Agreed, an interesting confluence of concepts, technologies and cross-venue analogies. Whoa, one has to be careful not to comment about any concept here without taking into consideration related, dependent and supporting technologies and their analogies;)
So, I'll be taking this slowly. I'd like to recommend an engineering read on VoIP titled: "Bulking Up for VoIP" from the July 2001 issue of Communications Systems Design Magazine csdmag.com , which I think serves as a good primer or refresher and ground leveler for discussion going forward. Go to the url for graphics. The text is copied below for posterity.
csdmag.com
Bulking up for VoIP
To build carrier-class, high-density VoIP gateways, engineers must consider stringent DSP and software requirements to bridge the analog and digital worlds.
By Dennis Gatens and Brian Glinsman
Communication Systems Design
(07/01/01, 04:48:14 PM EDT)
Evolutions in packet-based voice technology are evolving to enable a single network to deliver a new generation of business and consumer voice and data services. There is no doubt that the convergence of voice and data services on a single, next-generation packet-based network is on the horizon and will eventually replace circuit-switched networks.
For equipment manufacturers to be successful and meet the requirements of service providers, they must offer scaleable, full-featured carrier class gateway solutions that meet demands for high-density, low-power, and low-cost per voice channel. These platforms will be enabled by a new generation of programmable digital signal processor (DSP)-based solutions. These solutions make it possible to run many voice and fax channels on the same DSP, while significantly reducing the cost and power per port, thus increasing the overall channel density of the solution. But simply approaching the problem with solutions that are high in channel density is not enough.
A system-level approach is necessary to design a solution that will meet the requirements of a service provider. This approach takes into consideration the voice features required to support toll quality voice, the extent to which external devices are required to support the DSP resources, and the method for aggregating high volumes of voice channels, while moving the traffic without bottlenecks in the system.
In a few short years packet-based voice has evolved from a technology demonstration to an integral part of next-generation networks and services. The accelerated pace at which this technology has evolved can be attributed to several key factors, including voice-over-IP (VoIP) technology feasibility, DSP-based technology with low power consumption, and scaleable channel density, enabling transport technologies (such as DSL and cable). Finally, interoperability and emerging standards have allowed this technology to come to fruition.
Gateway Implementation
VoIP gateways are being implemented for a number of applications ranging from small two-port integrated access devices (IADs) for the consumer and telecommuter to large carrier-class gateways used as class 5 switch replacements.
Packet-based voice provides several advantages over circuit-switched voice. First, the bandwidth required for a voice call can be significantly reduced through the use of voice activity detection (VAD) techniques and low bit rate codecs. VAD removes silence, which accounts for as much as 40% of the voice information that is transmitted. Low bit rate codecs reduce the amount of bandwidth for a voice call from 64 kbps to as little as 5.3 kbps.
Second, based on open standards, a packet-based voice and data infrastructure allows faster time to market for new features and enhancements, with third-party developers offering products to service providers. And finally, capital investment for packet-based platforms is significantly less than the circuit-switched equivalent, while operating costs are significantly reduced through the consolidation of platforms and network management applications.
Large scale VoIP networks have become increasingly feasible as DSP-based technology has been refined to provide lower power and higher densities, with manufacturers meeting the challenge of interoperability. Silicon and software technology for VoIP has evolved to support hundreds of channels and will soon support OC-3 channel density on a single chip (greater than 2,000 channels).
Solution Density
Service providers have a great deal to gain through VoIP implementations. They envision new revenue opportunities in leveraging the open architecture of packet-based services to quickly bring to market new and enhanced services, while reducing operating costs through the deployment of packet-based networks. Service providers, however, will not sacrifice toll quality voice for the benefit of deploying a single converged infrastructure.
Solution density allows service providers and platform developers to more clearly understand and implement high-density VoIP solutions. It requires that VoIP solutions cannot be evaluated merely for the number of channels on a chip. From a system engineering perspective, a solution must be evaluated on how the combination of system elements deliver a complete solution with lowest power and smallest area without compromising voice quality and features.
Solution density thus refers to the optimization of channel density, power, architecture, integration of system functions, and I/O with the required software-based features for the targeted application in the case of high-density carrier class gateways. The solution density concept provides a framework to evaluate solutions based on the integration of all major system elements and features by considering factors beyond just the performance for any single component. A solution density checklist would include architectural considerations, voice and fax features, and other considerations such as development tools.
Quality And Reliability
Standards for voice quality and network reliability are the same for packet-based networks as they are for traditional telephone networks. Customers expect service quality to be consistent with traditional telephone networks. Therefore, the fact that their service is packet-based voice should be transparent. Only through the presence of new and enhanced services should a user realize that they are utilizing a packet-based network and not the traditional TDM-based infrastructure.
High-density VoIP implementations are often discussed in terms of carrier-class or toll-quality, which are universally associated with high-quality voice services. These terms set the expectation of quality for service providers and, more importantly, their customers. Features such as tone processing, packet playout, voice activity detection (with comfort noise generation), and variable echo cancellation are key in meeting quality expectations.
Scaling Up
Scaling VoIP networks to large traffic volumes without degradation of voice quality is critical for the deployment of a packet-based infrastructure. In most cases, channel density will be limited by the power requirements for the total system. Power and cooling levels must remain within industry guidelines (such as the Telcordia Network Equipment Building Standard [NEBS] that specifies a maximum of 1,275 W for a 23-ft. bay with forced air cooling). For platform developers striving to offer the highest level of scaleability, this criteria can only be met with a solution that is optimized for power and area while maintaining carrier-class features. Therefore, the most important specification is not channels per chip but power per channel.
Flexibility is an overused word in communications, but generally it means the ability to add new services and to remain current with evolving standards. VoIP gateway platforms based on solutions that support features beyond pulse code modulated (PCM) voice (features such as low bit rate codecs and fax relay) enable service providers to add services without the disruption and expense of replacing equipment. These software-based features allow platform developers to distinguish their gateways from the competition. From a solution density perspective, architectural considerations such as the sizing of DSP resources and memory will determine a solution's level of flexibility.
Carrier-Class Voice Features
Features and functionality that are critical to delivering toll quality voice and bundled voice and data service enhancements must be implemented in software. This includes a number of complex, computation-intensive algorithms that platform providers will need to consider with the ongoing emergence of new standards, the demand for new and enhanced services, and the requirement for global interoperability.
Software features critical for toll quality voice services include echo cancellation, voice compression, packet play-out software, tone processing, fax/modem support, packetization, signaling support, and network management.
One of the keys to high-quality VoIP is having a hardened line echo canceller that can properly cancel echo, which is present even in a conventional POTS network. In a POTS network, echo is acceptable because delay is less than 50 ms and the echo is masked by the normal side tone that every telephone generates. Echo becomes a problem in packet networks because the delay is almost always greater than 50 ms, thus requiring echo-cancellation techniques as part of the VoIP solution. The ITU defines echo cancellers performance requirements.
The original standard for echo cancellers was ITU Recommendation G.165, and a more stringent set of requirements is provided in ITU Recommendation G.168. These standards provide a series of objective performance tests but do not address implementation nor do they address subjective performance.
A good echo canceller must remove echo well, including at the start of a call as well as preventing any form of echo during a call. Also, it should handle double talk (when both sides talk simultaneously), and not clip the voice at the beginning or end of a double talk voice spurt. It should handle background noise well, including high background noise and variable background noise, and exceed G.165/ G.168 and will support future (more stringent) ITU echo canceller standards, such as G.168-2002. Field proven considering that compliance with G.165/G.168 alone is no guarantee that the echo canceller will work properly in real life situations.
Echo cancellers should provide fast convergence time, low residual echo (depth of convergence), reliable detection of double-talk without divergence or clipping, and proper handling of background noise and narrow band signals. They should support up to a 128-ms tail (often specified for carrier class solutions) including support for multiple reflections over the entire 128-ms tail. They should also support redundancy and be capable of dynamically tracking echo path changes resulting from conferencing, call transfers, and permanent off-hook connections. Finally, they should behave properly in the presence of a 4-wire connection and low hybrid attenuation, and have built-in configurability and instrumentation.
Packetization, Signaling, And Management
Packet encapsulation should be performed in DSPs to facilitate scaleability and flexibility, and should include VoIP, real-time protocol (RTP), real-time control protocol (RTCP), a ATM adaptation layer 2 (AAL2), and AAL1xN for video conferencing streams. These should be supported on a per channel basis to support hybrid ATM/IP networking equipment. Another important feature is network channel switching, or the ability to route from packet network to packet network. The routing can include the ability to transcode the voice payload and/or change the encapsulation format, for example VoIP (RTP) to AAL2 or G.726 to dial tone multifrequency G.729AB.
Signaling support is another essential DSP element, and should include full tone detection and generation capabilities, including DTMF, MF R1/R2, SS7 COT, call progress tones and bidirectional tone processing. The two primary signaling types that must be supported are channel associated signaling (CAS) and common channel signaling (CCS). A scaleable, high density solution requires support in the DSP to detect signaling changes to reduce the processing requirements of the signaling host. A gateway should also provide play out of service announcements (for time-division multiplexing or packet network direction).
Any communication system should be able to discover, isolate and remedy problems as quickly as possible to minimize or eliminate the degree to which users are impacted. Configuration on a per-channel basis is important (including configurable country code information), as are per-channel statistics/status reporting, real-time trace and diagnostics, support for redundancy, and Bellcore test line functions.
High-density VoIP architectures are driven by a number of critical elements, including power (mW) per channel, cost per channel (silicon and hardware), software and intellectual property licensing costs, and channel density per square inch (see see Figure 1). System partitioning, including packet aggregation and routing, is critical, as are network management capabilities to address high availability and accountability.
A Total System Concept
Cost, power, and area must be evaluated on a total system basis and must be a function of the features and capabilities supported. see Figure 2 illustrates a high-density design consisting of hardware modules for alarm monitor & control (M&C), call processing, PSTN, packet, backplane, and VoIP interfaces.
In this system, the M&C module performs the overall network management for the equipment, including per-channel configuration, status and statistics collection, call record reporting, and alarm processing. The call processing modules perform call establishment and call tear down for the system and performs interworking functions between the PSTN and packet network. Depending on the application and location of the equipment, the system could perform PSTN telephony signaling such as SS7, ISDN, TR08 and TR303, or VoIP network signaling such as H.323, media gateway control protocol (MGCP), Megaco, session initiation protocol (SIP), and ATM broadband local emulation services (BLES).
There are two basic architectures for call processing - the processing is either centralized or distributed with the VoIP modules performing lower levels of the signaling protocols.
Traditionally, PSTN interfaces for VoIP consisted of T1 (24 channels) and E1 (30 channels).Today's high-density VoIP systems typically have multiple DS3 (28 T1s or 672 channels) and even multiple OC-3 (2,016 channels) PSTN interfaces, as manufacturers offer equipment capable of handling in excess of 100,000 voice channels in a single rack of equipment.
The packet interface modules provide the interface to the packet switch network, the two most prevalent of which are ATM and IP. Depending on the application, VoIP equipment may be ATM-centric, IP-centric, or support a hybrid of both ATM and IP voice. In many cases, it is important for the equipment to support both voice over ATM (VoATM) and VoIP on a per-call basis to provide interworking between the ATM and IP worlds. Packet interfaces include OC-n (OC-3, OC-12, etc.) optical interfaces for ATM and packet over SONET (POS), as well as multiple 100 Base-T and Gigabit Ethernet interfaces. A switch fabric module performs the routing of cells or packets through the system. Line cards fill out the appropriate header information that is used by the switch fabric to direct cells (packets) to the appropriate line card or external interface.
A typical VoIP module consists of a farm of DSPs that perform the actual conversion of the voice streams between the PSTN and packet worlds. In the PSTN-to-packet network direction, the VoIP modules receive 64-kbps data streams from the PSTN interface modules and output packets or cells to the packet interface modules.
Similarly, in the direction of packet network to PSTN, the VoIP modules receive packets or cells from the packet interface modules and output 64-kbps streams to the PSTN interface modules. The DSPs are controlled by a host processor that is responsible for configuration and software download of the DSPs as well as assisting in call establishment and termination.
In order to offer a highly scaleable platform capable of concentrating a large number of VoIP channels, aggregation logic is required (see Figure 3). This logic is required to aggregate packet streams from multiple DSPs to the backplane and packet network interface(s), route incoming packets from the interface to the appropriate DSP, provide a standard interface to the backplane/packet network interface, and to filter network management and call setup and teardown information to a host processor.
Backplane Interfaces
There are many different backplane interfaces that are used in systems such as these. Most typical are cell bus variants as well as packet over SONET (POS). TDM samples from the PSTN can be relayed over an H.110 TDM bus or the PCM samples can be encapsulated in ATM cells to be sent over the same cell bus that is used for packet traffic.
Solutions based on the concept of solution density that offer high density low power and that provide full-featured performance will enable a new generation of high-density, carrier-class gateway platforms. These platforms will enable service providers to deploy a common packet-based infrastructure enabling a new generation of enhanced voice and data services for businesses and the consumer.
Brian Glinsman is executive director of product management for Telogy Networks, a Texas Instruments company. He holds a BSEE from Rensselaer Polytechnic Institute in Troy, NY and can be reached at bglinsman@telogy.com.
Dennis Gatens is the Director of Product Management for the Broadband Terminals division of Carrier Access. He holds a BSEE degree from Virginia Tech and a MBA from Radford University. He can be reached at dgatens@carrieraccess.com. |
