SoCs: DSP World, Cores -> Multiservices over broadband cable

Cable networks have traditionally been designed to deliver analog video to consumer homes. National and local television channels are received at the cable head-end, are rechannelized and multiplexed and then transmitted over a tree-branch cable plant. Each television channel is assigned a 6-MHz slot on the cable RF spectrum. To enable cable plants to carry digitally compressed video, many plants have been upgraded as hybrid fiber coax plants. These HFC plants are two-way capable, with downstream bandwidth in the 54-MHz to 800-MHz range and upstream bandwidth in the 5-MHz to 42-MHz range. A hierarchical distribution topology is created to allow more-efficient delivery of digital content. The head-end is connected to several distribution hubs over an optical fiber ring. Each hub is typically connected to 40 to 100 fiber nodes, with each fiber node servicing 500 to 1,000 homes.

An example allocation of the downstream RF spectrum is as follows:

• Analog video service: 54 MHz to 500 MHz

• Digital services: 500 MHz to 800 MHz

• Digital video service: 500 MHz to 650 MHz

• Internet Protocol services: 650 MHz to 800 MHz

Like the analog channels, digital services are also assigned a 6-MHz RF channel. For two-way Internet Protocol conn ectivity, the upstream spectrum in the 5-MHz to 42-MHz range is used. Upstream channel widths are not fixed a priori as this RF band consists of interference from other RF services such as AM/FM radio.

Downstream capacity of a 6-MHz channel is approximately 27 Mbits/second using QAM-64 modulation (and 38 Mbits/s using the QAM-256 modulation). Digital video services are delivered using 4-Mbit/s to 6-Mbit/s MPEG-2 compressed video streams and require a digital set-top box at the customer's home to decode the compressed stream. A 27-Mbit/s channel allows a multiplexed transmission of four to six MPEG-2 videos.

Generally, an MPEG transport mechanism transports multiplexed MPEG-2 streams. Both scheduled as well as on-demand digital video services are delivered using video servers at the hub or the head-end. The hierarchical distribution mechanism allows serving of popular video content from the hub using video servers with reasonable storage (top 20 movies of the month). A much larger library of v ideo streams is available at the head-end on an on-demand basis.

Internet Protocol services have traditionally included best-effort and premium high-speed data services. With proper quality of service (QoS) mechanisms in place, the Internet Protocol network can be made to support real-time multimedia services such as voice/video telephony, audio and video. The Docsis standard, a layer 1 and 2 protocol, specifies the mechanism for delivering IP services on a cable plant. It is the communication protocol between the head-end cable modem termination system (CMTS) and the client cable modem (CM). An updated Docsis 1.1 under consideration is aimed at providing necessary support to enable multimedia services.

A multiservice multimedia-enabled Internet Protocol network offers many advantages:

• New services and devices can be easily added over a unified network infrastructure that supports standards-based protocols;

• Statistical multiplexing of resources allows efficient networ k utilization;

• Maintenance and upgrade costs are reduced with one network infrastructure instead of multiple incompatible systems.

Essentially, Docsis specifies physical layer and data-link layer mechanisms to enable interactive (two-way) Internet Protocol services over cable. The physical layer consists of the transmission convergence sublayer (for downstream only) and physical media dependent sublayer. The data link layer consists of a logical link control sublayer, link-layer security sublayer and media access control sublayer. A cable modem termination system may also support the Ipv4 network protocol, in which case it is referred to as a cable router. The client Docsis-compliant device is called a cable modem. A cable modem may also be integrated with a digital set-top box.

A common choice is to use QAM-64 on the downstream and 1.28 kbits/s with QPSK on the upstream, giving 30.4 Mbits/s downstream capacity and 2.56 Mbits/s upstream capacity. This gives downstream:upstream b andwidth asymmetry of about 12:1.

Basically, a Docsis cable network is a point-to-multipoint topology. The downstream channel is shared by all modems connected to it. Upstream is a TDMA multiple-access channel where a cable modem transmits in assigned time slots that may be subject to collisions. The cable modem termination system schedules these time slots (called mini-slots) and transmits the mini-slot allocation MAP on the downstream channel. The mini-slots are identified as: initial maintenance slots, station maintenance slots, request slots and data grants.

To support real-time multimedia traffic that demands guaranteed network resources-in terms of bandwidth, latency and jitter-Docsis 1.1 defines unidirectional service flows. Each service flow has a unique 32-bit ID. In addition, upstream service flows also contain a unique 14-bit service-flow ID. Each cable modem is thus assigned two service flows: a primary downstream flow and a primary upstream flow. Additional flows are assigned, st atically or dynamically, in accordance with service requests such as telephony and/or streaming media. The standard specifies support for both unicast and multicast transmissions and uses Internet Group Management Protocol (IGMP) to allow cable modems (or Internet devices, such as a personal computer) to join a particular multicast group.

The cable modem termination system (CMTS) to cable modem access link may be secured using the Docsis baseline privacy plus (BPI+) standard. Basically, BPI+ supports the cipher block chaining (CBC) mode of the U.S. Data Encryption Standard (DES) algorithm to encrypt a single packet data. Cable modems employ baseline privacy key management (BPKM) protocol to obtain authorization and keys from the CMTS. This includes X.509 digital certificates, RSA public-key encryption algorithm and two-key triple DES for cable modem-CMTS key exchanges. Keys have limited lifetimes and must be refreshed by the CMTS.

Bandwidth asymmetry on a Docsis network, coupled with unique b andwidth requirements of real-time and non-real-time Internet Protocol services, leads to interesting implications for aggregated traffic on the Docsis network. Data traffic (file transfer, Web access, etc.) is quite bursty, and TCP sessions are established between Web servers and clients to guarantee reliable data transfer. Multimedia traffic, on the other hand, is persistent and requires a guarantee on bandwidth, latency and jitter. To reduce session latency, multimedia packets are transmitted over UDP. To reduce jitter and to allow synchronization between several media streams, such as audio and video, these UDP packets are marked with sequence numbers, time stamps and stream synchronization information. The real-time transmission protocol (RTP) is a standards-based protocol that allows such markings. Bandwidth guarantees are required from all network devices touched by the media streams (switches, routers, and gateways).

Three different traffic types are considered below: data-only, data-plus-vid eo and data-plus-voice.

• Data-only traffic: Clearly, the 80-kbit/s to 30-kbit/s rate requirement for data-only traffic implies approximately 3:1 asymmetry. This coupled with the 12:1 capacity asymmetry gives a 1:4 ratio for downstream:upstream channel requirements. In other words, a CMTS must be designed in such a way that it supports four upstream channels for each downstream channel.

• Data-plus-voice traffic: A voice conversation requires equal bandwidth in both directions. To understand the CMTS channel density requirement, we use the fact that voice-only traffic demands 1:1 ratio for bandwidth consumption. This, coupled with the 12:1 capacity asymmetry, gives a 1:12 downstream-to-upstream CMTS channel ratio. Clearly, voice traffic imposes significant additional requirements for upstream port density. If data and voice traffic are allowed on the network in equal proportion (that is, one voice session for each data session), then the bandwidth requirement becomes 160 kbits/s down stream and 110 kbits/s upstream-a ratio of 1.5:1. This results in a CMTS port ratio requirement of 1:8. When employing G.729 compression, data-plus-voice traffic demands 104 kbits/s downstream and 54 kbits/s upstream bandwidth, giving a bandwidth demand ratio of 2:1. In this case, the CMTS channel requirement improves to 1:6.

• Data-plus-video traffic: In a streaming video application, video streams are sent downstream from CMTS to the client cable modem. The client modem sends negligible upstream traffic to control the stream-play, stop, pause. Assuming equal number of data and streaming video sessions, the combined bandwidth requirement becomes 380 kbits/s downstream and 30 kbits/s upstream-a bandwidth consumption ratio of about 13:1. With 12:1 available capacity, it is clearly seen that data and video traffic demand a 1:1 ratio for downstream:upstream port density on the CMTS. More downstream channels are required with increase in video traffic and/or video stream rate.

• Data-voice-vi deo traffic: With combined data, voice (G.729) and video traffic (one session per service), bandwidth requirements become 404 kbits/s downstream and 54 kbits/s upstream, resulting in a bandwidth consumption ratio of 7.5:1. This gives 1:1.6 (or about 2:3) ratio for CMTS channels.

This first-order analysis gives a basic understanding of CMTS requirements under different traffic conditions. However, a detailed analysis is necessary to obtain optimal CMTS requirements.

What is a framework that provides multiple services over a cable Internet Protocol infrastructure? The cable head-end brings national and regional content, including satellite and off-the-air video, the Internet backbone, voice and dial access to the cable network. The head-end reformats this content to be sent to hubs over the high-speed optical fiber network. Head-ends are often connected over a national fiber network owned by the cable operator. Each hub serves approximately 10,000 to 50,000 homes. The hub combines regiona l programming with local content, sending the combined content to the HFC network-only Internet Protocol digital services will be considered. Analog and other out-of-band digital services such as digital television services using the MPEG transport mechanism are multiplexed with Internet Protocol services using RF combiners.

Data service is targeted to two different client access devices: a personal computer or a TV. In the case of a client PC, it is attached to the Ethernet port of the set-top box or the cable modem. An HTTP session is established by the PC browser and the origin web server for web browsing. Web caches are employed at the head-end and possibly at distribution hubs to reduce session latency and backbone bandwidth. The Web cache control protocol (WCCP) may be employed to intercept HTTP traffic at the cable router and sent to a WCCP-compliant Web cache. In the case of a cache hit, the cache serves the page to the client. In the case of a cache miss, the Web cache transparently communic ates with upper-layer caches or the origin server to obtain requested data.

Data service can also be provided via television. Browser and other application software such as e-mail are ported to the set-top box. The box has limited processing power to process HTML pages and render images on the television. For this reason, head-end set-top box servers act as proxy for the client set-top box for Web access. An HTTP request is sent by the client box to the set-top box server. This server communicates with the origin server or a Web cache to obtain the HTML page information. It then transforms this page into a set of objects easily interpreted by the lightweight client set-top box software.

To generate a media stream, analog content-captured using a camera, for instance-is digitized, compressed, packetized and then transmitted by a media server to a media client over a packet network. This packetized transmission is referred to as a media stream because the connection between the media server and the client is a persistent one with a continuous, regularized flow of packets between the two. Unlike HTTP sessions which use TCP/IP for connection-oriented, reliable transmission of packets, streaming media uses UDP/IP mechanism for connectionless, unreliable transmission to minimize transmission latency-low latency is desired over reliable transmission. To ensure proper reassembly of packets as well as to identify lost packets, sequence numbering, time stamping and stream synchronization information defined in the RTP standard is added to UDP/IP packets. To initiate a media session and to control the stream (start, play, pause, etc.) clients use the RTP standard protocol.

The regional head-end and local hubs form a two-level content distribution hierarchy that is well-suited for streaming media delivery. High-demand content should be located as close to the subscriber as possible to reduce the bandwidth demands of streaming media in the backbone network. The two-level cable network hierarchy provi des an efficient content-distribution hierarchy where high-demand content is served by each local hub, while less popular content is available at the head-end on an on-demand basis.

Streaming cluster

Each local hub contains a cluster of streaming media servers and cache storage devices, and can intelligently balance the load across these servers for optimum performance and reliability. Local caches store popular content and also have sufficient room to cache on-demand content when necessary. The head-end contains streaming servers and caches and typically serves the hubs for on-demand content. Live or scheduled programs are also streamed by the head-end servers using multicast transmission.

When a subscriber requests streaming content, such as a newly released music video, the request is intercepted by the cable router and sent to the edge media server. The user is authenticated and the stream request is authorized based on communications between edge media server, cable route r and billing/provisioning systems. The edge media server then streams the requested content to the user. As described earlier, if the requested content is not available at the hub, the edge media server communicates with the core cache at the head-end to retrieve this content. If it is not available at the head-end, then either the session is dropped or the content is retrieved from the origin server (this is dependent on the policies established by the cable operator and/or the type of service the user has subscribed to).

In the case of a multicast stream, the user sends an Internet group management protocol (IGMP) "join" command to a multicast-enabled router. The router adds the client to the IGMP group and routes the incoming multicast stream to the outbound interface to which the client is connected. To ensure a policy-based multicast session admission, multicast streams must be encrypted on the DOCSIS access network. A client wanting to view the stream must obtain a key to view the multicast pr ogram. A key exchange mechanism verifies the session admissibility prior to releasing the key to the client.

To provide voice services, the set-top box or the cable modem may provide telephony connections (RJ-11 interface) to which traditional analog phones are connected. A voice conversation requires a two-way connection between two end users. The calling party initiates the session by dialing the number. The called party may be on the same network or off it. The voice calling agent is the intelligent device located at the head-end that determines the on-net/off-net characteristic of the session. It also recognizes a fax transmission versus a normal phone conversation. In the case of an off-net call, the voice packets are directed to the PSTN gateway. The specific interactions between client device, cable router, voice call agent and voice gateway are specified in the PacketCable architecture.