Evaluation of single carrier and multi-carrier modulation techniques for digital ATV terrestrial broadcasting

Transcription

1 Evaluation of single carrier and multi-carrier modulation techniques for digital ATV terrestrial broadcasting by Richard V. Paiement, P.Eng. Radio Broadcast Technologies CRC REPORT NO. CRC-RP OTTAWA, DECEMBER 1994

2 Evaluation of single carrier and multi-carrier modulation techniques for digital ATV terrestrial broadcasting Abstract This report presents results of computer simulations and laboratory tests conducted at the CRC to investigate various channel coding techniques proposed for the terrestrial broadcasting of digital advanced television (ATV). The system performance of these techniques is evaluated for various impairments, such as Gaussian noise and co-channel NTSC and ATV interference. The performance of NTSC with interference from co-channel ATV is also evaluated. The channel coding techniques investigated include conventional single carrier per channel (SCPC) modulation using trellis-coded QAM constellations, and multi-carrier modulation (MCM), such as orthogonal frequency division multiplexing (OFDM) also with a trellis-coded QAM constellation. A measure of the bit error rate (BER) performance is given for comparing the various configurations. Results indicate that both single carrier and MCM offer similar performance in noise and NTSC interference, but OFDM can have a considerable advantage in multipath. Évaluation des techniques de modulation simple porteuse et multiporteuses appliquées à la radiodiffusion numérique de signaux télévision à l'aide d'émetteurs terrestres Résumé Ce rapport présente les résultats de simulations d'ordinateur et de tests de laboratoires effectués au CRC au cours d'une étude des techniques de codage canal proposées pour la diffusion terrestre d'un service de télévision numérique de pointe (Advanced Television: ATV). La performance des systèmes de ATV utilisant les techniques considérées est évaluée pour diverses situations, incluant du bruit Gaussien, et de l'interférence co-canal de signaux NTSC ou ATV. La performance d'un système NTSC subissant une interférence co-canal d'un signal ATV est aussi étudiée. Les techniques de codage canal considérées incluent la modulation conventionnelle, utilisant une simple porteuse modulée par une constellation QAM codée par treillis, ainsi que la modulation multiporteuse communément appelé OFDM (orthogonal frequency division multiplexing), qui utilise aussi une constellation QAM codée par treillis. Une comparaison des différentes configurations est présentée basé sur la mesure du taux d'erreur binaire obtenu pour diverses difficultés de transmission. Les résultats obtenus indiquent que les deux méthodes considérées, soit la modulation à porteuse simple et la modulation multiporteuse, offrent une performance semblable pour un environnement de bruit ou d'interférence dû à un signal NTSC, mais que le OFDM peut offrir un avantage considérable en présence de multi-trajet.

3 Table of Contents 1 Introduction 2 ATV Requirements 3 Channel Coding 3.1 Single Carrier Per Channel 3.2 Multiple Carrier Per Channel 3.3 Performance Comparison of Both Techniques 4 Simulation Procedure 4.1 TC32QAM Transmitter/Receiver 4.2 TCOFDM Transmitter/Receiver 4.3 Transmission Channel 5 Laboratory Test Procedure 5.1 Single Carrier Interference 5.2 OFDM Interference 6 Results - TC32QAM 6.1 Performance Results a) AWG Noise b) NTSC Interference c) ATV Interference d) Multipath 6.2 Analysis of Results a) AWG Noise b) NTSC Interference c) ATV Interference d) Multipath 7 Results - TCOFDM 7.1 Performance Results a) AWG Noise b) NTSC Interference c) ATV Interference 7.2 Analysis of Results a) AWG Noise b) NTSC Interference c) ATV Interference i

4 8 Results - NTSC 8.1 Performance Results a) 16QAM and 32QAM Interference b) OFDM 16QAM Interference 8.2 Analysis of Results a) 16QAM and 32QAM Interference b) OFDM 16QAM Interference 9 Comparing Conventional and TCOFDM Modulations 9.1 Noise into ATV 9.2 NTSC Interference into ATV 9.3 Multipath into ATV 9.4 ATV Interference into NTSC 9.5 Complexity 9.6 Flexibility 10 TCOFDM: Recommendations for Future Research 10.1 Carrier Separation 10.2 Code Linking Versus Spectral Shaping 10.3 Reliability Weighting by the Viterbi Decoder 10.4 Optimal Trellis Code 10.5 Perfect Interleaving 10.6 Perfect Interference Estimation 10.7 Perfect Channel Response Estimation 11 Conclusions 12 Acknowledgement A Preliminary Investigation of Multi-Layer Services for ATV References ii

5 1 Introduction Current advanced television (ATV) research for terrestrial broadcasting in the VHF/UHF bands is converging toward fully digital implementation. Selecting a scheme for digital channel coding, to deal with impairments such as noise, co-channel and adjacent channel ATV and NTSC interference, and multipath, is currently the subject of some discussions. Conventional single carrier per channel (SCPC) modulation schemes are being considered by many, while others prefer the alternative of multi-carrier modulation (MCM), implemented by the orthogonal frequency division multiplexing (OFDM) scheme. The requirements for a digital ATV system are presented in Section 2, followed by a brief description of the channel coding techniques of interest, namely conventional single carrier per channel and multi-carrier modulation, in Section 3. The computer simulation work and laboratory test procedure are described in Section 4 and Section 5. Performance results are presented and discussed in Section 6 for the single carrier system, in Section 7 for the multi-carrier systems, and in Section 8 for conventional analogue NTSC. Advantages and disadvantages of multi-carrier systems versus single carrier are considered in Section 9, while Section 10 discusses future work for ATV services, relating to the channel coding. Conclusions are presented in Section ATV Requirements In the generic ATV system block diagram shown in Figure 1, the original audio and video HDTV material, which represents well over 1 Gbit/s of data, is compressed to about 20 Mbit/s with the help of source encoding. This compressed data must then be channel encoded to fit within a 6 MHz bandwidth, to allow for terrestrial broadcasting of the information in conventional television channels within the VHF and UHF bands. HDTV MATERIAL SOURCE ENCODING CHANNEL ENCODING CHANNEL CHANNEL DECODING SOURCE DECODING HDTV DISPLAY Figure 1: ATV system block diagram. This modulated signal is transmitted over the terrestrial broadcast channel, which suffers from various impairments, including multipath, additive white Gaussian (AWG) noise, and interference from co-channel and adjacent channel NTSC and co-channel and adjacent channel ATV. Within the receiver, the channel decoder demodulates the incoming signal while compensating as much as possible for the channel impairments. A bit error rate (BER) on the order of 10-9 needs to be achieved in order for the source decoder to reconstruct audio and video data required to provide an acceptable distribution quality HDTV service. 1

6 Forward error correction (FEC) encoding is required to achieve this level of BER. Typically, two independent levels of coding are used: an outer Reed-Solomon (RS) code and inner trellis code modulation (TCM). Interleaving between both levels ensures proper de-correlation of the codes, therefore maximizing performance. The inner code is designed to offer a BER of between 10-3 and 10-4 in the receiver, which is reduced to about 10-9 by the outer code. These and other details of channel coding are shown in Figure 2 and Figure 3 for a single carrier per channel (SCPC) ATV system and a multi-carrier modulation (MCM) ATV system using trellis coded orthogonal frequency division multiplexing (TCOFDM), respectively. INNER FEC ENCODER (Reed-Solomon) OUTER INTERLEAVER INNER FEC ENCODER (TC-QAM) INNER INTERLEAVER MODULATOR CHANNEL DEMODULATOR ADAPTIVE EQUALIZER INNER DE-INTERLEAVER INNER FEC DECODER (Viterbi) OUTER DE-INTERLEAVER OUTER DECODER (RS) CARRIER RECOVERY Figure 2: Single carrier ATV transmission system. INNER FEC ENCODER (Reed-Solomon) TIME INTERLEAVER INNER FEC ENCODER (TC-QAM) FREQUENCY INTERLEAVER FFT -1 GUARD INTERVAL INSERTION MODULATOR CHANNEL DEMODULATOR GUARD INTERVAL FFT REMOVAL FREQUENCY DE-INTERLEAVER NORMALIZER - Channel Response Estimation - Interference Estimation INNER FEC DECODER (Viterbi) TIME DE-INTERLEAVER OUTER FEC DECODER (RS) Figure 3: TCOFDM ATV transmission system. For this study, the assumption is made that an HDTV service requires 19.5 Mbit/s of information after source encoding, and a 10-9 BER at the receiver after source decoding. The systems considered for this study are described in more detail in Section 4 and Section 5. 2

7 3 Channel Coding Two different channel coding techniques are considered in this study: conventional single carrier per channel (SCPC) M-QAM modulation, where the signal bandwidth is occupied by one full width carrier; and multi-carrier modulation implemented with TCOFDM, where the signal bandwidth is occupied by multiple fractional width carriers located side by side spectrally. 3.1 Single Carrier Per Channel The performance of conventional single-carrier modulation schemes is generally well known, and has been recently studied in an ATV context as part of the process undertaken by the Advisory Committee on Advanced Television Service (ACATS) of the Federal Communications Commission (FCC) in the United States of America. Most ATV systems proposed to the ACATS use either M-VSB or M-QAM as a singlecarrier scheme. For this study, only coded 32-QAM is considered, though some results for coded 16-QAM are presented for comparison. 3.2 Multiple Carrier Per Channel Unfortunately, the state of knowledge on MCM is very limited, though it has been the subject of important work over the last 30 years or so at Bell Telephone Laboratories [1 ], [2 ], [3 ], [4 ], [5 ], [6 ], and has been applied in military HF communication systems [7 ] such as the KINEPLEX system from Collins Radio Co. (USA) [8 ], [9 ], the ANDEFT/SC-320 system from General Dynamics Corp. (USA) [10 ], and the AN/GSC-10 KATHRYN system from General Atronics Corp. (USA) [11 ]. MCM has since been considered for other interesting applications such as high-speed voice-band data communication modems at NEC Corp. (Japan) [12 ], [13 ], where it serves to alleviate the degradations caused by an impulsive noise environment, and digital radio broadcasting at the CCETT (France) [14 ], [15 ], [16 ], the Institute for Communications Technology (Germany) [17 ], [18 ], the Communications Research Centre [19 ], work that is now leading to CCIR standardization [20 ]. With the arrival of affordable high-speed VLSI technology to implement the FFT and the adjuncts of interleaving and coding, OFDM can now be considered a serious contender for single carrier schemes [21 ]. Of more interest for this study, OFDM is currently being studied for digital terrestrial television broadcast (DTTB), as shown by recent work in various research groups, including the CCETT (France) [22 ], [23 ], Thomson-CSF/LER (France) [24 ], NTL (UK) [25 ], [26 ], [27 ], NHK (Japan) [28 ], HD-DIVINE (Scandinavia) [29 ], [30 ], and the Communications Research Centre. A description of OFDM is not offered in this report; the following papers are suggested as excellent sources of such information: [15], [21], [22] and [25]. 3.3 Performance Comparison of Both Techniques From theory, the performance of COFDM in noise is equivalent to that of SCPC modulation, assuming the same code and constellation are used in both systems, and compensating for the power loss due to the guard interval on one hand, and the cutoff filter on the other. 3

8 It is not as straightforward with frequency-selective interference such as co-channel NTSC or multipath. With SCPC modulation, the effect of such channel impairments can be minimized with an adaptive equalizer [38]. For COFDM, the trellis code and Viterbi decoder can be optimally designed to deal effectively with such impairments, as long as proper frequency interleaving is used [15] [22]. It should be noted that an OFDM system without some form of coding will result in very poor performance in the presence of frequency-selective impairments. Consequently, this study is only concerned with coded OFDM (COFDM), more specifically with trellis-coded OFDM (TCOFDM). Using a trellis code to link the interleaved carriers in the OFDM signal distributes the interfered symbols in a random fashion and offers considerable performance improvement. Whenever OFDM is mentioned in this report, it refers to uncoded OFDM. 4 Simulation Procedure The ATV systems implemented in the computer simulation work, including SCPC modulation using 32QAM and TCM, and TCOFDM using QAM, are shown in Figure 4 and Figure 5 respectively. PRBS INNER FEC ENCODER (TC-32QAM) INNER INTERLEAVER MODULATOR CHANNEL DEMODULATOR ADAPTIVE EQUALIZER INNER DE-INTERLEAVER INNER FEC DECODER (Viterbi) BIT ERROR COUNTER Figure 4: TC32QAM ATV simulation setup. A pseudo-random binary sequence (PRBS) is fed to the transmitter to generate the desired signal. The resulting coded, interleaved and modulated baseband signal goes through a simulated channel where impairments such as noise, interference, and multipath are injected. The resulting received signal is processed through demodulation, de-interleaving and decoding. An adaptive equalizer is used in the TC32QAM system. The (de)modulation blocks differ for both systems. The resulting decoded binary stream is analyzed to determine the performance. For this study, the outer RS code is not included in the simulation work, as computer simulations for BER values in the 10-9 range would be too time consuming. Suitable RS codes are readily available to reduce the bit error rate (BER) at the receiver as required for adequate protection of the source coded data. 4

9 PRBS INNER FEC ENCODER (TC-QAM) FREQUENCY INTERLEAVER FFT -1 GUARD INTERVAL INSERTION CHANNEL GUARD INTERVAL REMOVAL FFT FREQUENCY DE-INTERLEAVER INNER FEC DECODER (Viterbi) BIT ERROR COUNTER Figure 5: TCOFDM ATV simulation setup. Also, this study has not considered impairment from adjacent channel interference, because performance is more dependant on the design of the out-of-band rejection filters than it is on the choice of modulation and coding. 4.1 TC32QAM Transmitter/Receiver The single carrier system implemented for the simulation work uses a trellis coded 32QAM (TC32QAM) scheme with the characteristics presented in Table I, based on the Digicipher system proposed to the ACATS. The 4.9 Msymbol/s rate offers slightly more than the 19.5 Mbit/s throughput targeted for this study. The modulator block includes a cutoff filter with a 21% roll-off factor. Symbol rate ~ 4.9 Msymbol/s Roll-off factor 21% Constellation Interleaving depth Code Equalization CROSS-32QAM 32 symbols Rate 4/5 trellis code optimized for noise Adaptive 256 taps T/2 fractionally spaced Table I: TC32QAM system simulation parameters. The rate-4/5 8-state trellis code, taken from Ungerboeck [31 ], is designed for use in a noise channel. The encoder output is mapped to a CROSS-32QAM constellation, and the resulting symbols are interleaved over a 32 symbol depth. In the receiver, a T/2 fractionally spaced equalizer with 256 adaptive taps is used. 4.2 TCOFDM Transmitter/Receiver Two different TCOFDM configurations are considered for the simulation work. The first configuration, with parameters presented in Table II, offers a sufficient throughput for the study target of 19.5 Mbit/s. 5

10 Symbol rate Useful symbol duration Guard interval duration ~ 4.8 Msymbol/s 128 ms 20 ms Number of sub-channels 749 Constellation Interleaving depth Code Cross-32QAM 28 symbols Rate 4/5 trellis code optimized for noise Table II: TCOFDM 32QAM system simulation parameters. The useful symbol has a 128 ms duration chosen to minimize interference to and from NTSC (see Section 10). The resulting carrier separation is khz, with 768 carriers within the 6 MHz bandwidth. By forcing to zero 19 carriers at the band edges, adjacent channel interference is reduced, and 749 useful carriers remain. A guard interval duration of 20 ms serves to reduce inter-symbol interference from passive echoes due to multipath, and from active echoes generated by on-frequency repeaters, allowing for constructive power addition of these echoes. The rate-4/5 8-state trellis code used in the SCPC system is also used here with a CROSS-32QAM constellation. The resulting symbols are interleaved over a 28 symbol depth, ensuring statistical redistribution of burst errors occurring over adjacent subchannels. A second configuration was also studied. System parameters are presented in Table III. A trellis code optimized for wideband frequency-selective impairment, such as NTSC interference or multipath distortion, is used with a slightly different mapping approach. The trellis code is optimized for narrow-band flat-fading impairments, since the frequency-selective impairment over 6 MHz is relatively flat within one OFDM subchannel, with a bandwidth on the order of a few khz. The previous configuration is expected to achieve poorer BER performance in the presence of such frequency-selective impairments. To map the encoded data onto a constellation, the conventional method is to code/map both the in-phase (I) and quadrature (Q) components of the signal dependently as a QAM constellation. Here, the I and Q components are coded/mapped independently using a one-dimensional AM constellation on each of I and Q; the two orthogonal AM constellations are then combined to achieve an effective two-dimensional QAM constellation. 6

11 Symbol rate Useful symbol duration Guard interval duration ~ 4.9 Msymbol/s 127 ms 21 ms Number of sub-channels 730 Constellation (4AM) 2 Interleaving depth Code 28 symbols Rate 1/2 trellis code optimized for flat fading Table III: TCOFDM (4AM) 2 system simulation parameters. The main advantage of such a technique is the reduced Viterbi decoder complexity. And though in noise the performance will not be optimal, a performance "roughly similar to [that of] the more elaborate twodimensional (2D) Ungerboeck codes" is achieved [32 ]. Furthermore, using optimally designed codes in a flat fading channel, such an approach seems to offer slightly better performance than the conventional twodimensional technique [33 ]. This second configuration has a useful symbol duration of 127 ms with a guard interval duration of 21 ms. This results in a carrier separation of khz, with 762 carriers within 6 MHz. With the removal of 32 carriers at the band edges to deal with adjacent channel interference, 730 useful subchannels remain. The rate-1/2 64-state trellis code used in this system is designed for use in a flat fading channel with a (4AM) 2 constellation [34 ]. On each of I and Q, one bit is encoded with rate-1/2 and mapped to 4AM, resulting in an effective 16QAM constellation with a 2 bit/s/hz capacity. The symbols are interleaved at a depth of 28. The drawback of this second configuration is the low throughput: only about half of the target 19.5 Mbit/s throughput is achieved. A higher rate code, such as 2/3 applied to (8AM) 2, is thus required. This is discussed further in Section 10. To help determine TCOFDM's robustness to the presence of NTSC interference, a system that assumes perfect channel response estimation, perfect interference estimation, and perfect interleaving, is implemented. The trellis code, which may be the most critical block for such impairment, can then be studied independent of other system weaknesses. The channel response information, obtained at the receiver by analyzing the received signal associated with transmitted pilot tones, serves to normalize the received data symbols on a subchannel basis, to compensate for attenuation and phase shifts. The interference information, obtained by analyzing the received signal associated with a transmitted null symbol, provides the Viterbi decoder with reliability information on the data symbols on a subchannel basis. Finally, interleaving serves to redistribute error bursts so they appear as periodic single errors. 7

12 With the above perfect implementations, single erasures are introduced at regular intervals at the receiver to simulate binary weighting in the Viterbi decoder. This simulates a situation with actual NTSC interference where the decoder erases all interfered carriers using the perfect estimations, and redistributes evenly the erasures with the help of the perfect interleaver. Of course, multi-level weighting in the Viterbi decoder is expected to offer somewhat better performance, but always at the expense of increased complexity. 4.3 Transmission Channel The transmission channel is shown in Figure 6, with impairments including noise, co-channel analogue NTSC interference, and co-channel ATV interference WHITE GAUSSIAN NOISE CO-CHANNEL NTSC 75% COLOUR BARS GENERATOR W/STEREO AUDIO CO-CHANNEL ATV Figure 6: ATV transmission channel. The noise used in the simulations is modeled as additive white Gaussian noise. The co-channel NTSC interference signal comes from a 75% color bars software generator developed as part of this study. The signal also contains a sound carrier modulated by stereophonic audio according to the BTSC specification, using pink noise as the source. The co-channel ATV interference signal is generated the same way as the desired ATV signal, using a similar transmitter differing only in the seed used by the pseudo-random binary sequence generator; this ensures that the desired and undesired transmitted signals are uncorrelated. 5 Laboratory Test Procedure The laboratory test setup shown in Figure 7 consists of a waveform generator programmed to produce an ATV signal, which is then added at RF (channel 12) to an NTSC signal whose source includes both fixed test patterns and live video signals. The NTSC picture with resulting interference is then viewed by an expert viewer at a distance of 5H on a high-end consumer television receiver. The same viewer was used throughout the tests. The desired to undesired signal ratio is varied to determine the threshold of visibility (TOV) and threshold of audibility (TOA); that is, the points at which the interference starts to be perceptible in the video and audio programme respectively. 8

13 NTSC TRANSMITTER OFDM WAVEFORM GENERATOR RF MODULATOR + NTSC RECEIVER 5.1 Single Carrier Interference Figure 7: NTSC laboratory test setup. The SCPC signal generated is an uncoded 32QAM with a symbol rate of about 4.9 Msymbol/s. The absence of coding here is not important since it does not affect the 32QAM constellation, to which NTSC is sensitive. Interference from an uncoded 16QAM single carrier system with the same symbol rate as the uncoded 32QAM system is also considered for comparison. 5.2 OFDM Interference The signal generated is an uncoded OFDM 16QAM with parameters in Table IV. The absence of coding has again no effect on the performance of NTSC. Symbol rate Useful symbol duration Guard interval duration 5.15 Msymbol/s 100 ms 0 / 20 ms Number of sub-channels 512 Constellation Spectral holes: A " " B " " C 16QAM None 21 carriers 41 carriers Table IV: OFDM system laboratory test parameters. Two values of guard interval duration, namely 0 and 20 ms, are considered to study its effect on the TOV and TOA. Three spectral shaping configurations are considered to study the effects on TOV and TOA of removing carriers from the OFDM signal. The configuration labeled A has no hole; that is, no carriers are removed. The other two configurations have holes of differing widths centered at the NTSC vision carrier. The holes are created by forcing to zero several adjacent carriers in the OFDM spectrum. The B hole is 21 carriers wide, which is more than 200 khz, whereas the C hole is 41 carriers wide, or more than 400 khz. 9

14 6 Results - TC32QAM Performance results from computer simulation of the TC32QAM single carrier per channel system are presented and analyzed. A description of this system was given in Section Performance Results BER performance results obtained from the computer simulations of the TC32QAM single carrier system are presented for various impairments, including noise, interference and multipath. a) AWG Noise For a channel impaired by additive white Gaussian noise, the BER performance of TC32QAM is presented in Figure 8, which also includes performance results of the TCOFDM 32QAM configuration for comparison. Results for TCOFDM (4AM) 2 are also presented, but are discussed later. b) NTSC Interference For a channel suffering from analogue NTSC interference, the BER performance of TC32QAM is shown in Figure 9 with and without adaptive equalization. The SNR is set at 40 db so the effect of noise is negligible. The BER performance of TC32QAM when both noise and NTSC interference are present is shown in Figure 10, for different values of C/I ranging from 8 to 12 db. Figure 11 gives the performance threshold of TC32QAM for combined impairments of noise and NTSC interference, with and without equalization. c) ATV Interference For a channel suffering from co-channel ATV interference, the BER performance of TC32QAM is presented in Figure 12 with and without adaptive equalization and time and frequency offsets. Offset in time, with the interfering signal offset by a half symbol, and offset in frequency, with the interfering signal offset by 10 khz, are considered. d) Multipath Four multipath distortion models described in Table V are considered for the simulation work; these models were developed for NTSC ghost canceling tests [35 ]. The BER performance is shown in the table for a given SNR near threshold for two different values of adaptive equalization step size D. Performance threshold of TC32QAM with combined impairments of noise and NTSC interference, using adaptive equalization with step size D = 100 x 10-6 for fast convergence, is presented in Figure 13 for the different multipath models of interest. The effect of the adaptive equalization step size on the performance threshold of TC32QAM in combined impairments and multipath is considered in Figure 14, where multipath model D is used. 10

15 BER 1.0E E E E E E-06 TCOFDM (4AM) TCOFDM 32QAM TC32QAM SCM 2 1.0E SNR (db) Figure 8: Performance of ATV in noise. BER 1.0E E E E E E-06 SNR = 40 db w/eq w/o EQ 1.0E C/I (db) Figure 9: TC32QAM performance in NTSC interference. 11

16 BER 1.0E E E E-04 C/I (db) E SNR (db) Figure 10: TC32QAM performance in combined noise and co-channel NTSC interference impairments. C/I (db) w/o EQ w/eq Threshold SNR (db) Figure 11: TC32QAM performance threshold in combined noise and co-channel NTSC interference impairments. 12

17 BER 1.0E E E E E E E E-08 w/o EQ w/o offset w/o EQ w/time offset w/o EQ w/freq offset w/eq w/o offset w/eq w/time offset w/eq w/freq offset 1.0E C/I (db) Figure 12: TC32QAM performance in ATV interference. 6.2 Analysis of Results a) AWG Noise When the TC32QAM signal is impaired by additive white Gaussian noise, a BER performance on the order of 10-3 (which is close to the TOV in noise impairment) requires about 16.3 db of signal to noise ratio (SNR), as indicated by results from Figure 8. This is comparable to the results published as part of the FCC ACATS process, which suggests 16 db for TOV. b) NTSC Interference With the system suffering from co-channel analogue NTSC interference, a BER performance of 10-4 (which is close to the TOV for this type of impairment) requires about 7.8 db of C/I without equalization, according to Figure 9. With adaptive equalization, this reduces to about 1 db, for a gain of almost 7 db. The published FCC ACATS test results suggest a C/I between 7 and 8 db for the TOV. Of course, this simulation result assumes perfect coherent demodulation. Typically, a phase lock loop (PLL) circuit is used in a real implementation to recover the carrier. However, in the presence of strong NTSC co-channel interference, the receiver carrier regeneration might pose a problem. It has been shown [36 ] that a PLL circuit can recover the carrier only when the C/I is 6 db or more. When the system is in the presence of both noise and NTSC interference, Figure 10 shows the performance reduces for stronger noise or interference, as is expected. With the C/I ranging from 8 to 12 db, the SNR must range from 23 down to 18 respectively to achieve a BER performance of around These results 13

18 suggest that the system can still offer good performance with reasonable amounts of combined noise and interference impairments. The performance threshold of TC32QAM for combined impairments of noise and NTSC interference, shown in Figure 11, reaches a performance floor at C/I = 7 db without equalization, which is reduced considerably when equalization is added. Equalization provides a SNR improvement of more than 9 db (from over 30 db down to 21 db) with a C/I value of 6 db, and a C/I improvement of 3 db (from 9 db to 6 db) for a SNR value of 21 db. However, for SNR values below 19 db, the performance with equalization shows degradation. These results suggest that in combined noise and interference impairments, equalization is very useful for reducing the effect of these impairments. c) ATV Interference With co-channel ATV interference, performance results from Figure 12 show that equalization offers a gain of about 1 db at a BER of Multipath Model Delay (ms) Attenuation (db) Phase ( ) Total D/U (db) D = 25.0e-6 D = 100.0e-6 SNR (db) BER SNR (db) BER A (Typical) e e-4 B (Typical) e e-5 C (Microreflections) e e-4 D (Strong microreflections) e e-4 Table V: TC32QAM performance in multipath with adaptive equalizer step size D. 14

19 C/I (db) Multipath Model A B C D SNR (db) Figure 13: TC32QAM performance threshold in combined noise and NTSC interference and multipath using a large equalizer step size. C/I (db) EQ Step Size Large Small SNR (db) Figure 14: TC32QAM performance threshold in combined noise and co-channel NTSC interference impairments and multipath model D. 15

20 For this BER value of 10-3, offset in time of a half symbol provides a slight performance advantage with and without equalization. With equalization, this gain is in the order of 0.5 db. This is expected from the cyclo-stationary property of the QAM modulated signal [37 ]. When a time offset of between zero and a half symbol is considered, the performance falls between these two extremes, so the maximum advantage at the BER value considered is 0.5 db. When a frequency offset of 10 Hz is considered instead of a time offset, the advantage in performance appears more important for BER values smaller than 10-4, which correspond to high SNR values. For a BER value of around 10-3, TC32QAM in ATV interference with no offset requires 15.6 db of C/I, which is 0.7 db better than in noise. This is reduced to 14.8 db with equalization. These are comparable to published FCC ACATS test results, which suggest a SNR between 14.5 and 15.1 db for the TOV, with zero offset. The TC32QAM spectrum is somewhat noise-like, which explains the small difference between performance results in noise and in ATV interference. d) Multipath With TC32QAM suffering from multipath distortion, Table V shows that for small adaptive equalizer step size D (25 x 10-6 ), the multipath model D, characterized by strong micro-reflections, offers the best BER performance, where the SNR is 16 db, though it has the worst D/U ratio, at 4.6 db. This is because multipath model D has micro-reflections with relatively short delays and low D/U's. With a larger step size D (100 x 10-6 ), multipath model D gives the highest SNR threshold. For combined impairments of noise and NTSC interference in the different multipath models, Figure 13 shows that multipath model D offers the worst performance threshold. A C/I floor seems to exist at 14 db. For the results in Figure 13, the adaptive equalization process used the larger step size D (100 x 10-6 ) for fast convergence. This was a compromise between speed of convergence and performance in noise (residual errors): a larger step size provides faster convergence and faster response to airplane flutter, but it also increases the residual errors [38 ]. As shown in Figure 11, with a C/I value of 30 db, the noise enhancement due to the fast convergence adaptive equalizer is about 1.5 db. The best approach might be to use adaptive step sizes. The performance threshold for the other multipath models, A, B, and C, are similar relative to that for model D, as shown in Figure 13, and follow closely the performance threshold in combined impairments with no multipath from Figure 11. Model A, which contains the least echoes, offers the best performance in combined impairments and multipath. The adaptive equalizer is sensitive to the type of multipath, but seems to achieve good results for typical cases. The effect of the step size is shown in Figure 14 for TC32QAM in combined impairments of noise and cochannel NTSC interference, with multipath model D. The performance gain for the shorter step size is significant, but its disadvantage is the slow convergence. 16

21 Simulation results have shown that adaptive equalization can converge on a single echo for value of D/U as low as 3 db. 7 Results - TCOFDM Performance results from simulation of the two TCOFDM configurations are presented and analyzed. A description of these systems was presented in Section Performance Results BER performance simulation results are presented for the two TCOFDM configurations in a channel suffering from noise and interference. a) AWG Noise Figure 8 presented BER performance results for both the (4AM) 2 and the 32QAM configurations of the TCOFDM system. For comparison with uncoded 2 bit/s/hz modulation, the performance of (4AM) 2 is shown in Figure 15 with the theoretical performance of uncoded QPSK. BER 1.0E E-02 QPSK TCOFDM (4AM) 2 1.0E E E Eb/No (db) Figure 15: Performance of TCOFDM (4AM) 2 in noise. 17

22 b) NTSC Interference As explained in Section 4.2, the robustness of the TCOFDM code to frequency-selective interference, such as analogue NTSC, can be partly determined by observing the performance of the given code in the presence of erasures. The TCOFDM 32QAM system uses a code designed for noise, and performance results in interference are poor, so very little time was spent studying this code in interference. The performance of TCOFDM (4AM) 2 in erasure (the trivial binary case of weighting) is given in Figure 16, where two different implementations of the code are considered, both generated using the same paritycheck coefficients: a systematic structure with feedback, and a non-systematic structure with feed-forward. Combined impairments of erasure and noise are also considered, with results shown in Figure 17. c) ATV Interference Performance results of TCOFDM interfering with itself are not given graphically, but simulations have shown that in the BER range of interest, namely from 10-3 to 10-4, the performance is about 1 db better in ATV interference than it is in noise. 7.2 Analysis of Results a) AWG Noise Results from Figure 8 indicate that the TCOFDM 32QAM system achieves a BER performance of 10-3 with a SNR value of about 15.5 db. This is about 0.8 db better than results for TC32QAM, and is attributable to the different trellis codes, the guard interval versus the cutoff filter, and other simulation implementation differences. If the same code was used and implementation differences were considered, both single carrier and TCOFDM would achieve the same performance in noise. From Figure 8, TCOFDM (4AM) 2 achieves a 10-3 BER performance with a SNR of 8.8 db, or equivalently 5.8 db E b /N o, as presented in Figure 15. For this error rate, the coding gain is about 0.7 db, when comparing with theoretical uncoded QPSK, which offers the same data throughput. The use of trellis coding offers less than one db of performance gain compared to an uncoded system in noise, but its importance is much greater in frequency-selective interference, as is shown next. b) NTSC Interference In a noiseless channel, Figure 16 shows that the rate 1/2 trellis code designed for use in a flat fading channel can withstand periodic data erasure of as many as 10% of the received data symbols with no apparent effect on the BER performance. Periodic data erasure here means that, for example, with 10% erasure, every received tenth data symbol is forced to zero, in effect giving the Viterbi decoder no information on that particular symbol. 18

23 BER 1.0E E E E E E E-06 Encoder Structure Systematic Non-Systematic 1.0E Erasure (%) Figure 16: TCOFDM (4AM) 2 code performance in the presence of erasure. BER 1.0E E E E-04 Erasure (%) Zero E Eb/No (db) Figure 17: TCOFDM (4AM) 2 code performance in combined noise and erasure. 19

24 Below this 10% threshold, the Viterbi decoder can recover the data without error; however, if more than 10% of symbols are weighted to zero, the decoder gets lost and generates considerable errors. Trellis codes designed for use in noisy channels were tested with periodic erasure, and were found to degenerate very quickly in performance with erasure of less than 1% of symbols. The two code structures considered are generated from the same parity-check coefficients, but the systematic structure with feedback is prone to error propagation because of the feedback loop, and so offers slightly lower performance at higher BER compared to the non-systematic structure with feedforward. Results for the rate-1/2 code with erasure, in a channel suffering from co-channel NTSC interference with no noise, indicate that when 10% or less of the TCOFDM carriers are erased (binary weighted) to account for interference, then relatively error-free performance is achieved. This of course assumes erasure decisions using perfect interference estimation, and perfect interleaving of the symbols. Similar results are expected when non-binary weighting is used in the presence of stronger interference. With noise added to the TCOFDM system with erasure, Figure 17 shows that the performance slowly deteriorates as the erasure level increases up to 10%, where the signal power required for a 10-4 BER performance is about 6 db higher than with no erasure (0%). Beyond the 10% level, performance quickly degrades: with 12.5% erasure, a floor in the BER performance appears at 2 x The assumption of perfect interleaving can be closely implemented in practice, but the critical part is perfect interference estimation. This is discussed further in Section 10. These results apply to the rate 1/2 trellis code selected for use with (4AM) 2. However, a suitably designed higher rate trellis code is required to achieve the information throughput required for an ATV service. This is discussed further in Section 10. c) ATV Interference For a BER performance somewhere between 10-3 and 10-4, TCOFDM with ATV interference requires about 1 db more signal power than TCOFDM with noise. This similarity was expected because TCOFDM's spectrum is somewhat noise-like. 8 Results - NTSC NTSC performance results from laboratory tests with interference from uncoded single carrier 32QAM and 16QAM, and uncoded OFDM 16QAM systems, are presented and analyzed. A description of these systems was given in Section 5. 20

25 8.1 Performance Results Laboratory test results of NTSC performance, measured as the threshold of visibility (TOV) and the threshold of audibility (TOA), are given for interference from the various ATV systems considered. a) 16QAM and 32QAM Interference The C/I required to achieve the TOV in NTSC when being impaired by a single interference, such as noise, single carrier 16QAM or 32QAM, or NTSC 75% color bars, is presented in Table VI. The C/I value presented is an average for three different desired NTSC test pattern signals: 50 IRE multi-burst, 5 step staircase, and 100% color bars. The performance difference between these signals is within 1 db for all laboratory tests. Noise 16QAM 32QAM NTSC Colour Bars 46 db 47 db 48 db 50 db Table VI: Average C/I at TOV for single carrier QAM interference into NTSC. Similar tests were also performed to determine the C/I required to achieve the TOA in NTSC, with results presented in Table VII. For this case, however, only one of the three NTSC test patterns, namely the NTSC 100% color bars signal, is used as the desired signal, as the content of the video signal did not seem to have an effect on the audio signal. Noise 16QAM 32QAM NTSC Colour Bars 25 db 14 db 15 db 17 db Table VII: Average C/I at TOA for single carrier QAM interference into NTSC. b) OFDM 16QAM Interference The C/I required to achieve the TOV in NTSC with undesired OFDM 16QAM interference signals is presented in Table VIII, where four configurations of the undesired signal are given. Again, an average C/I is given because tests are done with three different desired NTSC signals. No Hole Small Hole Large Hole Guard Interval 48 db 47 db 46 db 49 db Table VIII: Average C/I at TOV for OFDM interference into NTSC. The four OFDM 16QAM configurations were described in Section 5. The first three do not use a guard interval, but have differing hole widths for spectral shaping, varying from no hole up to a large hole of 41 carriers (over 400 khz) centered on the analogue NTSC vision carrier. The fourth case has no holes but uses a 20 ms guard interval. 21

26 The C/I required for the desired analogue NTSC signal to achieve the TOA when being impaired by undesired OFDM 16QAM signals is presented in Table IX. As with single carrier interference earlier, the only desired NTSC test signal used is 100% color bars. No Hole Small Hole Large Hole Guard Interval 17 db 17 db 17 db 17 db 8.2 Analysis of Results a) 16QAM and 32QAM Interference Table IX: C/I at TOA for OFDM interference into NTSC. From Table VI, the effect of an undesired uncoded 32QAM single carrier signal is 2 db less offensive than undesired NTSC, but it is 2 db more offensive than noise, when considering C/I required to achieve TOV. Due to the modulation and demodulation of the desired NTSC signal, the 46 db C/I required in noise is somewhat less than the 53 db C/I required with a higher quality studio NTSC signal. For the effect of interference on the NTSC signal's audio data, Table VII reveals that uncoded 32QAM is again 2 db less offensive than NTSC. However, the noise is now the worst offender, requiring 10 db more C/I than with uncoded 32QAM; this is expected due to the roll-off at the channel edge. b) OFDM 16QAM Interference Comparing results from Table VI and Table VIII shows that TOV is achieved with a C/I of 48 db with interference from either OFDM 16QAM (Table VIII) without holes, or uncoded 32QAM single carrier. Adding holes to the OFDM signal offers some performance improvement, though small: 1 db for a small hole and 2 db for a large hole. This can hardly justify the use of spectral holes when considering the resulting decrease in channel throughput. The presence of the guard interval reduces performance by 1 db, which is explainable by the presence of ripple in the spectrum of the OFDM 16QAM signal when the guard interval is present. Results in Table VII and Table IX reveal that TOA is achieved with a C/I of 17 db with interference from OFDM 16QAM (Table IX) without holes, compared with 15 db for uncoded 32QAM single carrier. Adding holes to the OFDM signal offers no TOA performance improvement. The guard interval is also seen to have no measurable impact on the NTSC audio carrier. The 30 db C/I difference between TOV and TOA for OFDM interference suggests that the NTSC video information will collapse long before the audio carrier becomes affected, so the TOA is not a concern. 22

27 9 Comparing Conventional and TCOFDM Modulations When comparing these two modulation schemes, there are two main issues of interest: performance in different impairments, and complexity, which directly affects the cost. These are discussed here. 9.1 Noise into ATV From both theory and practice, single and multi-carrier techniques achieve the same BER performance in noise, assuming similar configurations, including the same trellis code. The TCOFDM (M-AM) 2 approach has a slight performance disadvantage in noise, which is typically less than 1 db, compared with conventional TCOFDM M-QAM, but it seems to offer considerable complexity reduction in the Viterbi decoder [34]. 9.2 NTSC Interference into ATV Due to the wide-band nature of the single carrier signal, the effect of NTSC interference seems comparable to that of noise, when considering the RMS power of the NTSC signal, instead of the peak power as is typical. In the TCOFDM case, some subchannels can be badly damaged by the NTSC interference, while others may not be affected at all. The effect of NTSC needs to be countered by estimating the level of interference, and using the resulting measure of carrier data reliability to weight the Viterbi decoder metrics associated with the received symbols. In the presence of co-channel analogue NTSC interference, it is not obvious that TCOFDM using the carrier erasure technique can offer considerable performance advantage over single carrier techniques. Though more work is required on this subject, it seems fair to say that a simple implementation of TCOFDM can achieve the same performance as single carrier does. However, more sophisticated receiving techniques, such as the reliability weighting mentioned in Section 10, are expected to offer further performance improvements. Assuming perfect interleaving, perfect interference estimation, and no noise, results from the TCOFDM (4AM) 2 configuration using erasure suggest that error-free performance may be achieved if less than 10% of the TCOFDM carriers are affected by the interference. If, instead of erasure, a multi-level reliability weighting is applied to the metric in the Viterbi decoder, it is likely that the decoder could afford to have more than 10% of the carriers affected somewhat. Given that the nature of the NTSC spectrum is well know, and it is somewhat stationary, especially at the peaks that are the biggest sources of interference, it is expected that careful design may lead to good practical implementations of the perfect interleaving and perfect interference estimation assumed in this study. Spectral shaping of the TCOFDM signal, by putting some carriers to zero, is seen by some as one of the advantages of multi-carrier modulation. In fact, single carrier modulation can also take advantage of such shaping by notching out the spectrum. However, as explained in Section 10, spectral shaping is not the best design alternative; optimal code design is preferable. 23

28 9.3 Multipath into ATV When in the presence of multipath, which cause frequency-selective interference, performance of a single carrier scheme is limited by the performance of the equalizer. This is not so for TCOFDM. Adaptive equalization resolves echoes and selects the strongest signal, ignoring the information in all others. For the adaptive process to converge on a single echo, the ratio between desired and undesired signals (D/U) must be at least 3 to 6 db. With TCOFDM, all echoes are used, and within the guard interval, the power of all echoes tend to add on average, with a resulting signal stronger than the main received signal without echoes. In a situation where echoes are of equal amplitude, adaptive equalization would not converge properly, but TCOFDM has no problem. The guard interval duration in TCOFDM can be selected to accommodate different echo delays. TCOFDM parameters can be properly selected to minimize the decrease in data capacity that would normally result from an increase in the guard interval duration. 9.4 ATV Interference into NTSC Both single carrier and TCOFDM schemes have an impact similar to noise on the performance of NTSC. One or two dbs can be gained by removing the most critical carriers in TCOFDM, but at the expense of a decreased channel capacity. 9.5 Complexity The issue of complexity has been considered by some as an important decision factor for one or the other modulation scheme, but the most expensive part of an ATV receiver is the tube, followed by the chassis and power supply. Therefore, the cost of electronics is not very critical. Besides, the complexity of the channel decoder hardware is much less than that of the source/video decoder hardware. When comparing the complexity of single carrier and TCOFDM ATV systems, only the blocks that are not common to both implementations should be considered. This means that the source encoder and decoder, the outer RS encoder and decoder, the outer and inner interleavers and de-interleavers and associated memory, the inner TCM encoder, and the Viterbi decoder, all should be excluded from the comparison. This of course assumes the same source encoding/decoding, RS code, and interleaving depth in both systems, which is quite possible, and the same trellis code and mapping scheme. Concerning this last item, it has been mentioned already in this study that (2M-AM) 2 seems to offer better performance than conventional M-QAM, and offers reduced Viterbi decoder complexity. This is to TCOFDM's advantage. From Figure 2 and Figure 3, blocks that are not common to both systems include the adaptive equalizer in the single system, and the FFT/FFT -1, guard interval insertion/removal, normalizer, channel response estimator (memory), and interference estimator (memory). The modulator and demodulator blocks may also be different for both systems. With the recent development of ghost canceling hardware, among other advancements, single chip adaptive equalizers are becoming more of a reality, and will be more affordable with time. 24