US005425050A.
`115)
`United States Patent
`5,425,050
`[11] Patent Number:
`
`[45] Date of Patent: Jun. 13, 1995
`Schreiberet al.
`
`Transactions on Communications, vol. COM-29, No.7,
`Jul. 1981, pp. 982-989.
`Chang, R. W., “Synthesis of Band-Limited Orthogonal
`Signals for Multichannel Data Transmission”, The Bell
`System Technical Journal, Dec. 1966, pp. 1775-1796.
`Pommier, D. et al., “A Hybrid Satellite/Terrestrial
`Approach for Digital Audio Broadcasting With Mobile
`and Portable Receivers”, NAB Engineering Confer-
`ence Proceedings, 1990, pp. 304-311.
`“Description of the COFDM System”, Groupement
`D'Interet Economique Regipar L’Ordonnance Du,
`Sep. 23, 1967.
`Alard, M.et al., “Principles of Modulation and Channel
`Coding for Digital Broadcasting for Mobile Receivers”,
`EBU Review-Technical, No. 224, Aug. 1987, pp.
`168-190.
`
`Primary Examiner—David C. Cain
`Attorney, Agent, or Firm—Fish & Richardson
`
`ABSTRACT
`[57]
`An apparatus for encoding a television production sig-
`nal for transmission, the television production signal
`includinga first input signal carrying a first class of data
`and a second input signal carrying a second class of
`data, the second class data requiring a higher-quality
`transmission the first class data. For each word of the
`first input signal, the first stage generates N serial sam-
`ples of the first output sample stream, each of the N
`serial samples being formed by a different combination
`of a set of more than one of the N samples of the each
`word. The secondstage includesan input stage combin-
`ing the second inputsignal stream with the first output
`sample stream to generate an intermediate input sample
`stream; a serial-to-parallel converter receiving the inter-
`mediate input sample stream and producing a second
`stream of words therefrom, each of the words of which
`being a parallel grouping of M successive samples of the
`intermediate input sample stream; a Discrete Fourier
`Transform (DFT) module producing a parallel output
`stream of wordsthatis the discrete Fourier transform of
`the second word stream; and a parallel-to-series con-
`verter generating the FDM outputsignal from the par-
`allel output stream of the DFT module.
`
`53 Claims, 11 Drawing Sheets
`
`CHANNEL
`(NOISE,ECHOES,
`INTERFERENCE)
`
`[54] TELEVISION TRANSMISSION SYSTEM
`USING SPREAD SPECTRUM AND
`ORTHOGONAL FREQUENCY-DIVISION
`MULTIPLEX
`
`[75]
`
`Inventors: William F. Schreiber, Cambridge;
`Michael O. Polley, Belmont, both of
`Mass.
`
`[73] Assignee:
`
`Massachusetts Institute of
`Technology, Cambridge, Mass.
`
`[21] Appl. No.: 149,264
`
`[22] Filed:
`
`Nov. 9, 1993
`
`[63]
`
`Related U.S. Application Data
` Continuation-in-part of Ser. No. 965,227, Oct. 23, 1992,
`Pat. No. 5,311,543.
`
`FSU] Vint, C16eeeecesssssessssesscerssensrersenens HOAN 7/167
`[52] UnS. CU. onessssscsscsesessssetsreneeetseoee 375/200; 380/10;
`380/13
`{58] Field of Search................. 375/1; 380/10, 20, 13,
`380/19
`
`[56]
`
`References Cited
`U.S. PATENT DOCUMENTS
`
`1/1970 Chang .
`3,488,445
`
`1/1984 Moses etal. ..
`4,425,642
`~ 4,821,120 4/1989 Tomlinson........
`
`4,890,283 12/1989 Tsinberg et al.
`.
`
`4,907,087 3/1990 Schreiber..........
`
`4,912,721
`3/1990 Pidgeonetal....
`5,073,899 12/1991 Collier et al.
`.....
`-
`§,127,021
`6/1992 Schreiber..........
`5,311,543
`5/1994 Schreiber et al...css 380/10
`
`OTHER PUBLICATIONS
`
`Weinstein, S. B. et al., “Data Transmission by Frequen-
`cy-Division Multiplexing Using the Discrete Fourier
`Transform”, IEEE Transactions on Communication
`Technology, vol. COM-19, No. 5, Oct. 1971,. pp.
`628-634.
`Hirosaki, B., “An Orthogonally Multiplexed QAM
`System Using the Discrete Fourier Transform”, IEEE
`
`
`
`1
`
`SAMSUNG 1015
`
`1
`
`SAMSUNG 1015
`
`

`

`U.S. Patent
`
`June 13, 1995
`
`Sheet 1 of 11
`
`5,425,050
`
`DEMODULATOR|
`
` FROM
`
`TO
`SOURCE
`DECODER
`
`2
`
`

`

`U.S. Patent
`
`June 13, 1995
`
`Sheet 2 of 11
`
`5,425,050
`
`
`
`d(n)=a(n)+jb(n)
`203
`
`COMPLEX |
`HYBRID
`
`!
`
`STREAM
`
` | | |
`
`|
`
`| CHANNEL
`
`|
`| rate(N*R/2)
`
`|
`|
`|
`|
`|
`1
`roNT
`S/P: Serial-to-Parallell
`FIG. 3
`OFDM Decode
`107
`CTTao —-——--— +5
`cost
`
`a(0)
`
`
`
`
`
`sin®yit 118=rate(R/2) 119
`
`— F
`
`IG. 4
`
`3
`
`

`

`U.S. Patent
`
`June 13, 1995
`
`Sheet 3 of 11
`
`5,425,050
`
`101
`
`
`
`100, ~anatoc)|°° 202 ©
`
`
`103
`VIDEO |
`ENCODER
`
`
`
`
`OFDM
`ENCODER
`
`
`
`
`
`
`1054 CHANNEL
`(NOISE,ECHOES,
`INTERFERENCE)
`
`104
`
`MODULATOR
`
`
`
`
` SOURCE
`+
`107
`VIDEO
`
`DECODER
`
`OUT 211
`
`4
`
`

`

`U.S. Patent
`
`June 13, 1995
`
`Sheet 4 of 11
`
`5,425,050
`
`
`
`@/(MeN)=9381
`
`XHIdNOD
`
`DOTVNV
`
`IOGWAS
`
`WVauls
`
`
`
`—--——-~~-9"SIs
`
`Z/(MaN)O81
`
`(u)qf=(u)e=(u)p
`
`IVLISIG
`
`WadOONA
`UaN978
`
`DOTVNV
`
` weeOOT
`
`\tor
`
`5
`
`
`
`
`

`

`U.S. Patent
`
`June 13, 1995
`
`5,425,050
`
`Sheet 5 of 11
`
`WULO0Id
`
`
`
`TVLIOIDATOHSHUHL
`
`NOISIDAd
`
`6
`
`

`

`U.S. Patent
`
`June 13,1995 Sheet 6 of 11
`
`5,425,050
`
`VARIATION OF SNR WITH RANGE
`
`(SNR=FIELD STRENGTH-PLANNING
`FACTORS)
`
`12d0B
`
`10
`
`20
`
`30
`
`40 MILES
`
`DIGITAL DATA
`
`7
`
`

`

`
`
`ei|toveulsa|.Lhe
`
`
`
`
`JapooeqpugqaysapooaqsrepeasrecesseceananeaaecsJepooegpejornujsuocey
`aceon“1ajeune!!6“Sis
`
`
`sjoquiAgpeaoiduyZthjeubiq
`a@oUEpUoy):+YORBUUOJU|‘+eaguapyuoggjeq
`en6ig|4apo9|102
`peg——-——————Le
`eeeneetteeeeeeeereee
`SIFOrrpejonjsuoaey
`
`601Bo
`
`SOL|‘seoyos'esjou)
`pOL|s0}8/NPOW
`90}|4018jnpowag
`e0lwaso
`
`(Qouasapayu
`sapooug
`I602resjoquiks
`(uoneuLojuy|'
`jBuo;NjoAUoD
`
`S0¢
`
`jauuByd
`
`90¢
`
`J
`
`U0}99105
`
`uley-+
`
`¥0
`
`pugsH
`
`
`
`JOpODpaieusywoueg
`
`LOL
`
`peaids
`
`winsyoeds
`
`Beg; eubia
`49pop66)
`
`Ook
`
`gauno
`
`syndu|
`
`O8plA
`
`yeuy
`
`LEE
`
`oepl,
`
`aoinog
`
`nding
`
`U.S. Patent
`
`June 13, 1995
`
`Sheet 7 of 11
`
`5,425,050
`
`401waso802jeuopnjonucg||60a
`
`
`
`
`
`6end40¢Breq1en6iqpepoo|pop
`
`payeuayeoucg|
`49poseqbbe
`
`8
`
`
`
`
`
`
`
`
`
`

`

`U.S. Patent
`
`June 13, 1995
`
`Sheet 8 of 11
`
`5,425,050
`
`H(f)
`
`-3MHz
`
`fc
`
`,
`
`+3MHz
`
`
`1.0
`a
`
`\
`|
`bt
`
`t F
`
`IG. 10
`
`TRANSMITTED
`SIGNAL
`
`RECEIVED
`SIGNAL.
`
`9
`
`

`

`U.S. Patent
`
`June 13, 1995
`
`Sheet 9 of 11
`
`5,425,050
`
`
` }13a03fe]1aa|
`
`LL‘Dis
`
`INOOJGIA
`
`
`
`y—{z_sa003q@3gd09030
`
`TINNVHD
`
`tL100930
`
`YaXaTdLINW
`
`
`
`4aXadiLINW
`
`NiO3dIA
`
`10
`
`10
`
`
`
`
`

`

`U.S. Patent
`
`June 13, 1995
`
`Sheet 10 of 11
`
`5,425,050
`
`
`
`JOWOJSUBL)1ej0d0}uBISELeD
`
`d<+9
`
`JOUUOJSUB}UBISBRD0}JE|O
`
`o<gEFDid
`
`FouBDL
`zeqe8h
`
`Wds0(*)
`
`
`
`ohhERCZiLL soy-mo7!bbb£|vel6Lz|SIO4Lreubiq|1@b@
`
`
`
`16pogpeyueyoucn=
`
`JoyeINpowOL
`
`G0z
`
`Jo
`
`
`122[ByvOIgyueweouRyuy
`
`d+9anusBled
`
`o9s/qWS>
`
`SarrTEl2reubiq
`
`jo——,1
`
`2eS/qWS>
`
`11
`
`O<d(x)
`
`6Sh.
`
`Zé}
`
`912ss00g—-~Roleu
` 9chaJEpoD-_OOS/SING
`
`
`LOL
`
`11
`
`
`

`

`U.S. Patent
`
`June 13, 1995
`
`Sheet 11 of 11
`
`5,425,050
`
`From Demodulator
`
`__confidenceinfo.|r
`
`Concatenated Decoder 1312,
`116
`117
`|
`
`Low-res
`209a Diai
`
` .
`
`Convolutional
`Decoder
`
`FIG. 14
`
`12
`
`12
`
`

`

`1
`
`5,425,050
`
`TELEVISION TRANSMISSION SYSTEM USING
`SPREAD SPECTRUM AND ORTHOGONAL
`FREQUENCY-DIVISION MULTIPLEX
`
`CROSS REFERENCE TO RELATION
`APPLICATIONS
`
`This application is a continuation-in-part of U.S. pa-
`tent application Ser. No. 07/965,227, filed Oct. 23, 1992,
`now U.S. Pat. No. 5,311,543.
`FIELD OF THE INVENTION
`
`The invention relates to television systems that make
`use of data compression and particularly to the trans-
`mission of coded TV signals in channels that are physi-
`cally analog such as the over-the-air transmission chan-
`nel. It also relates to communication systems making
`use of spread spectrum (SS) and orthogonal frequency-
`division multiplex (OFDM).
`BACKGROUND OF THE INVENTION
`
`10
`
`— 5
`
`20
`
`In U.S. Pat. No. 5,127,021 (the °021 Patent) and in
`U.S. patent application Ser. No. 07/965,227 (the ’227
`Application), both of which are incorporated herein by
`reference, TV systems using spread spectrum were
`disclosed. The present invention is an improvement of
`the earlier inventions, providing higher resistance to
`channel impairments as well as the ability to be em-
`ployed in single-frequency networks (SFNs).
`TheU.S.is in the process of developing standards for
`over-the-air broadcasting of high-definition television
`(HDTV),also called Advanced Television (ATV).
`Because substantial data compression is required in
`order to permit transmission in the same 6-MHz channel
`that is used for the current system (NTSC), system
`proponents have proposed all-digital systems, using
`both digital source coding and digital channel coding.
`The source-coding methods are now highly developed,
`but the proposed channel-coding methods, which were
`developed for wire-line transmission, are not. The over-
`the-air channelis a particularly hostile environment for
`digital transmission, plagued as it is by multipath trans-
`mission, noise, and interference. As a result, there is
`considerable doubt as to whether the goals of conve-
`nience,reliability, efficient use of spectrum, and provi-
`sion for a range of receivers of different price and per-
`formance will be achieved.
`Cable companies are also planning to use ATV tech-
`nology, but appear to be moreinterested in transmitting
`a multiplicity of programs of today’s definition in each
`cable channel rather than transmitting one HDTV sig-
`nal per channel. Cable channel characteristics are some-
`what different from those of the over-the-air channel,
`but many of the same problemsare present. The need to
`serve a range of receivers of different price/perfor-
`mance is commonto both kinds of service. The need to
`optimize performance for reception conditions is also
`common to both since cable does not provide equal-
`quality signals to all receivers.
`The overall object of the invention is to overcome the:
`limitations of existing and proposed TV transmission
`systems, particularly in connection with broadcasting of
`coded HDTV signals. In systems that use a single cen-
`tralized transmitter, the invention permits good perfor-
`mance in spite of multipath (ghosts or echoes), noise,
`and interference. It makes efficient use of spectrum and
`optimizes picture quality at each receiver in the light of
`its individual reception conditions. It permits the use of
`
`40
`
`45
`
`60
`
`13
`
`2
`relatively inexpensive receivers (typically with small
`screens) in applications where maximum image quality
`is not required.In single-frequency networks, in which
`the reception area of a station is served by a cellular
`array of low-power transmitters, all operating on the
`same frequency, its remarkable resistance to multipath
`permits satisfactory operation in spite of the presence of
`many strong ghosts. In all applications, the invention
`permits the use of less sophisticated antennas and sim-
`pler channel equalizers than are required for currently
`proposedall-digital systems.
`SUMMARY OF THE INVENTION
`
`In one aspect, the invention deals with source-coding
`systems that produce two classes of data, one that re-
`quires nearly perfect transmission and a second that
`tolerates some noise and distortion. In this aspect, the
`first class of data is transmitted digitally at a very low
`error rate while the secondclass is transmitted as analog
`or multilevel quantized (“digital”) samples that.gener-
`ally have noise added in transmission. Data samples of
`the second class are subjected to a SS operation and are
`combined with data samplesofthe first class, producing
`hybrid symbols.
`It should be noted that virtually all systems that have
`been proposed for HDTVareofthe type that produce
`two classes of data as mentioned above. All these sys-
`tems are based on transformation into the frequency
`domain. What is transmitted comprises selected trans-
`form coefficients plus identifying data, velocity vectors,
`etc. The coefficients do not need very precise transmis-
`sion, while the other data does.
`In anotheraspect, the said hybrid symbols are divided
`into a multiplicity of symbol streams (typically hun-
`dreds or even thousandsof such streams) each of which
`modulates a different carrier in an OFDM processor,
`said carriers being equally spaced throughout the width
`of the transmission channel.
`In a third aspect, the video information is encoded in
`such a way that images of different quality can be pro-
`duced bythe receiver by using different fractions of the
`received data, the channel coding being arranged in
`such a way that less than all the data can readily be
`extracted from the received signal by an inexpensive
`receiver.
`In a fourth aspect, the SS process permits control of
`the relative SNR of the various spatial-frequency com-
`ponents by appropriately weighting the spread compo-
`nents before transmission.
`In a fifth aspect, the OFDMprocess eliminates inter-
`symbol interference up to a given temporal spread of
`echoes by incorporating a guard interval into the trans-
`mitted signal.
`In a sixth aspect, a very simple channel equalizer is
`made possible by using such a large numberofcarriers
`that the gain and phase distortion of the channel is es-
`sentially constant across the band occupied by each
`carrier and its sidebands.
`In a seventh aspect, the effect of multipath distortion
`on the relative SNRs of the recovered data of the sec-
`ond class is equalized by choosing appropriate parame-
`ters for the SS and OFDM processors. All the recov-
`ered data has the same noise contentin spite of different
`CNR for each carrier. (In this application, the term
`SNR is reserved for image data recovered at the re-
`ceiver, while the term CNR refers to the carrier-to-
`noise ratio of transmitted data.)
`
`13
`
`

`

`5,425,050
`
`3
`In an eighth aspect, errors due to channel impair-
`ments are minimized and the error-producing effect of
`ghosts is lessened by using a concatenation of convolu-
`tional coding and Reed-Solomoncoding.
`In a ninth aspect, the effect of impulse noise in both
`the time and frequency domainsis minimized by using a
`combination of SS and OFDM.
`In a tenth aspect, a transmission constellation is used
`in which data of the first and second classes can be
`recovered independently.
`Other advantages and features will become apparent
`from the following description of the preferred embodi-
`ments and from the claims.
`
`BRIEF DESCRIPTION OF THE DRAWINGS
`
`10
`
`20
`
`30
`
`35
`
`40
`
`4
`mitted. The components are recovered at the receiver,
`as shownin FIG.2, by multiplying the received sum-of-
`productssignal, in turn, by each of the same sequences
`and integrating over the symbol time. The N sequences
`are orthogonal so that the N componentsignals are
`recovered without crosstalk. Various methods can be
`used to divide the original signal into N components. In
`the 021 patent, a quadrature-mirror filter bank was used
`in the described embodiment. In this invention, any
`method may be used, including, for example, a quadra-
`ture mirror filter bank or a spatial-frequency decompo-
`sition of the motion-compensated prediction error.
`An advantage of SS processing is that the relative
`SNRof the various components can be established by
`appropriately weighting the components when the sum-
`of-products signal is formed, at the same time preserv-
`ing the desirable properties of the transmitted signal,
`namely that it be of uniform spectrum and have noise-
`like statistical characteristics.
`An important property of SS signals is that each
`transmitted sample (corresponding in length to that of
`an element of the PN sequences)is a linear combination
`of all N components. Successive samples within the
`length of one sequence are independentlinear combina-
`tions of the N components. Demodulation to recover
`the sample values of the N componentsis, in effect, an
`inversion of the linear combinations used in the en-
`coder. This inversion amounts to forming the recovered
`samples by taking N linear combinations of N succes-
`sive received samples. This averages the noise perfor-
`mance over the N received samples, which is important
`when these N received samples are transmitted by
`OFDM with N subcarriers.
`Orthogonal Frequency-Division Multiplex
`OFDM was invented by Chang in 1965 (U.S. Pat.
`No. 3,488,445). It is an improvement over conventional
`frequency division in that no guard bands need beleft
`between the signals in separate subchannels; in fact, the
`spectra of adjacent-subchannel signals overlap. What
`makes the signals recoverable independently is their
`orthogonal nature, which is achieved by imposing the
`appropriate spectral shape onto the modulated carriers.
`This improvement means that no bandwidth is wasted.
`The total required bandwidth is simply the sum of the
`nominal bandwidth (i.e., half the sampling rate) of each
`of the separate subchannel signals. The operation of a
`straightforward OFDM coder and decoder are shown
`in FIGS.3 and 4.
`FIG. 3 shows the OFDM enccder 103. A stream of
`complex samples a +jb (a, b real numbers) 203 is input
`at NR/2 samples/sec to serial-to-parallel converter 114.
`The latter arranges each N successive samples into an
`N-wide word. The real parts are multiplied in multipli-
`ers 115 by the cosine subcarriers while the imaginary
`parts are multiplied by the sine subcarriers, all the prod-
`ucts being added in adder 116 to produce signal 204 at
`rate NR/2 complex samples/sec, which is input to the
`modulator for eventual
`transmission. The adder in-
`cludesfilters to shape the spectrum of each modulated
`carrier so as to achieve orthogonality according to the
`teachings of Chang.
`The OFDM decoder 107 is shown in FIG. 4. The
`signal 207, received from the demodulator,is multiplied
`by each ofthe sine and cosine subcarriers by multipliers
`117 and integrated over one symbol time by integrators
`118. The data is reconverted to a single complex data
`stream 208 by parallel-to-serial converter 119. If trans-
`
`FIG. 1 shows a spread-spectrum encoder in which
`the product signals are weighted to establish their rela-
`tive SNR when recovered at the receiver;
`FIG. 2 showsa spread-spectrum decoder appropriate
`to the encoder of FIG.1;
`FIG. 3 shows an OFDM encoder implemented with
`conventional modulators;
`FIG.4 shows an OFDM decoder implemented with —
`conventional demodulators;
`FIG.5 presents the basic principle of the invention, in
`which analog data is processed by a SS encoder and
`then combined with the digital data, the combined sym-
`bols then being processed by an OFDM encoder.It also
`shows how this data is transmitted, received, and de-
`coded;
`FIG. 6 shows the implementation of an OFDM en-
`coder by meansof the discrete Fourier transform;
`FIG. 7 shows the implementation of an OFDM de-
`coder by meansof the inverse discrete Fourier trans-
`form;
`FIG. 8 shows a typical example of the variation of
`CNR with range, in this case with an antenna height of
`1350 feet;
`FIG.9 showsa block diagram of the invention with
`error correction and channel equalization shown explic-
`itly.
`FIG.10 showsthe frequency response resulting from
`an echo;
`FIG.11 shows a pyramid coding scheme that can be
`used with the inventions when a numberof different
`performancelevels are desired at different CNR or for
`receivers of different price;
`FIG. 12 shows an 8-PSK constellation that can be
`used with the invention;
`FIG. 13 shows an encoderthat can be used to gener-
`ate the constellation of FIG. 12; and
`FIG. 14 shows a decoder for decoding a signal en-
`coded by the encoder of FIG. 13
`DESCRIPTION OF THE PREFERRED
`EMBODIMENTS
`
`55
`
`Spread Spectrum
`Transmission of video signals using SS is described in
`detail in the ’021 patent, and will be summarized here
`only for the sake of completeness. The general idea of
`the schemeis to divide the wideband video signal into a
`multiplicity, N, (typically N=256 to 2048,but higheror
`lower values may be used) of separate components of
`correspondingly narrower bandwidth and lower data
`rate. As shownin FIG.1, each sample of each compo-
`nent is multiplied by a different pseudorandom (PN)
`sequence of length N, thus being spread to full channel
`bandwidth. The N product signals are added andtrans-
`
`60
`
`65
`
`14
`
`14
`
`

`

`5,425,050
`
`15
`
`20
`
`25
`
`Note that the real part of the DFT is given by:
`
`Re(F) =
`
`=P—1P20 (crosaeBE + ipinar2), g=01,...P—1
`
`30
`
`5
`mission were perfect, data stream 208 would be identi-
`cal to input data stream 203 of FIG.3.
`The implementation of FIGS. 3 and 4 becomes un-
`wieldy for very large values of N. A very simple
`method of implementing OFDM was developed by
`Weinstein and Ebert in 1971 (“Data Transmission by
`Frequency Multiplexing Using the Discrete Fourier
`Transform,” IEEE Trans. on Comm., COM-19, 10,
`October 1971, pp 628-634). It is based on the similarity
`of the expression for the discrete Fourier transform
`(DFT) of a group of successive input samples to the
`expression for the desired OFDM channelsignal. Since
`the practicality of OFDM for video applications de-
`pends on the DFT implementation, it is important to
`understand how it works.
`Suppose we have a data stream comprising P com-
`plex samples every T seconds.
`do, di, ... dp, where dp=ap-+jb,(a, b real) Form the
`DFT of P successive samples.
`
`6
`are equal in performance to straightforward transmis-
`sion using amplitude modulation of a single carrier.
`However, they have different performance with respect
`to temporal
`impulses and narrow-band interference.
`(The latter can be considered as impulse noise in the
`frequency domain.) In SS, each input sampleis spread
`across the entire band, so that narrow-bandinterference
`is not catastrophic. OFDM is quite sensitive to narrow-
`band interference since the latter may obliterate data
`modulated onto a particular subcarrier. Both are rela-
`tively insensitive to temporal impulse noise since the
`transmitted symbols are of long duration. As weshall
`see below, the combination has important advantages in
`this respect.
`In the presence of narrow-band interference of
`known properties, such as a nearby NTSCstation, the
`sensitivity of OFDMin this respect can be used to ad-
`vantage. Subcarriers in the vicinity of the NTSC vision,
`sound, and color subcarriers are simply not used, thus
`greatly increasing the resistance of the OFDM signal to
`LL.
`the NTSC interference.
`Fqg=
`2 "be ( j2:
`)
`= 0,1
`P-1
`
`
`=! xp|—j2a »>g=0,1,...P—
`An outstanding feature of OFDM isits resistance to
`multipath distortion. The primary means by whichthis
`is accomplished is to append to the end (or beginning)
`of each transmitted symbol a guard interval at least as
`long as the temporal spread of the multipath—typically
`20 to 40 microseconds. During the guard interval, the
`initial (or the final) portion of the symbolis replicated.
`At the receiver, the decoding process involves integra-
`tion of the received signal for exactly one symbol dura-
`tion. If this integration period is anywhere within the
`total duration of one symbol plus its associated guard
`interval, then all of the received signals, regardless of
`their delay, are integrated for precisely one symbol
`time. The various subcarriers are mutually orthogonal
`even when shifted. This scheme completely eliminates
`one of the main effects of multipath, which is intersym-
`bol interference (ISI). Note that channel equalization
`need not be performed before OFDM decoding,as the
`use of the guard interval completely eliminates ISI.
`Whenthe temporal spread of the multipath happens
`to be much smaller than the duration of the guardinter-
`val, advantage can be taken ofthe fact that some data is
`transmitted twice to improve the CNR ofthe signal by
`averaging the correspondingsignals.
`.
`The other main effect of multipath is that the gains
`and phases of the many subcarriers are different. Multi-
`pathis a linear effect, in that the channel output at any
`instantis a linear combinationofscaled inputs occurring
`over sometime interval. Theeffectis precisely the same
`as that of a particular linear filter. This is because the
`frequency response corresponding to reception with a
`single ghost undulates across the band in an easily calcu-
`lated manner.
`FIG. 10 shows an example of multipath and the cor-
`responding frequency response. In this case, each trans-
`mitted sample gives rise to a main received sample plus
`a ghost of amplitude a, 7 secondslater. The amplitude of
`the frequency response is seen to vary between 1-+-a and
`1—a across the rf band at a frequency of 1/r. Note that
`this effect is shown at baseband. Inreality, the signal is
`always modulated on a high-frequency carrier. Very
`small shifts in + cause very large changes in therelative
`amplitude (but not the shape) of the decoded I and Q
`signals. Correction is always done at baseband for this
`reason.
`
`whereas the desired channel signal is:
`
`=P—1
`fo =?;2, pros2mfpt + bysin2mfpt)
`
`A sampled version of the channelsignal is thus identi-
`cal to Re(DFT) for sampling times of tz=qT/P and
`frequencies of f,=0, 1/T, ... p/T,... (P—1/T. The
`method of sampling produces the proper spectral shape
`so that the carriers are effectively orthogonal. Note that
`we have not generated the individual carriers by this
`method; we have only generated a sampled version of
`the sum of all the modulated carriers, which is adequate
`for obtaining the proper signal for transmission. Note
`also that we have omitted all multiplicative constants.
`At the receiver, the successive input samples can be
`recovered by using the inverse discrete Fourier trans-
`form (DFT). We sample the real channel signal to get
`2P real samples yz in time T, for k=0, 1,...P—1. We
`then take the IDFT of these 2P samples to recover the
`P complex input samples.
`
`—1
`GQ) = xno
`
`Skexp
`
`(20-8 p=01,...P—1
`
`G(p) = ay — jbo p = 1,2,...P—1
`GQ) =200, p = 0
`
`Note that the DFT implementation of OFDM pro-
`duces output at baseband. The modulator shifts the
`entire spectrum to the desired radio frequencyfor trans-
`mission.. Conventional implementation produces output
`at any frequency, depending on the carrier frequencies
`used in the OFDM processor.
`SS and OFDM have the same performance with
`respect to random channel noise. By themselves, they
`
`35
`
`45
`
`35
`
`15
`
`15
`
`

`

`7
`With multiple ghosts of various amplitudes, the fre-
`quency response can have many different shapes. With
`a large number of carriers, one can assume that the
`response is uniform across the band of each carrier, so
`that correction can be made with a single gain factor
`and a single phase factor for each carrier. These correc-
`tion factors can readily be obtained by observation of
`calibration signals inserted into the transmitted signal, a
`procedure well knownin the art.
`When the gain and phase of each channel are cor-
`rected, it is found that the CNR ofthe corrected signal
`is different for each carrier. The subcarriers with low
`gain will have higher noise than those with high gain.
`Forthe digital data, this is taken account of in the de-
`coding associated with the error-correction scheme.
`Forthe analog data, the SS parameters can be chosen so
`that the data recovered by the inverse SS operation has
`a uniform noise level in spite of the varying CNR of the
`OFDMcarriers.
`In both SS and OFDM,itis essential that the symbol
`time be longer than the temporal spread of the muiti-
`path. In this way,all the echoes associated with a single
`transmitted symbol arrive at worst within two succes-
`sive received symbols, thus reducing the intersymbol
`interference. For example,
`if the input signal has a
`5-MHzbandwidth,its symbol time is 200 nanosec. Ech-
`oesin terrestrial broadcasting typically extend to 10 or
`20 microsec, which is many symboltimes. By dividing
`the input signal into, say, 1000 contemporaneous com-
`ponents, each component then has a symbol time of 200
`microsec, which is much more than the temporal spread
`of the multipath. In OFDM with a guard interval, the
`guard interval itself must be longer than the temporal
`spread of the multipath. The guard interval is ‘““wasted”
`transmission time. (Some use can be made ofit to im-
`prove the CNR.) Therefore, to havehigh efficiency, the
`symbol time must be substantially longer than the guard
`interval. The symbol time of the componentsignals is N
`times that of the original input, where N is the number
`of subcarriers. In cases in which the temporal spread of
`multipath is large, a very large number of subcarriers
`may be needed.
`SS and OFDMareusually thoughtof as digital chan-
`nel-coding methods. However, they work equally well
`for analog or hybrid analog/digital input signals.
`Single-Frequency Networks
`An alternative to using a single centrally located
`high-power transmitter to serve a given reception area
`is to fill the area with a cellular array of low-power
`transmitters, all operating on the same frequency. This
`gives a much more uniform powerdistribution and also
`confines the usable signal to a well defined area ofal-
`most arbitrary shape. Different stations can be assigned
`to the same channel in adjacent areas. Except for a
`narrow “no-man’s land,” about the width of one cell,
`between the two service areas, there is little interfer-
`ence,
`thus allowing many more stations to operate
`within each locality within a given overall spectrum
`allocation. Directional antennas are required in the
`no-man’s land, either signal being recoverable by turn-
`ing the antenna.
`In order to operate a SFN successfully, the receivers
`must cope with multiple signals received from nearby
`transmitters, and which are identical to large ghosts.
`OFDMcan beusedfor this application because it com-
`pletely eliminates the ISI due to ghosts within a given
`temporal spread. While conventional single-carrier sys-
`tems can also be used in this application, much more
`
`8
`sophisticated channel equalizers and antennas are re-
`quired. As a practical matter, no known channel-coding
`method except OFDM can perform adequately in a
`SEN.
`A System Employing SS and OFDM
`The basic idea is shown in FIG. 5. A video signal 199
`is input to a source coder 100, of a type that produces
`“digital” signals 201 that require nearly error-free trans-
`mission, together with “analog” signals 200 that can
`suffer some noise and/or distortion without producing
`catastrophic degradation of the reconstructed image.
`The analog signals are subjected to a SS process in SS
`processor 101, after which they are combined with the
`digital signals in adder 102 to produce hybrid symbols
`203. The latter are input to the OFDM encoder 103 to
`produce the output 204, which is the sum of a large
`number of modulated subcarriers located at equal inter-
`vals over a given bandwidth. The output 204 of OFDM
`coder 103 is at baseband,i.e., it is centered at zero fre-
`quency. The modulator 104 shifts the signal up to the
`desired radio frequency (RF) for transmission and also
`implements. the desired transmission constellation, as
`discussed below. An example of a suitable rf modulator
`and demodulator is shown in FIGS.5 and 6 of the ’021
`patent. Note that audio, sync, and data signals as appro-
`priate are multiplexed with the low-frequency digital
`video signals using techniques well known in theart.
`The receiver reverses these operations. The received
`signal 206 is shifted to baseband by demodulator 106. It
`is then reconverted to a hybrid symbol stream 208 by
`OFDM decoder 107. Quantizer 108 produces a quan-
`tized version 209 of hybrid symbol stream 208. Signal
`209 is the desired reconstructed digital data for the
`decoder 111, while the analog data 210 for the decoder
`is found by subtracting digital signal 209 from hybrid
`signal 208, and processing the difference 210’ by SS
`decoder 110. Source decoder 111 then produces video
`output 211 which is an approximation to the input 199.
`More detailed diagrams of the encoder and decoder
`are shown in FIGS. 6 and 7. The DFT implementation
`of OFDM coding and decoding is used. Like numbers
`are used in FIGS.5, 6, and 7 for like parts. These sys-
`tems deal with signals appropriate for quadrature ampli-
`tude modulation (QAM)as in the ’021 Patent so that
`in-phase (EZ) and quadrature-phase (Q) baseband signals
`are used to modulate the sine and cosine forms of the
`carriers. The digital and analog encoders 111 and 112
`perform the function of accepting real data at rate NR
`samples/sec and producing complex data at rate NR/2
`samples/sec. It is convenient to think of pairs of sym-
`bols a and b as the real and imaginary parts of complex
`symbols. The real part is the in-phase signal and the
`imaginary part is the quadrature-phase signal.
`In FIG.6, as in the simplified FIG. 5, input video 199
`is encoded by source coder 100 to produce analog and
`digital data streams 200 and 201 at rate NR samples/sec.
`Each N successive analog samples are grouped in N-bit
`words at rate R words/sec by serial-to-parallel con-
`verter 113 and are then input to the SS encoder 101.
`(Theserial-to-parallel conversion is unnecessary if the
`source coder produces data as N-bit words.) In SS unit
`101, each sample of each of the N bit streams is multi-
`plied by a different pseudorandom (PN) sequence and
`then added together with appropriate weighting, as
`shownin greater detail in FIG. 1, to produce an analog
`data stream 202 at rate NR samples/sec.
`The analog coder produces complex data 202’ at
`NR/2 samples/sec, while the digital coder produces
`
`5,425,050
`
`rary0
`
`20
`
`40
`
`45
`
`30
`
`60
`
`16
`
`16
`
`

`

`9
`complex digital data 201’ at NR/2 samples/sec. These
`two are added to produce a complex hybrid symbol
`stream 203 at the same rate. At this point, the value of
`N, which is the PN sequence Jength and also the num-
`ber of samples in each word in the SS processor, is not
`apparent from the cha