1. Introduction. 2. The Payload

Transcription

1 Article Prepared for FERSEH UD KIO TECHIK (FKT) 15th April 1997 David Grieve (English Version) Unless you have lived in a cave for the last few years, you will know that the arrival of digital television is presumed imminent. The media are filled with news of big deals, optimistic interviewees, and market forecasts to excite even the most reluctant investor. But underlying all the hype there are natural concerns about the technical challenges ahead. These are perhaps most strongly felt by the technicians and engineers who have to specify, procure, install and maintain the new digital television infrastructure. The DVB measurement guidelines document ETR 290 is offered as an attempt to help the people faced with this difficult challenge. Mr. David Grieve is a Principal Engineer with Hewlett-Packard Ltd in Edinburgh. He has represented Hewlett-Packard in the DVB Project since 1994 and participated in the DVB Ad-Hoc Group on Measurements since its creation in January Introduction The European Project for Digital Video Broadcasting (DVB) has set clearly defined public standards for almost every aspect of digital television distribution and delivery [1]. The expected benefits of this work are well known and mostly rather obvious. One benefit of particular interest to system implementers is that it should be easier to specify and select equipment. However, once the equipment arrives there is still the difficult question of how to qualify both the existing network and the new equipment, as well as how to commission and maintain the system. As part of the larger DVB project, the DVB Measurements Group (DVB-MG), chaired by Dr J. Lauterjung, has attempted to assist in these areas by producing measurement guidelines for DVB Systems [2]. This article is an introduction to the measurements defined in the guidelines for the DVB digital cable television system (DVB- C) [3]. We start with a brief overview of the DVB-C system, then introduce some of the ways that a DVB-C signal can be displayed, before progressing to the central topic of measurements. For many readers, some of this information will be well known. We can only ask for their patience while everybody else catches up. 2. The Payload Each digital CATV channel typically carries a data transmission of around 38 Mbits/sec. The data is a multiplex of several compressed digital video and audio streams, together with system data (multiplex content and structure tables) and broadcast support data services such as subtitles or Teletext. The ISO/IEC MPEG-2 standards [4][5][6] define the audio and video compression methods and the structure of the multiplex. The multiplexed data stream is formally called the Transport Stream (TS). For the purposes of this article we do not need to understand the TS structure. It can be considered simply as a byte stream organised into 188 byte packets with the first byte of each packet being the fixed synchronisation byte (47 16). By definition, compressed video and audio data have had their natural redundancy removed, and the system data will have little redundancy anyway. The MPEG-2 system multiplex definition [4] adds very little error protection and so it would not require many transmission errors to corrupt a TS beyond repair. This is acceptable because, in DVB systems, there is a clear separation between the message (TS) and the transmission medium (cable, satellite, etc.). The responsibility for conveying TS data from broadcaster to viewer with as few errors as possible is placed entirely on the data transmission system. The term that is often used in connection with DVB systems is quasi error-free transmission, which is usually defined as less than one error event per transmission hour. 2. Channel Coding The first step towards achieving quasierror free transmission is to encode the TS data so that it is in a form suitable for modulation and has sufficient structured redundancy added to allow transmission errors to be corrected.

2 In common with other DVB transmission systems, DVB-C applies three coding processes; randomisation, Reed-Solomon error protection and convolutional interleaving. Randomisation is performed by adding a known pseudo-random binary sequence (PRBS) to the data in the transmitter, and subtracting it in the receiver to recover the original data. Randomisation is applied to break up any repetitive patterns in the TS data which might otherwise cause discrete tones in the modulated signal. It is not a form of encryption. Reed-Solomon error protection uses an RS(188,204) code. This means that each 188 byte TS packet has 16 check bytes added to make a coded packet size of 204 bytes. The transmitted data rate is increased to 204/188 times the TS data rate. For a typical TS data rate of Mbits/sec this gives a gross bit rate of 41.4 Mbits/sec. In the receiver, the Reed-Solomon decoder uses the redundancy provided by the 16 check bytes to reconstruct the original data from the errored data. The chosen RS code is capable of correcting up to 8 errored bytes per packet and detecting errors beyond that. Convolutional interleaving rearranges the time order of the data bytes over a 12 packet period to improve the performance of the Reed- Solomon code in the presence of burst errors. The receiver de-interleaving process disperses bytes that were time adjacent during transmission. This means that in the event of a burst error the errored bytes are spread over several packets which increases the probability of complete error correction. 3. Modulation DVB-C uses Quadrature Amplitude Modulation (QAM) of a single RF carrier to convey the TS data. The QAM signal bandwidth is determined by the transmitted data rate but typically it fully occupies an 8 MHz CATV channel. I Q cosine oscillator 90 phase shift ( ) RF = 2 I + Q 1 Q cosωt tan I 2 Figure 1. Basic Quadrature Modulator Figure 1. shows the general method of producing a quadrature modulated signal. A cosine wave (in-phase component) is amplitude modulated by part of the information and simultaneously a sine wave (quadrature component) is amplitude modulated by the rest of the information. The two components are then added to produce the final QAM signal. The instantaneous magnitudes of the In-phase (I) and Quadrature (Q) components determine the amplitude and phase of the resulting sinusoid by simple trigonometry. QAM is particularly suitable when the modulating information is easily split into independent parts. A familiar example from analogue TV is the two colour difference signals used to QAM modulate the PAL colour sub-carrier. In digital QAM the channel coded TS data is divided into groups of m bits which are then mapped to 2 m discrete (I,Q) amplitude pairs. DVB-C defines groups of 4, 5, or 6 bits leading to 16, 32 and 64 state QAM respectively. The mappings for 128 and 256 QAM are the subject of a forthcoming annex to ETS [3] Q Figure 2. Bit Mapping of 16 QAM As an example, figure 2 shows a mapping for 16 QAM. By mapping the I component on the X axis and the Q component on the Y axis (reflecting the quadrature relationship between the two components of the modulation) we create a representation of the amplitude/phase plane of the QAM signal. The dots mark the 2 m discrete I amplitude/phase states of the RF carrier. Each dot equates to a unique bit pattern. This is actually a slight simplification because in the DVB-C system the most significant 2 bits, the ones that define the quadrant, are actually differentially encoded, but the basic idea is still correct. Since each m-bit group causes one modulation state transition it is also referred to as a symbol. The symbol rate is equal to the gross bit rate divided by the symbol size. For example, a 64 QAM signal with a gross data rate of 41.4 Mbits/s has a symbol rate of 6.9 MBaud. The QAM signal does not change instantly from one amplitude/phase state to the next as each new symbol becomes ready for transmission. The I and Q components are low pass filtered in the transmitter so that the RF vector changes smoothly from one state to the next. The filters have a root-raised cosine frequency response, with a half-power bandwidth equal to half the symbol rate. They are matched by nominally identical filters in the receiver. A detailed explanation of the reasons for selecting this filter shape is beyond the scope of a short article but the basic purpose is to band limit the signal without changing its amplitude / phase value at the time centre of the symbol period. This is when a properly locked receiver will sample the signal. The defining parameter for this type of filter is the alpha. For DVB-C, the filter alpha is 0.15 (15%). This means that the filter roll-off starts 15% below the half-power frequency, and reaches maximum attenuation 15% above. 4. Signal Displays 4.1 RF Display A QAM modulated signal is a sinusoid varying rapidly in both amplitude and phase. On an oscilloscope you see a signal that looks like band-limited noise. This noise-like property also translates to the frequency domain. Figure 3 shows the RF spectrum of a 6.9 MBaud DVB QAM signal. ote the root raisedcosine shaping of the band edges and also that the signal half-power bandwidth equals the symbol rate.

3 Figure 3. RF Spectrum of QAM Signal The total bandwidth occupied by the signal is (1 + alpha) times the symbol rate. For 6.9 MBaud DVB-C QAM this gives an occupied bandwidth of MHz. 4.2 Baseband Modulation Display Earlier we referred briefly to PAL colour sub-carrier. The associated analogue vectorscope is also useful in a slightly modified form for displaying digital QAM. When a vectorscope is displaying the PAL colour sub-carrier of a live picture the display resembles a loose ball of string. You can almost imagine a plotter pen attached to the end of the sub-carrier vector, tracing out a record of its instantaneous amplitude and phase against time. This basic vector display is appropriate when the modulating signal is a continuous time function, but for digital QAM it contains too much unimportant information. For digital QAM we are only interested in the amplitude and phase of the signal at the time centre of the symbol period, which is when the signal should be passing through one of the 2 m state positions. This point in time is called the decision instant. The diagrams shown in figures 4 and 5 are a modified form of vector display where the signal is sampled at the symbol rate, and at the decision instant. A dot appears on the display at the (I,Q) co-ordinates corresponding to the sampled amplitude/phase value. This type of diagram is called a constellation display. Figure 4. Relatively noise free 64 QAM If the signal is relatively noise free then the sampled (I,Q) pairs will cluster tightly around the ideal positions (figure 4). Figure 5. oisy 64 QAM As noise and other transmission impairments start to affect the instantaneous amplitude and phase of the signal, the sample points become scattered (figure 5). It is evident from figures 4 and 5 that the constellation display can serve as a quick look subjective confidence check for digital transmissions, just as the waveform monitor and vectorscope have done for analogue TV. 5. Measurements For objective measurements we turn to the DVB measurement guidelines. The DVB-C related measurements in ETR 290 can be divided into four groups. i) Transmission parameters ii) System Margin iii) Monitoring iv) Troubleshooting 5.1 Transmission parameters These are the basic measurements necessary to establish and maintain a digital CATV service BER As already stated, the objective of a DVB transmission system, indeed, the objective of any digital transmission system, is to convey data from A to B with a low error rate. Preferably zero! It follows that the measurement of system performance is the transmission error rate over a range of expected operating conditions. Since the transmission is binary data, the usual currency of transmission errors is bit errors and the measurement is Bit Error Rate (BER). The Bit Error Rate (or Ratio), is the ratio of the number of bits received in error to the total number of bits received. However, in an RF transmission system, the BER cannot be considered in isolation. It only really has meaning when related to the transmitted power and prevailing noise level. We will return to this point after discussing the measurement of power and noise RF/IF Power A major concern when commissioning digital services, particularly when introducing digital channels into an established analogue CATV network, is to set the RF power of each digital signal correctly. The level must be chosen to minimise interference between analogue and digital services, while achieving an acceptable digital transmission performance (BER). A back-off of around 6 to 10 db in the digital power relative to the analogue power is becoming common practice. For historical reasons the RF signal is often referred to as carrier even though QAM is actually a suppressed carrier modulation scheme. This informality is acceptable as long as we are clear about what the digital carrier power is. The RF/IF power is the signal power in the occupied bandwidth which equals (1 + alpha) times the symbol rate (see figure 3). Another concept that should be introduced briefly is energy per bit (E b), the amount of energy expended in transmitting each bit. It is related to the RF power (C) by the equation; E b C = m f Where m is the number of bits per symbol and f S is the symbol rate oise power In a digital system we are primarily concerned by noise that falls within the receiver bandwidth because this noise is demodulated onto the I and Q components of the received samples and, if sufficiently large, will cause bit errors in the receiver. This statement suggests that the noise measurement bandwidth should equal the receiver bandwidth which, in turn, equals the symbol rate. However, for reasons given in annex G.3 of the measurement guidelines, the DVB-MG have chosen the occupied bandwidth of the signal as the most appropriate noise measurement bandwidth. S

4 The noise power is the non-signal power in the occupied bandwidth. For later use, we will also introduce here the concept of noise power density ( 0), this is the noise power present in a 1 Hz bandwidth. It is related to the noise power () by the equation; 0 = BW OCCUPIED Where BW OCCUPIED is the bandwidth over which the noise was measured BER versus E b/ 0 With the above definitions in hand we can measure the RF power (C), the noise power () and compute the C/ ratio. There is then the need, mentioned earlier, to relate the measured C/ to system performance (BER). oise on the demodulated I component scatters the constellation display dots along the X axis, while noise on the Q component scatters the dots along the Y axis. Random noise power at RF will be split on average 50:50 between the I and Q components so we get a cloud of scattered dots around each ideal position. For a fixed signal power, the amount of scatter increases with noise power. As the clouds get bigger, the probability that the receiver decision circuits will misinterpret a received (I,Q) pair as belonging to an adjacent state increases. This, in turn, causes one or more bit errors. For the idealised conditions of perfect, noise free modulation and demodulation with the only impairment being White Gaussian oise added in the channel, we can predict the relationship between C/ and BER mathematically. The details can be found in ETR 290 annex G. or in any good communications theory textbook [2][7]. In practice it is common to plot E b/ 0, rather than C/, versus BER because it removes the variability introduced by data rate and noise bandwidth, and allows direct comparison of different systems. The dashed line in figure 6 shows the theoretical relationship between E b/ 0 and BER for 64 QAM. BER 1.00E E E E E E E E-08 ED 1.00E Eb/o oise Margin Figure 6. Theoretical and Actual BER versus E b/ 0 Curves for 64 QAM Unfortunately, the Real World is not so kind. The demodulated noise that we see in a real constellation diagram comes from at least three distinct sources. The first is the intrinsic noise floor of the head-end QAM modulator (typically around 38 to 40 db down), the second is noise added to the signal in the transmission channel and the third is noise added in the receiver. The constellation will probably also be perturbed by other non-noise impairments. However, if we make the reasonable assumption that the dominant impairment is additive channel noise then we can still usefully plot the BER versus E b/ 0 curve, but we will get a graph similar to the solid line in figure 6. The curve is shifted to the right by system imperfections, and also flattens out at high E b/ 0 values as we reach the system noise floor. We will look at this graph again when we consider the system margin measurements BER before Reed-Solomon The BER computed for the E b/ 0 curve is the BER resulting from the demodulation process, it does not take into account the error correcting capacity of the Reed-Solomon decoder. This BER can be measured in two ways. If the system is out of service, then conventional known-pattern based measurements can be employed, the DVB-MG recommend a variant of the MPEG-2 null packet for this purpose which is defined in ETR 290 annex A.2. The second method is more generally useful. With some range limitations, the Reed-Solomon decoder can be used to make in-service BER measurements on live traffic. As long as the error correcting capacity of the Reed-Solomon code is not exceeded, the receiver decoder knows both the original and errored bit stream and can compute a bit error rate by simple comparison of the two bit streams. Many commercial set-top boxes provide this measurement as part of their test modes. One point worth mentioning is that the accuracy of the Reed-Solomon statistics based approach does degrade progressively as the noise approaches the error correcting capacity of the code because of the statistical nature of noise. The measurement typically reports slightly fewer errors than actually occurred, however, the benefit of in-service BER monitoring outweighs this minor inaccuracy Bit Error Count after RS With reference back to our original goal of near error free transmission, there is clearly also a need to monitor the endto-end error rate performance of the transmission link. Because we are more interested in the occurrence of any errors at all than we are in an exact error rate, the DVB-MG recommend Bit Error Count as the overall link performance measurement. From a practical perspective, this is also because it would take an unacceptably long time to accumulate a statistically valid result at typical post Reed-Solomon error rates. For example to measure a BER of 10-9 with a variance of 10% at Mbits/sec takes over 40 minutes Modulation Error Ratio We have already placed a question mark over the validity of only considering the noise added in the channel when evaluating the system performance. In practice the parameter that really determines the performance of a digital QAM system is the perturbation of the demodulated constellation at the input to the receiver decision circuits, and this is affected in many ways. oise is a major contributor, but the constellation is also affected by DC offsets and I/Q gain mismatch. Intersymbol interference due to imperfect filter implementations, channel nonflatness and reflections. Interfering tones (CSO, CTB) which the receiver equaliser may partially convert to noise, phase noise due to frequency conversions, AM/AM and AM/PM nonlinearity s due to active elements such

5 as amplifiers. The list is quite long! For this reason the DVB-MG have defined a new measurement, Modulation Error Ratio which is intended to provide a single figure of merit analysis of the received signal. It is computed to include the total signal degradation likely to be present at the input of a commercial receiver s decision circuits and so give an indication of the ability of that receiver to correctly decode the signal. [2] Before describing MER in detail we introduce the concept of the error vector. This is the vector between the ideal position of a constellation dot at the sampling instant and the actual position. Figure 7. shows one error vector in one cloud of a constellation. In practice, there is an error vector associated with every received (I,Q) sample pair. δq δi Figure 7. Error Vector error vector MER is the ratio of the mean signal power to the mean error vector power in db as given by the equation; MER= log ( I j + Qj) j= ( δi j + δqj) j= 1 db This definition of the constellation perturbation was chosen for the following reasons (quoted from [2]); - The sensitivity of the measurement, the typical magnitude of the measured values, and the units of measurement combine to give MER an immediate familiarity for those who have previous experience of C/ and S/ measurements. - MER can be regarded as a form of signal to noise ratio measurement that will give an accurate indication of a receiver s ability to demodulate the signal, because it includes, not just Gaussian noise, but all other uncorrectable impairments of the received constellation as well. - If the only significant impairment present in the signal is Gaussian noise then MER and SR are equivalent. 5.2 System Margin These measurements give an indication of how much margin the system has. This is particularly important in digital television systems where, as everybody is rapidly learning, there is no graceful degradation of picture quality to warn of impending failure. The service is basically either perfect or off Equivalent oise Degradation From a system quality perspective an important feature of the graph in figure 6. is the gap between the ideal and actual performance. Clearly the size of this gap varies with E b/ 0 so for comparative purposes it is usual to focus on the value at a specific BER. For DVB-C the preferred BER is 10-4 and the gap expressed in db is called the Equivalent oise Degradation (ED). A BER of 10-4 is chosen because this is generally accepted as being close to the point where the error rate of the data stream after Reed-Solomon error correction exceeds the quasi-error free criterion oise Margin Another, related measurement is oise Margin. In this case noise is artificially added to the live signal until a reference receiver reports a BER before Reed- Solomon of The number of db by which the C/ (or E b/ 0) ratio was degraded is the oise Margin. It is expressed in db. 5.3 Transmission monitoring These measurements treat the system as a black box service. It is expected that the availability measurements will eventually form the basis of contractual agreements between service providers and broadcasters, while the event logging measurements will serve as an audit trail when service interruptions occur System and Link Availability System Availability and Link Availability are two very similar system level monitoring measurements. The definition of availability is based directly on the criteria given in ITU-T G.826 annex A.1 [8]. Briefly, TS packets which are un-correctable are called Errored Blocks (EB) and are counted against real time. If more than a certain percentage of packets are corrupt within a one second period, then that second is considered a Severely Errored Second (SES). Ten consecutive severely errored seconds constitute the start of a period of Unavailable Time. There is also the concept of a Severely Disturbed Period, which is basically total sync loss. These measurements expressed as counts relative to system running time and are envisaged as long-term quality monitoring measurements. The difference between the two availability measures is in the different criteria for identifying an errored block. System availability is based on a flag bit in the TS which indicates that the packet has been corrupted at some point earlier in the transmission system, whereas link availability is based on the error indicator of the most recent Reed- Solomon decoder and thus reflects the availability of the last link before the measurement point Error Event Logging The third monitoring measurement is Error Event Logging. This, as its name suggests, is a real-time tagged log of at the most recent 1000 error events where an error event could be loss of sync, loss of signal, or reception of errored TS packets. In addition to the time of occurrence, error log entries contain diagnostic information such as the packet identifier (PID) of the corrupt packet. It is likely that test and measurement manufacturers will enhance the level of diagnostic information available, though ETR 290 does not require it. 5.4 Troubleshooting measurements In the event of problems occurring, these measurements are aimed at identifying the source of the problem IQ Signal Analysis We have already mentioned MER which is a form of IQ signal analysis, however it is more of a continuous catch all monitoring measurement than a diagnostic tool. Beyond MER, there is a great deal of diagnostic information that can be obtained from the constellation display. Geometric distortion, smearing and grouping of the dots that make up the constellation display can reveal a great deal about the signal impairments. For example, when impairments like interfering tone or phase jitter are applied individually, very characteristic grouping and smearing distortions of

6 the constellation display result. At this point we could fill several pages with pictures of distorted constellations but it would be of limited value, because in a real system, several impairments will be present simultaneously and their text book effect on the constellation will be masked by noise anyway. For this reason IQ analysis is best performed by statistical or mathematical analysis (for example FFT) of the time record of data that constitutes the constellation display. ETR 290 defines the recommended equations for extracting carrier suppression, amplitude imbalance, quadrature error, phase error, phase jitter, interfering tone and signal-tonoise ratio measurements from the constellation data Impulse, Amplitude and Phase Response of the Channel The final source of diagnostic and troubleshooting information is the receiver s adaptive equaliser. This is a standard and very important part of all DVB-C receivers. The adaptive equaliser is a digital filter enclosed in a feedback loop that automatically adjusts its impulse response to compensate for the amplitude and phase response of the channel. The purpose is to minimise the linear distortion of the received QAM signal which would otherwise cause spreading of the received constellation points due to inter-symbol interference. A by-product of this process is that after adaptation is complete, the equaliser impulse response reflects the impulse response of the channel. The impulse response can be displayed and should have one peak in it. Secondary peaks indicate the presence, relative amplitude and duration of reflections. including futher measurements, recommended test set-ups, and some more tutorial material in the annexes. We hope this article will encourage you to look at ETR 290, and that you will find it a helpful document. 7. References [1] DVB Document A020: A Guideline for the use of DVB Specifications and Standards. DVB February [2] ETR 290: Digital Video Broadcasting (DVB); Measurement guidelines for DVB systems. March 1997 draft. [3] ETS : Digital broadcasting systems for television, sound and data services; Framing structure, channel coding and modulation for cable systems. ETSI [4] ISO/IEC : Information Technology - Generic Coding of Moving Pictures and Associated Audio Information: Part 1: Systems. ISO [5] ISO/IEC : Information Technology - Generic Coding of Moving Pictures and Associated Audio Information: Part 2: Video. ISO [6] ISO/IEC : Information Technology - Generic Coding of Moving Pictures and Associated Audio Information: Part 3: Audio. ISO [7] Proakis, John G. Digital Communication, McGraw Hill [8] ITU-T G.826: Error performance and objectives for international, constant bitrate digital paths at or above the primary rate. ITU Also, a Fast Fourier Transform (FFT) of this data gives an estimate of the channel amplitude and phase response. 6. Conclusion This article has presented a necessarily brief overview of the DVB-C related measurements defined in ETR 290. The measurement guidelines document contains considerably more detail