W.F. Schreiber's Conclusions from the OFDM/COFDM Meeting at MIT, October 1992
Summary
Professor W.F. Schreiber summarizes findings from a two-day MIT workshop where European and Canadian researchers presented COFDM field test results, concluding the modulation scheme offers clear advantages over single-carrier systems in coverage, multipath performance, and single-frequency network support. The report notes COFDM's equalizer is simpler to implement than SCM equivalents, and that single-frequency networks could reduce required broadcast channels from 68 to approximately 20.
"From the presentations and discussions, a conclusion can be drawn that COFDM has important advantages in all these operational characteristics and is no harder -- perhaps easier -- to implement. Although more work is required to define the performance precisely and to optimize the design of SFNs, there appears to be no major drawback that might offset the advantages of COFDM".
Report on OFDM Meeting at MIT 26-27 October 1992
Conclusions of W.F.Schreiber
The opinions expressed herein are those of the author only.
Abstract
Representatives from all of the European and Canadian projects on coded orthogonal frequency division multiplex (COFDM) met to present their results and to exchange views. Work in France, UK, Germany, Scandinavia, and Canada was discussed, including the results of field tests. Coverage area, interference performance, multipath performance, antenna requirements, the ability to support single-frequency networks (SFNs), and the complexity of implementation, were discussed. Comparisons were made with single-carrier modulation (SCM), although no SCM proponents attended. SFN is attractive because it has the potential of reducing from 68 to about 20 the number of channels needed to provide today's amount of program choice to viewers.
From the presentations and discussions, a conclusion can be drawn that COFDM has important advantages in all these operational characteristics and is no harder -- perhaps easier -- to implement. Although more work is required to define the performance precisely and to optimize the design of SFNs, there appears to be no major drawback that might offset the advantages of COFDM.
1. Introduction
The purpose of the meeting was to bring together representatives of the principal laboratories working on orthogonal frequency-division multiplex (OFDM) and single-frequency networks (SFN) to present their work and to exchange views. This objective was certainly met and, as a result, the current theoretical and practical status of these developments was made more clear. However, there were no system proponents and only one person from the FCC. Thus no one attended who could forcefully present the technical case for single-carrier systems. A number of those present are connected with the ACATS and ATTC activities.
Before the meeting, the attached list of topics had been distributed to the European and Canadian speakers. All of the topics were touched on in the presentations and discussions. More work, both theoretical and experimental, is needed to explore a number of these topics, in particular the comparison with single-carrier modulation. (SCM) In what follows, I discuss performance with respect to a number of separate measures, as they were dealt with at the workshop. However, in a valid system design, these measures are found to be highly interrelated, so that careful tradeoffs are required.
Familiarity with COFDM is assumed in this report.
2. Equalization and SNR
It is widely agreed that the suppression of linear distortion by means of an automatic equalizer in SCM results in exactly the same SNR as the use of subchannel-by-subchannel gain and phase correction in OFDM. The corrected noise spectrum is the inverse of the uncorrected frequency response. However, the OFDM equalizer is very easy to implement, while the SCM equalizer requires a great deal of computation. The former requires a simple measurement, after which the correction parameters are computed in one step. Iteration is not needed and convergence of parameters is not an issue.
For lack of an advocate at the meeting, the limitations on the performance of automatic equalizers were not much discussed. For example, GI reports that its equalizer cannot deal with echoes larger than -6 dB. Since there is, in fact, a proper solution for echoes much larger than this, the GI limitation must be due to the specific algorithm used, and not to a theoretical obstacle. This is very important in the case of single-frequency networks with omnidirectional antennas, where 0-dB echoes may be expected.
3. Error-Rate Performance
It is sometimes said by OFDM proponents that echoes add constructively according to their power. No conclusive evidence that this is the case was presented at the meeting. (Efforts at MIT to demonstrate this behavior theoretically have so far not proved fruitful. However, there was impressive experimental evidence that echoes do combine in a positive manner, which is certainly not the case in SCM. It appears that the benefit of the combination depends on the coding that is used -- the "C'' in COFDM. Uncoded OFDM, as used in the Thomson CSF test at the Sarnoff Laboratories in1990, would not be expected to show this advantage. The 1990 test used equipment provided by Thomson/CSF-LER. In a recent letter to Renn McMann, K. Jonnalagadda of Sarnoff described the results as generally no better than the dual-carrier scheme advanced by ATRC. However, The Thomson representative at the meeting, whose equipment was used in the tests, does not agree with the assessment. He also pointed out that the equipment was OFDM, not COFDM, and would not be expected to show the superior error-rate performance in the presence of multipath. He distributed copies of a press release reporting the successful transmission of 60 Mb/sec in an 8-MHz video channel by the BBC in London using Thomson equipment. This experiment used both planes of polarization to double the normal capacity.)f
In the presence of echoes, the frequency response becomes nonuniform. Therefore, the SNR for some subcarriers is poorer than for others. The reliability of data transmitted on low-SNR subcarriers is low, and this \can be taken into account in the decoding process. COFDM involves scrambling (called interleaving) in time and frequency, so that transmitted data is dispersed at random with respect to subcarriers. Data from less-reliable channels is weighted less than data from more-reliable channels, and this results in a lower overall error rate, after correction, than with equal weighting. CCETT reported the experimental finding that the error rate after correction went down in the presence of echoes. In addition, they presented data from field tests showing that the service area of two transmitters on the same frequency was larger than the union of the service areas of the individual transmitters.
4. Interference to NTSC
Interference is always a two-way phenomenon. A signal that is relatively immune to interference can be transmitted at lower power for the same coverage area, thus interfering less with other signals. Nevertheless, it is useful to consider the specific aspects of signals that determine sensitivity to undesired signals separately from the tendency to interfere with other signals.
A signal such as emitted by the GI system has a nearly uniform spectrum and appears to be random. Its amplitude probability distribution is determined by the QAM. For reasons I must confess not to understand at this moment, this distribution is not uniform, but has more-or-less random peaks that require linearity of the transmitter in excess of the rms power. The OFDM signal is quasi-Gaussian. CCETT estimates that the excess linearity requirement is 3.7 dB for SCM and 5.0 dB for OFDM at a spectral efficiency of 5 bits/cycle.
I do not think that the probability distribution itself has much direct effect on noise visibility. (At least this was the result of an MIT thesis investigation many years ago.) However, the somewhat higher percentage of high peak values in OFDM might cause more interference to a specific kind of video signal. More investigation is needed.
To the extent that the emitted signal is nonrandom, its interference into cochannel NTSC for a given rms power level is increased. For example, switching off carriers in the vicinity of NTSC carriers (something like ATRC's 2-carrier system) increases the interference for a given power level. It seems unlikely that, for equal interference from NTSC, an ATRC 2-carrier system and an OFDM system will be very different with respect to interference to NTSC.
In OFDM, the spectrum is sharply confined because it is the sum of equally spaced subchannel signals, each of which is sharply confined. It does not require channel-limiting filters at the transmitter and has much less tendency to generate adjacent-channel interference than SCM. The Canadian simulation studies show that, as a result, 5.625 MHz of a 6-MHz channel can be used, which probably makes up for the communication loss due to the use of guard intervals in OFDM. (See below.)
5. Interference from NTSC
Switching off carriers, as discussed above, is a very effective method of boosting immunity to cochannel NTSC interference. If the pattern of requency interleaving were identified in the signal and downloaded into the encoder and decoder hardware (a feature not found at present in the OFDM experiments but that would be quite easy to add), then the pattern could be optimized for the particular interference situation faced by each station. When nearby NTSC stations eventually go off the air, the unused carriers could be reclaimed and picture quality improved without hardware changes. In addition, this scheme would allow general upgrading over time in a nondisruptive manner (a feature the FCC has asked for) and would also allow higher quality for compatible ``contribution'' systems.
6. ATV-ATV Interference, Other Interference
The Canadian simulations show that OFDM performance with respect to ATV-ATV interference is quite close to that of SCM. CCETT reports that OFDM has 10-dB better resistance to pulse interference and 4-dB better resistance to sine-wave interference. Its white-noise threshold is very similar to that of SCM.
7. Multipath Performance
OFDM was specifically designed to deal with multipath by dividing the signal into a large number of low-data-rate components whose symbol length is longer than the temporal spread of the echoes. If the integration period at the receiver is correct, all echoes related to a single transmitted pulse arrive within one pulse period and intersymbol interference (ISI) completely disappears.
The timing of the receiver is greatly simplified if a guard interval, itself a little longer than the multipath spread, is added to the transmitted symbol period. The effect of multipath is now simply a gain and phase change for each subchannel. These can easily be measured with the aid of a pilot signal so that the multipath disturbance is removed completely, except for its effect on the SNR. OFDM equalization is much simpler than for SCM, and has no limit as far as echo amplitude is concerned. In SFN operation (see below), 0-dB echoes can be tolerated. As a result, directional antennas are not required, although their use would generally result in higher SNR.
These characteristics of OFDM have been fully borne out in experiments carried out by a number of the organizations represented at the meeting, and no contrary evidence has surfaced. Note that SCM has no equivalent to the use of the guard interval.
Coded OFDM (COFDM) makes use of the echoes to reduce the corrected error rate compared to what it would be with the direct signal only. In SCM, the cancellation of echoes reduces the SNR and therefore increases the error rate. This is a very important difference in favor of COFDM.
8. Antennas
Service area calculations for the digital systems are based on 10 dB antenna gain and 14 dB front-to-back ratio. The antenna gain influences the noise-limited service area while the front-to-back ratio influences the interference-limited service area. The requirement of a high-gain, high-directivity antenna imposes cost and inconvenience on viewers, the acceptability of which is not clear. Current widespread use of "rabbit ears'' in city areas indicates that, at the very least, extensive consumer education will be required.
The main antenna issue is the limited ability of automatic channel equalizers, needed in SCM to deal with multipath, to cope with high-level and/or rapidly-changing echoes. Such echoes are suppressed easily in OFDM, as a result of which omnidirectional antennas proved satisfactory in the earlier tests of digital audio broadcasting to moving vehicles. Except in special circumstances, OFDM permits the use of omnidirectional antennas while SCM does not; this is a major advantage.
Note that, in current practice, the service areas of stations, where they are interference limited, is directly affected by the assumptions about receiving antennas. If better front-to-back ratio could be counted on, then NTSC stations on the same channel could be moved closer together, even without any new technical developments.
9. Single-Frequency Networks
A SFN is composed of an array of low-power transmitters operating on the same frequency. (This arrangement is sometimes referred to as ``distributed transmission.'') The multipath immunity of the receiver is relied upon to achieve satisfactory operation when a number of strong ghost-like signals are received. SFNs permit very high spectrum efficiency, since the service area of each station is precisely defined by the location of transmitters. Contiguous regions can be used by different stations on the same channel, so that there are no taboo channels due to cochannel interference. The "no-man's land'' between stations, in which reception is problematic, is about one cell-diameter in width. In this area, directional antennas are required, which would enable reception of either station, as desired) If SFNs were used in this manner, the total number of channels that would have to be allocated for TV would be no larger than the number of different programs to be made available in each city or market.
The SFN principle has been demonstrated at least in France and Canada for DAB with COFDM. The ability to support SFNs and their attendant high spectrum efficiency is a very important characteristic of COFDM.
When the SFN principle is implemented with omnidirectional receiving antennas, 0-dB echoes must be expected. General Instrument has proposed a form of SFN in which SCM and automatic equalizers are used. They propose using directional antennas to keep the echoes below -6 dB. This is the first time anyone has suggested that it might be possible to support SFN with conventional digital transmission systems. With sufficiently good antennas, properly oriented, such a scheme might well work, although it would be more costly and less convenient for viewers. It would be facilitated by colocation of the low-power transmitters. (It was not discussed in detail at the meeting, although the paper by GI in IEEE Trans. Bcasting 9/92 was circulated.)
Another aspect of SFNs that was discussed is the manner of feeding the low-power transmitters. Merrill Weiss pointed out that, if rented communication facilities were used, this would add greatly to the expense. In most of the experiments done to date, however, each transmitter was fed from the central transmitter or from a neighbor, so that there was no such expense. The difficulty of feeding the small transmitters is also affected by their relative timing. This matter needs further discussion.
A distributed scheme is very attractive to broadcasters because it can provide reliable coverage in odd-shaped areas with moderate transmitted power in spite of obstacles such as hills and tall buildings. It also permits low-cost extension of service to specific outlying areas without causing unintended interference to other stations.
A final aspect of SFNs is the size of the cells, which was discussed by Merrill Weiss and Tony Uyttendaele. Some broadcasters have expressed a preference for rather large cells, which would require a very large number of subcarriers. As pointed out by Tony, the signal halfway between cell sites may decrease below the value required to protect against cochannel interference. He believes that this phenomenon indicates that the separation of the distributed transmitters must be no more than 30 kilometers, which he thinks is too small. My own opinion is that, since so little knowledge about OFDM has yet permeated the broadcasting community, it is too early to worry about this. We do not as yet have a firm idea of how the total cost would vary with cell size. (The best results would be achieved with cell sizes so small that the signal strength were virtually uniform within the intended coverage area.) I believe that cost alone is likely to favor very low-power transmitters and quite small cells, particularly if they are implemented by third parties who can provide sites, equipment, and communication facilities for all broadcasters in each city.
10. Adjacent-Channel Interference
A problem pointed out by another attendee is that, even though the average field strength in a distributed transmission scheme is much less than with a centralized transmitter, it is still quite high close to the low-power cell sites. If there is an NTSC station in an adjacent channel, the HDTV signal may interfere with it. (Note that this would be as true with SCM/SFN as with COFDM/SFN.)
This problem is present only during the transition period. However, it must still be dealt with, since a viable transition scenario is a vital part of any new TV system. I believe there are several ways to avoid the interference. One is to use high-gain transmitting antennas at the sites. These would have a narrow beam angle in the vertical plane and would be aimed at the horizon. If located safely above the receiving antennas, they greatly reduce the close-in field strength. Another scheme is to use a different polarization for the HDTV and NTSC signals. NTSC is already vertically and right-hand polarized. Using horizontal and/or left-hand polarization for HDTV would greatly improve the separation. After NTSC goes off the air, the alternate polarization could be added and quality greatly improved. (See earlier footnote on Thomson/BBC demonstration.)
Another solution is to require current NTSC users in the affected areas to get better antennas. This is a last-resort solution, only to be used where there is absolutely no other choice, as it is bound to be very unpopular.
11. Complexity of Implementation
Although there have been some objections to OFDM on account of the complexity that comes with large numbers of carriers, its implementation with the Fourier Transform is a great simplification. The consensus of the speakers was that, since OFDM offers easy equalization in the presence of many strong echoes, the proper tradeoff between SCM and OFDM relates to the relative complexity of the SCM automatic equalizer (1/3 of the circuitry in the GI receiver) and the DFT required in OFDM. The CCETT speakers presented a chart indicating that, in terms of area of silicon, OFDM was much less complex, particularly as the number of subcarriers was increased so that larger cell sizes could be used in SFN.
Several of the speakers showed pictures of their hardware, which amounted to no more than a handful of PCBs in one or two card cages.
12. Conclusion
Although more investigation is needed (and it is always possible that advocates of SCM may turn up some unexpected results), it seems clear that COFDM offers significant advantages over SCM in terrestrial service. It seems to have no major disadvantages, and it could readily be used by any of the four proposed digital systems. The ease with which large and rapidly changing echoes can be suppressed, as well as the error-rate improvement that results from the presence of echoes, point to a fundamental advantage.
With respect to the difficulty of implementation, it appears from the presentations that COFDM is no more complex than SCM and, in fact, may be considerably simpler.
In my opinion, the three most important aspects of COFDM are improved reliability under difficult transmission conditions, the ability to use omnidirectional antennas, and the improved spectrum efficiency that results from the use of single-frequency networks. Although it appears that SFNs could be implemented with SCM as well, that would require carefully positioned high-gain high-directivity antennas that are likely to be much less acceptable to the viewing public.
William F. Schreiber
Cambridge Mass 02139
brought to you by
Copyright 1999
|Home| |E-MAIL|
"From the presentations and discussions, a conclusion can be drawn that COFDM has important advantages in all these operational characteristics and is no harder -- perhaps easier -- to implement. Although more work is required to define the performance precisely and to optimize the design of SFNs, there appears to be no major drawback that might offset the advantages of COFDM".
Report on OFDM Meeting at MIT 26-27 October 1992
Conclusions of W.F.Schreiber
The opinions expressed herein are those of the author only.
Abstract
Representatives from all of the European and Canadian projects on coded orthogonal frequency division multiplex (COFDM) met to present their results and to exchange views. Work in France, UK, Germany, Scandinavia, and Canada was discussed, including the results of field tests. Coverage area, interference performance, multipath performance, antenna requirements, the ability to support single-frequency networks (SFNs), and the complexity of implementation, were discussed. Comparisons were made with single-carrier modulation (SCM), although no SCM proponents attended. SFN is attractive because it has the potential of reducing from 68 to about 20 the number of channels needed to provide today's amount of program choice to viewers.
From the presentations and discussions, a conclusion can be drawn that COFDM has important advantages in all these operational characteristics and is no harder -- perhaps easier -- to implement. Although more work is required to define the performance precisely and to optimize the design of SFNs, there appears to be no major drawback that might offset the advantages of COFDM.
1. Introduction
The purpose of the meeting was to bring together representatives of the principal laboratories working on orthogonal frequency-division multiplex (OFDM) and single-frequency networks (SFN) to present their work and to exchange views. This objective was certainly met and, as a result, the current theoretical and practical status of these developments was made more clear. However, there were no system proponents and only one person from the FCC. Thus no one attended who could forcefully present the technical case for single-carrier systems. A number of those present are connected with the ACATS and ATTC activities.
Before the meeting, the attached list of topics had been distributed to the European and Canadian speakers. All of the topics were touched on in the presentations and discussions. More work, both theoretical and experimental, is needed to explore a number of these topics, in particular the comparison with single-carrier modulation. (SCM) In what follows, I discuss performance with respect to a number of separate measures, as they were dealt with at the workshop. However, in a valid system design, these measures are found to be highly interrelated, so that careful tradeoffs are required.
Familiarity with COFDM is assumed in this report.
2. Equalization and SNR
It is widely agreed that the suppression of linear distortion by means of an automatic equalizer in SCM results in exactly the same SNR as the use of subchannel-by-subchannel gain and phase correction in OFDM. The corrected noise spectrum is the inverse of the uncorrected frequency response. However, the OFDM equalizer is very easy to implement, while the SCM equalizer requires a great deal of computation. The former requires a simple measurement, after which the correction parameters are computed in one step. Iteration is not needed and convergence of parameters is not an issue.
For lack of an advocate at the meeting, the limitations on the performance of automatic equalizers were not much discussed. For example, GI reports that its equalizer cannot deal with echoes larger than -6 dB. Since there is, in fact, a proper solution for echoes much larger than this, the GI limitation must be due to the specific algorithm used, and not to a theoretical obstacle. This is very important in the case of single-frequency networks with omnidirectional antennas, where 0-dB echoes may be expected.
3. Error-Rate Performance
It is sometimes said by OFDM proponents that echoes add constructively according to their power. No conclusive evidence that this is the case was presented at the meeting. (Efforts at MIT to demonstrate this behavior theoretically have so far not proved fruitful. However, there was impressive experimental evidence that echoes do combine in a positive manner, which is certainly not the case in SCM. It appears that the benefit of the combination depends on the coding that is used -- the "C'' in COFDM. Uncoded OFDM, as used in the Thomson CSF test at the Sarnoff Laboratories in1990, would not be expected to show this advantage. The 1990 test used equipment provided by Thomson/CSF-LER. In a recent letter to Renn McMann, K. Jonnalagadda of Sarnoff described the results as generally no better than the dual-carrier scheme advanced by ATRC. However, The Thomson representative at the meeting, whose equipment was used in the tests, does not agree with the assessment. He also pointed out that the equipment was OFDM, not COFDM, and would not be expected to show the superior error-rate performance in the presence of multipath. He distributed copies of a press release reporting the successful transmission of 60 Mb/sec in an 8-MHz video channel by the BBC in London using Thomson equipment. This experiment used both planes of polarization to double the normal capacity.)f
In the presence of echoes, the frequency response becomes nonuniform. Therefore, the SNR for some subcarriers is poorer than for others. The reliability of data transmitted on low-SNR subcarriers is low, and this \can be taken into account in the decoding process. COFDM involves scrambling (called interleaving) in time and frequency, so that transmitted data is dispersed at random with respect to subcarriers. Data from less-reliable channels is weighted less than data from more-reliable channels, and this results in a lower overall error rate, after correction, than with equal weighting. CCETT reported the experimental finding that the error rate after correction went down in the presence of echoes. In addition, they presented data from field tests showing that the service area of two transmitters on the same frequency was larger than the union of the service areas of the individual transmitters.
4. Interference to NTSC
Interference is always a two-way phenomenon. A signal that is relatively immune to interference can be transmitted at lower power for the same coverage area, thus interfering less with other signals. Nevertheless, it is useful to consider the specific aspects of signals that determine sensitivity to undesired signals separately from the tendency to interfere with other signals.
A signal such as emitted by the GI system has a nearly uniform spectrum and appears to be random. Its amplitude probability distribution is determined by the QAM. For reasons I must confess not to understand at this moment, this distribution is not uniform, but has more-or-less random peaks that require linearity of the transmitter in excess of the rms power. The OFDM signal is quasi-Gaussian. CCETT estimates that the excess linearity requirement is 3.7 dB for SCM and 5.0 dB for OFDM at a spectral efficiency of 5 bits/cycle.
I do not think that the probability distribution itself has much direct effect on noise visibility. (At least this was the result of an MIT thesis investigation many years ago.) However, the somewhat higher percentage of high peak values in OFDM might cause more interference to a specific kind of video signal. More investigation is needed.
To the extent that the emitted signal is nonrandom, its interference into cochannel NTSC for a given rms power level is increased. For example, switching off carriers in the vicinity of NTSC carriers (something like ATRC's 2-carrier system) increases the interference for a given power level. It seems unlikely that, for equal interference from NTSC, an ATRC 2-carrier system and an OFDM system will be very different with respect to interference to NTSC.
In OFDM, the spectrum is sharply confined because it is the sum of equally spaced subchannel signals, each of which is sharply confined. It does not require channel-limiting filters at the transmitter and has much less tendency to generate adjacent-channel interference than SCM. The Canadian simulation studies show that, as a result, 5.625 MHz of a 6-MHz channel can be used, which probably makes up for the communication loss due to the use of guard intervals in OFDM. (See below.)
5. Interference from NTSC
Switching off carriers, as discussed above, is a very effective method of boosting immunity to cochannel NTSC interference. If the pattern of requency interleaving were identified in the signal and downloaded into the encoder and decoder hardware (a feature not found at present in the OFDM experiments but that would be quite easy to add), then the pattern could be optimized for the particular interference situation faced by each station. When nearby NTSC stations eventually go off the air, the unused carriers could be reclaimed and picture quality improved without hardware changes. In addition, this scheme would allow general upgrading over time in a nondisruptive manner (a feature the FCC has asked for) and would also allow higher quality for compatible ``contribution'' systems.
6. ATV-ATV Interference, Other Interference
The Canadian simulations show that OFDM performance with respect to ATV-ATV interference is quite close to that of SCM. CCETT reports that OFDM has 10-dB better resistance to pulse interference and 4-dB better resistance to sine-wave interference. Its white-noise threshold is very similar to that of SCM.
7. Multipath Performance
OFDM was specifically designed to deal with multipath by dividing the signal into a large number of low-data-rate components whose symbol length is longer than the temporal spread of the echoes. If the integration period at the receiver is correct, all echoes related to a single transmitted pulse arrive within one pulse period and intersymbol interference (ISI) completely disappears.
The timing of the receiver is greatly simplified if a guard interval, itself a little longer than the multipath spread, is added to the transmitted symbol period. The effect of multipath is now simply a gain and phase change for each subchannel. These can easily be measured with the aid of a pilot signal so that the multipath disturbance is removed completely, except for its effect on the SNR. OFDM equalization is much simpler than for SCM, and has no limit as far as echo amplitude is concerned. In SFN operation (see below), 0-dB echoes can be tolerated. As a result, directional antennas are not required, although their use would generally result in higher SNR.
These characteristics of OFDM have been fully borne out in experiments carried out by a number of the organizations represented at the meeting, and no contrary evidence has surfaced. Note that SCM has no equivalent to the use of the guard interval.
Coded OFDM (COFDM) makes use of the echoes to reduce the corrected error rate compared to what it would be with the direct signal only. In SCM, the cancellation of echoes reduces the SNR and therefore increases the error rate. This is a very important difference in favor of COFDM.
8. Antennas
Service area calculations for the digital systems are based on 10 dB antenna gain and 14 dB front-to-back ratio. The antenna gain influences the noise-limited service area while the front-to-back ratio influences the interference-limited service area. The requirement of a high-gain, high-directivity antenna imposes cost and inconvenience on viewers, the acceptability of which is not clear. Current widespread use of "rabbit ears'' in city areas indicates that, at the very least, extensive consumer education will be required.
The main antenna issue is the limited ability of automatic channel equalizers, needed in SCM to deal with multipath, to cope with high-level and/or rapidly-changing echoes. Such echoes are suppressed easily in OFDM, as a result of which omnidirectional antennas proved satisfactory in the earlier tests of digital audio broadcasting to moving vehicles. Except in special circumstances, OFDM permits the use of omnidirectional antennas while SCM does not; this is a major advantage.
Note that, in current practice, the service areas of stations, where they are interference limited, is directly affected by the assumptions about receiving antennas. If better front-to-back ratio could be counted on, then NTSC stations on the same channel could be moved closer together, even without any new technical developments.
9. Single-Frequency Networks
A SFN is composed of an array of low-power transmitters operating on the same frequency. (This arrangement is sometimes referred to as ``distributed transmission.'') The multipath immunity of the receiver is relied upon to achieve satisfactory operation when a number of strong ghost-like signals are received. SFNs permit very high spectrum efficiency, since the service area of each station is precisely defined by the location of transmitters. Contiguous regions can be used by different stations on the same channel, so that there are no taboo channels due to cochannel interference. The "no-man's land'' between stations, in which reception is problematic, is about one cell-diameter in width. In this area, directional antennas are required, which would enable reception of either station, as desired) If SFNs were used in this manner, the total number of channels that would have to be allocated for TV would be no larger than the number of different programs to be made available in each city or market.
The SFN principle has been demonstrated at least in France and Canada for DAB with COFDM. The ability to support SFNs and their attendant high spectrum efficiency is a very important characteristic of COFDM.
When the SFN principle is implemented with omnidirectional receiving antennas, 0-dB echoes must be expected. General Instrument has proposed a form of SFN in which SCM and automatic equalizers are used. They propose using directional antennas to keep the echoes below -6 dB. This is the first time anyone has suggested that it might be possible to support SFN with conventional digital transmission systems. With sufficiently good antennas, properly oriented, such a scheme might well work, although it would be more costly and less convenient for viewers. It would be facilitated by colocation of the low-power transmitters. (It was not discussed in detail at the meeting, although the paper by GI in IEEE Trans. Bcasting 9/92 was circulated.)
Another aspect of SFNs that was discussed is the manner of feeding the low-power transmitters. Merrill Weiss pointed out that, if rented communication facilities were used, this would add greatly to the expense. In most of the experiments done to date, however, each transmitter was fed from the central transmitter or from a neighbor, so that there was no such expense. The difficulty of feeding the small transmitters is also affected by their relative timing. This matter needs further discussion.
A distributed scheme is very attractive to broadcasters because it can provide reliable coverage in odd-shaped areas with moderate transmitted power in spite of obstacles such as hills and tall buildings. It also permits low-cost extension of service to specific outlying areas without causing unintended interference to other stations.
A final aspect of SFNs is the size of the cells, which was discussed by Merrill Weiss and Tony Uyttendaele. Some broadcasters have expressed a preference for rather large cells, which would require a very large number of subcarriers. As pointed out by Tony, the signal halfway between cell sites may decrease below the value required to protect against cochannel interference. He believes that this phenomenon indicates that the separation of the distributed transmitters must be no more than 30 kilometers, which he thinks is too small. My own opinion is that, since so little knowledge about OFDM has yet permeated the broadcasting community, it is too early to worry about this. We do not as yet have a firm idea of how the total cost would vary with cell size. (The best results would be achieved with cell sizes so small that the signal strength were virtually uniform within the intended coverage area.) I believe that cost alone is likely to favor very low-power transmitters and quite small cells, particularly if they are implemented by third parties who can provide sites, equipment, and communication facilities for all broadcasters in each city.
10. Adjacent-Channel Interference
A problem pointed out by another attendee is that, even though the average field strength in a distributed transmission scheme is much less than with a centralized transmitter, it is still quite high close to the low-power cell sites. If there is an NTSC station in an adjacent channel, the HDTV signal may interfere with it. (Note that this would be as true with SCM/SFN as with COFDM/SFN.)
This problem is present only during the transition period. However, it must still be dealt with, since a viable transition scenario is a vital part of any new TV system. I believe there are several ways to avoid the interference. One is to use high-gain transmitting antennas at the sites. These would have a narrow beam angle in the vertical plane and would be aimed at the horizon. If located safely above the receiving antennas, they greatly reduce the close-in field strength. Another scheme is to use a different polarization for the HDTV and NTSC signals. NTSC is already vertically and right-hand polarized. Using horizontal and/or left-hand polarization for HDTV would greatly improve the separation. After NTSC goes off the air, the alternate polarization could be added and quality greatly improved. (See earlier footnote on Thomson/BBC demonstration.)
Another solution is to require current NTSC users in the affected areas to get better antennas. This is a last-resort solution, only to be used where there is absolutely no other choice, as it is bound to be very unpopular.
11. Complexity of Implementation
Although there have been some objections to OFDM on account of the complexity that comes with large numbers of carriers, its implementation with the Fourier Transform is a great simplification. The consensus of the speakers was that, since OFDM offers easy equalization in the presence of many strong echoes, the proper tradeoff between SCM and OFDM relates to the relative complexity of the SCM automatic equalizer (1/3 of the circuitry in the GI receiver) and the DFT required in OFDM. The CCETT speakers presented a chart indicating that, in terms of area of silicon, OFDM was much less complex, particularly as the number of subcarriers was increased so that larger cell sizes could be used in SFN.
Several of the speakers showed pictures of their hardware, which amounted to no more than a handful of PCBs in one or two card cages.
12. Conclusion
Although more investigation is needed (and it is always possible that advocates of SCM may turn up some unexpected results), it seems clear that COFDM offers significant advantages over SCM in terrestrial service. It seems to have no major disadvantages, and it could readily be used by any of the four proposed digital systems. The ease with which large and rapidly changing echoes can be suppressed, as well as the error-rate improvement that results from the presence of echoes, point to a fundamental advantage.
With respect to the difficulty of implementation, it appears from the presentations that COFDM is no more complex than SCM and, in fact, may be considerably simpler.
In my opinion, the three most important aspects of COFDM are improved reliability under difficult transmission conditions, the ability to use omnidirectional antennas, and the improved spectrum efficiency that results from the use of single-frequency networks. Although it appears that SFNs could be implemented with SCM as well, that would require carefully positioned high-gain high-directivity antennas that are likely to be much less acceptable to the viewing public.
William F. Schreiber
Cambridge Mass 02139
brought to you by
Copyright 1999
|Home| |E-MAIL|