Input, as the User Interface has long sent
   the stimulus.  This is a situation where there is a <pre>
   specification, but the "suppressed" attribute will not be "true" in
   the notification.  If there is a <pre> tag that the User Interface
   matched and the User Interface is unable to suppress the User Input,
   it MUST set the "suppressed" attribute to "false".

   A KPML User Interface MAY perform suppression.  If it is not capable
   of suppression, it ignores the suppression attribute.  It MUST set
   the "suppressed" attribute to "false".  In this case, the pattern to
   match is the concatenated pattern of pre+value.

   At some point in time, the User Interface will collect enough User
   Input to the point it matches a <pre> pattern.  The interdigittimer
   attribute indicates how long to wait for the user to enter stimulus
   before reporting a time-out error.  If the interdigittimer expires,
   the User Interface MUST issue a time-out report, transmit the
   suppressed User Input on the media stream, and stop suppression.
Top   ToC   RFC4730 - Page 14
   Once the User Interface detects a match and it sends a NOTIFY request
   to report the User Input, the User Interface MUST stop suppression.
   Clearly, if subsequent User Input matches another <pre> expression,
   then the User Interface MUST start suppression.

   After suppression begins, it may become clear that a match will not
   occur.  For example, take the expression


   At the point the User Interface receives "*8", it will stop
   forwarding stimulus.  Let us say that the next three digits are
   "408".  If the next digit is a zero or one, the pattern will not

      NOTE: It is critically important for the User Interface to have a
      sensible inter-digit timer.  This is because an errant dot (".")
      may suppress digit sending forever.

   Applications should be very careful to indicate suppression only when
   they are fairly sure the user will enter a digit string that will
   match the regular expression.  In addition, applications should deal
   with situations such as no-match or time-out.  This is because the
   User Interface will hold digits, which will have obvious User
   Interface issues in the case of a failure.

3.5. User Input Buffer Behavior

User Interfaces MUST buffer User Input upon receipt of an authenticated and accepted subscription. Subsequent KPML documents apply their patterns against the buffered User Input. Some applications use modal interfaces where the first few key presses determine what the following key presses mean. For a novice user, the application may play a prompt describing what mode the application is in. However, "power users" often barge through the prompt. User Interfaces MUST NOT provide a subscriber with digits that were detected prior to the authentication and authorization of that subscriber. Without prohibition, a subscriber might be able to gain access to calling card or other information that predated the subscriber's participation in the call. Note that this prohibition MUST be applied on a per-subscription basis. KPML provides a <flush> tag in the <pattern> element. The default is not to flush User Input. Flushing User Input has the effect of ignoring key presses entered before the installation of the KPML subscription. To flush User Input, include the tag
Top   ToC   RFC4730 - Page 15
   <flush>yes</flush> in the KPML subscription document.  Note that this
   directive affects only the current subscription dialog/id

   Lock-step processing of User Input is where the User Interface issues
   a notification, the Application processes the notification while the
   User Interface buffers additional User Input, the Application
   requests more User Input, and only then does the User Interface
   notify the Application based on the collected User Input.  To direct
   the User Interface to operate in lock-step mode, set the <pattern>
   attribute persist="single-notify".

   The User Interface MUST be able to process <flush>no</flush>.  This
   directive is effectively a no-op.

   Other string values for <flush> may be defined in the future.  If the
   User Interface receives a string it does not understand, it MUST
   treat the string as a no-op.

   If the user presses a key that cannot match any pattern within a
   <regex> tag, the User Interface MUST discard all buffered key presses
   up to and including the current key press from consideration against
   the current or future KPML documents on a given dialog.  However, as
   described above, once there is a match, the User Interface buffers
   any key presses the user entered subsequent to the match.

      NOTE: This behavior allows applications to receive only User Input
      that is of interest to them.  For example, a pre-paid application
      only wishes to monitor for a long pound.  If the user enters other
      stimulus, presumably for other applications, the pre-paid
      application does not want notification of that User Input.  This
      feature is fundamentally different than the behavior of Time
      Division Multiplexer (TDM)-based equipment where every application
      receives every key press.

   To limit reports to only complete matches, set the "nopartial"
   attribute to the <pattern> tag to "true".  In this case, the User
   Interface attempts to match a rolling window over the collected User

   KPML subscriptions are independent.  Thus, it is not possible for the
   current document to know if a following document will enable barging
   or want User Input flushed.  Therefore, the User Interface MUST
   buffer all User Input, subject to the forced_flush caveat described
Top   ToC   RFC4730 - Page 16
   On a given SUBSCRIBE dialog with a given id, the User Interface MUST
   buffer all User Input detected between the time of the report and the
   receipt of the next document, if any.  If the next document indicates
   a buffer flush, then the interpreter MUST flush all collected User
   Input from consideration from KPML documents received on that dialog
   with the given event id.  If the next document does not indicate
   flushing the buffered User Input, then the interpreter MUST apply the
   collected User Input (if possible) against the digit maps presented
   by the script's <regex> tags.  If there is a match, the interpreter
   MUST follow the procedures in Section 5.3.  If there is no match, the
   interpreter MUST flush all of the collected User Input.

   Given the potential for needing an infinite buffer for User Input,
   the User Interface MAY discard the oldest User Input from the buffer.
   If the User Interface discards digits, when the User Interface issues
   a KPML notification, it MUST set the forced_flush attribute of the
   <response> tag to "true".  For future use, the Application MUST
   consider any non-null value, other than "false", that it does not
   understand to be the same as "true".

      NOTE: The requirement to buffer all User Input for the entire
      length of the session is not onerous under normal operation.  For
      example, if one has a gateway with 8,000 sessions, and the gateway
      buffers 50 key presses on each session, the requirement is only
      400,000 bytes, assuming one byte per key press.

   Unless there is a suppress indicator in the digit map, it is not
   possible to know if the User Input is for local KPML processing or
   for other recipients of the media stream.  Thus, in the absence of a
   suppression indicator, the User Interface transmits the User Input to
   the far end in real time, using either RFC 2833, generating the
   appropriate tones, or both.

3.6. DRegex

3.6.1. Overview

This subsection is informative in nature. The Digit REGular EXpression (DRegex) syntax is a telephony-oriented mapping of POSIX Extended Regular Expressions (ERE) [13]. KPML does not use full POSIX ERE for the following reasons. o KPML will often run on high density or extremely low power and memory footprint devices.
Top   ToC   RFC4730 - Page 17
   o  Telephony application convention uses the star symbol ("*") for
      the star key and "x" for any digit 0-9.  Requiring the developer
      to escape the star ("\*") and expand the "x" ("[0-9]") is error
      prone.  This also leads DRegex to use the dot (".") to indicate
      repetition, which was the function of the unadorned star in POSIX

   o  Implementation experience with MGCP [11] and H.248.1 [12] has been
      that implementers and users have a hard time understanding the
      precedence of the alternation operator ("|").  This is due both to
      an under-specification of the operator in those documents and
      conceptual problems for users.  Thus, the SIPPING Working Group
      concluded that DRegex should not support alternation.  That said,
      each KPML <pattern> element may contain multiple regular
      expressions (<regex> elements).  Thus, it is straightforward to
      have pattern alternatives (use multiple <regex> elements) without
      the problems associated with the alternation operator ("|").
      Thus, DRegex does not support the POSIX alternation operator.

   o  DRegex includes character classes (characters enclosed in square
      brackets).  However, the negation operator inside a character
      class only operates on numbers.  That is, a negation class
      implicitly includes A-D, *, and #.  Including A-D, *, and # in a
      negation operator is a no-op.  Those familiar with POSIX would
      expect negation of the digits 4 and 5 (e.g., "[^45]") to include
      all other characters (including A-D, R, *, and #), while those
      familiar with telephony digit maps would expect negation to
      implicitly exclude non-digit characters.  Since the complete
      character set of DRegex is very small, constructing a negation
      class using A-D, R, *, and # requires the user to specify the
      positive inverse mapping.  For example, to specify all key
      presses, including A-D and *, except #, the specification would be
      "[0-9A-D*]" instead of "[^#R]".

   The following table shows the mapping from DRegex to POSIX ERE.

                          | DRegex | POSIX ERE |
                          | *      | \*        |
                          | .      | *         |
                          | x      | [0-9]     |
                          | [xc]   | [0-9c]    |

                   Table 1: DRegex to POSIX ERE Mapping
Top   ToC   RFC4730 - Page 18
   The first substitution, which replaces a star for an escaped star, is
   because telephony application designers are used to using the star
   for the (very common) star key.  Requiring an escape sequence for
   this common pattern would be error prone.  In addition, the usage
   found in DRegex is the same as found in MGCP [11] and H.248.1 [12].

   Likewise, the use of the dot instead of star is common usage from
   MGCP and H.248.1, and reusing the star in this context would also be
   confusing and error prone.

   The "x" character is a common indicator of the digits 0 through 9.
   We use it here, continuing the convention.  Clearly, for the case
   "[xc]", where c is any character, the substitution is not a blind
   replacement of "[0-9]" for "x", as that would result in "[[0-9]c]",
   which is not a legal POSIX ERE.  Rather, the substitution for "[xc]"
   is "[0-9c]".

      NOTE: "x" does not include the characters *, #, R, or A through D.

   Users need to take care not to confuse the DRegex syntax with POSIX
   EREs.  They are NOT identical.  In particular, there are many
   features of POSIX EREs that DRegex does not support.

   As an implementation note, if one makes the substitutions described
   in the above table, then a standard POSIX ERE engine can parse the
   digit string.  However, the mapping does not work in the reverse
   (POSIX ERE to DRegex) direction.  DRegex only implements the
   normative behavior described below.

3.6.2. Operation

White space is removed before parsing DRegex. This enables sensible pretty printing in XML without affecting the meaning of the DRegex string. The following rules demonstrate the use of DRegex in KPML.
Top   ToC   RFC4730 - Page 19
   | Entity  | Matches                                                 |
   | c       | digits 0-9, *, #, R, and A-D (case insensitive)         |
   | *       | the * character                                         |
   | #       | the # character                                         |
   | R       | The R (Register Recall) key                             |
   | [c]     | Any character in selector                               |
   | [^d]    | Any digit (0-9) not in selector                         |
   | [r1-r2] | Any character in range from r1 to r2, inclusive         |
   | x       | Any digit 0-9                                           |
   | {m}     | m repetitions of previous pattern                       |
   | {m,}    | m or more repetitions of previous pattern               |
   | {,n}    | At most n (including zero) repetitions of previous      |
   |         | pattern                                                 |
   | {m,n}   | At least m and at most n repetitions of previous        |
   |         | pattern                                                 |
   | Lc      | Match the character c if it is "long"; c is a digit 0-9 |
   |         | and A-D, #, or *.                                       |

                              DRegex Entities

   For ranges, the A-D characters are disjoint from the 0-9 characters.
   If the device does not have an "R" key, the device MAY report a hook
   flash as an R character.

       | Example      | Description                                |
       | 1            | Matches the digit 1                        |
       | [179]        | Matches 1, 7, or 9                         |
       | [2-9]        | Matches 2, 3, 4, 5, 6, 7, 8, 9             |
       | [^15]        | Matches 0, 2, 3, 4, 6, 7, 8, 9             |
       | [02-46-9A-D] | Matches 0, 2, 3, 4, 6, 7, 8, 9, A, B, C, D |
       | x            | Matches 0, 1, 2, 3, 4, 5, 6, 7, 8, 9       |
       | *6[179#]     | Matches *61, *67, *69, or *6#              |
       | x{10}        | Ten digits (0-9)                           |
       | 011x{7,15}   | 011 followed by seven to fifteen digits    |
       | L*           | Long star                                  |

                              DRegex Examples
Top   ToC   RFC4730 - Page 20

3.7. Monitoring Direction

SIP identifies dialogs by their dialog identifier. The dialog identifier is the remote-tag, local-tag, and Call-ID entities defined in RFC 3261 [4]. One method of determining the dialog identifier, particularly for third-party applications, is the SIP Dialog Package [17]. For most situations, such as a monaural point-to-point call with a single codec, the stream to monitor is obvious. In such situations the Application need not specify which stream to monitor. But there may be ambiguity in specifying only the SIP dialog to monitor. The dialog may specify multiple SDP streams that could carry key press events. For example, a dialog may have multiple audio streams. Wherever possible, the User Interface MAY apply local policy to disambiguate which stream or streams to monitor. In order to have an extensible mechanism for identifying streams, the mechanism for specifying streams is as an element content to the <stream> tag. The only content defined today is the <stream>reverse</stream> tag. By default, the User Interface monitors key presses emanating from the User Interface. Given a dialog identifier of Call-ID, local-tag, and remote-tag, the User Interface monitors the key presses associated with the local-tag. In the media proxy case, and potentially other cases, there is a need to monitor the key presses arriving from the remote user agent. The optional <stream> element to the <request> tag specifies which stream to monitor. The only legal value is "reverse", which means to monitor the stream associated with the remote-tag. The User Interface MUST ignore other values. NOTE: The reason this is a tag is so individual stream selection, if needed, can be addressed in a backwards-compatible way. Further specification of the stream to monitor is the subject of future standardization.

3.8. Multiple Simultaneous Subscriptions

An Application MAY register multiple User Input patterns in a single KPML subscription. If the User Interface supports multiple, simultaneous KPML subscriptions, the Application installs the subscriptions either in a new SUBSCRIBE-initiated dialog or on an existing SUBSCRIBE-initiated dialog with a new event id tag. If the User Interface does not support multiple, simultaneous KPML
Top   ToC   RFC4730 - Page 21
   subscriptions, the User Interface MUST respond with an appropriate
   KPML status code.

   Some User Interfaces may support multiple key press event
   notification subscriptions at the same time.  In this situation, the
   User Interface honors each subscription individually and

   A SIP user agent may request multiple subscriptions on the same
   SUBSCRIBE dialog, using the id parameter to the kpml event request.

   One or more SIP user agents may request independent subscriptions on
   different SIP dialogs, although reusing the same dialog for multiple
   subscriptions is NOT RECOMMENDED.

   If the User Interface does not support multiple, simultaneous
   subscriptions, the User Interface MUST return a KPML document with
   the KPML status code set to 533 on the dialog that requested the
   second subscription.  The User Interface MUST NOT modify the state of
   the first subscription on account of the second subscription attempt.

(page 21 continued on part 2)

Next Section