From Voice Processing To Data Routing And Beyond: The Evolution Of Residential VoIP At The Processor Level
2004 was a turning point for the residential VoIP market. In North America, we witnessed independent service providers like Vonage lead the charge, while other suppliers, such as AT&T and Primus, began rolling out new VoIP services as well. Traditional telecom service providers and cable multi-service operators (MSOs) also launched testing and early deployment phases. As the focus now shifts from the feature-rich and lucrative enterprise market into the consumer space, a new series of challenges faces VoIP providers. Retail channels such as Best Buy and Staples are offering VoIP products from manufacturers like Linksys and Motorola, resulting in commoditization and pricing pressures that will challenge providers, OEMs, and solution providers across the board.
As the industry takes steps to move VoIP deployments across a broader range of the market, a variety of end product configurations are emerging, including solutions that not only integrate the voice functionality but the combination of the voice gateway and home router as well. The early adoption of terminal adapter products, providing only basic analog to IP conversion for VoIP service, is being replaced by a much more feature rich voice gateway solution that encompasses both voice and data router functionality in a single solution. At the end of the day, service providers will live or die by features and capabilities — and many of the processors that are out there today cannot handle this. Price wars and entry into uneducated retail channels are setting the performance bar too low.
Residential Processor Performance Requirements
While some residential voice gateways are single POTS or FXS (RJ11) channels, the most common configuration in the residential VoIP space today is two full-featured voice ports. These ports must support voice and fax relay (the ability to transmit fax information reliably). Fax is most commonly characterized by the T.38 protocol. Voice encoding requirements always include the G.711 PCM vocoder but most service providers also require low bit rate vocoders (G.729ab, G.723.1) for low- bandwidth broadband connections, such as DSL lite. For full voice quality, each channel should have the complete, robust voice processing system that includes excellent echo cancellation, voice activity detection, adaptive jitter buffer/voice playout system, tone detection and generation, RFC 2833 for DTMF relay, a variety of caller ID variations and support for supplementary services such as call forwarding and call transfer.
From an architectural perspective, a residential voice gateway sits behind a broadband modem or fiber/ethernet device. While it provides interfacing for two full voice capable channels, a large number of other functional requirements exist. Among the most significant is WAN to LAN routing. At a minimum, the speed requirements must be the maximum rate capable by the modem. Key routing related features such as Network Address Translation (NAT), device firewall, Quality of Service (QoS) mechanisms that support transmit prioritization of real time related packets (such as voice), and authentication and voice security features must be incorporated in the voice gateway design. The inability to handle these requirements will result in poor quality voice.
Data routing directs data from the external WAN network to a properly addressed computer IP address on the internal LAN. For a residential router, another function included is a firewall to protect the internal LAN from corruption or sabotage through the WAN.
Network Address Translation (NAT) is an Internet standard that enables IP addresses on private LANs to be separate or hidden from corresponding public IP addresses. The NAT function provides the necessary address translations so that data can pass back and forth from the LAN to the WAN and vice versa, while acting as a protection mechanism by shielding the internal IP addresses from public view. IPv4 is the most common type of Internet protocol in use today. There are a limited number of IP addresses available in IPv4 as the allocation of these address are mostly reserved, leaving little for private consumption. NAT has been the enabler, allowing multiple PCs or devices on a home network to appear as a single IP address to the public network, therefore consuming only a single address. This provides a layer of security by isolating the internal addresses from public access and also allows for internal addressing schemes and management without conflict to the public IP addressing model.
A firewall in the residential voice device is an element that serves as an enforcer of security between two networks. It determines which traffic to block and which to allow access to in an internal or private network. Firewalls are configured to protect against unauthenticated logins from the “outside’’ world and keep internal network segments secure.
Some of the features supported by a firewall include:
- Protection against remote login without approval
- SMTP session hijacking
- Operating system bugs allowing remote access
- Denial of service attacks
- E-mail bombs
- Hacker designed macros
- Dubious source routing
Depending on how vulnerable a network is, designers may want to enable protection against all external attacks. However, this maximum protection will take up extra CPU cycles and therefore, reduce performance of the data router.
QoS implementations can vary, although they all typically include some type of service tagging or queuing to allow the prioritization of packets in high traffic conditions. In the case of voice gateway routers, the primary purpose is to ensure that voice packets have priority over data packets since voice packets are delay sensitive. If they arrive too late, or with significant delay in transmission, voice quality suffers.
Security, with respect to device authentication and service provider provisioning, is a key requirement in residential VoIP applications as well. It is critical that only qualified (paying) customers are allowed access to a service provider’s network and services. In addition, providers are now starting to request security features for the voice payload and the voice call setup as well. These new features obviously require additional processing horsepower to execute.
Wireless LAN access point router functionality is becoming an increasingly important requirement in residential voice gateway devices. This level of integration allows customers to essentially have multiple “boxes” in one device. WLAN requirements usually include 802.11b/g functionality, as well as the access point software to properly route data packets between wireless home computers and the broadband modem. Of course, the performance of the access point solution plays a key role in the voice processor performance as well.
The processor selected for residential VoIP applications must handle a large number of simultaneous operations and functional real-time requirements, as detailed previously. The majority of processors in use for VoIP today are System-on-a-Chip (SOC) processors that have integrated additional circuitry such as ethernet, TDM, and memory, with direct relevance to the application.
The processor architecture must be able to handle four to five simultaneous data streams, including:
- The wide-area network, which is typically the broadband interface;
- The local-area network interface, which can be a single PC connection or a three- to five-port Ethernet connection;
- Two channels of telephony (voice); and
- [Often], a WLAN interface.
At the same time, the processor must also handle routing and application level functionality. Ideally, the processor should only inspect and tag for action these data streams, otherwise, it risks becoming loaded with data movement and therefore, is unable to perform data manipulation and processing.
The majority of the VoIP processors utilize a dual processor approach with both RISC and DSP integrated to share the processing load. These devices typically include the DSP function for voice/telephony processing and the RISC processor for network and telephony protocol processing (along with general device management), in a single, integrated solution. The RISC processor will also be used to perform routing functionality. These integrated devices are available in a variety of speeds and performance levels; some designed for basic voice gateway functionality, while others have the capacity to perform routing as well. This architecture is optimized for VoIP applications and often has the DSP and the RISC processor running at different frequencies, offering peak performance for robust voice and data routing.
Some SOCs utilize a single high-speed RISC processor that executes not only the telephony, network protocol, and router functions, but also the voice processing operations as well (which typically run on a DSP). This requires higher speed, and hence, more expensive and higher heat dissipating processors. To run the voice function on the RISC, it usually takes three extra instructions in RISC versus DSP for voice processing. Conservatively, it typically takes at least twice the amount of RISC MHz than it does DSP MHz. Significant care must be taken in sizing the overall solution functionality. If there are not enough MIPs at any given time, voice quality will be affected.
In analyzing the overall processor capability, the speed of the processor clock is only one key criterion. The other critical piece resides in the internal architecture configurations relating to data flow. It is important to minimize choke points in the architecture by relieving the RISC or DSP from the unnecessary task of data movement. Therefore, a distributed DMA architecture retrieves and delivers data to the processor without intervention. When the processing is complete, it moves the data to memory for queue to other peripherals. High-speed switching architectures, bus widths, data bursting capabilities, and intelligent peripherals that can directly move and process data without constant CPU monitoring, play a critical role in overall system performance. This function allows the processor to perform the processing, command and control functions that it is designed for.
Some of the architectural features supporting optimized data flow include:
External Memory Interfacing — External memory interface can affect performance and amount required if it is not designed well. Memory bus width of 16 or 32 bit, clock speed, and bank interleaving capability — features offered on all PC chipsets — can improve performance by up to 40 percent if combined with optimized software.
Processor Cache Sizing — Cache running at processor speed reduces slower external memory fetches and increases performance. Presence and sizing for instruction and data cache can be optimized for the typical processing and traffic requirements.
Internal Bus Switching — Switched central resource architectures with cross-bar functionality can remove blocking effects that allow multiple data streams to flow simultaneously, as well as allow concurrent control register accesses. This enables data to move between any two peripherals that are not involved in data transfer.
Peripheral Configurations — Distributed DMA control and data bursting capabilities can maximize throughput by autonomously moving data with CPU intervention. Dedicated peripheral interfaces can be configured to require no CPU support after initialization, and maximize efficiency by efficiently performing specific tasks without CPU intervention.
Implementation of the capabilities mentioned above can significantly improve processing speed and reduce CPU loading. Optimization of the voice and data processing flows will lead to lower delay and jitter effects, therefore resulting in optimized voice quality.
Looking at the near-term horizon, voice related security requirements are one of the most critical aspects of widespread residential deployment. Security mechanisms for the voice payload, which typically include encryption, authentication, and key exchange, are particularly important. A leading candidate for this functionality is the Secure RTP (SRTP) standard. In addition, the signaling portion of the voice functionality requires secure mechanisms as well. Since most residential voice gateways are SIP-based, SIP TLS is the most popular solution for voice call setup security. Both of these functions require additional processing power to execute.
As wireless carriers begin their rollout of VoIP with connectivity to existing wireless networks, the likely trend will be to include support for wireless-based vocoders, such as GSM-FR and EFR, and EVRC, in addition to traditional VoIP vocoders, such as G.729ab and G.723. Since wireless vocoders require more processing horsepower to execute than the traditional VoIP vocoders, care must be taken to include processor overhead for future growth.
As consumer mindshare and interest in VoIP continues to grow, new equipment configurations and applications will continue to emerge. OEMs will require optimized processor configurations that can be utilized across an assortment of product configurations. The R&D investment in platform and application software, as well as investments in field hardening and interoperability, must be leveraged across future product lines. Selecting optimized VoIP-enabled SOCs, which are designed for the requirements of residential VoIP gateways, will lead to reduction of overall BOM costs and provide the necessary flexibility for new and emerging VoIP applications. IT
Fred Zimmerman is the executive director of CPE solutions at TI’s VoIP business unit. Debbie Greenstreet is the product management director at TI’s VoIP business unit. For more information, please visit the company online at www.ti.com/voip.
[ Return To The February 2005 Of Contents ]