Sei sulla pagina 1di 17

◆ IMS Network Signaling Peering:

Challenges and Proposal


Jean-Philippe Joseph

IP Multimedia Subsystem (IMS) network peering is a key enabler that will


help accelerate deployment of next-generation IMS-based networks. Today’s
early deployments of dispersed IMS networks require public switched
telephone network (PSTN)/public land mobile network (PLMN) bridges for
network interconnection between IMS islands. The PSTN/PLMN bridging
arrangement is inefficient, however, in that it results in unnecessary
settlements for the carriers. It further impedes the implementation of rich
multimedia and Voice over IP (VoIP)-related services that require end-to-end
Internet Protocol (IP) connectivity. Last, it perpetuates the reliance on the
existing PSTN/PLMN network for voice calls among subscribers served by
different IMS-based carriers. This paper analyzes in detail the IMS peering
challenges from the perspective of Session Initiation Protocol (SIP) signaling
peering. It discusses issues related to the routing of SIP messages, and
addressing, address resolution, and discovery of peering points for IMS
signaling peering. It further establishes that a new routing algorithm is
needed that will allow signaling peering points to dynamically discover the
“best” transit network among others for reaching a destination. In closing,
it presents a high-level IMS signaling routing process that includes, among
other benefits, support for number portability as a key function for
inter-carrier IMS peering. © 2008 Alcatel-Lucent.

Introduction
Globally, many service providers (SP), wireline voice, IMS networks essentially rely on the PSTN to
and wireless alike, are planning or already starting route voice calls among the IMS islands. That is, calls
to deploy next-generation IP Multimedia Subsystem among IMS subscribers served by different SPs are
(IMS) [1] networks to evolve from the embedded most likely routed through the PSTN, resulting in less
base public switched telephone network (PSTN). The efficient routing, an inability to support end-to-end
IMS networks will, among other things, support rich VoIP related services (e.g., wideband voice), unnec-
and robust multimedia services including Voice over essary settlements, and the undesirable impact of
Internet Protocol (VoIP) and, hence, enhance user perpetuating the PSTN.
experience. Various standards organizations have recently
These initial IMS networks, however, are being begun to tackle crucial aspects of IMS peering,
deployed in islands with no direct interconnections including policy, signaling, network, and security
among themselves. For instance, for narrowband peering. The International Telecommunication Union,

Bell Labs Technical Journal 12(4), 33–48 (2008) © 2008 Alcatel-Lucent. Published by Wiley Periodicals, Inc. Published
online in Wiley InterScience (www.interscience.wiley.com) • DOI: 10.1002/bltj.20265
Panel 1. Abbreviations, Acronyms, and Terms
3GPP—3rd Generation Partnership Project NP—Number portability
AAA—Authentication, authorization and NPA—Numbering plan area
accounting NPDB—Number portability database
BGCF—Breakout gateway control function OPEX—Operational expenditures
BGP—Border Gateway Protocol PbUI—Public user identity
CdP—Called party P-CSCF—Proxy CSCF
CSCF—Call session control function PDA—Personal digital assistant
DDDS—Dynamic delegation discovery system PF—Policy function
DNS—Domain name system PLMN—Public land mobile network
E2U—E.164 to URI Resolution PP—Peering point
ENUM—Electronic number mapping PrUI—Private user identity
GPRS—General packet radio service PSTN—Public switched telephone network
GRX—GPRS roaming exchange QoS—Quality of service
GSM—Global System for Mobile RACF—Resource and admission control function
Communications RFC—Request for comments
HSS—Home subscriber server S-CSCF—Serving CSCF
IBCF—Interworking border control function SCTP—Stream Control Transmission Protocol
I-CSCF—Interrogating CSCF SF—Signaling function
IETF—Internet Engineering Task Force SIP—Session Initiation Protocol
IMS—IP Multimedia Subsystem SLA—Service level agreements
IN—Intelligent network SLF—Subscriber location function
IP—Internet Protocol SM—Session manager
ISDN—Integrated services digital network SP—Service provider
ISP—Internet service provider SPEERMINT—Session Peering for Multimedia
ISUP—ISDN user part Interworking
ITU—International Telecommunication Union SPNP—Service provider number portability
ITU-T—ITU Telecommunication Standardization SRV—Service record
Sector SS7—Signaling System 7
LD—Long distance TAS—Telephony application server
LERG—Local exchange routing guide TCP—Transport Control Protocol
LF—Location function TEL—Telephone
LIA—Location-info-answer THIG—Topology hiding interface gateway
LIR—Location-info-request TLS—Transport layer security
LRN—Location routing number UA—User agent
MGCF—Media gateway control function UDP—User Datagram Protocol
NAI—Network access identifier URI—Uniform resource identifier
NAPTR—Naming authority pointer VoIP—Voice over IP
NG—Next-generation VPN—Virtual private network
NGN—Next-generation network

Telecommunication Standardization Sector (ITU-T) [18, 20]. The 3rd Generation Partnership Project
next-generation network (NGN) defined resource (3GPP*) defines in Release 7 (R7) a new signaling
and admission control functions (RACF) require- peering entity: the interworking border control func-
ments [15] for effective resource management in the tion (IBCF) [2]. The GSM Association (GSMA) defined
transport layer. The Internet Engineering Task Force guidelines for IMS roaming and interworking [10].
(IETF) recently created the Session Peering for The European Telecommunications Standards Institute
Multimedia Interworking (SPEERMINT) working (ETSI) addressed issues related to interconnection
group to define a framework for layer 5 peering and routing, and their implications on numbering,

34 Bell Labs Technical Journal DOI: 10.1002/bltj


naming, and addressing for NGN Telecommunication In addition, the entire process is supported by com-
and Internet Converged Services and Protocols for mon databases that are shared among carriers, such
Advanced Networks (TISPAN) [6]. The content of this as the local exchange routing guide (LERG), and by
paper reflects the work conducted, thus far, by these pre-defined routing tables that associate switches with
organizations. telephone numbers.
This paper analyzes in detail the IMS peering Call routing in IMS, however, is completely
challenges from the perspective of Session Initiation different from that in the PSTN. First, the current
Protocol (SIP) signaling peering. It focuses primarily IMS systems have not deployed a routing capability
on voice applications and delves into other critical that is fully equivalent to that of the existing PSTN.
networking issues that further complicate the discov- Second, routing in IMS is based on a different entity:
ery of signaling peering points for inter-carrier rout- SIP uniform resource identifiers (URI) [23]. SIP URIs
ing. Finally, it presents a high-level signaling routing are used to identify IMS users and resources (e.g.,
process for IMS signaling routing that includes sup- servers and databases) and have a format similar to
port for number portability as a key function for inter- an email address with a user name and a domain name
carrier IMS peering, and identifies key areas for (e.g., sip:bob.smith@domain.com). The domain por-
further investigation. tion identifies the host responsible for handling the
The paper is organized as follows: We first pro- request for the user. For instance, the domain portion
vide background material on why the PSTN is used may identify the IMS operator serving the particular
today for IMS network interconnection and the user. Hence, during call processing, SIP URIs need to
impacts of PSTN bridging. Next, we describe a generic be resolved using domain name system (DNS) proce-
case of IMS peering to discuss the critical issues dures, described in [5], in order to determine the
related to address resolution and discovery of peer- IP address and port number of the corresponding
ing points. We follow by presenting other factors— domain name server for the domain portion of the
such as inter-domain IP infrastructure, identification, SIP URI.
and address format—that impact IMS peering. We In the particular case where an E.164 telephone
describe a proposal for a signaling routing process for number is used to establish a call, IMS invokes tele-
IMS peering, summarize our conclusions, and outline phone number mapping (ENUM) [7], an application
key areas for further investigation. of the DNS, to perform the translation of the E.164
number into a SIP URI. The ENUM protocol uses a
Background specific naming authority pointer (NAPTR) DNS
Today, when placing a Voice over IP call to reach resource record, the E.164-to-URI resolution (E2U),
a destination endpoint served by a different VoIP as defined in [19], to identify and map the E.164
operator, most likely, the call is routed through the number with associated communications services
PSTN. In some cases, even calls between two VoIP and resources (e.g., call server, email). This mecha-
subscribers served by the same SP are hairpinned over nism allows a user to reach a destination by dialing
the PSTN. The next section examines why it is so. a telephone number, regardless of the device (e.g.,
Why Are IMS-to-IMS Calls Routed Through the phone, PDA) the destination party might happen to
PSTN Today? be using. In short, in IMS, ENUM/DNS provides an
One of the key issues that impede direct peering essential location function—equivalent to the LERG
among IMS networks is address resolution in support in PSTN—for reaching a user identified with an
of routing of SIP signaling messages, particularly dur- E.164 number.
ing session/call establishment or setup. In the PSTN This location function in IMS is at the heart of
network, call routing is based on global addressing the problem that forces routing through the PSTN for
using ITU-T E.164 [13] numbers (e.g., NPA-NXX- IMS-to-IMS calls. Currently, there are several pro-
XXXX in North America) to uniquely identify the posals for ENUM/DNS implementation in support of
users, and on point codes to identify switching points. address resolution in IMS networks. However, there is

DOI: 10.1002/bltj Bell Labs Technical Journal 35


no agreement in the industry on a definitive solution. expenditures (OPEX) by retiring the PSTN more
These proposals are as follows: quickly.
• Global (public) ENUM is dedicated to making First, many IP and multimedia services (e.g., active
ENUM data publicly accessible through the phone book, video telephony) requiring end-to-end
E164.arpa domain at the public DNS root. IP connectivity will not work over the PSTN. As a
• Private ENUM is owned and controlled by a single result, an SP might be able to offer a particular serv-
SP, which determines whether it wants to share ice to its own subscriber base; however, it will not be
any information with another SP. able to guarantee that the service will work if the
• Infrastructure ENUM, also known as carrier ENUM, other endpoint belongs to a different SP to which it is
allows a group of carriers to share routing data only connected via the PSTN. Hence, there is a poten-
among themselves and to look-up the destina- tial for lost revenue for the SPs. For instance, a VoIP
tion addresses of one another’s customers. call initiated with a non-E.164 address (e.g., SIP URI)
Note that both the private and the infrastructure to another IP endpoint will not work over the PSTN.
ENUM solutions are outside the scope of global ENUM Support for such a service will require the initial SIP
in that they do not use the E164.arpa domain. INVITE message containing the destination SIP URI
However, they do share the same technological basis to be encapsulated in ISDN user part (ISUP) and car-
and conventions as global ENUM. ried over to the destination IMS domain via the SS7
Meanwhile, current implementations of IMS network. SIP message encapsulation in ISUP messages
networks by early adopters leverage private ENUM is not supported today and would require further
solutions for the most part. That is, the private development in the PSTN, which is unlikely to occur.
ENUM database is mainly populated with NAPTR In general, applications requiring end-to-end IP
records about the SP’s own subscribers. Hence, for fall into these categories:
voice applications, if the called party (CdP) belongs to • Peer-to-peer applications initiated with non-E.164
a different IMS SP domain for which there is no identifiers (e.g., initiating a voice call with the email
associated NAPTR record in the DNS server of the address of the called party).
originating network, the SIP URI address resolution • Applications requiring SIP signaling.
process results in an ENUM failure indicating that the • Client/server applications that require end-to-end
IMS domain does not know how to locate and route IP connectivity (e.g., multiparty gaming).
the call to its destination. Today, when such a situa- • Applications requiring broadband access and
tion occurs, by default, the IMS network simply higher bandwidth than the 64 Kbps narrowband
routes the call to the PSTN. This default routing ends used in the PSTN (e.g., video telephony).
up creating a PSTN bridge in between two IMS Second, keeping a call between two IP endpoints
islands, particularly when the destination party is an in the IP domain will help SPs avoid unnecessary set-
IP endpoint. tlements and reduce charges associated with PSTN in-
The technical challenges associated with IMS sig- terconnections. Moreover, relying on the PSTN to
naling peering are explored in more detail in subse- support calls among IP endpoints ends up deferring
quent sections. For now, suffice it to say that the PSTN the decision to retire the PSTN more quickly, and de-
bridging model is not financially sustainable over the lays the opportunity for SPs to start realizing OpEx
long term, and it is not a technically viable option for savings for managing one packet network.
non-VoIP-only applications. The next section provides Last, PSTN bridging results in quality of service
an assessment of the impacts of PSTN bridging. (QoS) degradation caused by the cascading effect of
Impacts of PSTN Bridging bearer conversions at the edges of the network. That
PSTN bridging in support of next-generation is, the bearer signal needs to be converted per ITU-T
applications will have serious financial implications G.711 specifications, thus incurring additional pro-
for IMS carriers in terms of both lost revenue oppor- cessing delays in the media gateways placed on both
tunities and the opportunity to reduce operational the ingress and egress sides of the PSTN.

36 Bell Labs Technical Journal DOI: 10.1002/bltj


IMS domain A IMS domain B

IMS IMS
Application
application application

DNS/ DNS/
ENUM L5 peering ENUM

Other Other Call/


core IMS P/I/S-CSCF P/I/S-CSCF core IMS session
elements elements control

RACF RACF

Managed IP Managed IP
PSTN/PLMN
MG core PEP PEP core PEP

PEP Media/
access
IMS peering Wireless
PEP scope access network
Wireline
access network

CSCF—Call session control function MG—Media gateway


DNS—Domain name system PEP—Policy enforcement point
ENUM—Electronic number mapping P/I/S-CSCF—Proxy/interrogating/serving CSCF
IMS—IP Multimedia Subsystem PLMN—Public land mobile network
IP—Internet Protocol PSTN—Public switched telephone network
L5—Layer 5 RACF—Resource and admission control function

Figure 1.
IMS peering scope.

Need and Scope protocols such as Border Gateway Protocol (BGP)


As SPs continue to deploy IMS networks and are well defined to support layer 3 peering. Much
introduce next-generation (NG) applications to enh- work has been done in the RACF area, which de-
ance the end user experience, the need for peering fines a framework for policy peering in support of
among IMS networks and across SP boundaries will QoS [15].
become critical. However, peering at the session/signaling layer
As shown in Figure 1, IMS peering will occur (or layer 5) is not yet standardized. IETF SPEERMINT
at different layers including network, policy, sig- recently defined a peering architecture [21] between
naling, and security. Layer 3 peering is well under- two SIP SPs, as depicted in Figure 2. The architec-
stood—IP networks are interconnected today and ture identifies the following logical functions:

DOI: 10.1002/bltj Bell Labs Technical Journal 37


The next section discusses in detail the steps
LF
associated with these phases.

LF LF
Signaling Steps
In this example, each carrier implements a pri-
Originating PF PF Receiving vate ENUM/DNS solution in its network, and domain
SP‘s SP‘s
network SF SF network A and domain B are the home networks for user
agent A (UAa) and user agent B (UAb), respectively.
MF MF
For simplicity, the registration and authentication
steps, which would involve the use of an authentica-
IETF—Internet Engineering Task Force
LF—Location function tion, authorization, and accounting (AAA) server, and
MF—Media function the home subscriber server (HSS) are not shown. We
PF—Policy function
SF—Signaling function further assume that each domain contains only one
SP—Service provider HSS server, and therefore the subscriber location
SPEERMINT—Session Peering for Multimedia Interworking
function (SLF) is not required. We also combine the
three call session control functions—interrogating
Figure 2.
CSCF (I-CSCF), serving CSCF (S-CSCF), and proxy
IETF SPEERMINT reference architecture.
CSCF (P-CSCF)—into a session manager (SM). Further-
1. Location function (LF) provides the routing data for more, we do not represent any element at the appli-
the purpose of the discovery of the signaling and cation layer (e.g., a telephony application server or
policy functions, and reaching the end user. The TAS) that might be invoked during the call setup. Last,
LF, in fact, uses the ENUM/DNS infrastructure to we do not show the breakout gateway control func-
support the peering process. tion (BGCF) and media gateway control function
2. Policy function (PF) provides a framework for (MGCF) components, which would be invoked if the
authentication and the exchange of policy param- call were destined for the PSTN/PLMN.
eters to be used by the signaling function. The signaling routing steps for a voice call between
3. Signaling function (SF) handles the routing of SIP UAa and UAb, shown in Figure 3, are as follows:
messages and assists in the discovery of param- 1. UAa initiates a SIP INVITE with UAb’s telephone
eters to be used by the media function. number, as the called party. The CdP number is
4. Media function (MF) performs media-related func- mapped into the request URI of the SIP INVITE.
tions including media transcoding between two 2. Upon receiving the SIP INVITE, SMa queries the
SIP-based SPs’ networks. ENUM server on the CdP: UAb.
The scope of our work, in what follows, is lim- 3. Two cases might occur:
ited to the LF and SF, as described above. a. If the ENUM server does not contain a NAPTR
record for the CdP, and the carrier-of-record
IMS-to-IMS Signaling Peering for a Basic associated with the CdP cannot be found or
Voice Call reached via SIP peering, then the call is routed,
In general, the inter-carrier signaling routing by default, to the PSTN. This case is not shown
process in IMS can be divided into the following on Figure 3.
major phases: b. Otherwise, the ENUM server returns a NAPTR
1. Initialization and address analysis and resolution, record to SMa with the domain name asso-
2. Determination of a peering point in the orig- ciated with Uab.
inating network, 4. SMa requests support from a local policy infra-
3. Discovery of a peering point in the terminating structure (Pa) to determine the peering point in
network, and domain A (PPa) where it should forward the SIP
4. Routing in the terminating domain. INVITE. PPa is an IBCF, which provides, among

38 Bell Labs Technical Journal DOI: 10.1002/bltj


E D D E

IMS IMS
domain A 3 7 10 domain B
2 6 9 12 13

1 8 11 14 15
Peering IP
UAa SMa PPa network PPb SMb UAb

4
5

Pa Pb

D—DNS server IMS—IP Multimedia Subsystem PP—Peering point


DNS—Domain name system IP—Internet Protocol SM—Session manager
E—ENUM server P—Policy UA—User agent
ENUM—Electronic number mapping

Figure 3.
Signaling routing steps.

other things, firewall functions for signaling, shown on the diagram, might occur in order to
policing of signaling, and topology hiding. locate SMb. If the SIP INVITE message does not
5. Pa applies its local policy rules to determine the contain a route header field that points directly to
appropriate peering point to handle the call, and SMb, PPb sends a location-info-request (LIR) over
then forwards the SIP INVITE to SMa with the the Cx interface to the HSS to locate the appropriate
URI associated with PPa. SM. HSS replies with a location-info-answer (LIA)
6. SMa queries the DNS server in domain A to command that indicates the SIP URI of the server,
retrieve the IP address for PPa. in this case SMb, that servesUAb.
7. The DNS server in domain A returns the PPa’s IP 13. PPb queries the DNS server in domain B to
address to SMa. obtain the IP address for SMb.
8. SMa sends the SIP INVITE to PPa with the domain 14. PPb forwards the SIP INVITE request to SMb.
name serving UAb. 15. SMb may invoke a TAS to apply further call logic
9. PPa queries the DNS in domain A to resolve the and forwards the SIP INVITE request to UAb.
address of a peering point in domain B (PPb). At 16. UAb accepts the call and replies with a 200 OK
this stage, resolving the address of UAb results in message to UAa, also not shown.
finding the IP address, port number, and transport
protocol for PPb. Issues Related to the Signaling Routing Steps
10. DNS server returns the IP address of PPb to PPa. In this section, we analyze the signaling routing
11. PPa , as a gateway to external networks, may steps discussed above to identify the salient rout-
encrypt part of the SIP message to perform the ing issues that might occur in IMS network implemen-
topology hiding interface gateway (THIG) tations. We group these issues under three categories
function and forward the SIP INVITE to PPb. and discuss them separately.
12. PPb, upon determining that the domain associated CdP address resolution. As indicated in step 2, the
with the request URI is domain B, needs to forward CdP address resolution is performed using the ENUM
the SIP INVITE to SMb. The following steps, not service offered on the DNS server. In the example we

DOI: 10.1002/bltj Bell Labs Technical Journal 39


presented, the SP uses a private ENUM infrastructure, level agreements (SLA). Engineering factors such as
which might not contain a NAPTR record for the CdP. call rate, and current utilization of the proxy server
This might happen if a different SP serves the CdP. In performing the peering function can lead to a distri-
this case, the decision to route to the PSTN upon an bution of the calls among the available peering points
ENUM failure response, as described in step 3a, cre- in support of load balancing requirements. Last, this is
ates a routing dilemma because the CdP could be an an area that requires further investigation and would
IP endpoint in domain B, and not a PSTN user. The most likely require support from a policy infrastructure.
routing decision is based on not having the capability We will discuss further aspects of this selection process
to map the CdP telephone number with a SIP URI in later in the paper.
domain A, but does not take into consideration the Discovering peering point in terminating domain.
fundamental question as to whether the CdP is an IP Step 10 discusses a DNS-based mechanism for the peer-
or a PSTN endpoint. In the latter case, it is appropri- ing point in the originating network, which allows the
ate to route to the PSTN. In the former case, how- discovery of a peering point in the terminating net-
ever, the call should simply be maintained in the IP work. The IETF has addressed in detail the discovery
domain. process of a proxy server (i.e., the peering point in our
Finally, we believe that private ENUM implemen- context) in a different domain in RFC 3263 [22]. In
tation will not satisfy the long term need to remain essence, the peering point in the originating network
in the IP domain for an inter-domain call served by makes use of DNS procedures, using both service
different SPs. A long term solution should be based record (SRV) [11] and NAPTR records, to determine
on a common directory of telephone numbers and the IP address, port, and transport protocol (e.g.,
associated domain names that are restricted to the SPs’ Transport Control Protocol (TCP), User Datagram
use and not accessible by public Internet users. Such Protocol (UDP), Stream Control Transmission Protocol
a common directory can cover many SPs across a (SCTP), Transport Layer Security (TLS)), for the peer-
geographic area (e.g., regions in country) and be ing point in the terminating network. Additionally, RFC
administered by a trusted third party on behalf of the 3263 describes the procedures to allow the peering
SPs. This solution is conceptually similar to existing point in the terminating network to discover a backup
administrative domains and database infrastructures peering point in the originating network, in case the
available in the PSTN, such as the telephone number- latter fails in the middle of the call setup transaction.
ing plan or LERG. In short, the originating network will find all required
Determining peering point in originating domain. In information to establish a call on IP to the terminating
step 4, we discussed another important aspect of the network, including the border elements of the termi-
signaling routing process: selection of a peering point nating network.
in the originating IMS domain. The originating IMS The premise of RFC 3263, however, is based on a
network will need to make intelligent decisions in the global address resolution scheme that is accessible
selection of its own signaling peering point in support through the E164.arpa domain at the public DNS root.
of call setup, as it will likely have multiple signaling That is, SPs will have to publish SIP URIs of their peer-
peering points to interconnect with other networks. ing points in the public DNS so that they know where
This decision is dynamic in nature and varies with the to signal one another. While such architecture works
target destination network of the CdP. SPs may employ reasonably well today for handling email services over
caching methods to minimize call setup delays of sub- the Internet, its implementation in SP networks will
sequent calls in the selection of a peering point. end up exposing their critical signaling resources to
However, several other conditions in the originating Internet users. Thus, we believe that SPs deploying
IMS network might influence such a selection as well. IMS networks will be reluctant to embrace this solu-
For instance, the choice of a peering point can be tion until they can be assured that great concerns
made based on quality of service policies (e.g., call about security risks or attacks on key signaling assets
setup delay) defined to meet peering and user service are fully addressed.

40 Bell Labs Technical Journal DOI: 10.1002/bltj


Meanwhile, a way forward is for SPs to create fed-
erations and use infrastructure ENUM/DNS for peering
point discovery within a federation. A federation is a Public
ENUM/
group of SPs that agree to accept calls from one another DNS
via SIP, according to a set of administrative rules and
contractual financial terms established for such calls. In
the next section, we analyze other networking condi- Domain A Internet Domain B

tions that impact the peering process as well.

Other Factors Influencing Signaling Peering


In addition to the issues identified in the signaling Domain Z
steps, there are other important factors that will
influence the signaling peering process. In general,
DNS—Domain name system
these factors are related to the nature of the peering ENUM—Electronic number mapping
infrastructure between the originating and terminat-
ing domain, as well as the handling of the CdP address Figure 4.
format and translation at the peering points. Peering via public Internet.

Inter-Domain IP Infrastructure for Signaling


The inter-domain IP infrastructure between the peering process between the originating and termi-
IMS originated and terminated networks can be one nating IMS domains. We evaluate these alternatives
of the following cases: from the points of view of address resolution and
1. The public Internet. discovery of the peering point in the terminating
2. Direct connections over leased lines, IP-virtual network.
private network (VPN) or tunneling over the The public Internet. This scenario, as shown
public IP network, among the SPs’ peering points in Figure 4, leverages an open interconnection
to create a fully meshed interconnect signaling architecture where the SPs’ peering points’ addresses
network. are public IP addresses and are directly accessible by
3. A managed IP network commonly owned by the Internet users. This also implies that inter-carrier IMS
SPs for the purpose of transferring IMS signaling- signaling messages are routed over the public Internet
related traffic among the different SP networks. without any guarantee that the signaling require-
This case is similar to the proposed GPRS roaming ments (e.g., delay, priority of traffic) will be met. This
exchange (GRX) [9] of the GSM Association for scenario has some key advantages. It potentially offers
PLMN network interconnections. a more open solution to the peering issue and enables
4. A third-party-managed IP network similar to an ad-hoc peering among SPs who do not have an
existing VoIP clearinghouse, an SP that offers explicit peering agreement.
termination, financial services, and enhanced However, this scenario requires support from a
applications to other SPs (e.g., Internet service global ENUM/DNS system with global mapping of
provider (ISP), VoIP) but serving the IMS SPs E.164 numbers to DNS NAPTRs containing the SIP
signaling interconnection needs. URIs of all the SP peering points. For the reasons dis-
The inter-domain IP infrastructure need not be cussed earlier and related to universal connectivity
an IMS network, but should ensure that key require- through global DNS, we do not believe this scenario
ments—such as QoS, reliability, security, and address will address the SPs’ security concerns, and hence it is
resolution—are met. Issues related to IPv4-to-IPv6 not recommended.
interwork- ing are outside the scope of this document. Fully meshed interconnect network. This scenario
In what follows, we evaluate each inter-domain archi- provides direct links among the SPs’ signaling peering
tecture alternative and its impacts on the signaling points and results in an n2 problem, where n is the

DOI: 10.1002/bltj Bell Labs Technical Journal 41


Public Domain X
Domain A Fully meshed Domain B
ENUM/
DNS

Carrier
Domain Z ENUM/
DNS

Managed IP
Figure 5. Domain A Domain B
network
Peering via fully meshed network.

number of peering points, as illustrated in Figure 5. Domain Z


SPs may use public IP addresses for their peering
points; however such addresses are not accessible
by public Internet users. The supporting DNS system DNS—Domain name system
ENUM—Electronic number mapping
resolves the terminating peering point address into an IP IP—Internet Protocol
address that is routed over a particular link. In general,
this solution may prove not to be scalable for a large Figure 6.
number of peering points. However, SPs might decide to Peering via managed IP network.
selectively implement it to support high volume traffic
in between large communities of interests. name into the IP address of the terminating peering
Common managed IP network. In this scenario, point.
depicted in Figure 6, the common managed IP net- Clearinghouse model. This scenario, also shown in
work is a private network. It may use public IP Figure 6, is functionally the same as the one depicted
addresses to route SIP requests to the SPs’ peering for the common managed IP network, except that a
points; however, these IP addresses are not accessible clearinghouse provides the administration of the
by public Internet users. inter-domain IP infrastructure. A variant of this case
In addition, this scenario requires an infrastruc- is that the clearinghouse is only used as a trusted third
ture ENUM/DNS system that is used only to resolve party registry administering the infrastructure ENUM/
SIP URIs associated with SP peering points, which are DNS hierarchy and providing SPs with the routing
attached to the common managed IP network. In information needed for address resolution.
order to peer with a public domain (i.e., one not In summary, the nature of the inter-domain IP
attached to the managed IP network), the SP would infrastructure impacts the address resolution process
use a public Internet DNS for address resolution. for the discovery of the terminating peering point. It
Hence, the public Internet DNS system is used to resolve further determines what network knowledge is requi-
the domain names of all destinations not served red to reach the terminating domain. For instance, in
by the common managed IP network. the first scenario, a global directory ENUM/DNS is
Finally, for traffic among the SPs served by the com- required, whereas in the second case, static routing
mon managed IP network, the discovery of the ter- tables might be sufficient to reach the terminating
minating peering point always results in finding the IP network. Hence, the call processing algorithm for SIP
address of an access server in the common managed peering should take the inter-domain infrastructure
IP network. Once a SIP URI request is sent to the into consideration.
common managed IP network, the infrastructure Transit network/federation. In the absence of a
ENUM/DNS server resolves the associated domain global directory ENUM/DNS, SPs will likely group

42 Bell Labs Technical Journal DOI: 10.1002/bltj


Domain X

3 Fe
n de
io ra
at A –X–Z tio
der n
Fe 4

Domain A Federation 1 Domain B Federation 2 Domain Z

A–B–Z

Figure 7.
Peering via transit.

themselves into federations to satisfy their intercon- right transit network to use to reach a destination
nection needs. Figure 7 illustrates a multi-federation becomes a critical part of the signaling peering routing
arrangement among four SP networks. Moreover, it process and must be taken into consideration by the
shows that the relationship with a federation might routing algorithm during call processing.
not be exclusive. An SP might belong to more than
one federation for various financial and technical rea- Identification and Address Format
sons. For instance, domain A belongs to federations IMS, as for any type of network, provides mecha-
1 and 3. Furthermore, an SP which belongs to multi- nisms to uniquely identify users so that calls can be
ple federations might offer to become a transit SP for appropriately routed to them. IMS uses public user
other SPs. In our example, domains B and X can be identities (PbUI) and private user identities (PrUI) to
such transits for domains A and Z. Likewise, domains identify its users. PbUI are SIP URIs or TEL URIs [24]
A and Z can be used as transits for traffic between do- and are used to route SIP signaling in IMS. As discussed
mains X and B. As a result, domain A can use either earlier, SIP URIs are in the form of sip:⬍user⬎@oper-
domain X or domain B to reach domain Z. Therefore, ator.com, where the user part can be an E.164 tele-
a key question a routing algorithm needs to address is phone number. A TEL URI, however, is in the form of:
which federation to use to reach a target destination. tel:⫹1-NPA-NXX-XXXX, to represent, for example, a
In addition, [16] and [12] discuss the need for fed- telephone number in the North America Numbering
erations to use fully qualified domain names as their Plan. The TEL URI is used when the domain of own-
identifiers, and that members of a federation might ership of the phone number is unknown. SIP, how-
need to publish federation membership as domain ever, requires that the URI under registration be a SIP
attributes. In other words, to route a call to its destina- URI [3]. Hence, it is not possible to explicitly register a
tion, the originating network might need to retrieve the TEL URI in SIP. However, PrUIs do not use SIP URI
set of federations to which the terminating network or TEL URI formats. Instead, they use a network access
belongs. If there is a shared federation, which can be identifier (NAI) in the form of ⬍user⬎@operator.com.
determined through the domain policy dynamic They are exclusively used for subscription identifica-
delegation discovery system (DDDS) application as spec- tion and authentication purposes [3].
ified in [17], it is used to support the call. Otherwise, the The rest of the paper focuses on analyzing the SIP
originating network might route the call to a transit SP. URI-based PbUI formats and their impacts on IMS
Thus, matching a common federation or selecting the signaling peering.

DOI: 10.1002/bltj Bell Labs Technical Journal 43


SIP URI formats for the PbUI. IMS users will use a of the routing steps and issues discussed above, we pro-
variety of devices (e.g., PDA, IP phone) to trigger NG pose a high level signaling peering process, which
services and to initiate calls to other users. Likewise, provides a comprehensive approach to signaling
SPs may assign more than one PbUI to their customers peering and takes into account multiple constructs
so they can be reached through a variety of means (e.g., IP networks, federations, and transits) that might
(e.g., email, business phone). As a result, SIP URIs as- exist between the originating and terminating domains.
sociated with PbUIs can take various forms, as de- Moreover, the proposed signaling peering process
scribed in [6], including the following: addresses the need to perform address translation in
1. sip:⫹⬍E.164⬎@⬍service provider⬎;user ⫽ phone support of service portability as an essential step for
2. sip:⬍user⬎@⬍service provider⬎ inter-carrier communications. A key assumption, in
3. sip:⬍user⬎@⬍domain name⬎ this regard, is that “all call query,” a method by which
4. sip:⬍ip address⬎ a number portability query is performed on all origi-
Impacts on address resolution for SIP peering. The nating calls as described in [8], is used as the service
format of the SIP URIs, as presented above, might in portability implementation option.
some cases render impossible the resolution of the
destination address of the CdP. For instance, Signaling Process for IMS Peering
⬍user⬎@⬍domain name⬎ (case 3) might be impos- Figure 8 illustrates the signaling peering process
sible to resolve if used to initiate a call. First, assuming for a basic voice call initiated with an E.164 number,
the domain name can be resolved to an associated which we describe as follows:
SP domain, as discussed above, ⬍user⬎@⬍service 1. The calling party initiates the call with a dial
provider⬎ might not be unique in the SP domain. string, which the originating network converts
Second, different users belonging to a domain name into a fully qualified E.164 number.
might be served by different SPs. For example, johndoe 2. Map the CdP’s E.164 number into SIP headers
@alcatel-lucent.com might be served by a different SP (e.g., request URI, ‘To ⫽ ’) and send the app-
than janedoe@alcatel-lucent.com. Therefore, the res- ropriate INVITE request to the session manager
olution of a ⬍user⬎@⬍domain name⬎ SIP URI in the originating network.
should identify the specific SP serving each particular 3. Analyze the CdP address for service portability
user. That is, the ⬍domain name⬎ server might need (e.g., SP portability).
to be a redirect server pointing to the carrier-of-record 4. If the CdP is not ported, proceed with 6.
that serves the user. This can be an implementation 5. If the CdP is ported:
nightmare in that all domain names will be required – Perform the portability database lookup to
to support the redirect function. Last, considerations retrieve the location routing number (LRN)
should be given to the additional delay such a process associated with the CdP. Note that this paper
will introduce on key performance parameters such as does not specify an infrastructure for perfor-
call setup time. ming a number portability (NP) database (DB)
In summary, the format of the SIP URI used to query. Options range from reusing existing
initiate a call will impact the signaling routing process NPDBs in the intelligent network (IN) to ret-
in the originating domain as well as in any transit net- rieving a NAPTR record from an ENUM/DNS
work that might be used to support the call. In addi- query pointing to the LRN.
tion, some formats might be impossible to resolve in – Append the ‘npdi ⫽ yes’ field to the request
support of inter-carrier communications. URI header in the subsequent SIP URI request
created with the LRN.
Proposed Signaling Peering Process 6. Perform address resolution of the CdP or LRN
In order to define an appropriate peering and through the ENUM/DNS infrastructure to de-
routing approach for a given scenario, one needs to termine which destination network the CdP
consider and evaluate the end-to-end process. In light belongs to. The output of this step results in a

44 Bell Labs Technical Journal DOI: 10.1002/bltj


Identifier CdP ID mapping RFC 3261
Request URI

PbUl:SIP URI or CdP address


TEL URI analysis ENUM/DNS queries for NP or
query NPDB via IM-SSF

All call query Y Perform NP lookup and


Perform
Is CdP open to portability? NPDB? address translation

What‘s the domain name of CdP SP? N


Is CdP served by initiating domain?
obtain ENUM/DNS NAPTR Address resolution
Invoke policy peering function for
How to determine that the establishing the nature of peering
CdP belongs to the PSTN? network (e.g., federation, transit)
Y CdP in N
PSTN

Is RFC 3263 sufficient?


Invoke BGCF Peering scenario Determining/locating PP?
(originating/terminating)

Selection based
on NPA-NXX and Select MGCF/MGW Discover peering point ENUM/DNS with
local policy rules local policies

Obtain PSTN routing Resolve PP address SIP address translation


data (e.g., from LERG) Perform SIP/ISUP
conversion and mapping
Generate ISUP IAM
Translate CdP
Determine S-CSCF serving
Route to PSTN CdP and route to CdP

Terminating Y Route SIP invite to CdP


PSTN endpoint
domain in terminating domain

N
IP endpoint
Route to next domain

BGCF—Breakout gateway control function NAPTR—Naming authority pointer


CdP—Called party NP—Number portability
CSCF—Call session control function NPA—Numbering plan area
DNS—Domain name system NPDB—Number portability database
ENUM—Electronic number mapping PP—Peering point
IAM—Initial address message PSTN—Public switched telephone network
ID—Identifier RFC—Request for comments
IM—IP Multimedia S-CSCF—Serving CSCF
IP—Internet Protocol SSF—Service switching function
ISDN—Integrated services digital network SIP—Session Initiation Protocol
ISUP—ISDN user part SP—Service provider
LERG—Local exchange routing guide TEL—Telephone
MGCF—Media gateway control function URI—Uniform resource identifier
MGW—Media gateway

Figure 8.
Proposed signaling peering process.

DOI: 10.1002/bltj Bell Labs Technical Journal 45


SIP URI that represents an address of record for helps determine which network to route a call to for
the CdP. inter-carrier communications. Various options are im-
7. If the call is destined to the PSTN, invoke the plemented for NP. For instance, in the United States,
BGCF to determine which MGCF to provide SPs implement the “all call query” option for service
the ISDN user part and SIP interworking func- provider number portability (SPNP), which allows a
tion, as defined in [4], and then route the call to user to change the SP while keeping the same
the PSTN. telephone number. SPNP leaves the responsibility for
8. Otherwise, determine whether a common fed- performing the NPDB query to the penultimate net-
eration exists between the originating and ter- work. This decision is somewhat straightforward
minating domains: in the PSTN in that for a long distance (LD) call, the
– If so, consult the policy peering infrastructure carrier of record for LD services is the penultimate
and apply common federation rules to dis- network, and, hence, it performs the query. For a toll
cover the peering point in the terminating local call, the local SP is the penultimate network and,
network. as such, does the query.
– If not, determine a transit network to support IMS, however, introduces new challenges in sup-
the call and apply corresponding federation porting NP. First, as discussed above, the CdP identifi-
rules for the transit. cation might not be an E.164 number. Obviously, in
9. Determine the peering point in the originating such cases, existing E.164-based NP databases cannot
network according to the local policy infrastruc- be leveraged. Therefore, translation between SIP URIs,
ture and procedures defined in RFC 3263 [22]. and IP addresses via DNS in support of NP, becomes a
10. Perform the SIP header translation at the critical function in determining the next domain/
peering point and according to the SIP URI network towards a destination.
format used, if necessary. Moreover, the decision as to which SP should per-
11. Route the SIP INVITE to the next hop (e.g., form the NPDB becomes more complex in IMS, if we
proxy/access server in federation) towards the ought to follow the penultimate model discussed
destination above. How to determine which network is next to
12. When the terminating domain is reached, the last before performing address translation of the
determine the serving CSCF for the CdP, locate CdP? In today’s deployment of VoIP networks, these
the CdP, and route the INVITE to the CdP. demarcation lines are not clearly defined. As a case in
13. Establish the session and transfer of media. point, VoIP service offerings assume mobility of end-
points. For instance, one can be living in California
Proposal Benefits and may get assigned a phone number with a New
The end-to-end signaling peering process Jersey numbering plan area (NPA) (e.g., 732). This
described above introduces new decision points and proposal simplifies the decision process by recom-
routing steps during call setup for inter-carrier com- mending that the originating network always perform
munications. The benefits of such a process include: the appropriate translation of the CdP address on all
1) support for number portability, 2) more efficient calls, once it determines that such an address is open
routing among IP endpoints, and 3) avoidance of to portability.
unnecessary delays during call setup. The following Determining the nature of CdP endpoint. Another
discusses these benefits. important decision point during call processing,
Support for number portability. In many countries, summarized in step 7 and step 8 above, is for the
number portability (NP) is a regulatory requirement originating network to determine whether the CdP
that SPs must comply with. As presented in step 4 is a PSTN or IP endpoint, so the call is handled appro-
and step 5 in the section above, support for NP priately. Otherwise, SPs will continue to use PSTN
requires SPs to perform user identification translation bridging for IP-to-IP inter-carrier calls. At this point, it
for the CdP. The outcome of the translation process is not clear whether the mechanism to identify the

46 Bell Labs Technical Journal DOI: 10.1002/bltj


nature of an endpoint (i.e., PSTN or IP) would exist This paper discusses these new challenges and
outside the ENUM framework. Hence, as discussed proposes a high-level signaling process that identifies
earlier, we recommend that a common directory of key peering decision points and provides a compre-
telephone numbers and associated domain names hensive approach to solving them. Such an approach
(e.g., carrier ENUM) be in place to support that is necessary for accelerating the deployment of IMS
decision point. Such an infrastructure should be networks and reducing the reliance on PSTN/PLMN
restricted to SP use and would be inaccessible to network for inter-carrier communications among IP
public Internet users. endpoints.
This proposal takes advantage of the recom- Finally, it is clear that more detailed work is
mended common infrastructure to help determine needed to address some specific technical challenges
whether a call terminates in the PSTN or in another identified in this paper. However, in priority order,
IMS domain and to route it accordingly. the following areas are critical for further investiga-
Avoiding additional call setup delay. A net benefit of tions. First, more work is needed to define the re-
determining the nature of the CdP endpoint is to avoid quirements for a policy peering infrastructure in
the additional call setup delay, as defined in ITU-T support of intelligent selection of peering points in an
E.127 [14], which might occur in the PSTN when SP IMS network. This will invoke the PF in both the
bridging two IMS islands. For instance, for calls in the originating and terminating IMS networks and apply
PSTN, E.127 recommends an average call setup delay policy rules for the selection of the peering points.
of no more than 3 seconds and 5 seconds, respectively, Second, a new routing algorithm for the selection of
for local and toll calls. The 95th percentiles are federation/transit in a multi-transit/federation sce-
6 seconds and 8 seconds, respectively. nario is needed. This will invoke the LF and PF in
both originating and terminating networks.
Conclusions and Areas for Further Investigations
IMS network signaling peering is a very complex pro- Acknowledgments
blem to tackle for the deployment of next-generation The author would like to acknowledge several
IMS networks. Unlike the embedded base PSTN/PLMN, colleagues for sharing their insights on various aspects
where common databases and established routing prin- of this work. In particular, he is grateful to Deirdre
ciples allow SPs to securely share call routing data and Doherty, Mohamed El-Sayed, Igor Faynberg, Paul
identify one another’s signaling points, there are new Gagen, Vijay Gurbani, Volker Hilt, Markus Hofmann,
challenges facing the IMS SPs. First, SPs need to per- Vanita Katkar, Eunyoung Kim, Lyle Kipp, Humberto
form address resolution of users’ identifiers (e.g., SIP La Roche, Tom Morawski, Bob Moul, Morgan Stern
URI) and other resources (e.g., call servers) at almost and Carlos Urrutia-Valdés.
every step during call setup. Second, the format of the *Trademarks
identifiers may vary depending on what service is being 3GPP is a trademark of the European Telecommunica-
initiated. Accordingly, the address resolution process tions Standards Institute.
requires different mechanisms to identify the users and
their serving domains. Third, the nature of the infra- References
[1] 3rd Generation Partnership Project,
structure between two SP networks impacts the peer-
“Architectural Requirements,” 3GPP TS 23.221,
ing process in that it influences the discovery of the June 2004, ⬍http://www.3gpp.org/ftp/
peering points for a given call. Last, SPs will create fed- Specs/html-info/23221.htm⬎.
erations and transit networks for network intercon- [2] 3rd Generation Partnership Project, “IP
nect, as achieving universal connectivity is an Multimedia Subsystem (IMS), Stage 2 (Release
unrealistic goal. Hence, the selection of what federa- 7),” 3GPP TS 23.228, v7.2.0, Dec. 2005,
⬍http://www.3gpp.org/ftp/Specs/html-
tion/transit network to use to reach a destination be-
info/23228.htm⬎.
comes an integral part of the signaling peering decision [3] G. Camarillo and M. A. Garcia-Martin, The 3G
process. IP Multimedia Subsystem (IMS): Merging the

DOI: 10.1002/bltj Bell Labs Technical Journal 47


Internet and the Cellular Worlds, 2nd ed., John [15] International Telecommunication Union,
Wiley and Sons, Chichester, England, Hoboken, Telecommunication Standardization Sector,
NJ, 2006. “Resource and Admission Control Functions in
[4] G. Camarillo, A. B. Roach, J. Peterson, and Next Generation Networks,” ITU-T Rec. Y.2111,
L. Ong, “Integrated Services Digital Network Sept. 2006, ⬍http://www.itu.int⬎.
(ISDN) User Part (ISUP) to Session Initiation [16] O. Lendl, “Publishing SIP Peering Policy,”
Protocol (SIP) Mapping,” IETF RFC 3398, IETF Internet Draft, Dec. 2005,
Dec. 2002, ⬍http://www.apps.ietf.org/rfc/ ⬍http://www.ietf.org⬎.
rfc3398.txt⬎. [17] O. Lendl, “The Domain Policy DDDS
[5] R. Daniel and M. Mealling, “Resolution of Application,” IETF Internet Draft, Feb. 2006,
Uniform Resource Identifiers Using the Domain ⬍http://www.ietf.org⬎.
Name System,” IETF RFC 2168, June 1997, [18] R. Mahy, “A Minimalist Approach to Direct
⬍http://www.apps.ietf.org/rfc/rfc2168.txt⬎. Peering,” IETF Internet Draft, June 2006,
[6] European Telecommunications Standards ⬍http://www.ietf.org⬎.
Institute, “Interconnect and Routing Issues [19] M. Mealling, “Dynamic Delegation Discovery
Related to Numbering Naming and Addressing System (DDDS) Part Three: The Domain
(NN&A) for Next Generation Networks (NGNs),” Name System (DNS) Database,” IETF RFC 3403,
DTR/TISPAN-04006-Tech, ETSI TR 184 006, Oct. 2002, ⬍http://www.apps.ietf.org/
v0.0.6, Feb. 2006, ⬍http://www.etsi.org⬎. rfc/rfc3403.txt⬎.
[7] P. Faltstrom and M. Mealling, “The E.164 to [20] J.-F. Mule, “SPEERMINT Requirements for SIP-
Uniform Resource Identifiers (URI) Dynamic Based VoIP Interconnection,” IETF Internet
Delegation Discovery System (DDDS) Application Draft, June 2006, ⬍http://www.ietf.org⬎.
(ENUM),” IETF RFC 3761, Apr. 2004, [21] R. Penno (ed.), “SPEERMINT Peering
⬍http://www.apps.ietf.org/rfc/rfc3761.txt⬎. Architecture,” IETF Internet Draft, Aug. 2006,
[8] M. Foster, T. McGarry, and J. Yu, “Number ⬍http://www.ietf.org⬎.
Portability in the Global Switched Telephone [22] J. Rosenberg and H. Schulzrinne, “Session
Network (GSTN): An Overview,” IETF RFC Initiation Protocol (SIP): Locating SIP Servers,”
3482, Feb. 2003, ⬍http://www.apps.ietf.org/ IETF RFC 3263, June 2002, ⬍http://www.apps.
rfc/rfc3482.txt⬎. ietf.org/rfc/rfc3263.txt⬎.
[9] GSM Association, “Inter-PLMN Backbone [23] J. Rosenberg, H. Schulzrinne, G. Camarillo,
Guidelines,” Doc. IR.34, Revision 3.7, Apr. 2006. A. Johnston, J. Peterson, R. Sparks, M. Handley,
[10] GSM Association, “IMS Roaming and and E. Schooler, “SIP: Session Initiation
Interworking Guidelines,” Doc. IR.65, Revision Protocol,” IETF RFC 3261, June 2002,
3.5, Aug. 2006. ⬍http://www.apps.ietf.org/rfc/rfc3261.txt⬎.
[11] A. Gulbrandsen, P. Vixie, and L. Esibov, “A DNS [24] H. Schulzrinne, “The tel URI for Telephone
RR for Specifying the Location of Services Numbers,” IETF RFC 3966, Dec. 2004,
(DNS SRV),” IETF RFC 2782, Feb. 2000, ⬍http://www.apps.ietf.org/rfc/rfc3966.txt⬎.
⬍http://www.apps.ietf.org/rfc/rfc2782.txt⬎.
[12] M. Haberler, M. Hammer, and O. Lendl, “A (Manuscript approved August 2007)
Federation Based VoIP Peering Architecture,”
IETF Internet Draft, Sept. 2006, JEAN-PHILIPPE JOSEPH is a member of technical staff in
⬍http://www.ietf.org⬎. the Advanced Network Planning and
[13] International Telecommunication Union, Modeling group at Alcatel-Lucent in Murray
Telecommunication Standardization Sector, Hill, New Jersey. He holds a B.S. in
“The International Public Telecommunication electronics engineering from La Faculté des
Numbering Plan,” ITU-T Rec. E.164, May 1997, Sciences de l’Université d’Etat d’Haïti in
⬍http://www.itu.int⬎. Port-au-Prince, and an M.S. degree in electrical
[14] International Telecommunication Union, engineering from Polytechnic University of New York in
Telecommunication Standardization Sector, New York City. Mr. Joseph’s current research interests
“Network Grade of Service Parameters and include IMS-based architectures and solutions for
Target Values for Circuit-Switched Services in network evolution and service integration, IMS
the Evolving ISDN,” ITU-T Rec. E.721, May signaling peering, network modeling, and
1999, ⬍http://www.itu.int⬎. optimization. ◆

48 Bell Labs Technical Journal DOI: 10.1002/bltj

Potrebbero piacerti anche