RF Measurements. in DVB-T and DVB-H

RF Measurements in DVB-T and DVB-H

2

Table of Contents Introduction....................................................................................................4 Power Measurements............................................................................................ 5 Noise Measurement and Carrier/Noise (C/N) Ratio..................................................................... 6 DVB-T Spectrum Details and Mode Identification...................................................................... 8 Central-Frequency Measurements.................................................................................10 RF Channel Width Measurement...................................................................................12 Guard Internal Measurements.....................................................................................15 Emission Masks................................................................................................16 Analyzer Sensitivity and Dynamic Margin..........................................................................20 Distortion-free input level......................................................................................20 Dynamic Range.............................................................................................21 Linearity Measurements.........................................................................................23 CCDF Measurements............................................................................................28 SFN Synchronization............................................................................................32 Continuous-Signal Interference....................................................................................36 Carrier Suppression.............................................................................................37 Oscillator Phase Noise...........................................................................................38 Conclusions...................................................................................................40 3

Introduction Digital modulation of radio frequency (RF) channels is characterized by the efficient use of available spectrum. These modulations can be classified in several ways, and one of them is to consider modulation as being over a single carrier or over several orthogonal carriers. In both cases, an attempt is made to obtain an extended spectrum that will use all, or most, of the spectrum available in the communication channel in question. In the case of digital TV, both systems are used. Only one carrier is used in DVB-S (Digital Video Broadcast Satellite), DVB-C (Digital Video Broadcast Cable) and land-based digital TV broadcasting using the ATSC (Advanced Television System Committee) standard. With single-carrier modulation, the number of bits transmitted per symbol depends on the type of digital modulation employed. Constellations from four points (two bits per symbol), as in QPSK, to 256 points (eight bits per symbol), as in some QAM modulations over cable, are used, where the occupied spectrum is inversely proportional to the duration of the symbol transmitted. OFDM (Orthogonal Frequency Domain Multiplex) modulation with several orthogonal carriers is used in various land-based broadcast systems, including the medium wave system called DRM (Digital Radio Mondiale), VHF emission for high quality audio such as DAB (Digital Audio Broadcast), as well as UHF emission for land-based digital TV defined by DVB-T (Digital Video Broadcast Terrestrial). In these OFDM systems, the number of bits transmitted per symbol depends on the constellation type employed in each carrier, which can be Bi-Phase, QPSK or QAM, and the number of carriers, which can vary in each system between several (from 103 to 461) for DRM, or many thousand (1,705, 3,409 or 6,817) in the case of DVB-T and DVB-H (Digital Video Broadcast Handheld). In OFDM systems, the occupied spectrum is obtained by multiplying the frequency separation between carriers by the number of active carriers of the system. The inverse of the useful duration of the symbol determines the separation between carriers. An important OFDM modulation feature, and a principal reason for its use, is that symbol duration is much longer than for single carrier modulations, which permits system behavior to be improved when dealing with multiple-path reception, echo, etc. With DVB-T/H and DRM, this is achieved thanks to the inclusion of a guard interval, the length of which can be selected as a function of the topography and coverage desired for each region. The following list contains some of the RF measurements specific to the DVB-T and DVB-H system, which can be obtained with a spectrum analyzer. The section numbers come from the TR 101 209 document, Measurement Guidelines for DVB Systems. Following the list is a commentary on some of the measurements, with practical examples made with the Tektronix Real Time Spectrum Analyzer, model number RSA3408A. 9.1.1 RF frequency accuracy (Precision) 9.1.2 RF channel width (Sampling frequency accuracy) 9.1.3 Symbol length measurements at RF (Guard interval verification) 9.4 Phase noise of local oscillators (LO) 9.5 RF/IF signal power 9.6 Noise power Carrier to Noise (C/N) 9.7 RF and IF spectrum (Emission mask) 9.12 Coherent interferer 9.10 Linearity characterization (Shoulder attenuation) 9.18.4 Carrier suppression 9.20 SFN synchronization 4

Power Measurements In contrast with analog AM modulations, digital AM modulations have an average power value that is independent of the content of the modulating signal. However, the instantaneous power varies randomly and may exceed the average power by several decibels. The statistical distribution of instantaneous power is similar to the white noise for the width of the assigned channel and is Gaussian; so a 15 db peak above the average power is expected with a probability of 1x10-8. The DVB recommendation in the TR 101 290 document indicates the following: 9.5 RF/IF signal power Purpose: Signal power, or wanted power, measurement is required to set and check signal levels at the transmitter and receiver sites. Method: The signal power of a terrestrial DVB signal, or wante d power, is defined as the mean power of the signal as would be measured with a thermal power sensor. In the case of received signals (by an antenna, for example), care should be taken to limit the measurement to the bandwidth at the wanted signal. When using a spectrum analyzer or a calibrated receiver, it should integrate the signal power within the nominal bandwidth of the signal (n x f SPACING ) where n is the number of carriers. Figure 1 represents a DVB-T digital TV signal in an 8 MHz nominal bandwidth channel (UHF channel 58), in which the measurement is taken according to the DVB recommendation: integrating the spectrum into an effective bandwidth of 7.61 MHz. The Real-Time Spectrum Analyzer (RTSA) is being used in conventional spectrum analyzer emulation mode, using a 300 khz bandwidth Gaussian filter. The use of a RTSA rather than a conventional swept-spectrum analyzer allows for a very high speed refresh of the traces, which in this case are continuously averaged, showing the result of the last 100 traces. Figure 1. Channel power measurement integrating the assigned spectrum. Given that in all the current cases of DBV-T and DVB-H, the spacing between carriers is inversely proportional to the number of carriers, the nominal bandwidth for power measurement is independent of the mode employed, whether this is 2k, 4k or 8k. However, the nominal bandwidth does depend on the specified channel width: In 8 MHz channels the nominal bandwidth is n x f SPACING = 7.61 MHz. It is 6.66 MHz for 7 MHz channels, 5.71 MHz for 6 MHz channels and 4.75 MHz for 5 MHz channels. The average power measured at the input of the RTSA for the digital TV channel (Channel 58) is -40.23 dbm. It can also be seen, looking at the marker situated on the video carrier of channel 59, that the nominal power received from the analog channel is -33.67 dbm. It has always been recommended to use the standard filter of 300 khz to measure the peak power on analog TV channels, both terrestrial and CATV. This is because the peak power of the video carrier at the sync pulse top has to be measured with a filter whose bandwidth is greater than the inverse of the sync pulse duration (4.7 µs); therefore, a filter greater than 1 / 4.7 µs = 212 khz. However, for digitally modulated signals this is not the case. 5

Noise Measurement and Carrier/Noise (C/N) Ratio Noise measurement of a DVB-T/H channel is defined as follows: 9.6 Noise Power Purpose: Noise is a significant impairment in a transmission network. Figure 2. Spectrum obtained with 8,192 point FFT. Figure 2 shows two traces. The lightest one was acquired in the same conditions as those in Figure 1, using filters equivalent to those of conventional digital analyzers, and then frozen. Another trace, acquired using 8,192 point FFT processing for the 36 MHz span used, corresponds to a noise equivalent resolution bandwidth of 12.52 khz (NBW) using a 4B Blackman-Harris weighting window. The improvement over conventional equivalent filters can be seen in Figure 2. The equivalent resolution filter is narrower, 12.5 khz instead of 300 khz, allowing for better spectral resolution, not to mention the faster acquisition speed while maintaining the correct average power value measurement (-40.35 dbm, in this occasion). This allows for the steepness of the spectrum occupied by the digital channel to be checked, and whether it invades the analog channel can be determined. Similarly, the non-invasion of the analog channel over the digital channel can be checked, with better resolution than that of the 300 khz filter. The valley can be found at -106 dbm (~70 dbc), while the valley of the frozen trace is found at -77 dbm (~43 dbc). By using narrower filters, such as 12.5 khz, both noise and dense signals equivalent to noise exhibit lower levels, in this example some 14 db below the level shown with the 300 khz resolution filter. This is to be expected, since 10 log (12.5/300) = -13.8 db. Method: The noise power (mean power), or unwanted power, can be measured with a spectrum analyzer (out of service). The noise power is specified by using the occupied bandwidth of the OFDM signal (n _ f SPACING ) where n is the number of carriers. Note: The term C/N should be calculated as the ratio of the signal power, measured as described in subclause 9.5, to the noise power, measured as described in this subclause. This measurement can only be made when there is no signal in the desired channel. It is useful in measuring collective antenna systems, evaluating receptor sensitivity and measuring the transmitter power system output when not emitting the modulated signal. Figure 3 shows the desired power measurement in channel 69 and the measurement of unwanted noise power, received from the broadband amplifier of the collective antenna used. Both are measured with the same channel width parameters used to measure desired power and shown compared to the background noise of the spectrum analyzer itself. The DVB measurements document indicates that the C/N ratio can be calculated as the ratio between each measurement and, as shown on the right side of Figure 3, the result would be C/N = 19.39 db. However, in the practical case shown here, as well as in many other cases, it is advisable that care should be taken with the results obtained under certain conditions, where the power level or measured noise is very close to the background noise of the measuring instrument itself. These precautions should be taken whenever the difference between measured noise and measuring instrument noise is less than approximately 10 db. 6

Figure 3. Noise measuerment and C/N calculation. Figure 4a. Background noise measurement setup. Figure 4b. Calculation of the Correction Factor and example of use. Figure 4a shows, at 1 db / Div., the two noise traces and the method for detecting both traces by measuring at point B (noise floor of the RTSA) and at point A (external noise received and measured). This is the noise observed on the screen, which is the sum of external noise (which is to be measured) and instrument noise. The external noise value can be determined by measuring the difference indicated (D in Figure 4b), and thus finding a correction factor in db (CT in Figure 4b), which must be applied to the obtained measurement of noise or power. This correction should only be applied when the difference D is greater than approximately 2 db, since the correction 7

Figure 5. Complete UHF spectrum from 470 MHz to 862 MHz. Figure 6a. Details with five DVB-T channels and relative power measurement. factor becomes less precise the lower the difference D is. It can be seen in the table that small variations in the measured difference D value imply large differences in the CT factor when D < 2 db. It could be said in this case that the C/N is 19.39 + 7.5 = 26.9 db, or, given the low value of D, that the C/N ratio is better than 24 db in round figures. This value is obtained applying a more conservative correction factor corresponding to D = 2 db (factor of 4.33 db), with the result being 19.39 + 4.33 = 23.7 db. The difference D value would be improved to 10dB or better,thus significantly reducing the uncertainty, if the pre-amplifier option for the RTSA is used for this measurement. DVB-T Spectrum Details and Mode Identification Figure 5 shows the complete UHF spectrum plagued with analog TV signals of various strengths. Six DVB-T digital TV channels can be identified as well as a test DVB-H channel (channel 27). Thanks to the narrow 12.5 khz filter (narrow considering a 400 MHz span), it is possible to identify each and every one of the 49 UHF channels which occupy the 392 MHz assigned (from 470 MHz to 862 MHz). Figure 6a shows some details by enlarging the zone that includes the four national SFNs (Single Frequency Networks) in channels 66 to 69, as well as the regional Madrid, Spain SFN (channel 63). The power measurement of a channel, in this case centered on channel 66, and the relative power of the adjacent channels, three on each side of the selected channel, can be used to measure simultaneously all of these channels. Therefore, channel 66 is received in this case with -60.93 dbm, and the other digital channels are received with powers of -70.28 dbm for channel 63 (-60.93 9.35 = -70.28), -62.8 dbm for channel 67, -63.07 dbm for channel 68 and -64.12 dbm for channel 69. The measurements for channels 64 and 65 must be discarded since this method is not recommended for analog channels. Figure 6b shows only one channel, 58, enlarged to occupy the whole screen and with a filter resolution close to 200 Hz over a span of 8 MHz. This shows that the DVB-T spectrum presents some details which illustrate that the channel is not purely Gaussian and has some static frequencies which appear when resolution filters are narrow enough. 8

Figure 6b. Details of one channel (58) with high resolution. Figure 7. Channel 58 center detail. The type of carriers that can be seen in Figure 6b are called continuous pilots and serve as the reference for signal receivers to latch on to the frequency and phase. They also allow rough estimation of the state of a channel s reception, with regard to frequency and phase. A precise determination can be made using the scattered pilots, which are shown with some detail in Figure 7. The OFDM channel mode, which in this case is 8k, can be determined by examining Figure 6b and observing that the continuous pilot carrier position pattern is repeated four times. In the 4k mode it occurs twice, and in the 2k mode only once. However, it is faster and more accurate to check the spacing between carriers, which in this example is 1,116 Hz. Observe in Figure 7 that with a 10 khz width there are 9 carriers and that 10 / 9 = 1.111Hz. This is an approximate measure but enough for mode detection. In case 4k or 2k modes are used, the carrier spacing will be 2,232 Hz and 4,464 Hz respectively. In Figure 7, samples of each of the different carrier types used in DVB-T modulation can be seen. These have been identified by indicating some of the k carrier indices from the 6817 carriers, which, in the 8k mode used in channel 58, extend from k = 0 to k = 6816. In this part of the spectrum two continual pilots can be distinguished, one with index k = 3387 and the central frequency with index k = 3408. It is useful to recall that in the 8k and 4k modes the central carrier is always continual, while in the 2k mode it is a data carrier. The question may arise as to why the continual pilot k = 3387 is not seen as a pure spectral line, something that does occur with the central carrier k = 3408. The answer is that the continual pilot term refers to the fact that these carriers exist in all symbols for their corresponding k index, and each one has a fixed and permanent phase at the beginning of each symbol, which is determined by a pseudo-random algorithm. However, this does not imply that they are continuous in phase through the complete symbol; that is, the number of cycles of each carrier is a whole number throughout the duration of the symbol. It is indeed a whole number for all the carriers through the useful part of a symbol, but not necessarily during the duration of each guard interval. 9

The k = 3408 carrier, central frequency of the 8k mode, has an integer number of cycles for all of the guard intervals, since its Fourier q index, different from the carrier index k, is 4096, which is a multiple of the inverse of all of the guard intervals. In other words, it is a multiple of 4, 8, 16 and 32, while the carrier k = 3387 with a Fourier index of 4075, is not divisible by any of those numbers and, therefore, does not have an integer number of cycles for any of the guard intervals allowed by the standard in the 8k mode. The Fourier index is obtained by adding 688 to the carrier indices in the 8k mode, adding 344 in the 4k mode and adding 172 in the 2k mode. Figure 7 also shows one of the transmission parameter signal carriers (TPS), specifically the TPS with k = 3391. The data carriers can also be perfectly seen, as well as their principal lobes created by the digital modulation, which generates a dense averaged spectrum. Every three carriers a few spectral lines can be seen, due to the presence of scattered pilot carriers (SP). In each position or index corresponding to a scattered carrier, for example k = 3411, a four symbol sequence is repeated, where during one symbol the SP reference carrier is present, and in the following three it s the data carrier. This is why a mix of data lobes and SP carrier spectral lines are present. The measurement of phase and amplitude, during the useful part of each symbol, of each one of the scattered carriers allows the receiver to compensate the channel response, whereby echo, multi-path reception, DVB-H Doppler effects and so forth can be canceled or reduced. Figure 7 shows the spectrogram display, which clearly depicts the various carrier types and their constancy in the time domain. The vertical axis of the spectrogram indicates time, the horizontal axis represents the frequency within the spectrum and signal amplitude is represented by different colors. Figure 8. Channel central frequency measurement (Channel 58 in 8 k mode). Central-frequency Measurements In the 4k and 8k work modes, a continual pilot carrier is present in the center of the channel, as well as the usual outer carriers in all work modes. In addition, the central frequency has, in all modes, a Fourier index which is a multiple of 32 (the Fourier index q is equal to 1024, 2048 and 4096 for the 2k, 4k and 8k modes respectively). Therefore, besides being continual they are also continuous, meaning they have an exact number of cycles with any of the possible guard intervals. The DVB recommendation for frequency accuracy measurement indicates the following: 9.1.1 RF frequency accuracy (Precision) Purpose: Successful processing of OFDM signals requires that certain carrier frequency accuracy be maintained at the transmitter. Specific network operation modes such as SFN require high accuracy of the carrier frequency (central frequency). Method: The 8k (and 4k) modes of the DVB-T (and DVB-H) always have a continual pilot, with continuous phase along successive OFDM symbols, exactly at the channel center (k = 3408 and k = 1704, respectively). 10

Figure 9. Unreferenced central frequency measurement (left) and with internal refference (right). Its frequency may be directly measured by any spectrum analyzer that has an integrated counter and at least a resolution filter of 300 Hz or less (if necessary by utilizing an external reference source of sufficient accuracy). Note: Should more accuracy be needed, the two outer continual pilots may be measured as indicated under 9.1.2 RF channel width, and the mean of the two values be calculated. In Figure 8, the measurement was performed over the central carrier of channel 58 and a value of 770 MHz was obtained, using a span of 10 khz and a 12.5 Hz resolution filter. Therefore, the measurement resolution is restricted to ±6 Hz. However, since the analyzer is referenced with a high-accuracy external frequency supplied by a GPS receiver, the measurement is obtained with the full precision expected. Figure 9 supplies clues about the accuracy that can be expected if a high-performance RTSA is used, but without an external reference available to measure central channel frequencies in SFN networks, which are usually referenced with GPS in all transmitters. The left side of the figure shows the result obtained when the analyzer loses the external reference. It can be seen that the measured value of the central frequency is 770,002,350 Hz, representing an error of 2.35 khz with respect to the actual frequency of the channel, which in relative terms is 3.05 x 10-6, an acceptable value for a crystal oscillator with no external reference. The right side of the figure shows the result of measuring with the same analyzer, but using its internal reference with temperature stabilization, in which case the frequency value obtained is 769,999,991.125 Hz, representing an error of 8.875 Hz, far lower than that obtained previously but still not precisely the expected value. The relative error is 1.15 x 10-8, which is the value expected for a temperature stabilized oscillator, as well as being within the accuracy specified for this RTSA model. Figure 9. Unreferenced central frequency measurement (left) and with internal reference (right). 11

know that the guard interval is higher than 1/32, is to use the k = 804 carrier, which is closer to the center of the channel k = 852. However, there is another procedure suggested in TR 101 290 that consists of measuring the frequency of the two border pilots of the channel and calculating the arithmetic mean between both values, which results in the central frequency of the channel. This procedure can also be used to calculate the precision of the IFFT processor sampling frequency of the transmitter modulator. RF Channel Width Measurement Figure 10. Central frequency measurement with external GPS reference. Figure 10 shows the result of the measurement once the RTSA is referenced with an external source more stable than its internal one. In this case, a 10 MHz GPS-stabilized generator is used, the GPS3000 from Albalá Ingenieros, which is a model specifically designed to reference DVB, DAB transmitters and TV studio master generators. In this example, the measured frequency error is only 0.125 Hz, showing that the transmitter is also latched on to its GPS reference, which is what was being checked. These measurement results show that to measure the behavior of a transmitter, or any other device, the measurement should be made with an instrument that is at least as precise as the device being measured. In this case, the precision is 1.62 x 10-10. When the 2k mode is used, where the central carrier carries data and therefore does not show up as a spectral line, the measurement must be taken indirectly. One way to do so is using a continual pilot carrier with an integer number of cycles. This could be, for instance, the carrier with index k = 1140, which is the only one in this mode that is phase-continuous, meaning it has an integer number of cycles for all guard intervals. Another way, if we The width of a channel can be measured directly over the extreme continuous carriers when the selected modes permit this direct measurement. This is possible in the 8k mode with guard intervals of 1/4, 1/8 or 1/16, or in the 4k mode if the guard interval is 1/4 or 1/8 and in the 2k mode with a guard interval of 1/4. 9.1.2 RF channel width (Sampling Frequency Accuracy) Purpose: Channel width measurements are convenient for verifying that sampling frequency accuracy is maintained at the modulator side. Method: The occupied bandwidth of a COFDM modulated channel depends directly on the frequency spacing and thus from the sampling frequency. The outermost carriers in a DVB-T/H signal are continual pilot carriers. Their frequencies are measured (see Appendix E.1) and the difference between them should be compared to the nominal channel width. Channel Width Nominal Bandwidth 8 MHz 7,607,142.86 Hz 7 MHz 6,656,250.00 Hz 6 MHz 5,705,357.14 Hz 5 MHz 4,754,464.28 Hz Table 1. 12

Figure 11. Upper frequency measurement (Channel 58 in 8k mode and 1/4 GI). Figure 12. Lower frequency measurement (Channel 58 in 8k mode and 1/4 GI). Figures 11 and 12 show the direct measurement of the extreme carriers of the channel, which are continuous in frequency, amplitude and phase, and also have an integer number of cycles in the complete symbol, including the guard interval. On the left side of Figure 11, the measurement was performed with a span of 10 khz and a resolution filter of 12.5 Hz, so it is perfectly clear that this is the last continuous carrier from the upper end of the measured channel. On the right side, the span was changed to 50 Hz (with a filter resolution of 125 mhz) to achieve a higher-precision measurement, and a frequency value of 773,803,571.725 Hz was obtained. On the left side of Figure 12, the measurement was repeated for the last continuous carrier of the lower end of the spectrum. On the right side of the figure, with a resolution of 125 mhz, a frequency value of 766,196,428.35 Hz was obtained. 13

12.5 Hz Filter Resolution 0.125 Hz Filter Resolution Upper frequency 773,803,577.9 773,803,571.725 Lower frequency 766,196,422.4 766,196,428.350 Channel width (frequency difference) 7,607,155.5 7,607,143.375 Nominal value for 8 MHz 7,607,142.857 Sampling frequency error 12.643 0.518 Sampling frequency precision 1.662 x 10-6 6.81 x 10-8 Table 2. The difference between these values can be calculated in order to obtain the width of the channel and the sampling frequency precision: Note that the precision obtained for the sampling frequency has an order of magnitude of 7 x 10-8 and is not in the order of 10-10 or 10-11 as could be expected if it were latched to GPS. A reason for this is that the sampling frequency is linked with the data rate. The data rate depends on the precision of the multiplexer used, or in the case of single frequency networks, on the precision of the SFN adapter, which in this type of network must be placed on the modulating signal generation side; that is, on the multiplexer side. The multiplexer signal passes through the SFN adapter to generate the MIP packets for frame synchronization timestamps in all transmitters, and it can also adjust the frame speed to its nominal value with the desired precision. In this case, the resulting precision apparently was 6.81 x 10-8. Notice that the RTSA allows for filters which are <1Hz wide. In this example, the resolution of the filter used within the analyzer was 125 mhz, which improved the resulting precision to 0.125 / 7,607,142.857 = 1.64 x 10-8. These carriers can also be measured, even if this must be performed indirectly, in the cases where the external continual pilots are not continuous; that is, where the transmitter is using a guard interval of 1/32 in the 8k mode, a guard interval of 1/16 or 1/32 in the 4k mode or one of 1/8, 1/16 or 1/32 in the 2k mode. To carry out these measurements, it is useful to analyze which of the various cases described in Appendix E of the TR 101 290 document, paragraphs E.1.1 to E.1.3, applies. Due to their extension and detail, these cases are not included in this paper. Other parameters may be derived from the measurement of the frequencies of the outer carriers. Defining these as F h and F l (high frequency and low frequency respectively), it is possible to determine the following values (calculated for the 8k mode) with the required precision, using the formulas shown below. In this case, values obtained with a 0.125 Hz filter resolution were used. Occupied Bandwidth, OBW = F h F l, and using the example from Figures 11 and 12, we obtain 773,803,571.725 766,196,428.350 = 7,607,143.375 Hz, which represents an error of 0.515 Hz or 6.76995 x 10-8 with respect to the nominal value of 7,607,142.86 Hz. Frequency Spacing, FSPACING = (F h F l ) / (k 1) = 7,607,143.375 / 6,816 = 1,116.0715 Hz, which represents an error of 0.0001 Hz or 9.3675 x 10-8 with respect to the nominal value of 1,116.0714 Hz. Useful symbol duration, TU = (k 1) / (Fh Fl) = 6,816 / 7,607,143.375 = 895.999939 µs, which represents an error of -61 picoseconds or 6.808 x 10-8 with respect to the nominal value of 896 µs. 14

Center channel first Intermediate Frequency, IF 1 = (q/2) x (F h F l ) / (k -1) = 4,096 x 7,607,143.375 / 6,816 = 4,571,428.8826 Hz, representing an error of 0.31123 Hz or 6.808 x 10-8 with respect to the nominal value of 4,571,428.5714 Hz. Sampling Frequency of the first Intermediate Frequency, IF S = 4 x IF 1 = 18,285,715.5304 Hz, which logically represents the same 1.2447 Hz error or 6.808 x 10-8 with respect to the nominal value of 18,285,714.2857 Hz. Channel Central Frequency, CF = (F h + F l ) / 2 = (773,803,571.725 + 766,196,428.350) / 2 = 770,000,000.0375 Hz, which represents an error of 37.5 mhz or 4.87 x 10-11 with respect to the nominal value of 770,000,000.00 Hz. Note that this value is more accurate than that obtained by measuring the channel central frequency directly, which resulted in a precision of 1.62 x 10-10 (Figure 10). Precision is also higher than if the 12.5 Hz resolution filter is used: (773,803,577.9 + 766,196,422.4) / 2 = 154,000,000.3 / 2 = 770,000,000.15, which represents an error of 150 mhz, or 1.948 x 10-10 with respect to the nominal value. The following conclusions can be drawn from the results of these frequency measurements: 1. The measurement of the channel central frequency is more accurate if the average of the outer carriers is used, but it is quicker to measure the central frequency directly. 2. The analyzer with internal reference yielded values according to the manufacturer s specifications. In this case, the Tektronix RSA3408A analyzer has a specification of 2 x 10-8 and the results obtained are in accordance with this value. However, the use of an external reference can improve precision to the degree given by the external source, as was seen when using the external reference supplied by Albalá Ingenieros s GPS3000 device. 3. The occupied bandwidth and the directly associated values, such as frequency spacing, useful symbol duration and values of the first IF and its sampling frequency can be obtained from the precise measurement of the outer pilot carriers of a DVB-T or DVB-H channel spectrum. This is in accordance with the recommended procedure from the DVB Measurement Group, as indicated in the TR 101 290 document. Guard interval measurements As was seen in Figure 7 and in more detail in Figure 8, scattered carriers are located every three data carriers. It is useful to remember that all the spectra taken with analyzers show the spectral average over a given time interval: either the acquisition interval in RTSAs or the time it takes for the sweep to analyze each zone of the spectrum in conventional analyzers. In the spectrum of Figure 7, for a span of 50 khz and an FFT of 8,192 points, the analyzed time interval is 256 ms, and therefore the spectrum is the average of some 236 symbols. Also, 98 spectra have been averaged in Figure 7, which results in the average value of roughly 25 seg. (256 ms x 98) of the spectrum being observed instead of its instantaneous value (taking into account that instantaneous here means a time interval of 256 ms). Therefore, the frequency components that are regularly repeated appear stable. The TR 101 290 document indicates an appropriate method to determine the total duration of the OFDM symbol. In the previous section, the useful duration of the symbol was deduced as the inverse of the separation of the measured carrier frequencies. What has not been measured yet is the total length of the symbol. The time difference between each result produces the length of the guard interval employed. It will be shown that it is relatively easy to measure the total length of four consecutive OFDM symbols. 15

9.1.3 Symbol length measurements at RF (Guard interval verification) Purpose: Verification of the guard interval used in a received DVB-T/H signal may be carried out at RF level by careful frequency measurements. This measurement is valid in cases where there is an uncertainty on whether a modulator is correctly working and producing a signal with the expected or assigned Guard Interval. Method: The scattered pilots produce a pulsed-like spectrum every third carrier in a DVB-T/H spectrum due to their repetition presence at the same phase and location every fourth symbol. The frequency difference between two contiguous spectral lines representing a scattered pilot represents the inverse of the time length of four consecutive DVB-T/H symbols. Measuring such frequency difference and dividing its inverse by four will provide the total symbol length TS of the measured signal. By subtracting the nominal useful symbol duration TU, the length of the GI is found. See Annex E.1 for details on the measurement procedure and symbol lengths. The method indicated by TR 101 290 can be carried out with great precision with a RTSA. The table shown includes all the separation values between spectral lines for the four guard intervals corresponding to 8 MHz channels in the 8k mode. Table Valid for 8 MHz channels and 8k mode For other cases see TR 101 290 Table 3. Guard interval used Spectral line separation, in Hz 1 / 4 223.2 Hz 1 / 8 248.0 Hz 1 / 16 262.6 Hz 1 / 32 270.6 Hz In Figure 13, a measurement of 225 Hz is shown on the left side and it can be immediately deduced that it corresponds to a guard interval of 1/4, since this is the closest value according to the table. However, on the right side the measurement has been repeated with a span allowing for higher resolution, and a better approximation to the table value is observed with a measurement of 223.125 Hz. In general terms, the intent is to verify that the programmed guard interval for the transmitter is the one that is actually being broadcast. It can also be used to identify the value of the guard interval, in case it is not known. The precision of the guard interval is subject to the precision of the sampling frequency, which was seen earlier, since they are dependent on one another. Figure 14, in a manner similar to Figure 13, shows the measurement in another channel, 27, which is using a guard interval of 1/8 and where the separation between lines is 248 Hz. In the 8k mode, where the symbol duration is 1,120 µs, 1,008 µs, 952 µs and 924 µs for 1/4, 1/8, 1/16 and 1/32 guard intervals respectively, the repetition periods (and therefore the separation between spectral lines) is 4,480 µs where 1/4,480 = 223.21 Hz for GI = 1/4; 4,032 µs where 1/4,032 = 248.02 Hz for GI = 1/8; and so on for other GI values. Emission Masks It is desirable that the energy transmitted in a TDT channel be limited to the assigned channel and not disturb adjacent channels. Once more, the TR 101 290 document indicates the principles that should be followed: 9.7 RF and IF spectrum (Emission mask) Purpose: To avoid interfering with other channels, the transmitted RF spectrum should comply to a spectrum mask, which is defined for the terrestrial network. If the spectrum at the modulator output is defined by a spectrum mask, the same procedure can be applied to the IF signal (with no pre-correction active). 16

Figure 13. Guard interval measurement (Channel 58 in 8k mode and 1/4 GI). Figure 14. Guard interval measurement (Channel 27 in 8k mode and 1/4 GI). Method: This measurement is usually carried out by using a spectrum analyzer. The spectral density of a terrestrial DVB signal is defined as the long-term average of the time-varying signal power per unity bandwidth (i.e. 1 Hz). Values for other bandwidths can be achieved by proportional increase of the values for unity bandwidth. To avoid regular structures in the modulated signal, a non-regular, e.g. a Pseudo-Random Binary Sequence (PRBS), or a program type digital transmitter input signal is necessary. Care has to be taken to ensure the input stage of the selective measurement equipment is not overloaded by 17

Figure 15a. The first phases of critical mask construction. Figure 15b. The first phases of critical mask construction. the main lobe of the signal while assessing the spectral density of the side lobes, i.e. the out-of-band range. Especially in cases with very strong attenuation of the side lobes, non-linear distortion in the measurement equipment can produce side lobe signals that mask the original ones. Selective attenuation of the main lobe has proven to be, in principal, a way to avoid these masking effects. However, as the frequency response of the band-stop filter has to be included in the evaluation, this procedure may become somewhat complex. For the resolution bandwidth, the recommended values should not exceed 30 khz. Preferred values are approximately 4 khz. The measurement should be Noise-normalized to 4 khz. The masks are described in several tables of the foundational DVB-T specification document, EN-300744. The diversity of masks takes into account whether the adjacent channel, upper or lower, is of analog type such as PAL, PAL-I, with NICAM sound, etc. To illustrate an example of mask compliance measurements, a symmetrical mask for use in critical cases has been selected; that is, when the TDT transmitter is situated in an area where there are also low power analog channel transmitters. This is shown in Figures 15a, 15b, 15c, 15d and 15e. In Figure 15a, the measurement of the level (total mean power) is shown. In this example, it is approximately -50 dbm. It must be noted here that due to lack of access to an actual transmitter, this example simulates the measurements at the output of a transmitter where the total power is naturally far superior. The EN-300744 document indicates that this measurement must be carried out with a spectrum analyzer equipped with a 4 khz resolution filter. This value is not normally found in modern analyzers, but it was selected for historical reasons. The analyzer used for this article features an ample selection of filters in various work modes. One filter which approximates the nominal specification with an error of less than 20 percent typically specified in conventional analyzers is the 4.166 khz filter with an error of only 4.15 percent with respect to 4 khz. Due to the use of that filter, the flat part of the spectrum is shown on the screen with a level of: 10 log(4 khz / 7.61 MHz) = 32.79 db below total power 18

Critical mask EN-300744 Total channel power = 0 db Channel 58 from example at 770 MHz Total channel power = -50 dbm (0 db) -Freq. Offset (MHz) Level (dbc) +Freq. Offset (MHz) -Freq. Offset (MHz) Level (dbc) +Freq. Offset (MHz) -3.8-32.8 3.8 766.2-82.8 773.8-4.2-83 4.2 765.8-133 774.2-6 -95 6 764-145 766-12 -120 12 758-170 782 Table 4. Figure 15c. The following phases of critical mask construction. Figure 15d. The following phases of critical mask construction. The reference value of the measured average signal power is taken as 0 db, and the mask is traced nominally at -32.8 db, which represents that value measured with a 4 khz resolution filter. From this point on, we highlight the various compliance points as a function of frequency offset and the desired attenuation for the emission spectrum. The left side of the table is labeled in dbc, which indicates power below carrier and wears implicitly the minus sign. Figure 15b shows the relative power level of three channels, two of which are analog, in a 24 MHz span. It can be seen that the average power received from channel 58 is similar to the power of the video carrier of the adjacent upper channel, and is approximately 4 db below that of the adjacent lower channel. Figures 15c and 15d show the continuation of the mask that has been drawn over several printed screen captures of the analyzer. Auxiliary lines available in this analyzer model have been used (magenta colored vertical and horizontal) to illustrate the levels in this example. 19

Analyzer sensitivity and dynamic margin The spectral density of the thermal noise power transmitted to a spectrum analyzer by a nominal load of 50 _ connected to the input, and maintained at a temperature of 290 K, is -173.975 dbm/hz. This is approximately -174 dbm/hz and, given that an analyzer such as the one used in this article shows a spectral density of some -151 dbm/hz (see Figure 4), it can be said that the analyzer s noise factor raises the noise level shown by 23 db. Figure 15e. Critical mask according to EN-3000744. In Figure 15e, the complete mask for this transmitter-verification simulation has been drawn. It may appear that the lower points of the mask cannot be obtained with a RTSA since they are situated at levels which are so low that they will always remain hidden due to the analyzer s own background noise. However, recall that this simulation was performed using a signal received with an antenna and a very low level compared to the signal strength available at the transmitters. Therefore, if the measurement is taken at the transmitter output and after the band-limiting filter, with a power level some 60 db higher than that used here (i.e. with some +10 dbm of power), the upper level of the spectrum would end up at -22.8 dbm, while the -83 db and -95 db points would be found at -73 dbm and -85 dbm, well above the -115 dbm where the noise level is situated using a 4 khz resolution filter. The -120 db points would be found at -110 dbm, too close to the noise. Even though this may seem sufficient to measure those points, there is still one aspect which must be considered, and it is related to the characteristics of spectrum analyzers. The middle trace of the thermal noise analyzed with a 4 khz filter would be -174 + 10 log 4,000 = -174 dbm + 36.02 db -138 dbm, and the screen of the analyzer would show 23 db above that. In other words, it would be a noise of approx. -115 dbm. This is the level shown for Figure 15e, and would appear to corroborate the fact that the -110 dbm values corresponding to the ±12 MHz points of the mask can be seen and measured. However, as will be demonstrated later, this is not as easy as it may seem. Distortion-free input level This is a characteristic that does not usually appear in spectrum analyzer specifications because it is very difficult to define what distortion free means. Instead, what is usually specified is the value of the point at which compression is 1 db. The analyzer used in this example is specified as +2 dbm (with 0 db input attenuation). However, this specification is far from adequate for the purpose described above, because what it is actually indicating is that a sine wave signal with a power level of +3 dbm at the input will compress by 1 db and will be measured as +2 dbm. The intermodulation distortion under these conditions would be so large that the shoulders of the spectrum would look highly amplified. Another usual specification is the TOI (Third-Order Intermodulation Distortion) level, which is typically specified in many different ways depending on the manufacturer, or even on the particular analyzer model. In the example used here, TOI distortion is specified for two sinusoidal 20

tones of the same level with a total combined power of -7 dbm, and with an input attenuation of 20 db, the TOI distortion will be better than 78 dbc. This means that the total power applied to the first mixer is -27 dbm (each tone has a -10 dbm level and reaches the mixer with -30 dbm), and the third order intermodulation products (IMD) generated by the analyzer itself will be situated at 78 db below each tone, meaning they will be found below (-10 dbm 78 dbc) = -88 dbm. So they would be, for a DVB-T signal, above the mask levels. It is usually accepted that the distortion-free level for most of the measurements made with a conventional RF spectrum analyzer with a diode-bridge mixer is obtained with power levels in the first mixer that do not exceed -30 dbm. In OFDM type signals, the number of carriers is huge. In the case of the DVB-T standard in 8k mode, there are 6,817 carriers. If they are considered to all have the same power, for the sake of simplifying the math, when the total signal power is -30 dbm the average power of each one of the carriers would be (-30 dbm 10 log 6,187 db) = -68.34 dbm. The IMD products generated by the analyzer would then be below its own noise level when considering the IMD produced by each pair of carriers, but the sum of IMD products created by so many carriers are at much higher levels than that. Therefore, a +10 dbm signal as shown in Figure 15e should be attenuated by 40 db so it reaches the mixer with -30 dbm. Also, all the values should be reduced by the same amount, so the mask points end up at -62.8 dbm for the upper level of the spectrum at ±3.8 MHz, -113 dbm for the points at 4.2 MHz, -125 dbm for the points at ±6 Mhz (i.e. below the noise level of the analyzer when a 4 khz filter is used, which is -115 dbm in this example) and -150 dbm for the points at ±12 MHz, which is well below the noise level. Dynamic range The dynamic margin with respect to the background noise, considering a power level of -30 dbm at the first mixer and using a 4 khz filter, is [-30 - (-115)] = 85 db. The -115 dbm value is the DANL (Display Averaged Noise Level) for 4 khz. Of course, with 1 Hz noise bandwidth filters the dynamic margin would be 122 db (since the DANL in this case would be -152 dbm), or even higher if filters lower than 1 Hz are used. But the measurement standard specifies 4 khz filters and the mask parameters specified are adequate for that filter value. The critical mask from document EN-300744 for DVB places a substantial demand on spectrum analyzers, both in terms of sensitivity and intermodulation dynamic margin. If an analyzer with a DANL of -128 dbm at 4 khz is considered, it would require that the analyzer s own noise factor be NF = 10 db and the signal level be adjusted to reach the mixer at -30 dbm in order for it to work distortion free. A situation as depicted in Figure 15f would result, where new scale factors are simulated on the right. As can be seen, the -120 db critical mask points cannot be measured because they would still lay some 22 db below the hypothetical noise level of the analyzer. In other words, the signal would have to be raised by more than 22 db in order to be able to measure the mask, which imposes the requirement that the distortion free intermodulation level of the input stage of the analyzer be higher than -8 dbm. This implies a dynamic margin of 120 db free of intermodulation distortion. It is worth observing that preamplifiers will not help to solve the problem, because if an amplifier were used with a gain of, for example, +30 db and a noise figure of +5 db NF, what is increased is not the dynamic range, but the sensitivity. 21

With an amplifier that has a lower noise figure, the sensitivity can be improved even further without losing dynamic margin. Using an amplifier with a 2 db noise factor and 20 db gain, a 20 db increase in sensitivity would be obtained. But as may be assumed, measurements at the transmitter base do not require increased sensitivity since a lot of power is available and the problem usually is how to attenuate the signal in order to feed it to the analyzer without damaging it or distorting it. Figure 15f. Simulation with -30 dbm signal and aalyzer with NF = 10 db Under these conditions, the preamplifier input signal should be -60 dbm so that the analyzer s mixer still receives it at -30 dbm. The thermal noise equivalent to the amplifier input would be -174 dbm + 5 db = -169 dbm, which, when amplified by 30 db, would result in an output level of -139 dbm being sent to the analyzer. To use a 4 khz filter, 36 db should be added, resulting in -103 dbm, a noise level 13 db worse than the -115 dbm that was obtained without using the preamplifier. The dynamic range would have fallen by 13 db while the sensitivity would have increased by 30 db. In fact, preamplifiers reduce the spectrum analyzer dynamic range and they are normally only used when sensitivity boosting is required, even to the detriment of dynamic range. In case a preamplifier with a noise figure of 5 db is used, its gain should be less than +30 db in order to boost sensitivity without reducing dynamic range. For instance, a gain of 17 db would be an ideal value to increase sensitivity by that amount while maintaining the noise level at -116 db. Reducing the filter resolution to a lower value does not solve the problem, since the signal would also be reduced proportionally to the noise, given that both the desired and undesired signals are of the dense spectrum kind, and behave like thermal noise. This section was written to explain the problems that will be encountered when making the emission mask measurement with any spectrum analyzer and the reasons why this happens. The mask is very stringent at ±12 MHz. The problem arises from the fact that the total power of the channel appears at the front end of the instrument, and this power is 32.8 db higher than the top of the spectrum when measured at 4 khz RBW. If the main channel is filtered out with a filter of about 40 db attenuation that does not attenuate significantly from ±8 MHz through ±12 MHz, the out of band emissions can be easily seen and measured with the RTSA, depending on the characteristics of the filter. Another method, perhaps easier, is to measure the passband filter to be implemented at the output of the transmitter and characterize its attenuation along the measured channel up to ±12 MHz of center. Next add these attenuation values to the amplitude correction table available in the RTSA, and then measure the spectrum before the filter. 22

Linearity Measurements Linearity measurements aim to determine the linearity of transmitters and repeaters. With thousands of carriers present (6,817 in the 8k mode), many intermodulation products result from a lack of linearity and are distributed within and beyond the channel. The products that fall within the channel degrade its demodulation and greatly lower the MER (Modulation Error Ratio) value at the transmitter output, which cannot be solved once produced. This represents the most harmful effect caused by intermodulation. Even though the products that fall outside the channel, which are seen as an expansion of the shoulders of the spectrum, can affect adjacent channels, the mask filter placed on the transmitter s output limits interference on those channels. Therefore, the main concern is related to the degradation of the MER as explained above. The effect of the intermodulation products within the channel, once produced, cannot be separated from random noise or interference of any other kind; therefore, it is not possible to measure it in isolation from these interferences. It is only possible to measure the global effect, and those results are included in the MER measurement, which is taken by analyzing the points of the constellation. In order to transmit with nearly perfect linearity, it would be necessary to work with a peak/average power ratio as high as 15 db, which would yield a probability of occurrence of peak power clipping equal to 9.4x10-9. Since OFDM peaks occur once per symbol, this probability indicates that clipping occurs once every 106 million symbols, and if the transmission is made in 8k, 8 MHz and _ GI, each symbol lasts 1,120 microseconds (i.e. there are 893 symbols per second). Therefore, on average, symbol clipping should occur every 119,149 seconds (i.e. every 1,986 minutes or every 33 hours). A back-off is frequently applied to transmitter power output, and work is usually done with peak/average power ratios in the order of 10 db to 12 db. This implies accepting that they work in quasi-linear mode. This occasional distortion is accepted by trusting that the error correction systems will be able to compensate the lack of linearity. Even though distorting the output signal with occasional clipping reduces the theoretical coverage of the transmitter, this is compensated by the coverage increase caused by the transmitter s power increase. Making a similar calculation, for 10 db we obtain a probability of 7.8x10-4 (i.e. once every 1,282 symbols or once every 1.44 seconds), a value easily correctable by FEC systems at the cost of loss of coverage in border areas. It is useful to recall that an analog TV transmitter with a peak power output of 1 kw (analog TV transmitters are specified based on peak power during the sync pulse instant), would provide an average power of 100 W if working with digital TV with a 10 db back-off, while with only a 7 db back-off the power would rise to 200 W. If, on the other hand, a 13 db ratio is desired, the average transmitted power would be 50 W. Given the occasional clipping produced by reducing the linear margin of the transmitter, a DVB-T/H spectrum shows instantaneous shoulder growth, which occurs with higher or lower frequency depending on the degree of transmission power clipping. In the lower part of Figure 16 (page 24), there is a spectrogram corresponding to the acquisition of a block of consecutive spectra in real time. Each spectral frame corresponds to a time interval of 800 µs (value determined by the selected span, which in this case is 1 MHz), and these are shown from frame 0 at the bottom of the spectrogram to frame -87 at the top. A spectrogram is a collection of spectral frames, called a block, stacked as they are being generated such that the most recent frame is added to the bottom. Upper frames have negative numbers because they are added at a previous time, which can be seen in the time scale on the vertical axis of the spectrogram. Frequency is represented 23

Figure 16. Real-time mode acquisition. Normal spectrum. Figure 17. Real-time mode acquisition. Instantaneous intermodulation spectrum. on the horizontal axis and corresponds linearly to the frequency scale of a conventional spectrum (the upper part of the figure), while the amplitude scale is represented by the color encoding shown on the left side of the spectrogram. The spectrogram shows that in some spectral frames the spectrum of the land-based TV channel signal extends toward the right. These are the green lines that cross the vertical dividing line between the OFDM signal spectrum (green zone on the left) and the guard valley (central blue zone of the spectrogram), which is found between two consecutive channels, the digital channel being measured (58 UHF in this example) and the analog channel above it (59), with a lower lateral band that can be identified by the green zones on the right side of the spectrogram. Color represents the intensity of the signal received and, in this case, green corresponds to the strongest signals. As a side note for the curious, and given that the spectrogram has been taken in real-time mode, it is possible to identify part of the structure of the analog channel and identify it in the figure as the duration of two TV image fields (20 ms per field), and the corresponding vertical blanking intervals. In Figure 16, spectral frame number 67 has been selected from the block, identified by a marker and a white horizontal line in the spectrogram. Its spectrum is shown in the upper part of the screen. Given that each spectral frame corresponds to an 800 µs interval, the marker is set at the time instant -68 x 0.8 = - 54.4 milliseconds, as indicated by the red box in Figure 16 (there are 68 frames including frame 0). This spectral frame corresponds to a spectrum with no intermodulation distortion as can be seen in the upper window. However, in Figure 17, frame 68 from the same block is shown, and it does exhibit distortion, which is manifested by the shoulder of the spectrum. This is due to the fact that the phase transition between symbols is very abrupt, and is inherent to the COFDM system. Each symbol is completely independent of the adjacent ones, so the instantaneous voltage value in the time domain at the end of one symbol may be far away from the instantaneous value of the beginning of the next symbol. This sudden voltage change happens in a time interval corresponding to a single sample in the D/A converter of the transmitter. 24

This poses no problem for the correct working of the system, however, as it is part of the orthogonallity concept and the receivers are not sensitive to these transitions. Since the receivers are time synchronized with the received signal and demodulate only the useful part of each symbol, the MER is not affected by this phenomenon. In other words, the receivers recover the lost orthogonallity due to the Guard Interval insertion and do not respond to the rapid transitions between symbols. The fact that the demodulators do not respond to this phenomenon does not mean that it does not exist; it manifests itself in the spectrum analyzers because the analysis window taken in the time domain for calculating the spectrum is not synchronous with the beginning and length of the actual received symbols. Swept spectrum analyzers do not have their sweeps synchronous with the symbols either, but in this kind of analyzer it is more difficult to see such spectral broadening. The other reason for this broadening is that occasionally the phases and amplitudes of many carriers are coincident in such a manner that a big voltage peak appears, and it is eventually attenuated or cut in a non-linear region of the power stages of the transmitter. This happens more often if the back-off applied is too much. When the broadening is due by these cuts of such peaks, the intermodulation produced does affect all carriers, and, in this case, the MER is degraded by the phenomena. These spectral frames can only be obtained by acquiring a complete block in real time. The duration of each spectral frame from Figures 16 and 17 is 800 µs, and the spacing between them is also 800 µs (i.e. continuous). This is due to the real-time spectrum analysis mode which enables a continuous analysis of the signal spectrum and its evolution over time. Each time a shoulder growth occurs, the constellation of all the carriers making up the symbol is distorted. Thanks to the scattering and assignment of different carriers to different bits of information, this distortion is distributed among many Transport Stream (TS) packets instead of being concentrated among a few. This allows the Forward Error Correction (FEC) algorithms, using Viterbi demodulation and Reed-Solomon decoding, to correct the errors and provide an error-free TS. However, it is not prudent to abuse the error correction systems. It is advantageous to reduce the effects of intermodulation, leaving enough margin for the FEC algorithms to correct other errors and noise that may appear. The phenomenon of occasional intermodulation also appears in collective-antenna preamplifiers. This happens if they are ignored in the design of the receiving head and/or in the distribution, and no adequate margin for amplifier linearity is left. This margin can be found to be between 10 and 15 db in order to avoid greatly increasing the effects caused by back-off at the transmitters, which are sending the signal being received, or can be of fewer decibels if the signal received is free from defects such as those described above. This is a good reason to avoid abusing back-off at the transmitter, since the distortion effect accumulates between transmitters and collectiveantenna amplifiers. It is important to observe instantaneous shoulder growth from the vantage point that this is a good indicator of the intermodulation phenomenon, which reduces the MER of the system and for that reason must be carefully measured as a health and quality indicator of the OFDM signal at the transmitter output. Even though the band-limiting filter will prevent interference on adjacent channels, the internal effect on the channel itself may be unacceptable. Besides, it should be taken into account that if intermodulation distortion is produced in collective-antenna systems, the fact that these lack effective filters means that shoulder growth will clearly affect adjacent channels, whether they are analog or digital. In principle, linearity measurements are unrelated to emission mask measurements, since the band-limiting filter may very well limit the invasion of the spectrum of adjacent 25

Figure 18a/18b. Shoulder measurement simulation, showing points A and B at 300 and 700 khz. channels. Even so, non-linearity distortion can be so disastrous that it may not be possible to demodulate the channel itself, which is why it is a good idea to carefully measure linearity, both at the power amplifier output as well as in collective-antenna installations. These measurements are done on the shoulders of the spectrum and in proximity to each of its ends; what is measured are the values situated at only ±500 khz from those ends. Measurement 9.10 from TR 101 290 was included to aid the verification and adjustment of transmitters during their design, manufacturing and maintenance processes. It is based on measuring shoulder growth in the spectrum, due to intermodulation products. The measurements are made at 300 khz, 500 khz and 700 khz from each end of the channel, and it is best to take the spectrum in several successive sweeps while setting the analyzer s recording mode to maximum-value mode, given that intermodulation products are generated differently in each of the emitted symbols. 9.10 Linearity characterization (shoulder attenuation) Purpose: The spectrum shoulder attenuation can be used to characterize the linearity of an OFDM signal without reference to a spectrum mask. Method: Apply the following procedure on the measured RF spectrum of the transmitter output signal: (a) Identify the maximum value of the spectrum by using a filter with a resolution of approximately 10 times the carrier spacing. (b) Place declined, straight lines connecting the measurement points at 300 khz and 700 khz from each of the upper and lower edges of the spectrum. Draw additional lines parallel to these, so that the highest spectrum value within the respective range lies on the line. (c) Subtract the power value of the centre of the line (500 khz away from the upper and lower edge) from the maximum spectrum value of (a) and note the difference as the shoulder attenuation at the upper and lower edge. (d) Take the worst case value of the upper and lower results from (c) as the overall shoulder attenuation. Note: for a quick overview, the value at 500 khz can be measured directly provided that coherent interferers are not present. Figures 18a and 18b show a simulation of shoulder measurement to indicate how the measurement method suggested in document TR 101 290 for DVB would be performed in practice. It was simulated due to the fact that an actual TDT transmitter was not available. 26

Figure 18c/18d. C point indication at -500 khz on the measureing line. An irregular green line has been drawn to show a simulated spectral response for a hypothetical transmitter, and over it the A and B points have been marked, indicating where this spectral line crosses the -300 khz (A) and -700 khz (B) frequency values. A straight blue line passing through both points has been plotted. In Figure 18c, a straight orange line parallel to the previous blue line and tangent to the highest point on the green spectral trace between points A and B (which is point C) has been plotted. Note that there is a higher point at approximately -210 khz, but it has been ignored because it does not lie between points A and B. Then, in Figure 18d, the value of the signal level considered for the measurement is taken as the value at the orange line at the -500 khz frequency point, even though at that point there is no real value for the green spectral trace. With that value referenced to the maximum value of the spectrum, a 62 dbc value is obtained in the example shown in Figure 18d, which would be the value resulting from the measurement and logged in the result report for the linearity measurement of the tested transmitter. In transmitters, it is advantageous to make these shoulder measurements by taking the spectrum before the band-limiting filter is applied to the transmitted signal. This is because the filters attenuate shoulder growth and therefore the emissions outside the channel, yet they do not eliminate simultaneous intermodulation produced within the channel by the power stages before these filters. Measurements made after the filters would give better results than if made before them, since spectrum expansion would be attenuated by the filters. However, the results would not render a correct view of power amplifier linearity because the intermodulation products within the channel will remain unaltered, while the whole point of measuring linearity is to quantify these effects. If for the sake of curiosity one should wish to check how these values correspond to the emission mask, the maximum value of the spectrum should be converted to the value it would have had if measured with a 4 khz filter. Since the measurement was taken with a 12 khz filter, the maximum value should be reduced by 10 log(12/4) = 4.77 db, resulting in a maximum value of -78.03 dbm - 4.77 db = -82.80 dbm and a zero level for the mask measurement at -82.8 db + 32.8 = -50 dbm. The mask is also shown in each of the Figure 18 images. As specified for the mask, the spectrum measured at -4.2 MHz from the center of the channel should be at -83 db with respect to zero, which in this example is -133 dbm. The -4.2 MHz point, which would correspond to a measurement at -400 khz from the edge of the channel (where 7.61 MHz / 2 = 3.805 MHz is the edge of the channel with respect to the center), is at a value of 27

approximately 56 db below the maximum value of the spectrum; that is, at -82.8 dbm - 56 db = -138.8 dbm or 5.8 db below the value specified for the mask. It should be taken into account, however, that these are fictitious values, since a hypothetical spectral trace was simulated. CCDF Measurements Measurement 9.10 from TR 101 290 (shoulder attenuation) has been designed for application during manufacture, and even though it can also be used for transmitter acceptance (commissioning), it is not very practical for monitoring transmitters in service. On the other hand, measurement 9.7 (mask) is useful for checking the absence of emissions in adjacent channels, but not for linearity or intermodulation measurements. There is another traditional measurement for digital modulation systems. In addition to being useful for monitoring transmitters in service it also serves, for example, in evaluating the linearity of collective-antenna systems where the signal available, supplied by the antenna and amplified by distribution amplifiers, contains all of the channels that are captured at the location being measured. In large cities, these collective-antenna systems may contain a highly populated spectrum. On the other hand, it is understood that in some cases both transmitters and collective-antenna distribution systems exhibit some degree of linearity distortion. As indicated above, this distortion is more or less controlled since errors introduced by it can be compensated by FEC error-correction systems. However, this distortion should not exceed certain values since it would then saturate the FEC systems and possibly render them ineffective. Under these conditions, one may ask: How often do these occasional errors occur? And more important: How can one determine if a broadening of the shoulders has been caused by the transition between symbols, where MER is not affected, or if it has been caused by excessive back-off or other non-linearity that indeed affect the transmitted MER? Figures 16 and 17 (page 24) show that spectrum expansions, perhaps due to occasional compression of peak signals that exceed the average power value, are happening with high frequency; some 5 ms if only those of the largest amplitude are considered. But they are of different amplitudes in each case and are therefore difficult to quantify. In this run, shoulder growth as large as those mentioned has not been detected, but a number of smaller values have been found, which occur more frequently. CCDF (Complementary Cumulative Distribution Function) measurements are useful for quantifying the amount by which instantaneous power values exceed the average value, and for how long. More importantly, this measurement is not affected by the spectral growth caused by the transitions between symbols. To obtain this, the instantaneous power value is measured during relatively long periods of time, and is tabulated with respect to the average power measured during that time period. It is advantageous to refresh some statistical concepts, and Figure 19 may be useful for this. Consider a collection of instantaneous power point measurements. Digital power meters, digital oscilloscopes and digital spectrum analyzers of one kind or another (sweep or FFT) that are able to carry out this function usually present a collection of instantaneous power samples in a graphical form similar to those seen in Figure 19. Spectrum analyzers generally have the advantage of being able to select the bandwidth over which the measurement is made and, in general, are also more sensitive than other instruments. That is to say, they are selective and sensitive. 28

the result will be the absolute distribution of instantaneous power values over the total range of available power values. If the value of each interval is normalized by dividing by the total number of samples recorded, the relative distribution is obtained. The normalized value of each interval calculated this way is called interval frequency of occurrence. Figure 19. Basic Probability Density Function concepts. Figure 20. PDF and CCDF function basic concepts. If the instantaneous power values obtained are ordered in small and regular intervals, and the number of times a measurement has fallen within each interval is recorded, As the intervals are narrowed, the relative frequency value also drops and this can be compensated by dividing the relative frequency values by the width of the interval, which results in the histogram. If the intervals are made very small (tending to zero) and a very large number of samples is taken (tending to infinity), the histogram becomes the Probability Density Function (PDF), as seen in Figure 19. To know the probability that a given sample has a power value between zero and the x1 and y1 values respectively, the probabilities of each of the small intervals between zero and x1 or between zero and x2, respectively, should be summed. In analog terms, the PDF function integral is calculated up to x1 and x2 respectively, as can be seen in Figure 20. 29

The sum of all the values of the curve, from zero to the peak power value, is always one because the density function is obtained from the normalized histogram. The probability that a given sample has a power value greater than xj is easy to calculate if the probability that the power value is below xj is known, since the total is Figure 21. Logarithmic representation of CCDF function. always one. And since this is the case in calculations of probability of occurrence for a ratio between peak and average power, then the Complementary Cumulative Density Function (CCDF) is used. It is defined as CCDF = 1 - CDF. Figure 21 shows a CCDF function graph on a linear scale and another one on a log scale. The instantaneous power samples are shown on the horizontal axis expressed in db, and the probability is shown on the vertical axis in percentage deciles. Figure 22. CCDF measurement with the spectrum analyzer on TDT channel 58. The curve formed by the integral values of each and every one of the integration results from zero to each value on the x axis is called the Cumulative Density Function (CDF). Spectrum analyzers that include these measurements can be used to verify if the power distribution of a transmitter follows the expected pattern. In the case of digitally modulated signals, the CCDF curve depends on the characteristics of the modulation employed and the spectrum conforming filters used. In the case of OFDM signals, the number of carriers is extremely large and without spectrum conforming filters, except the adjacent channel limiter. Through the flat part of the spectrum is a distribution quite similar to thermal noise, called Additive White Gaussian Noise (AWGN). Therefore, the curve measured for a transmitter can be compared to a theoretical Gaussian curve corresponding to white noise. 30

Figure 23a/23b. CCDF measurements on 10 MHz and 5 MHz spans, respectively. Figure 22 shows the result of measuring 10.5 million instantaneous power values in a channel center (channel 58 in this case) using a 500 khz bandwidth (the analyzer s span was set to 500 khz, centered at 770 MHz). The green trace represents the result of that measurement, and overwrites the gray reference trace corresponding to the AWGN white noise curve. It should be considered that if, for the measurement, a span wider than the real width of the spectrum is used, as is shown in Figure 23a with 10 MHz, the curve will not follow the Gaussian reference curve and an excessive back-off may be assumed in the transmitter being analyzed. This is due to the fact that the average power value calculated is distorted by the low level noise spectrum values, which exist on both ends of the analyzed spectrum. Figure 23b shows that taking a MHz span centered in the spectrum results in the curve being Gaussian. To observe this, it suffices to record 620,544 samples instead of reaching 10.5 million as was the case in Figure 22. It can also be seen that with a reduced span, the curve is also Gaussian, and it is still possible to record a relatively large number of carriers so that the result is similar to AWGN noise. This is shown in Figures 23c and 23d with 2 MHz and 1 MHz respectively, and where the number of points taken is greatly inferior to that used in Figure 22. As can be seen, a span of only 500 khz still shows the Gaussian characteristic. The number of points accumulated for the graph of the CCDF curve affects the total result, which establishes the value at the crest of the signal. The crest value is 12.08 db with 10,536,960 points accumulated in the case of Figure 22, and 10.89 db with 215,040 samples and 10.98 db with 600,064 samples for Figures 23c and 23d respectively. To reach values of 15 db in Gaussian noise requires taking many more samples, because the probability of reaching that value is lower than 9.39 x 10-9, while the graph of Figure 22 shows values of 1 x 10-7 with more than 10 million samples. 31

Figure 23c/23d. CCDF measurements on 2 MHz and 1 MHz spans, respectively. SFN Synchronization One of the characteristics of COFDM modulation is that it has a great protection capability against signal reception through different routes, either due to multiple reflections or to the eventual reception of the same signal originating from two transmitters or two repeaters emitting in the same RF channel. To clarify, three cases may be considered. Case A corresponds to the reception of reflected signals (i.e. echoes) and is frequent in all cities and even in rural areas with mountainous geography. These echoes are undesired, and in the case of analog signals such as PAL or NTSC transmissions, they are manifested as double images or ghosts. Note that this can also occur within collective-antenna systems where impedance mismatches may exist in the distribution network. Case B corresponds to multiple signal reception in the same RF channel and occurs in digital broadcast networks such as DVB-T and DVB-H. In taking advantage of the system s robustness, signal repeaters are used to cover certain zones with the same channel as the main transmitter. This was not possible in analog networks due to the creation of echoes or ghosts, but it is possible in COFDM modulated digital networks. The two previous cases do not conform to what is known as a Single Frequency Network (SFN), but are examples in which multiple reception can be caused both intentionally and unintentionally. Case C corresponds to SFNs where a given zone is covered by a network of two or more transmitters, each with independent modulators, which emit on the same RF channel. In this case two conditions must be met: 1. It must be ensured that every transmitter in the network emits the same COFDM signal symbol in the same instant of time (time synchronization). 2. It must be ensured that the central frequency of each channel is exactly the same for all of the transmitters (frequency synchronization). 32

The first condition of time synchronization requires that the base band signal the Transport Stream (TS) be the same; same content, bit rate and modulator sampling frequency in all modulators. This first condition is normally met with a Pulse per Second (PPS) signal supplied by a GPS receiver, and with the information included in a data package from the TS itself known as MIP (Mega-frame Initialization Packet). The MIP is generated in a SFN adapter at the head of the entire transmission network. The second condition of frequency synchronization requires that the local RF channel oscillators of every transmitter be synchronized at the same frequency. This second requirement is met using a 10 MHz synchronization reference signal, also available in GPS receivers. In cases A and B, the first condition is met by the nature of the phenomenon itself (i.e. the reflection of the same signal and the repetition of the same signal respectively). In case A, the second condition is also met by the very means it is generated, and in case B it may automatically be met if the repeater implements the Gap Filler concept either in professional or domestic models. Gap Fillers receive the RF signal with a properly aimed antenna. They amplify it and re-send it with another antenna that has been appropriately isolated from the receiving antenna. Logically, in this case, the signal is the same, and the central frequency is the same as well. Case B can also be carried out with re-emitters which convert the received signal to IF, which later is again converted to RF on the same channel. In these cases, the design of the synthesizers used in the local oscillators, which affect the conversions, must use the same reference frequency and be stable with respect to the central frequency received in order to assure that the second condition is met. It is not normal for a re-emitter to be designed with independent oscillators for conversion to IF and then RF, and thus it can be assumed that this case will not occur in practice. In case C, compliance with the first condition is enforced as was described above with the SFN adapters and the common PPS reference signal for the whole broadcast network. Also, compliance with the second condition is enforced by using a reference signal common to the whole network, as is the case of the 10 MHz GPS signal. Given these helpful explanations divided into arbitrary cases A, B, and C, and the two conditions that must be met so a reception will be considered good, another division may be considered for static reception or reception under motion. These two concepts are important for DVB-T reception, which may be carried out under static conditions or under motion with a velocity range from that of a pedestrian to that of vehicles with speeds higher than 150 km/h. To achieve greater design freedom in SFN networks where good reception under motion is desired, a 4k mode has been designed that expands the range of options to 2k, 4k and 8k. A network operator can choose one mode or the other according to the coverage required, taking into consideration geographic reception characteristics such as highways, railways, etc. The DVB-H mode (for portable handheld reception) is independent of the choice between 2k, 4k and 8k modes. This mode is also noteworthy because, beyond the interest in reception under motion, it features energy saving requirements at the receivers. Figure 24 shows multiple reception, which can occur in any of the cases described previously as A, B and C. It is a static reception, and the fact that it is receiving one signal delayed with respect to another has an effect which can be described as corresponding to a comb filter (sum of two or more signals, with a given delay). The lobes that can be seen have a frequency domain width that is inversely proportional to the difference in reception times from each signal, which is to say they indicate the delay between 33

Figure 24. Multipath reception in UHF channel 27. Figure 25. Multipath reception in UHF channel 27, with corresponding spectrogram. the two signals. (It can be noticed that the lobes on Figure 24 are not due to the data carriers seen previously; the lobes shown here are much wider and in this example are equivalent to 56 carriers @ 62,5 khz / 1,11 khz = 56 carriers). The calculations corresponding to the measurements taken with both cursors have been placed on the figure itself. They indicate a lobe width of 62.5 khz. This corresponds to a 16 microsecond delay, and the measurement indicates this value is lower than the emission guard interval (set at 1/8 during the tests, and in the 8k mode, corresponds to 112 µs). In addition to the delay measurement, it is possible to obtain the distance difference between the two emitters with respect to the receiver. Considering the speed of electromagnetic signals is 300,000 km/s, a distance of approximately 4.8 km is obtained. In other words, if one transmitter is at a distance of x km from the receiver, the other will be at a distance of x + 4.8 km, though not necessarily in the same direction. The above conclusion is valid whether the lobes are stable in position or not; successive spectra show the same image, meaning the average spectrum can be obtained over a small time interval in order to get a more detailed graph. However, this was not the case for figure 24, as the lobes were shifting toward the right. This phenomenon indicates one of two things: one of the transmitters was moving, which was known not to be the case, or there was a difference in central frequency between the two transmitters. This implies that the second condition was not met, from which we deduce that this is an SFN network, since noncompliance with the second condition can only happen in this type of network. The frequency difference between the two transmitters can be measured in two ways: directly with very narrow (1 Hz or less) resolution filters, or by using the spectrogram available in some digital spectrum analyzers. Both methods will be illustrated. Figure 25 shows the spectrogram, which, as indicated previously, displays the older spectra on the top, while the latest spectrum measured is at the bottom. An approximate measurement indicates that in a 1.8 second interval, the lobes have shifted by a frequency distance equal to the width of a lobe. In other words, the repetition cycle of the lobe position is 1.8 seconds, or 1 / 1.8 = 0.5555 Hz, which means that the frequency difference between both transmitters of the FSN network is only one half of a Hertz. Another way to interpret Figure 25 is to consider a carrier set at the upper yellow-red point of Figure 25, which varies in amplitude as time progresses, dropping through the spectrogram to the lowest spectrum and completing the 34

Figure 26. Multipath reception in UHF channel 27, with real-time spectrogram. Figure 27. Multipath reception in UHF channel 27, central frequency measurement. cycle in 1.8 seconds. Recall that the beat between two sine waves corresponds to a frequency that will be the frequency difference between the two sine waves. This small error is the cause of moving lobes and therefore affects the receiver since the spectrum changes during reception of each symbol (i.e. it is not the same before and after the signal is received). Thus, the reception and the MER of the signal received will be worse with increasing frequency difference between the two transmitters. This phenomenon, which in the illustrated example is due to a slight frequency error between the two transmitters, also occurs in reception under movement even if only one transmitter is used. It is more pronounced if two or more transmitters are used. This is due to the Doppler Effect. Another way of measuring the frequency difference between two SFN network emitters using the spectrogram is to register the signal in the analyzer s real-time mode, as shown in Figure 26. In this figure, an 80 ms interval between spectra has been selected, with a spectrum length of 80 ms as well. In other words, the signal is continuously registered, with no lost intervals between spectra. Observe that the selected span is only 10 khz, and thus the spectral length corresponds to 80 ms. It is much longer than that obtained in figure 24, where there was a 1 MHz span with a spectral length of only 800 µs. In order to see a 1.8 second interval, 2,250 spectra were required, which wouldn t fit on the screen. In order to illustrate lobe width, Figure 25 was not obtained in real-time mode. The multiple reference line mode has been used, and these lines have been adjusted to the point where they match in temporal distance the repetition of the amplitude increase of the spectrum. From this we obtain a direct reading of 1.76 µs between lines; that is, between periodic spectrum variations. Again, the inverse supplies the frequency difference between each transmitter (in this case, 568.1818 mhz), which is also shown in the measurement window. It can be seen that this measurement is somewhat more precise than the previous one. However, the frequency difference can also be measured directly in the spectral mode and with a resolution filter narrower than the frequency separation to be measured; that is, with a filter narrower that 0.5 Hz. Figure 27 shows the measurement of the central frequency of UHF channel 27, where it can be seen that there are two carriers instead of one. With a 125 mhz resolution filter, they can be separated and measured as two spectral lines 500 mhz apart, which is the width corresponding to two resolution filter widths. 35

The delay measurements shown are possible only when the principal and secondary signals have similar amplitudes, as is the case illustrated here. These delay measurements may represent those produced in the three cases A, B and C shown before, and are therefore not specific to SFN networks. In all the cases where the signals may differ in amplitude by a large amount, it is advantageous to carry out these measurements with an OFDM signal demodulator. This is performed by measuring the reference pilot carriers to find the impulse response of the channel from the frequency response obtained. However, the measurement of central carrier differences can easily be made even if the amplitude between the two received signals is quite different. This is a typical SFN measurement. Continuous-signal Interference Continuous Wave (CW) interference is produced by analog channel carriers, which, even being located far away, may be reaching the point where reception signal quality measurements are being made. It is generally measured with a spectrum analyzer following the method described in the measurements document: 9.12 Coherent interferer Purpose: To identify any coherent interferer which may influence the reliability of the I/Q analysis or the BER measurements. Method: The measurement is carried out with a spectrum analyzer. The filter resolution is reduced stepwise so that the displayed level of the modulated carriers (and of the unmodulated pilots, due to the influence of the guard interval) is reduced. The CW interferer is not affected by this process and can be identified after appropriate averaging of the trace. These measurements require a lot of patience with a conventional analyzer if the interfering signal is of small Figure 28. Interface in UHF channel 27. amplitude yet still presenting problems for COFDM signal demodulation. This is due to the fact that if the interfering signal is of low amplitude, it is necessary to use a very narrow resolution filter with which it will not be possible to cover the entire spectrum of the channel. This implies that the measurement must be taken in several steps, the number of which will depend on the maximum span that can reasonably be used. In the case of FFT digital analyzers, the process is facilitated thanks to the possible use of 65,536 point transforms, which allow for a reasonably wide span and a relatively narrow resolution while keeping a great update speed for the spectral frames. Figure 28 shows an example of interference in channel 27. The interference is so great that it does not permit illustrating smaller interference cases. However, it does allow illustration of the idea presented. The 12 MHz span allows the complete channel to be seen, as well as the audio and video carriers from the lower and upper adjacent channels, respectively. These channels are not the ones creating the problem. 36

The problem is in the analog signal, which is in the same channel that was being tested (in this case, these were tests of a QPSK modulated DVB-H channel with a guard interval of 1/8). This interference was so strong that even with a 300 khz resolution filter (upper trace) it could be seen and measured. However, the lower trace, with a 6.24 khz filter, shows how the level of the carriers of the analog signal has barely dropped, while the random signal of the digital modulation has been greatly reduced. It behaves as if it were white noise. Note, however, that the reduction in the analog signals is due to the fact that they are not pure carriers, but modulated. When the wide resolution filter is used, the carrier level appears, including some of the modulation sidebands, but the digital signal is reduced much more. This permits a clearer identification of the carriers present within the DVB-H channel. What Figure 28 does show is that besides the analog channel interfering with the digital signal, there is another interfering signal. This signal was not clearly seen when using the 300 khz filter, but can be perfectly appreciated by applying a filter 50 times narrower. In this case, the long FFT resource was not used; rather, the standard 1,024 point length was used. It should also be noted that by measuring this signal with a demodulator (the MTM400 model from Tektronix was used), it was possible to synchronize this signal. The signal is so strongly interfered and demodulated, it measured a MER value of 3.3 db. Carrier Suppression The measurement of carrier suppression or residual carrier can conceptually be made with a spectrum analyzer. This is performed by using the same method explained previously for the measurement of interfering signals. By simply selecting a reduced span (i.e. 50 Hz) and a very narrow filter, the digital components will remain greatly reduced and the residual carrier will appear. The residual carrier is generally produced by deficient suppression in balanced modulators, which transform the I/Q signals modulating directly to an RF channel. It should be considered that if a digital modulation is performed directly at an intermediate frequency and the IF is transposed to the channel frequency, it is possible to better control the modulation residues. However, doing it in one way or the other is a question for modulator designers, and each method has advantages and disadvantages. What is covered here is the measurement to detect if such a residue exists. There is, however, a significant stumbling block. As was said previously, in the 4k and 8k modes the central carrier is always a continuous pilot carrier. Due to the presence of a residual carrier, it is very difficult to determine whether the continuous pilot carrier s amplitude is altered with respect to the other pilot carriers. In the 2k mode, the central carrier has data, so perhaps this procedure can be used to check for continuous wave residues in the position of this data carrier. The most advisable option is to perform the measurements as described in TR 101 290 based on the constellation, but finding this only for the central carrier. In the case of the 4k and 8k modes, the central carrier has a perfectly defined position, and hence the systematic deviation due to the presence of a residual carrier can be measured. In the 2k mode, the systematic error of the different constellation points will have to be calculated, as indicated by the measurement document. Another indirect method for checking for carrier residues is to compare the MER measured only for the central carrier with the MER measured for the rest of the pilot carriers. In either case, this is not a measurement ideally suited for a spectrum analyzer; it is best analyzed using a MPEG Transport Stream Monitor such as the Tektronix MTM400. 37

operating margin (noise margin) of the system and may directly increase the BER. The effects of ICI are peculiar to OFDM and cannot be corrected. This has to be taken into account as part of the total noise of the system. Method: Phase noise can be measured with a spectrum analyzer, a vector analyzer or a phase noise test set. Figure 29. Mask and table of values suggested by document TR 101 290. Oscillator Phase Noise The set of DVB measurements that was created by the ETR 290 document and later the TR 101 290 document, which expands on the first, defines measurements for oscillator phase noise in DVB-T systems. The measurements suggest some limits that correspond to a determined mask. However, no study has been carried out that allows for recommendations to be made; hence the values remain undefined. 9.4 Phase noise of local oscillators (LO) Purpose: Phase noise can be introduced in the transmitter, at any frequency converter, or by the receiver, due to random perturbations in the phase of the oscillators. In an OFDM system, the phase noise can cause Common Phase Error (CPE) which affects all carriers simultaneously, and which can be minimized or corrected by using the continual pilots. However, the Inter-Carrier Interference (ICI) is noise-like, and cannot be corrected. The effects of CPE are similar to any single carrier system and the phase noise, outside the loop bandwidth of the carrier recovery circuit, leads to a circular smearing of the constellation points in the I/Q plane. This reduces the Method for CPE: Phase noise power density is normally expressed in dbc/hz at a certain frequency offset from the local oscillator signal. It is recommended to specify a spectral mask with at least three points (frequency offsets and levels); for an example see the figure. Method for ICI: For the measurement of ICI, the use of multiples of the carrier spacing is recommended for the frequencies fa, fb and fc. The measurement is conducted by taking the noise power at certain distances from the oscillator frequency and expressed as normalized noise to 1 Hz in each sideband (SSB, or Single Side Band, method). Figure 29 shows the mask recommended in the TR 101 290 document and the table with the values that may be used for ICI measurements. The points A, B and C are defined, but not the corresponding values of frequency and noise level allowed. To illustrate these measurements made with a spectrum analyzer, the 10 MHz signal supplied by a GPS receiver has been taken and the phase noise has been measured with different frequency offsets closer or farther from the carrier. Figure 30 shows the oscillator signal, and its power has been measured at +3.68 dbm according to the principal marker. Given that the dynamic range with respect to noise is in the order of 112 db, the analyzer s scale has been 38

Figure 30. Mask and possible of values for ICI measurements. Figure 31. Mask and possible of values for CPE measurements. shifted in order to get the signal above the reference amplitude level. (Note that this does not involve saturating either the mixer or the A/D converter). The secondary marker is set at a 1 khz offset and measures -111.9 db with respect to the principal marker, and for the resolution filter band, which is 12.5 Hz. The Delta Marker is obtained and normalized for 1 Hz, with a value of 122.9 dbc/hz. In this hypothetical example, it could be said that the analyzed signal is within the mask with a wide margin. Note that the mask is specified in terms of spectral density (i.e. in dbc/hz units) and recall that the term dbc implies db below the carrier. Therefore, these are positive values since they indicate how many db below the carrier the noise can be found. Figure 31 shows another hypothetical mask to specify phase noise closer than in the previous case. The secondary marker is set at -110.9 db below the first marker. The noise bandwidth (NBW) is 2.5 Hz, and when normalized to 1 Hz, a 104.9 dbc/hz value is obtained at a distance of only 50 Hz from the carrier. Figure 32 shows yet another hypothetical mask to specify phase noise even closer than in the previous case. The secondary marker is -68.9 db below the primary marker. The noise bandwidth is 0.5 Hz, and when normalized to 1 Hz, a 65.9 dbc/hz value is obtained at a distance of only 6.75 Hz from the carrier. Figure 32. Mask and table of possible of values for CPE measurements. Note that the 1 Hz normalized value is 3 db lower than the value measured by the filter, when it is usually the reverse. This is because, in this case, the filter s NBW is lower than 1 Hz (it is half), and that yields the 3 db. If this example mask were desired by a manufacturer, this oscillator should not pass the test. But it should be recalled once again that these mask values are fictitious, and have been employed here with the sole purpose of illustrating the procedure and the interpretation of measurements. 39

Conclusions The RF measurements recommended by the TR 101 290 document, and which are appropriate for the use of Real Time Spectrum Analyzers (RTSAs), have been reviewed. This document was created by the DVB Measurement Group and is applicable to TDT (Terrestrial Digital Television) according to DVB-T and DVB-H specifications. Even though the second mode was defined after the year 2000 revision of the TR 101 290 document, the RF measurements are applicable without restrictions. The RTSA adds significant value to RF measurements on DVB-T signals by allowing more flexible and narrow filtering, seamless capture and analysis over any desired portion of the signal capture. There are other TDT signal quality measurements according to the DVB, but these measurements are made with demodulators specifically designed for the task, and with Transport Stream (TS) analyzers which represent the base band signal that is applied to the modulators. Contact Tektronix: ASEAN / Australasia (65) 6356 3900 Austria +41 52 675 3777 Balkan, Israel, South Africa and other ISE Countries +41 52 675 3777 Belgium 07 81 60166 Brazil & South America 55 (11) 3741-8360 Canada 1 (800) 661-5625 Central East Europe, Ukraine and the Baltics +41 52 675 3777 Central Europe & Greece +41 52 675 3777 Denmark +45 80 88 1401 Finland +41 52 675 3777 France & North Africa +33 (0) 1 69 86 81 81 Germany +49 (221) 94 77 400 Hong Kong (852) 2585-6688 India (91) 80-22275577 Italy +39 (02) 25086 1 Japan 81 (3) 6714-3010 Luxembourg +44 (0) 1344 392400 Mexico, Central America & Caribbean 52 (55) 56666-333 Middle East, Asia and North Africa +41 52 675 3777 The Netherlands 090 02 021797 Norway 800 16098 People s Republic of China 86 (10) 6235 1230 Poland +41 52 675 3777 Portugal 80 08 12370 Republic of Korea 82 (2) 528-5299 Russia & CIS +7 (495) 7484900 South Africa +27 11 254 8360 Spain (+34) 901 988 054 Sweden 020 08 80371 Switzerland +41 52 675 3777 Taiwan 886 (2) 2722-9622 United Kingdom & Eire +44 (0) 1344 392400 USA 1 (800) 426-2200 For other areas contact Tektronix, Inc. at: 1 (503) 627-7111 Updated 28 February 2006 For Further Information Tektronix maintains a comprehensive, constantly expanding collection of application notes, technical briefs and other resources to help engineers working on the cutting edge of technology. Please visit Copyright 2006, Tektronix. All rights reserved. Tektronix products are covered by U.S. and foreign patents, issued and pending. Information in this publication supersedes that in all previously published material. Specification and price change privileges reserved. TEKTRONIX and TEK are registered trademarks of Tektronix, Inc. All other trade names referenced are the service marks, trademarks or registered trademarks of their respective companies. 11/06 DM/ xxx 37W-20027-0