There are a number of ways to transmit a multiplexed MPEG-2 transport stream. In some cases, where a dedicated synchronous transmission facility is available (such as SONET), the digital payload can simply be transmitted over a physical channel. In other cases, where it is necessary to transfer the payload over a shared network, higher-level network functions, such as routing and re-multiplexing, are supported by the network.
There are also two main categories of transmission networks: baseband and broadband. In baseband transmission, the entire physical media (twisted-pair, coaxial, laser-link, and so on) is dedicated to the transmission of a bit-stream. This makes transmission very robust but limits overall transmission rates to the maximum speed of a single transmitter or receiver. In broadband networks, each channel is modulated onto a carrier frequency in such a way as not to interfere with any of the other channels. Although broadband networks are not as inherently robust as baseband networks, they achieve a higher transmission capacity at lower cost than baseband networks.
This section covers baseband and broadband transmission as follows:
Baseband transmissionATM, SONET, SDH, IP, and DVB ASI provide a baseband transmission facility that can be used to carry MPEG transport streams. This section discusses each of them in turn and compares their suitability.
Broadband transmissionQPSK, QAM, and VSB are alternative broadband modulation schemes that are commonly used to carry MPEG transport streams. This section discusses their application in cable systems.
Although baseband and broadband transmission are discussed separately, it is possible to layer multiple baseband channels onto a broadband system, and these techniques are by no means exclusive. For example, in the Orlando Full Service Network (see Chapter 11, On-Demand Cable System Case Studies), multiple ATM links were modulated and combined into a broadband cable system for delivery to the set-top.
In baseband networks, the MPEG-2 transport packets must be adapted so they can be carried by the network. It was understood by the architects of the Grand Alliance system that it would be necessary to distribute MPEG-2 transport streams over a number of different network types, and considerable effort was spent in choosing an efficient mapping of MPEG-2 packets into ATM.
Asynchronous Transfer Mode
Asynchronous transfer mode (ATM) is a connection-oriented network protocol that can be used to build wide-area network (WAN) switched networks. ATM can provide guaranteed QoS metrics, which are sufficient to support MPEG transport streams. However, ATM adds an additional 12% overhead over native MPEG-2 transmission.
The following are characteristics of ATM:
Multiplexing structureATM is a data-link protocol that can carry video, audio, or data in fixed-length cells that carry a 48-byte payload. Each cell has a 5-byte header that is used to identify the connection. ATM supports bandwidth-on-demandby varying the cell rate, connections of any required bandwidth may be created.
MPEG-into-ATM mappingThe mapping of MPEG-2 packets into ATM packets was studied by the Grand Alliance (GA), and the MPEG-2 packet size was set at 188 bytes to facilitate mapping into ATM adaptation layer type 1 or 2, as shown in Figure 4-5.
Figure 4-5 Grand Alliance Mapping for AAL Type 1 or 2Since that time, an alternative adaptation layer, AAL type 5, has become popular in ATM applications, as illustrated in Figure 4-6. AAL type 5 mapping has been nicknamed straight-8 mapping because it packs two transport packets into eight ATM cells.
Figure 4-6 Straight-8 Mapping for AAL type 5
Error detection and recoveryThe header of each ATM cell is protected by a header error check sequence to protect the connection identifier from becoming corrupted. However, error detection of the payload is the responsibility of the ATM adaptation layer. AAL type 1 and 2 do not support error detection, whereas AAL type 5 computes a 32-bit cyclic redundancy check (CRC) across the payload. This provides a valuable error detection mechanism for data, but it is not much use for streaming (that is, video or audio) applications because there is no time to retransmit a corrupted packet.
TimingATM specifies a timing mechanism for synchronous payloads called Synchronous Residual Time Stamp (SRTS). However, this mechanism is designed for telephone carrier emulation (for example, T1 or E1 emulation) and is not applicable to MPEG-2 transmission. Studies have shown that it is possible to transmit MPEG-2 across ATM networks without the need for system timestamp correction if the ATM cell delay variation is tightly controlled (see Testing Digital Video, by Dragos Ruiu and others). In addition, it is possible to use timing recovery mechanisms to de-jitter" an MPEG-2 transport stream carried over ATM before delivering it back to a constant delay transmission system (such as QAM).
SwitchingThe ATM cell structure was designed so that switches can be implemented entirely in hardware making multigigabit switches possible. ATM switches are extremely cost-effective for this reason. In video-on-demand applications, a switching function is required to deliver a program stream from a server to a particular set-top. Because MPEG-2 switches are not commercially available, ATM switches are often used to build video-on-demand networks (see the section Time Warner Full Service Network in Chapter 11).
LimitationsATM adaptation requires additional hardware to map MPEG-2 transport packets into cells at source and then to reassemble the cells into packets at the destination. This adaptation overhead is justified only if a true wide-area switching service is required. Few distribution applications have been deployed that require a switching function, and for broadcast systems the cost of ATM is not justified [Adams].
Synchronous Optical Networks
Synchronous Optical Network (SONET) is a North American standard specified by Bellcore for digital optical transmission. There is an equivalent European standard called the Synchronous digital hierarchy (SDH), which is specified by the International Telecommunications Union (ITU). SONET is a link-layer protocol that carries synchronous payloads in multiples of 50.112 Mbps (within a 51.84 Mbps STS-1). Similarly, SDH carries synchronous payloads in multiples of 150.336 Mbps (within a 155.52 Mbps STM-1). The SONET STS-3c is identical to the SDH STM-1.
The following are characteristics of SONET/SDH:
MappingSONET and SDH are ideal candidates for the carriage of MPEG-2 transport streams, but no direct mapping exists for MPEG-2 into SONET or SDH payloads. An alternative approach is to map a multi-program transport stream (for example, the entire payload of a DVB ASI link) into a single ATM Virtual Channel. The ATM Virtual Channel is then mapped into a SONET payload. This approach has the advantage that only a single ATM segmentation and reassembly (SAR) process is required.
Error detection and recoveryOptical networks run error-free if properly maintained, and failures are usually catestrophic due to equipment failure of a fiber cut. For this reason, SONET includes mechanisms for error monitoring and protection switching. A block check is used to monitor the error rate on each SONET link (which is usually zero). If the error rate exceeds a set threshold, a protection switch will be made to a spare link (assuming it is available and its error rate is below threshold). SONET protection switches take less than 50 milliseconds to complete but, despite this, are quite noticeable to a customer watching MPEG-2 compressed video.
IP is widely used as a data communications protocol. Recently, there has been considerable interest in using IP to carry telephony, audio, and video services. Although IP was not designed with QoS in mind, there has been considerable effort to provide QoS over IP networks (see Quality of ServiceDelivering QoS on the Internet and in Corporate Networks by Paul Ferguson and Geoff Huston).
The following provides a brief description of IP:
Multiplexing structureIP is a network protocol that can carry arbitrary data in variable-length packets. IP supports bandwidth-on-demandby varying the packet rate, flows of any required bandwidth may be created.
Error detection and recoveryLike ATM, IPs approach to error detection and recovery is datacentric. That is, errors may be detected by computing a CRC-32 across the packet, and if an error is detected the Transmission Control Protocol (TCP) is used to retransmit the packet.
RoutingIP routers were first constructed from general purpose minicomputers. However, router design has evolved to the point that routers can be implemented almost entirely in hardware. By placing the packet-forwarding function in hardware, the forwarding delay can be reduced by orders of magnitude, and high-performance routers can approach ATM switches in delay performance.
LimitationsIPs ability to support connectionless networks is its Achilles heel for any application requiring QoS guarantees. Connectionless networks support dynamic reconfiguration by rerouting around congested or failed links in the network. This makes packet delay variation very difficult to control and, as we have seen, MPEG-2 is extremely sensitive to variations in delay. Moreover, IP networks currently do not provide admission control mechanisms, so the only way to obtain bandwidth guarantees is by overprovisioning.
IPs variable-length packet is a disadvantage in delay-sensitive applications, because larger packets can delay shorter packets by taking considerable time to traverse a link. However, as link speed increases, this effect is less noticeable. By using a technique called packet over SONET (PoS) to map IP packets directly into high-bandwidth SONET payloads, delays due to large packets are greatly reduced. (For example, at OC-12 rates of 622 Mbps, a 4 KB packet occupies the link for only 51 microseconds.)
For streaming applications, retransmission is not useful, so user datagram protocol (UDP) is used because it has no payload error checking and requires less overhead than TCP.
However, IP routing is fundamentally very different from ATM switching in architectural terms. Classical IP routing is completely connectionless, which means that there is no state knowledge in the IP routers about the packet flow. This makes IP routing very flexible and obviates any need for connection management; however, it also means that by nature it is impossible to predict the load on any particular link or router. Thus, there is a statistical probability that a particular route may become congested, and this can interfere with a particular IP flow. The effect to the user is that video may freeze, or voice may become garbled for some period of time until the congestion clears.
DVB Asynchronous Serial Interface
DVB asynchronous serial interface (ASI) was developed for the interconnection of professional MPEG-2 equipment and is a native baseband transmission facility for MPEG-2 transport streams. DVB ASI uses 8b/10b coding at a line rate of 270 Mbps yielding a maximum payload of 216 Mbps. DVB ASI is designed to use two physical media:
Coaxial cableCoaxial cable is less expensive than optical fiber and ideal for interconnecting racks of equipment in the headend. However, coaxial cable attenuation limits the reach to about 5 meters.
Optical fiberOptical fiber is more expensive than coaxial cable but supports considerably greater reach. Multimode fiber, which has a reach of several kilometers, is typically used, but there is no physical reason why single-mode fiber transceivers (with a reach of up to 100 kilometers) could not be used.
A brief description of DVB ASIs characteristics follows:
Transmission formatThe DVB ASI transmission format is shown in Figure 4-7. Each byte is encoded as 10 bits using 8b/10b coding, and each MPEG transport packet is preceded and trailed by at least 2 synchronization bytes. A packet may be interspersed with an arbitrary number of stuff bytes.
Error detection and recoveryDVB ASI has no mechanisms for error detection or recovery because it is designed to be for interconnection of equipment over short distances.
Figure 4-7 DVC ASI Transmission Format
TimingDVB ASI, as its name suggests, is asynchronous. That is, there is no relationship between the line clock and the MPEG system timing. Therefore, each piece of equipment using a DVB ASI input performs timestamping using a high-fidelity local 27 MHz counter. (see the section MPEG-2 Systems Layer, later in this chapter). When MPEG-2 transport packets are output onto the ASI link, they are timestamped again to compensate for any jitter introduced by re-multiplexing.
LimitationsDVB ASI is limited to the interconnection of equipment over short distances because of its lack of physical reach and error protection mechanisms.
Comparison of Baseband Transmission Alternatives
Table 4-3 compares ATM, SONET/SDH, IP, and DVB ASI for the baseband transmission of MPEG-2 transport streams. Although it is unfair, strictly speaking, to compare data link protocols with network protocols, this distinction is lost on the engineer who needs to decide how to move an MPEG-2 transport stream from one location to another.
Table 4-3 Comparison of Baseband Transmission Alternatives for MPEG-2 Transport
Baseband Transmission Interworking
As discussed in the previous sections on baseband transmission, it is quite common to layer one or more protocols on top of another. Figure 4-8 summarizes the protocol layerings that are, or are in the process of being, standardized.
Figure 4-8 Standard Protocol Layering
From left to right, Figure 4-8 illustrates the following mappings:
An MPEG-2 multi-program transport stream is mapped into a single ATM Virtual Channel, providing transport of an entire DVB ASI payload over an ATM/SONET network.
An MPEG-2 single-program transport stream is mapped into a single ATM Virtual Channel for transport over an ATM/SONET network.
An MPEG-2 single-program transport stream is mapped into IP packets. The IP connection is supported by a single ATM Virtual Channel for transport over an ATM/ SONET network.
An MPEG-2 single-program transport stream is mapped into IP packets. The IP connection is supported by a direct packet over an SONET (POS) adaptation layer to provide transport over an SONET network.
In networks, simplicity is usually most cost-effective, and DVB ASI is the simplest and least expensive approach for local interconnection of equipment. The mapping of an MPEG-2 MPTS into a single ATM virtual channel is also gaining favor with cable companies that need to deliver MPEG-2 transport streams over some distance.
Transmission in a broadband system uses modulation to separate each channel into a given frequency band. This technique is often called frequency-division multiplexing (FDM). This section discusses the three common modulation techniques used for MPEG-2 transport in North America: QPSK, QAM, and VSB.
Error correction and protection techniques are typically employed in broadband transmission systems to reduce the number of errors introduced by analog transmission; these are also discussed in this section.
Quaternary Phase Shift Keying
Quaternary phase shift keying (QPSK) modulation is very robust in the presence of noise, so QPSK is used for satellite transmission links and for control channel modulation in cable systems (see the section Out-of-Band Communications in Chapter 5).
Figure 4-9 shows how QPSK modulation is applied to a baseband signal. Two bits are encoded per baud. The 2-bit symbol is divided into one in-phase (I) bit and one quadrature-phase (Q) bit, which are each converted to an analog level. These levels are used to modulate the amplitude and phase of a carrier.
Figure 4-9 QPSK Modulator Block Diagram
Figure 4-10 shows the constellation diagram for QPSK and the symbol mapping for each phase angle and amplitude vector. QPSK is very robust because the detector needs to detect only two levels and two phase angles to determine the symbol. (QPSK is equivalent to 4-QAM.)
Figure 4-10 QPSK Constellation Diagram
Quadrature Amplitude Modulation
Figure 4-11 shows how 64-QAM modulation is applied to a baseband signal. Six bits are encoded per baud, which is three times as efficient as QPSK. The 6-bit symbol is divided into three in-phase (I) bits and three quadrature-phase (Q) bits, which are each converted to an analog level. These levels are used to modulate the amplitude and phase of a carrier.
Using the North American standard for 64-QAM modulation (ITU J.83 Annex B), a payload of approximately 27 Mbps is achieved within a 6 MHz channel. This is an efficiency of 4.5 bits per baudconsiderably less than the theoretical maximum of 6 bits per baud. There are two main reasons for this:
The entire 6 MHz bandwidth cannot be used, and guard-bands need to be introduced on either side of the signal to prevent interference between adjacent channels.
Error correction and protection mechanisms introduce some overhead (see the section Forward Error Correction, later in this chapter).
Figure 4-11 64-QAM Modulator Block Diagram
Figure 4-12 shows the constellation diagram for 64-QAM. There are 64 different phase angle and amplitude vectors and, for this reason, 64-QAM is less robust because the detector needs to differentiate between these to determine the symbol. In practice, 64-QAM requires a carrier-to-noise ratio in excess of 22 dB to work (see Chapter 4 of Modern Cable Television Technology; Video, Voice, and Data Communications by Walter Ciciora and others, for more details).
Figure 4-13 shows how 256-QAM modulation is applied to a baseband signal. Eight bits are encoded per baud, and this is 33% more efficient than 64-QAM. The 8-bit symbol is divided into four in-phase (I) bits and four quadrature-phase (Q) bits, which are each converted to an analog level. These levels are used to modulate the amplitude and phase of a carrier.
Using the North American standard for 256-QAM modulation (ITU J.83 Annex B), a payload of approximately 38.8 Mbps is achieved within a 6 MHz channel. This is an efficiency of 6.47 bits per baudconsiderably less than the theoretical maximum of 8 bits per baud. A payload of 38.8 Mbps was chosen because it is sufficient to carry two HDTV channels.
Figure 4-12 64-QAM Constellation diagram
Figure 4-13 256-QAM Modulator Block Diagram
Figure 4-14 shows the constellation diagram for 256-QAM. There are 256 different phase angle and amplitude vectors, making the points on the constellation closer together; for this reason, 256-QAM is less robust than 64-QAM. In practice, 256-QAM requires a carrier-to-noise ratio in excess of 28 dB to work in practice (6 dB more than 64-QAM).
Figure 4-14 256 QAM Constellation Diagram
There has been some discussion of still higher orders of QAM modulation512-QAM, 768-QAM, or even 1024-QAM. However, there are diminishing returns1024-QAM increases the payload by only 25% over 256-QAMand it is unlikely that these will be used in the near future.
Vestigial Side Band
VSB-8 modulation has been adopted for use in terrestrial digital broadcasting (see A.53 Annex D). VSB-8 has a payload of approximately 19.4 Mbps and was designed to carry a single HDTV channel. VSB-8 is a one-dimensional modulation scheme because it uses only amplitude modulation of the carrier (in contrast, QAM is two-dimensional modulation technique because it uses both I and Q components). In VSB-8 modulation, the baseband signal is coded as an 8-level value, so 3 bits are encoded per baud.
Error Correction and Protection
Analog transmission systems are subject to noise, distortion, and interference from other carriers. Therefore, error correction and protection techniques are used to maintain an acceptable error rate.
Three commonly used techniques are discussed in this section: forward error correction, interleaving, and trellis coding.
Forward Error Correction
Forward error correction (FEC) uses a mathematical function to generate a check sequence across the payload data. The check sequence is transmitted with the data, and the same mathematical function is used at the receiver to check for payload errors and to correct errors. This technique is also used in error correcting code (ECC) memory.
The Reed Solomon (RS) function is used in conjunction with QPSK, QAM, and VSB modulation. RS (204,188) t=8 describes a scheme where 16 check bytes are generated for each MPEG-2 packet. This represents an overhead of 8.5% but allows 1- and 2-byte errors to be corrected by the receiver.
Noise pulses can obliterate a signal for several microseconds, and at a 256-QAM payload rate of 38.8 Mbps, this represents hundreds of bits. By itself, FEC is incapable of correcting such long error sequences.
Interleaving effectively spreads the payload data over time. Figure 4-15 shows an example of a block interleaver developed by Scientific Atlanta for the Time Warner Full Service Network (see Chapter 11).
Figure 4-15 Block Interleaver Used in the Time Warner Full Service Network
In this implementation, a noise pulse affects only 1 byte in each row, a situation that can be rectified by the RS correction. The main disadvantage of interleaving is buffer memory and delay, which are 32 KB and 11.5 milliseconds, respectively, in this implementation.
Convolutional interleavers require only half the memory and introduce only half the delay of block interleavers. For this reason, both ITU J.83 Annex B (QAM) and ATSC A/53 Annex D (VSB) use convolutional interleaving. Figure 4-16 shows an example of a convolutional interleaver with an interleave depth of six. The blocks (labeled J) buffer the payload and operate as a shift register. The interleaver (at the modulator) and the deinterleaver (at the demodulator) are synchronized so that the payload is reassembled in its original form.
The effect of interleaving is to distribute the errors due to a noise burst over a period of time so that the errored bits are no longer adjacent, which makes FEC more effective.
Figure 4-16 Example of a Convolutional Interleaver
As noise is introduced into a QAM or VSB signal, it perturbs the points on the constellation diagram so that ultimately they overlap and the decoder sees the wrong symbol. Trellis coding adds some redundancy and uses sophistical mathematics at the receiver to determine the best fit of the constellation to a symbol. The trellis coding specified by ITU J.83 Annex B QAM uses a 14/15 coding rate to improve noise immunity by approximately 2 dB. VSB-8, which is designed for more challenging broadcast applications, specifies a 2/3 trellis coding rate, improving noise immunity but reducing payload rate by 33%.