[November 15, 2001]

Voice Over DSL: Success With BLES

BY COLE ERSKINE AND BERT DAVENPORT

Continuing growth in the worldwide digital subscriber line (DSL) market is opening the opportunity for broadband suppliers to augment or replace traditional voice services to residential and small office customers without adding infrastructure wiring. Packet-based voice over DSL (VoDSL) technology allows broadband data connections to carry multiple channels of toll-quality voice along with the network traffic. The Broadband Loop Emulation Service (BLES), defined in Annex A of the DSL Forum TR-039 standard, explains how [download a .doc of the standard].

To appeal to both provider and customer, packet-based voice must interoperate with the Public Switched Telephone Network (PSTN) without requiring wholesale equipment replacement. Providers need to be able to continue offering value-added services such as custom calling and caller identification, without replacing their existing infrastructure. Customers need to be able to easily connect their existing Plain Old Telephone Service (POTS) or ISDN phones, faxes, modems, POS terminals, and the like to the new system. This has been one of the inherent design philosophies of the BLES specification since its inception.

Properly implemented, a DSL channel can become the sole physical carrier for data and multiple voice channels while appearing completely transparent to customer and provider equipment. The DSL Forum has recently defined such an implementation in Annex A of Technical Report TR-039, "Requirements for Voice over DSL." Annex A defines not only the requirements, but the alternatives and options for implementing BLES.

Figure 1 below shows the basic structure of the BLES architecture. It uses two additions to a traditional DSL system: the central office interworking function (CO-IWF) and the customer premise interworking function (CP-IWF). These two blocks perform the translations between signaling and bearer methods used by existing telephony equipment, and those used by BLES. These blocks can be stand-alone devices or integrated into the traditional equipment as opportunity and architecture dictate. For example, the availability of digital signal processor ICs with BLES functionality pre-programmed makes the integration of the CP-IWF with the DSL modem an inexpensive option.

Figure 1: Sample network model illustrating how packetized
voice and data are combined onto a single subscriber loop.
Courtesy of the DSL Forum.

Annex A acknowledges the vital importance of voice quality and service transparency in the acceptability of packetized voice. The BLES service transparency requirement includes support of Custom Local Area Signaling Service (CLASS) and Centrex features, as well as custom calling services and analog phone, fax, and modem services. The voice channels, specified in the parent document TR-036, use G.726 ADPCM to offer a perceived voice quality that is indistinguishable from a POTS line. Channels are selectable for �-law or A-law encoding as required by the installation.

The BLES implementation promises to be more successful for providers, and permit more rapid deployment, than earlier packetized voice approaches using voice over Internet protocol (VoIP).

BLES operates over an Asynchronous Transfer Mode (ATM) network, and is based largely on the ATM Forum's Voice Telephony over ATM (VToA) standard [download a .pdf of the standard]. This standard includes ATM adaptation layer 2 (AAL2), a protocol extension specifically designed to provide real-time voice service, and an underlying infrastructure that supports the quality of service (QoS) functions necessary for customer satisfaction.

ATM offers a lower delay in the voice channel than IP-based systems. ATM's cell size, 53 bytes of which 44 bytes may be voiceband information, results in a relatively small 5.5 ms packet size at the standard telephony system-sampling rate of 8 KHz. The typical VoIP system, which utilizes Real-Time Protocol (RTP), has much larger packet sizes resulting in longer delay. ATM's use of a Virtual Channel (VC) concept, as opposed to a purely "connectionless" IP network, suffers less packet loss and virtually no out-of-order packet arrival, eliminating another source of delay. Further, IP/RTP-based systems don't allow a voice message to interrupt a data message. If a high-priority voice packet becomes available immediately after transmission of a lower-priority data packet has begun, the voice packet suffers queuing delay. All these delays impose longer jitter compensation requirements on the system.

High-quality voice cannot come at the expense of the customer's use of broadband for data traffic, however. To prevent that compromise, ATM supports dynamic bandwidth allocation. Bandwidth is reserved for (high-priority) active voice channels, but unused voice bandwidth can be allocated to (lower-priority) data on a cell-by-cell basis. AAL2 also supports variable bit rate voice services, where a cell not being used to send voice packets when the conversation is not active can be allocated to data packets. This permits the system to use silence suppression to augment data traffic bandwidth.

VoIP networks often use high-compression coding algorithms, such as G.723.1/A and G.729A/B, to reduce the impact of voice on data traffic. But this approach has voice quality penalties. The G.711 A-law/�-law and G.726 ADPCM speech coding algorithms specified by VoDSL BLES provide higher perceptual speech quality than the high-compression voice coders. Also, because G.711 and G.726 are sample-based algorithms that can fully utilize any cell or packet size, they can offer significantly lower delay than G.729A/B and G.723.1/A, which operate on 10 msec and 30 msec fixed frame sizes, respectively.

Service transparency issues are also affected by the choice of speech coder. Dual Tone Multi Frequency (DTMF); CLASS features such as caller ID and call waiting; and other signals pass in-band through the G.711 and G.726 coders. G.723.1/A and G.729A/B do not reliably pass DTMF tones or CLASS feature signaling tones in-band. As a result, VoIP systems must utilize separate "tone relay" algorithms running in parallel with the speech coders to detect these events and pass them as out-of-band messages. If not implemented carefully, these algorithms may cause parts of the voice signal to be lost, lowering speech quality.

G.711 and G.726 also provide efficient fax and modem support. Low bit rate fax calls pass through G.726 transparently. High-bit-rate fax calls are preceded by a 2100 Hz echo canceller disabler tone, which is passed in-band transparently. When this tone is detected while G.726 is active, logic can be incorporated that supports falling back to G.711, which passes both high- and low-bit rate fax calls transparently. The same scenario is true for modem support.

By offering toll-quality voice and telephony service transparency while maintaining a high data traffic bandwidth, BLES avoids many current sources of customer dissatisfaction with packetized voice over broadband. The service transparency also allows customers and telephony service providers to continue using existing equipment, keeping the cost of switching to packetized voice down. At the same time, this switch opens the doorway for adding more voice channels without additional wiring, so customers get more services, and vendors have more revenue opportunities. Together, these factors spell success with BLES.

Cole Erskine is chief technical officer and founder of VoicePump, Inc.; Bert Davenport is manager, applications engineering. VoicePump, which is a wholly owned subsidiary of DSP Group, Inc., develops integrated silicon solutions that incorporate DSP Group's DSP cores and mixed-signal technology.