|
VoDSL: The Facts Behind AAL2
By George H. Dobrowski and Ajay Sharma
Voice over DSL (VoDSL) is changing the traditional telecom landscape
in radical ways. For the first time, it has taken relatively high-speed ATM into the home, eliminating the last-mile bottleneck.
Referred to as the “killer application,” VoDSL is spreading like wildfire. The DSL equipment market is estimated to be worth $3.2 billion, with approximately 3 million subscribers in service this year, and is expected to reach 28 million by 2003.
1
So what’s making it all fly: Asynchronous Transfer Mode Adaptation Layer 2 (AAL2) and its
allied protocols.
The AAL2 story
ATM is a cell-based technology that blends both circuit- and packet-mode communication mechanisms. For traffic to be carried over an ATM network, some refinement is needed to adapt the application (circuit or packet) to ATM transport. This refinement is called ATM Adaptation Layer (AAL), and several types have been standardized by the ITU-T: AAL1,
AAL2, and AAL5.
Voice over ATM (VoA) has been around for some time but has remained in the background because a feature-rich and robust voice network already existed. Its absence from the mainstream became apparent as rival technologies like Voice- over-Frame Relay (VoFR) and Voice-over-IP (VoIP) solutions started to gain ground. By using spare bandwidth available on WAN links, the adaptation of packet-based voice principles was devised as a means to reduce costs in corporate or enterprise
networks. The cost savings were significant. Savings were not only attributed to compressed packet-based voice requiring less bandwidth, but also to consolidation, which eliminated the parallel overlay networks. While the industry recognized the potential of carrying VoA, the embedded carriers/service providers were not economically motivated to migrate it — until VoDSL hit the scene.
2
However, unlike VoIP or VoFR, the initial VoA specification never gained substantial ground among users because of
its limited applicability, that of trunking only.
VoA made its debut in the form of the Circuit Emulation Service (CES) in 1995. This service allowed T1 or T3 to be mapped statically to an ATM virtual channel connection (VCC). The service allowed for two modes of operation: unstructured, where the whole T1 can be mapped to an ATM VCC; and structured, which allowed users to hand pick the individual DS0s of a T1/T3 span. The service was based on converting T1 or E1 into a stream of ATM cells that is
transmitted on a constant bit rate (CBR) basis, consistent with circuit-mode communications. No matter what the actual traffic, the bandwidth used for that circuit connection always remained the same. Because it is a CBR connection, the customer pays for the entire bandwidth, irrespective of whether any active voice calls are being carried. CES uses ATM Adaptation Layer 1 (AAL1) protocol and is widely deployed today. This protocol is not for voice, however, but is used to transport circuits such as T1/T3
and E1/E3 in bulk transparently over the ATM network as a private-line service for business enterprise networks. While this strategy may be effective for large corporations wanting to interconnect their existing private branch exchanges (PBXs) transparently and for LAN interconnect, it is not effective for individual voice connections because CES was designed for a trunking application.
To alleviate the problem, the ATM Forum introduced an efficient implementation of CES, known as Dynamic Bandwidth
CES, (DB-CES). This service allows the dynamic addition and deletion of the individual DS0 channels in reaction to on-demand call traffic. Around the same time that DB-CES was announced, the ATM Forum also introduced its first specification for voice and telephony over ATM (VTOA) using CES/AAL1 for end-to-end VoA calls. VTOA provides a complete architecture (including signaling) to perform TDM-to-ATM interworking over a public or private network. Meanwhile, packetized voice was already hitting the
mainstream.
Two technologies — VoFR and VoIP — were becoming increasingly popular. VoFR, which was the de facto standard for WAN connectivity, provided consolidation cost benefits by putting data and voice onto a single connection between enterprise users. VoIP, on the other hand, was already riding on the ubiquitous nature of IP and avoiding voice access charges and had begun to enjoy phenomenal growth for certain constrained environments. But it could never be completely satisfactory as a general
solution unless the Internet could provide quality of service (QoS) guarantees. In addition, you could only communicate with users that had an IP address and the same equipment end-to-end. Common to both technologies, however, was a standard way to transport packetized voice and other signaling information: Real Time Protocol (RTP) for VoIP applications, and FRF.11 for VoFR applications.
AAL2 to the rescue
The lack of an effective variable bit rate (VBR) adaptation layer, which supported real-time dependent applications requiring guaranteed QoS, hindered the transport of compressed VoA. The AALs that existed at that point — AAL1 and AAL5 — were developed based on different application criteria. As a result, ATM’s role in carrying packetized voice was seriously limited. But with the standardization of AAL2, this shortcoming was resolved.
The AAL2 Common
Part Sublayer (CPS: ITU-T Recommendation I.363.2) protocol defines a packet header for voice packets that can be used for other real-time media such as video. It contains a channel identifier (CID), a length indicator, a way of identifying the type of voice encoding and/or compression the packet contains, and header error-check protection. In addition to defining the packet format, the protocol defines the method by which a stream of AAL2 packets may be assembled into ATM cells. AAL2’s flexible yet
powerful infrastructure can deal with low-rate, short- and variable-length data in a real-time sensitive environment dominated by voice and facsimile applications. Some of AAL2’s unique capabilities, which were previously missing from traditional VoA, are as follows:
Transfers short- and variable-length packets containing compressed-voice payload and signaling information
Allows packets with variable interarrival times to
accommodate packets from different media sources, such as different rate codecs or silence suppression descriptors
Allows packets to cross ATM cell boundaries for improved packing efficiency of an ATM cell
Allows for partially filled cells needed to minimize jitter and guarantee real-time response
Allows multiple connections to be multiplexed on one virtual channel (VC).
AAL2 also
introduced another level of switching in the ATM virtual path identifier/VC identifier (VPI/VCI) hierarchy — the concept of a channel. An AAL2 VC can have up to 255 channels, each one identified by a CID. The data payload consist of at most 45 bytes, which in turn is protected using a header error check (HEC) field.
Figure 1
shows the details of an AAL2 cell.
Almost there
Compared to VoIP or VoFR, however, the challenges for VoA were not over yet. One crucial element was still missing: a standard way to carry voice, in-band tones, dialed digits, and signaling messages over an AAL2 channel. Without such a standard, any large-scale deployment of AAL2-based VoA was inconceivable. The industry realized this void and soon started working to fill the gap. The effort resulted in the AAL2 Service Specific Convergence Sublayer for Trunking
(ITU-T Recommendation I.366.2), which enables an AAL2 VC to behave like PSTN line and to provide all the functionality associated with it. A complimentary specification, the AAL2 Service Specific Convergence Sublayer for Frame Data (ITU-T Recommendation I.366.1) was also introduced to carry large chunks of data and out-of-band signaling information over AAL2 channels. The combination of I.366.1 and I.366.2 completed the VoA transformation.
Figure 2
presents a
high-level view of the solution.
The I.366.1 and I.366.2 duo, in collaboration with the AAL2 CPS, presents a comprehensive solution for carrying a variety of media, as well as information payloads such as:
Uncompressed/compressed voice
Silence insertion descriptor
Circuit-mode digital data
Narrowband ISDN messages (as frame-mode data)
Dialed digits and tones
Channel associated signaling (CAS) bits
Demodulated facsimile
Alarms and local management information (as frame mode data).
All media information and signaling payloads are carried using two types of packet formats referred to as Type 1 and Type 3, respectively. The Type 1 packet is unprotected in terms of a cyclic redundancy check (CRC) and is not
retransmitted in case of loss. It is the default packet format used on an AAL2 VC. The information payload is typically used for raw data transfer, such as a fax image or audio. The Type 3 packet, is not only protected with CRC-10, but is also transmitted three consecutive times to prevent loss, ensuring that the control information reaches the destination in one piece. This feature is referred to as triple redundancy. This packet format is ideal for the transport of such control data as digits, signaling
bits, and fax-control information, where incorrect communication results in misrouted connections and/or interoperable connections.
Information is transmitted in the form of primary and secondary streams. Information like compressed voice, facsimiles, and circuit mode data are classified as primary stream candidates, and only one of them can be active at a time per channel. However, information like digits, signaling bits, alarms, and frame mode data are considered to be secondary stream types and may
be sent along with a primary stream in the same channel.
Before voice communications can commence, both the sender and receiver vocoders must agree on a profile. A profile is simply a list of capabilities of the different vocoders available at each end. If, for example, the type of voice codec to be used is already known, then these procedures are not necessary before every call. By transporting the profile index, each side can effectively convey the necessary information about the vocoders that
are supported. This simple mechanism also makes it possible to change a vocoder in mid-call without the intervention of any external procedure. This capability is extremely useful in modulating the voice quality, depending on the network condition and/or service being provided.
To conserve bandwidth, a silence descriptor is used any time a silence interval is detected throughout the life of the call. The receiving entity can generate comfort noise when information is received so that the user does not
think the connection has been dropped.
The transport of large chunks of data including alarm, performance, and configuration data is done using the services of the I.366.1 protocol. This protocol has almost all the features of the traditional Signaling Adaptation Layer (SAAL) protocol and, in fact, reuses much of the specification from SAAL (SSCF-UNI and SSCOP).
3
\
Since an AAL2 channel can occasionally carry large chunks of data, some kind of prioritization is needed between the
delay-sensitive data (such as voice) coming from the I.366.2 application, and the non-real-time data coming from the I.366.1 application. Without prioritization, a data burst could severely effect the quality.
The ATM Forum’s existing VTOA specification (af-vtoa-113), the Loop Emulation Service (LES) over AAL2 (nearing completion), and BLES heavily exploit these capabilities to provide a robust, scaleable solution.
Why the new attention on voice?
Now that there is a more efficient mechanism to carry voice, why all this sudden attention in emulating POTS lines when the Internet is growing at such tremendous rates? In 1999, the volume of data traffic hit parity with the volume of voice traffic. However, with data-traffic growth rates ranging from 45 to 300%, depending on what point you are in the network (for example, access/edge versus core backbone), and voice growth in
the 5 to 6% range, it is clear that data will significantly dominate all network traffic by 2004.
The answer is simple — follow the green. Today, voice accounts for approximately 80% of all communication revenue, and voice and voice-enabled data services associated with Web-based e-commerce and call distribution services are forecasted to continue to account for the lion’s share of the revenue in 2004 — up to 70% by most estimates. A carrier or service provider without voice may
not be economically viable over the long run.
The implication of this shift in traffic from circuit-based voice to packet-based has resulted in the realization that rather than design networks optimized for voice and do the best you can to carry data traffic, it’s better to deploy a network architecture based on packet technology and adapt voice to it. With a packet infrastructure, greater efficiencies can be gained. The next question is which packet technique — ATM based or IP based? AAL2
provides the real-time VBR (rt-VBR) ATM packet solution. Currently, VoA with a digital subscriber loop physical layer is emerging as the preferred solution (The ATM and Digital Subscriber Group forums have interdependent specifications). The VoIP approach duplicates some of the functions provided by the lower ATM layer, but at present, it does not provide QoS guarantees and has additional protocol overhead. Although efforts like MPLS and diffserv are trying to address this issue, an end-to-end QoS aware
IP network is still in the future. However, a lack of QoS is one thing you need not worry about in an ATM network.
Welcome VoDSL
Every day we hear about terabits of public backbone network capacity (fiber-based), while the last mile throttles the user applications to only kilobits-per-second using modem technology over local copper access loops. Digging up streets to replace
the local copper loops with fiber is not only expensive but would take decades to achieve. Seeing this situation more as an opportunity than an obstacle, a number of vendors have applied their digital signal processing expertise to the problem of mining the copper already in the ground to provide more bandwidth. Couple this with regulatory changes and competitive threats from cable systems and, suddenly, exciting things are happening with digital subscriber line (DSL) solutions. Placing voice over DSL is
known as VoDSL.
The existing VoDSL architecture is built around the concept of the integrated digital loop carrier (IDLC) — but with a twist. The traditional local loop is replaced with an xDSL loop (the x indicates the several types of DSL technologies that can be used). The introduction of xDSL enables broadband speeds over copper-wire pairs. However, for existing narrowband services, the upgrade to xDSL loops must be transparent, and the user still needs connections to the existing PSTN. To
achieve this connection, LES (using AAL2 for narrowband services) was developed by the ATM Forum. LES fulfills the need for an effective transport mechanism able to carry voice, voice-band data, and fax traffic over xDSL.
The architecture involves three key components:
AAL2-enabled customer premises equipment (CPE)
The digital subscriber loop access multiplexer (DSLAM) providing ATM cell multiplexing
A VoDSL gateway.
The solution delivers toll-quality voice and includes the conventional wire-line services such as call waiting and caller identification.
The CPE, essentially an integrated access device (IAD), packetizes the analog/digital voice traffic into AAL2 packets. These packets, along with signaling events, are then sent along a single xDSL to a DSLAM. From there, the voice traffic is aggregated upstream onto a permanent virtual connection (PVC), passing the voice
traffic to the VoDSL, as shown in
Figure 3
. (Another function of the DSLAM is to aggregate data traffic and deliver it to the appropriate ISP. A connection is made from the ATM network directly to the ISP on a nailed-up or SVC basis.)
The gateway completes the call by performing not only the ATM-to-TDM conversion, and vice versa, but also provides the GR303/TR08-based signaling interworking that is needed between the xDSL interface and existing PSTN switch
interfaces.
While the advantage of this approach is that it is transparent to the existing users, today’s VoDSL DSLAM model has a limited application: the packet aspect of the voice is constrained by the VoDSL gateway.
Let’s look at the example of making a call from one VoDSL subscriber to another, as would be the case in a small enterprise or a small residential community. The connection must traverse from the CPE through the DSLAM to a gateway that performs interworking functions
needed to make the connection to an existing PSTN based on GR-303. Finally, the Class 5 switch, after performing its routing and switching, completes this “hair pin” call by creating an outbound call destined towards the terminating subscriber.
Completing a call from one VoDSL subscriber to the other is impossible without going via PSTN and without performing an ATM-to-TDM and TDM-to-ATM conversion.
Tomorrow and beyond
Tomorrow’s architecture will allow much more flexibility by exploiting the capabilities of the distributive switching models that are based on such emerging standards as MEGACOP, being developed by the IETF, or the equivalent ITU-T version, designated H.248. MEGACOP/ H.248 is a new, packet-based con-trol protocol being developed as an alternative to traditional, proprietary, telecom switch-control logic. This protocol achieves a decoupling of
call-processing logic from media switching. The development of this protocol is motivated by the philosophy that the intelligence — the switching and routing — of the network should be controlled via a packet interface and should be readily accessible to an edge de-vice (residential gateway).
In the MEGACOP/H.248 architecture, a call agent (CA) provides all such network intelligence to its gateways. In other words, the CPE itself has access to switching and routing intelligence, as well as
other value-added service-dependent functions. MEGACOP/H.248 enables the control and management of gateways from the CA.
An external intelligence point, the CA is responsible for creating the connection in the gateways, performing signaling internetworking with the PSTN, and supporting packet voice and data over a single, integrated xDSL line. VoDSL, in this new architecture, is illustrated in
Figure 4
.
The gateway uses MEGACOP/H.248 to communicate with a
CA to place a packet voice call. One CA can control many distributed gateways. For redundancy purposes, a single media gateway can communicate with multiple CAs. If the destination (a called number) is a PSTN user, the VoDSL gateway interworks with the PSTN, using GR-303 signaling, and completes the call. Whereas, if the called number resides on the packet network, such as the IP, the VoDSL gateway interworks with the softswitch to enable an end-to-end packet call.
4
Within this architecture, a
VoDSL subscriber calling the other VoDSL subscriber is connected through the PSTN.
This architecture will promote a true end-to-end packet voice network, transforming VoDSL into a complete and independent network. As more and more subscribers start to appear on the packet voice network, the packet network will only expand, and the need for interworking between packet voice-based clouds, such as IP/ATM and PSTN, decrease.
The emerging network architecture opens new combinations of
connections that will require interworking, and will also provide new avenues of revenue generation for equipment and service vendors alike. Table 1 contains a list of possible interworking scenarios using AAL2.
Besides the previously mentioned interworking possibilities, the AAL2 channel itself presents some interesting likelihoods. Because the traditional ATM network element switches only at the virtual path or virtual channel level, the AAL2 channel introduces another layer to this switching
hierarchy. The introduction of this new switching entity has allowed equipment manufacturers and silicon vendors the opportunity to provide features like switching and multiplexing at the channel level.
VoDSL is changing the traditional telecom landscape in radical ways. For the first time, it has taken relatively high-speed ATM all the way to the residence, eliminating the last-mile bottleneck. With this bandwidth freedom comes a new universe of information and entertainment for businesses and the home.
VoDSL is paving the way for traditional ILECs and new CLECs to gradually move towards the next generation of multiservice — the converged packet network.
Ajay Sharma
is the director of software systems at GlobeSpan, Inc. He received his BS in physics from Delhi University and his BSCS and MSCS from City University of New York, NY. He can be reached at
asharma@globespan.net
.
George H. Dobrowski
is the
director of strategic initiatives at GlobeSpan, Inc. He obtained his combined MSEE/CS from Northwestern University, Evanston, IL and a BSEE from IIT, Chicago, IL. He can be contacted at
gdobrowski@globespan.net
.
| 
Puzzled by a network processing design issue?
