Architectural considerations for an Internet Protocol (IP) phone processor

102 International IC ? Korea ? Conference Proceedings Architectural considerations for an Internet Protocol (IP) phone processor Trefor Davies Young Soo Lee Mitel Semiconductor Abstract This paper begins by discussing the IP-based voice systems tech- nology, market, and the factors which influence the quality of voice IP data transmission. This paper then explains the basic requirements of a VoIP system and the corresponding interna- tional standards. Finally, an IPphone design example from Mitel Semiconductor is given, followed by the introduction of the solution of TDM trunk to IP network, and TDM to ATM net- works. Introduction During the mid-'90s, data and voice began to merge, propelled by advances in compression technology. The ubiquitous IP networks, and the desire to cut telephony costs are the major driving forces for the deployment of voice over IP (VoIP). One of the major advantages of VoIP technologies is that they pro- vide a way to use existing network resources and to dramati- cally reduce, or even eliminate, telephony costs. Internet-based telephony is an exciting prospect for every company running a network. For example, Company A has a LAN network and branches both locally and overseas. At each of its sites, there could be a voice PBX as well as network rout- ers. If the voice traffic could be carried both on the LAN, in local area, and on WAN between the cities, the immediate ad- vantages are huge. There will be substantial cost savings in equipment and in call rates, being essentially local rates, for long distance and international calls. Today, the transition from circuit-based to packet-based networks provides vendors of data communications equipment with entrance into a market traditionally served by telephone and PBX companies. The increased competition will acceler- ate the emergence of new services such as integrated email and voice messaging, and videoconferencing over the Internet. In 1999, voice and data converged over the Wide Area Net- work, bringing to fruition Voice over Internet Protocol (VoIP) technology. According to Dataquest, sales of carrier class VoIP gateways reached US $250 million worldwide in 1999, and they are expected to maintain a nearly 100% compound annual growth rate through the year 2003.* Problem Statement Unlike traditional switch networks however, IP networks are connectionless, band limited, and packages from a source are sent to a destination with no guarantee of timely delivery. It poses entirely different engineering challenges to achieve the quality and performance from the IP telephone equal to that of a traditional system. The following are basic requirements for VoIP : 1. Compression Reduce the amount of bandwidth of voice, while maintain- ing high quality. 2. Silence suppression During periods of silence, no bandwidth is used. 3. Signaling for voice traffic Support traditional PBXs and associated signaling 4. Echo control Echo is annoying and disruptive, and must be suppressed in the applications. 5. Quality Of Service (QoS) for network Minimum tolerance for delay, delay variable loss and error rate. 6. Voice switching Follows the QoS network protocol to route a call over the data network or has the ability to route it to the public switched telephone network. Most of these issues can be overcome through a systems- level approach in the development of silicon. There are several considerations that integrated circuit manufacturers can make related to the building blocks, and integration of software and firmware, into the solution. The winning proposition will fulfill the requirements above while achieving the lowest cost and bringing end products to market faster. Performance enhancements to equal POTS systems The most important function of IP phone silicon is to achieve quality and performance equal to that of a POTS system. Introducing voice into the data stream raises challenges in timing, jitter and echo cancellation. Since these effects degrade the quality of voice and performance of the system, they must be eliminated for the market to realize its potential . The International IC ? Korea ? Conference Proceedings 103 minimum tolerance requirement for delay, delay variable, packet error for VoIP system is termed quality of service (QoS). Quality of service (QoS) of network QoS is critical in facilitating the growth of the IP Phone mar- ket. customers used to typically 100% reliability of POTS sys- tems will not settle for an alternative offering less quality, any IP Phone silicon solution needs to support industry QoS initia- tives. Most layer 2 switches deployed today do not use VLAN tagging, thus the best way to prioritize voice traffic is to use Differential Services. Asimple RSVPclient could also be imple- mented as the market demands. QoS mechanisms vary based on the network layer: 1. At physical layer, if PC is behind the phone, the Ethernet bridge will prioritize phone traffic before the PC traffic to ensure that voice calls occur in real-time. 2. At layer 2, the Ethernet bridge driver will set the 802.1p tag for the voice packets in the Ethernet frame to ensure that voice gets priority over data. 3. At the layer 3 network level, the IP silicon will implement a diff-serv code point that can be configured via signaling protocols (like MEGACO/SDP, H323, SIP/SDP), RSVP, management protocols like SNMP, or static configuration, using any standard mechanism. The signaling protocols like SIP and MEGACO do not currently have a way to pass the code points but there is work in progress to add that to the SDP protocol. The H323 standard currently relies on the RSVP mechanism. As layer-3 switches become more com- mon, the value added by layer-2 switches, in terms of VLAN priority, will diminish. Echo control Packet-based VoIPnetworks require voice capabilities that with- stand long delays without introducing echo into the system. To overcome this problem, engineers can use echo suppression and echo cancellation technology. This eliminates residual echoes without removing near-end signals or background room noise, providing extremely good voice quality even when long delays are introduced. Privacy Early versions of the IP phone do not support privacy. Because of the ubiquity and open of the IP network, DES and IP Sec have both been considered as standard methods for encryption of voice traffic in the LAN. Power consumption Power source and power consumption plays a major role in the performance of an IP telephone. A normal POTS telephone powers itself from the telephone line. Current IP phones for the desktop require separate power cables. To gain broad ac- ceptance of IP telephones, it is necessary to move towards al- ternative power methods. One alternative is called Phantom feed, a developmental technology whereby Ethernet segments carry power to enable phones to be powered directly from the line. Optimizing the silicon for low-power operation requires mini- mizing component count and separating analog and digital func- tions, both of which serve to lower the power consumption. In addition, designers can equip the silicon with power down mode, to ensure that all circuitry is powered down when not in use. Example of the VOIP design The first generation of IP Phones demonstrated for the desktop were very expensive. They included several components includ- ing discrete RISC processors, DSPs and an Ethernet hub, to- gether with a codec and other peripheral devices such as flash memory and DRAM. Today, Mitel Semiconductor has a cost-effective solution offering optimal performance. The Mitel Semiconductor solu- tion effectively partitions an IP phone chipset into three sepa- rate components: a processor, a codec and an Ethernet PHY. Separating the analog and digital functions eases implementa- tion and allows system manufacturers to interchange specific devices with their own components that may have already been qualified, such as Ethernet PHYs. As the most critical component, the processor integrates the ARM7 RISC CPU, the OAKDSP and a full 801.1p 10/100 Ethernet Bridge. This complete phone system-on-a-chip also integrates software including the RTOS, signaling stack, and drivers. The ARM7 runs any proprietary signaling protocol together with H323, MGCP(MEGACO in future) and SIP. The OAKDSP handles the signal streaming, tones, and any com- pression algorithms that might be required. Most of the com- monly required vocoders such as G723.1 and G729A are al- ready integrated. The Echo control function is implemented in the DSP and echo canceller ( MT93L15) will be built in op- tionally. In addition to the processor, the phone chipset includes an analog front end with a codec and Ethernet PHY. The MT92303 is a full-featured codec that incorporates drivers for handsfree, speakerphone and handset operation, to provide functionality for the required of next-generation IP phones. The audio inter- faces designed into the MT92303 minimize component count with a direct connection from the loudspeaker to the on-chip amplifier. The device also incorporates differential drivers to achieve high power outputs from a 3.3V supply, about 20 per- cent more power than current 5V designs. Finally, a 10/100 Ethernet PHY completes the chipset. The NWK933 is a single-chip 3.3V solution that is compatible with the auto negotiation section of IEEE802.3u, and provides all of the functionality required of full-duplex operation. In addition, on-chip hardware acceleration allows imple- mentation of DES or IP Sec, and provides high performance over software-based solutions. Power down mode is supported in all the chips, to ensure that all circuitry is powered down when not in use. Addition- ally, implementing presence and mute detect further lower sys- tem power consumption, because it make the processor can enable or disable the circuitry and save power at the system level. Some other Mitel Semiconductor solutions for VOIP The MT90256/91024 PAR (Packetization and Resynchronization) is a device that packetizes up to 1024 TDM (E1,T1, and PDH )voice channels for transmission over IPnet- work, ATM or POS (Packet on SONET). Conversely, the PAR also extracts data packets on the network and synchronizes it for transmission in the TDM domain. It is well-know that ATM supports extensive QoS and ser- vice class capabilities, allowing time-sensitive traffic, such as voice, to be transported across the network in a reliable, jitter- free manner. The Mitel Semiconductor MT90500/2/3 SAR(Segmentation and Reassembly) solution allows systems based on a TDM bus to be interfaced to ATM networks using 104 International IC ? Korea ? Conference Proceedings Presentation Materials ATM Adaptation Layer 1 (AAL1), Layer 2 (AAL2), Layer 5 (AAL5),and Layer 0 (AAL0). The MT90220/1 IMA(Inverse Multiplexing for ATM) de- vice provides ATM system designers with a flexible architec- ture when implementing ATM access over existing TDM trunk interfaces. Conclusion The key to being successful in this emerging market is to pro- vide flexibility and support in systems integration and set-up. Using the chipset as described optimizes flexibility and gets customers to market faster with cost-effective, high-performance solutions at this early stage of the emerging market. Author's contact details Trefor Davies Marketing Manager Mitel Semiconductor Phone: 44 1522 502 894 Fax: 44 1522 502 893 Email: trefor_davies@mitel.com Website: www.mitelsemi.com Mitel Corporation 360 Legget Drive, P.O.Box 13089, Kanata, ON Canada K2K 1X3 Telephone (613) 592-2122 Fax (613) 592-6909 www.semicon.mitel.com International IC ? Korea ? Conference Proceedings 105 106 International IC ? Korea ? Conference Proceedings International IC ? Korea ? Conference Proceedings 107 108 International IC ? Korea ? Conference Proceedings International IC ? Korea ? Conference Proceedings 109