Le Floch, er 31.: Digital Sound Broadcasting to Mobile Receivers
`
`493
`
`DIGITAL SOUND BROADCASTING T0 MOBILE RECEIVERS
`
`Bernard Le Floch, Roselyne Halbert-Lassalle, Damien Castelain
`CCETT (Centre Commun d‘Eiudes de Télédiffusion et Telecommunications)
`35512 Cesson Sévigné, France
`
`ABSTRACT
`
`Many European countries have shown an increasing
`interest in the development of a new audio broadcasting
`service with a view to providing an improved sound quality
`on portable and mobile receivers.
`
`frequency—non-selectivity concerns only low bit«rate
`transmissions (a few tens of kbit/s) and can under no
`circumstances constitute a valid hypothesis for high-
`quality sound broadcasting.
`
`Digital techniques have progressed over the past few years
`in the areas of sound programme production and source bit
`rate reduction as well as in the field of channel coding and
`modulation. Within the framework of the DAB (Digital
`Audio Broadcasting) EUREKA 147 project, our research
`institute has designed a new broadcasting system, fully
`digital from studio to user.
`
`This paper deals with a promising modulation and channel
`coding system. suitable for digital broadcasting through
`the particularly hostile urban radio channel.
`
`A successful field demonstration has been organized under
`the auspices of the European Broadcasting Union during the
`WARC~ORB 88 in the town of Geneva,
`to validate these
`new concepts of digital sound broadcasting. combining
`spectrum and power efficiency.
`
`1. INTRODUCTION
`
`From the technical point of view. the development of such
`a new digital audio broadcasting system is related to a
`twofold problem :
`
`o
`
`-
`
`it requires the development of a sophisticated source
`coding system. It
`is generally admitted that.
`in this
`respect. a monophonic sound Willi a quality equivalent
`to that produced by compact disks can be reduced to
`128 kbit/s or less.
`
`it also requires the design of a channel modulation and
`coding system that can be used by mobile receivers up
`to 1 GHz or more.
`
`This article deals exclusively with this second problem.
`
`impulse
`the channel
`For reception in a built-up area.
`response usually extends over a few microseconds, and in
`some cases as much as 10 its or more. This being so.
`
`Given the high selectivity of the channel, conventional
`equalization techniques are very difficult
`to implement.
`More
`sophisticated techniques,
`such
`as Viterbi
`equalization, do not apply because of the excessively high
`number of states required in the Viterbi decoder.
`
`The methods generally adopted to resolve this type of
`problem are based on the use of M-orthogonal alphabets [l]
`(2] which increase the duration of symbols and allow for
`paths discrimination. Unfortunately intercorrelation
`constraints between the various alphabet sequences lead to
`a major increase in the bandwidth which is
`far
`from
`compensated by the increased number of bits transmitted
`per symbol. The spectral efficiency of this type of system
`is therefore totally incompatible with the broadcasting
`constraints.
`
`The system described in this article, combining spectrum
`and power efficiency, is mainly based on the conjunction of
`the Orthogonal Frequency Division Multiplexing technique
`(already proposed for HF data transmission in an
`ionospheric channel or
`for
`transmissions
`through
`telephone networks
`[3]
`[4]), and a coding strategy
`associated with diversity in the frequency domain.
`
`Section 2 gives a general overview of the urban mobile
`radio channel and its effects on data transmission. The
`general principles of the COFDM (Coded Orthogonal
`Frequency Division Multiplex) are explained in section 3.
`Section 4 gives a detailed representation of the signals and
`the decoding procedure. The performances of the system in
`the so-called selective Rayleigh channel are then presented
`in section 5. Section 6 describes the behaviour of the
`system when the conditions of temporal and frequential
`coherence of the channel are not fully met (very high speed
`reception and anormally extended channel delay spread). In
`section 7.
`the use of such a system in a broadcasting
`network is discussed. The realization aspects are finally
`presented in section 8.
`
`Manuscript received June 9, 1989
`
`0098 3063/89/0200 049350100 © 1989 [EEE
`
`APPLE 1012
`
`APPLE 1012
`
`1
`
`

`

`
`
`494
`
`IEEE Transactions on Consumer Electronics. Vol. 35, No. 3, AUGUST 1989
`
`the
`it can be shown that
`is large enough,
`sub—waves
`modulus of the term Ai(t) follows a Rayleigh distribution.
`The power spectrum of mm can be easily deduced from the
`distribution of the incidence of the sub-waves with respect
`to the direction of displacement of the vehicle. Assuming
`this distribution to be uniform and continuous over [at , rt],
`the power spectrum of the process Ai(t).
`translated to
`baseband and associated with each path,_ can be written :
`p.l
`
`v
`
`.
`
`v
`
`yAi(v)=
`
`if-Efu<v<:f°
`
`(gnu—v2
`
`f0 is the carrier frequency, v is the
`In this expression.
`vehicle speed and c is the velocity of the light.
`
`Moreover. it can be easily understood that the variations of
`the various Ai(t) are not correlated. because they arise from
`distinct origins.
`
`Considering the problem of data transmission through this
`type of channel, a large number of authors [6]
`[7]
`[8]
`pointed out that classical modulation systems (eg 4-PSK or
`FSK) lead to an irreductible bit error rate, even for large
`signal-to-noise ratio.
`caused by
`two independent
`phenomena :
`
`-
`
`'
`
`Selectivity. due to the spread of the channel impulse
`response. will cause intersymbol interference. as soon
`as the data rate tends to increase.
`
`Time variation of the channel characteristics. as a result
`of the changing environment of the receiver. will cause
`degradation in the phase estimation of the receiver. as
`soon as the data rate tends to be too low. for a given
`vehicle speed and carrier frequency.
`
`Even in the case where the channel is neither frequency-
`selective nor time—selective. the bit error rate as a function
`of the signal-to-noise ratio is known to decrease very
`slowly. More than 35 dB of Eb/No is needed for a 4-DPSK
`modulation scheme,
`to achieve a 10‘4 bit error rate (see
`section 6).
`
`This makes evident the need of sophisticated modulation
`and channel coding schemes for data broadcasting through
`the mobile urban channel.
`
`As it has already been said. frequency selectivity must be
`considered as an unavoidable phenomenon for high quality
`sound broadcasting. Moreover. the diversity provided by
`the use of wide-band transmission must be considered as an
`adVantage if the communication system is designed to make
`use of multipaths rather than be restricted by their presence.
`Because of the spread of the channel response, it is highly
`unlikely that fading would simultaneously affect a frequency
`band covering a few megahertz.
`
`The system described below exploits this property. In
`addition. it offers an excellent spectral efficiency, which is
`essential in sound broadcasting.
`
`2. DIGITAL TRANSMISSION IN THE URBAN
`RADIOMOBILE CHANNEL
`
`The broadcasting channel used for transmission to mobile
`receivers in a built-up area is a particularly hostile
`transmission environment.
`Industrial
`interference. and
`mainly multipath propagation caused by natural obstacles.
`necessitate
`the
`implementation of
`sophisticated
`modulation devices in order to provide an excellent—quality
`transmission. One of the major difficulties results from the
`constant evolution of the channel characteristics as the
`receiver is moving.
`
`Theoretical studies summarized in CCIR [5]
`established a twofold channel model :
`
`reports
`
`-
`
`-
`
`the first part estimates the value of the mean energy
`received in small areas (a few hundred wavelengths) ;
`
`the second part takes into account the combination of
`various paths originating from discrete reflections and
`received after scattering on objects (such as trees. other
`vehicles. etc...) that cannot be considered as mere
`reflective surfaces.
`
`With regard to the first part of this model. results of
`experimental research in an urban environment showed that
`the distribution of mean energy passed from one small area
`to another follows a log-normal law with a mean value
`related to free-space propagation.
`
`The second part of the model can be represented by the
`block diagram of figure 1 :
`
`Additive goussiun
`
`noise
`
`nltl
`
`
`E-iuicn l
`Reception
`
`
`
`tlI
`
`iAMlt]
`
`--
`
`r,“ _._.___..__1
`
`Wave of constant amplitude
`(my not exist)
`
`'___...._
`
`Figure 1
`Outline at the transmission channel
`
`specular
`The delays Ti originate from the various
`reflections. while the multiplying factors mm are the
`consequence of local scattering. If the number of scattered
`
`2
`
`

`

`Le Floch, et 211.: Digital Sound Broadcasting to Mobile Receivers
`
`495
`
`3.1 OFDM modulation technique
`
`The first principle consists in splitting the information to
`be transmitted into a large number N of modulated carriers
`with a low bit rate,
`in order to reduce the effect of the
`channel frequency selectivity. The OFDM technique permits
`the time—frequency domain to be split into small surfaces
`(fig. 2) with different dimension Ts and l/Ts on the time
`and the frequency axes respectively.
`
`The modulated OFDM signal can be expressed on an
`orthogonal base of elementary signals with complex
`coefficients, where each elementary signal is defined as one
`of the N emitted carriers during a symbol time T5, and each
`coefficient.
`taking its value from a finite alphabet.
`represents the modulation applied to each elementary
`signal.
`
`The main difference between conventional Frequency
`Division Multiplexing and this technique,
`is
`that
`the
`spectrum of the different carriers mutually overlap, giving
`therefore an optimum spectrum efficiency (asymptotically
`2 bills/Hz for a 4-PSK modulation of each carrier).
`Nevertheless,
`the signal verifies orthogonality conditions,
`so that it is possible to extract the information modulating
`each carrier without suffering interference due to the
`presence of the other carriers. Morever. the modulation and
`the demodulation processes can be undertaken by means of
`Fast Fourier Transform algorithms.
`
`Regarding the multiple paths of the transmission channel,
`the condition of perfect orthogonality between carriers is
`no longer maintained at the receiver input. due to residual
`intersymbol interference.
`
`This being so. a certain number of sound programmes (e.g.
`12 to 16) have to be multiplexed, forming a signal in which
`each basic data source will benefit from the "wide‘band"
`character of the transmission.
`
`3. GENERAL PRINCIPLES OF THE COFDM
`SYSTEM
`
`is
`it
`the above channel modelling.
`Taking into account
`the effects of the transmission by
`possible to represent
`combining the channel
`frequency response and time
`variation (fig.2). This
`two-dimensional
`function
`characterizes the so-called "selective Rayleigh channel".
`and admits a decomposition in surfaces of different sizes :
`
`.
`
`-
`
`frequency-time
`the
`represent
`surfaces
`the small
`areas where the channel can be considered as locally
`invariant ;
`
`the large surfaces indicate the minimum separation for
`which two small surfaces are statistically independent.
`
`Considering where the channel is invariant on the one hand
`and
`statistically independent on
`the other.
`this
`decomposition constitutes
`the basis of
`the channel
`modulation and coding method described in the following
`sections [9].
`
`
`Ivan
`y link
`
`/lllll[‘
`.
`independence
`frequent-y
`independence
`
`
`
`frequency
`.-
`
`Figure 2
`
`frequency
`channel
`response
`
`l l
` 1
`
`Symbol duration Ts
`
`useful
`
`period
`
`l
`guard
`interval 3
`
`i
`
`Elementary signal
`
`
`
`Channel
`
`impulse response
`
`Figure 3
`
`to suppress
`Use of a guard interval
`the intersymbol
`interference
`
`3
`
`

`

`
`
`496
`
`IEEE Transactions on Consumer Electronics, Vol. 35 . No. 3, AUGUST 1989
`
`An asymptotic solution to solve this problem would
`consist in increasing indefinitely the number of carriers and
`consequently the symbol duration. But this method is
`irrealistic taking into account the limitations dictated by
`the time coherence (see section 6) of the channel (Doppler
`effect). The retained solution comprises the addition of a
`safeguard interval
`before each useful symbol. If the
`safeguard interval duration is chosen sufficiently longer
`than the spread of the impulse response of the channel. the
`useful period of the signal remains free of intersymbol
`interference and the orthogonality remains perfect (fig. 3).
`
`3.2 Coding scheme
`
`Although the OFDM technique resolves the problem of
`transmission channel selectivity,
`it does not suppress
`fading. The amplitude of each carrier is usually affected by a
`Rayleigh law (or a Rice law when there is a direct path).
`This is why an efficient channel coding system is essential.
`
`The second principle of the COFDM system consists on
`linking by a coding procedure. elementary signals (small
`squares)
`transmitted at distant
`locations of the time-
`frequency domain (fig.2). This
`is
`achieved by
`convolutional coding (associated to soft decision Viterbi
`decoding)
`in conjunction with frequency and time
`interleaving. The interleaving depth is relative to the
`dimensions of
`the
`large surfaces of
`the channel
`representation.
`
`The diversity provided by interleaving plays a vital role in
`this system. The Viterbi decoder cannot function correctly
`unless successive samples presented at its input are affected
`by independent Rayleigh laws. In practice. the distortions
`to which these samples are subjected have a strong
`time/frequency correlation. When the receiver
`is not
`moving, the diversity in the frequency domain is sufficient
`to ensure that the system functions correctly. Due to the
`spread of the channel response (a few microseconds), flat
`fadings over a few megahertz are very unlikely. From this
`point of View,
`the existence of multipaths is a form of
`diversity and should be considered as an advantage.
`
`4. SIGNALS REPRESENTATION AND DECODING
`PROCESS
`
`The elementary modulation symbol transmitted during the
`time interval Ts = [5 +A is given by :
`
`x(t) = E Re (ckezm't)
`k=0
`
`te[0,T,]
`
`with
`
`fk = fo + k/ts
`
`where the equation parameters are the following :
`
`t5
`
`2
`
`useful symbol duration, on which the
`demodulation will be processed.
`
`A :
`
`guard interval duration.
`
`f0 :
`
`N :
`
`Ck :
`
`transmitting frequency.
`
`number of carriers of the multiplex.
`
`complex element of the modulation alphabet.
`
`is constituted by the juxtaposition in
`The emitted signal
`time of the elementary symbols defined above.
`
`The transmitted message fixes the Values of the modulation
`elements Ck. belonging to an alphabet that specifies the
`type of modulation.
`
`the alphabet corresponding to a 4-PSK
`For example.
`modulation is the following :
`
`{1+i,l-i.-l+i,-1-i]
`
`On the assumption that the guard interval is longer than the
`channel
`impulse response, and that
`the channel varies
`slowly as compared to the symbol duration (invariance of
`the channel on small surfaces),
`the elementary symbol
`received in the
`time interval
`free of
`intersymbol
`interference. can be expressed by :
`
`N-1
`y(l.)=2 Re Hkae
`km
`
`2mg:
`
`‘9k
`
`is the channel response at the frequency
`where Hk- pkel
`fk. The signal is translated in baseband by means of a local
`oscillator at the frequency f0 + 1/2 T where T 2 mm. The
`corresponding complex signal thus obtained and sampled at
`the rate l/T can be written :
`
`n N"
`k=o
`z(nT)=(—1). szCr e
`
`2mg
`N
`
`(n=0 to N-l)
`
`{ Hk Ck: k = {0...., N-l] }
`It appears that the set
`Discrete Fourier Transform of the set :
`
`is the
`
`{(.1)n z(nT)/N : n = {O.....N-1] }.
`
`the
`the FFT realizes in the discrete-time domain.
`In fact,
`equivalent of a bank of N filters. each of them being
`matched to a given carrier of the multiplex.
`
`receiver using
`The representation of the equiValent
`correlation is given in figure 4. The conditions of
`orthogonality mentioned above clearly appear in the fact
`that. at the input of the integrator of the row 1, the carriers
`fk (lc :- 1) exhibit an integer number of periods llc - 11 during
`the integration time.
`
`4
`
`

`

`Le Floch, et a1 .: Digital Sound Broadcasting to Mobile Receivers
`
`497
`
`The a posteriori maximum likelihood criterion consisu in
`maximizing with respect to the elements iCjJri linked by
`the convolutional coding, the expression :
`
`22H in-er Cir ii 2/ 20]}
`j
`k
`
`As far as the modulation and demodulation processes of
`each carrier are concerned. a 4-PSK modulation has been
`considered at the present time. In the first step of the study.
`other modulations with a larger number of states (8-PSK.
`l6-QAM....) have been discarded. taking into account their
`reduced power efficiency. Considering the coherent
`demodulation a 4-PSK-COFDM signal
`using
`a
`convolutional code of rate 1/2. and free distance 10.
`simulation results point out that a bit error rate of 10‘3 can
`be achieved for a mean value of the Eb/No ratio equal to
`5 dB in a selective Rayleigh channel. It means that the
`instantaneous value of this ratio can highly decrease
`leading to a particularly difficult implementation of the
`coherent demodulation.
`
`The differential demodulation is an alternative solution of
`which the essential
`interest resides in the very great
`simplicity of its implementation and its absence of inertia
`after a deep fade. The price to be paid is obviously a
`degradation in performance which nevertheless remains
`acceptable and which is in reality minimal
`if account is
`taken of the practical limitations of coherent demodulation.
`
`As far as the estimation of the channel frequency response
`is concerned. differential demodulation consists in using at
`the time j a simplified estimator of the channel deduced
`from the time j—l
`
`: “
`
`it
`
`” Yr 4:: / Cj-u
`
`Considering a preceding at the emission following the law,
`
`ajk+ibjk=il+iicjk/Cj-lkt
`
`ajk=tl.bj’k=t1
`
`the
`where a“ and bj‘k constitute the outputs of
`convolutional coder.
`it appears that
`the corresponding
`weightings to be used at the level of the Viterbi decoder
`are :
`
`ij Yr M
`Re —————2—-—
`(1-1}on
`
`Y- Y7.
`and Im -—"k—’lk
`(1-il oi:
`
`Moreover,
`
`these formula indicate that
`
`if the signal
`
`is
`
`the term 0'2”. in the
`affected by a narrow-band interferer,
`weightings has the effect of "deleting" the corresponding
`carriers in the same way as a fading of the same carriers.
`Taking into account that the spectral analysis of the noise
`is particularly simple with a COFDM system. this property
`makes this system extremely attractive for channels which
`are highly disturbed by noise or industrial interference.
`
`tWe? HG]“it‘ll lit-o in N[__‘
`
`LIT[in—{6361 v-—-—
`
`
`
`
`‘—-—‘TJ —2il‘lt,t
`
`-2il‘lt.
`
`>——~~-
`
`Second
`order
`““"mi
`
`Channel
`mtmng
`
`
`
`Figure 4
`
`Theoretically equivalent receiver
`using a bank of N correlators
`
`In absence of noise. the emitted symbols can be recognized
`without error. without taking into account the problem of
`Hk estimation that will be examined later. The phenomena
`of residual bit error rate due to the frequency selectivity
`have disappeared. Nevertheless, the decrease of the bit error
`rate as a function of the ratio Eb/No is extremely low in a
`Rayleigh channel. As previously said.
`this property
`justifies the use of an efficient channel coding system.
`
`By introducing the temporal dimension (index j) and the
`noise. the Fourier Transform process returns the samples :
`
`Yif ijcix+ Nu-
`
`where N”. is a complex gaussian noise, each component of
`which having a variance 02ij .
`
`5
`
`

`

`498
`
`IEEE Transactions on Consumer Electronics, Vol. 35, No. 3, AUGUST 1989
`
`distance separable (dmin = n — k + 1). The essential interest
`of these codes is that they do not require processing over
`the Galois field extension GF(28). but only over GF('2). The
`implementation of the decoding is considerably simplified.
`The
`codes
`that we
`considered
`have
`the
`form
`(n , n - 2t). where t is the number of corrected symbols ofj
`bits. For the simulations described hereafter, we have taken
`j=12,n=336andt=24.
`
`Accordingly. the performances given in figure 6 are those
`of a 4<PSK£OFDM system using a convolutional code of
`rate 1/2. constraint
`length 7, and free distance 10.
`eventually concatenated with a CSRS code (336. 288).
`
`BER
`
`4PSK - COFDM
`using coherent
`
`demodulation
`
`
`O : no coding
`
`A : convolutional
`+ algebraic
`coding
`
`: convolutional
`coding
`
`_
`: Gaussian channel
`
`.
`: Selective Rayleigh
`channel
`
`5. THEORETICAL RESULTS IN A SELECTIVE
`RAYLEIGH CHANNEL
`
`As far as the performance of the COFDM system is
`concerned. a wide choice of carrier modulations and channel
`coding parameters allows a trade-off between bit error rate
`performance and spectrum efficiency.
`
`For our application. a 4-PSK modulation scheme has been
`applied to each COFDM carrier. associated to a coherent or
`differential demodulation.
`
`curves may be
`Additionally. extremely abrupt error—ratio
`achieved by using an outer code of Reed Solomon type.
`concatenated with the inner convolutional code at the price
`of a slight reduction of spectrum efficiency.
`
`Accordingly. simulation results have been plotted in
`figure 5 in comparison with the Shannon limit.
`In
`particular. for different convolutional code rates lying from
`1/4 to 8/9 with constraint
`length 7. concatenated with
`Reed-Solomon codes of parameter (n,k). with n = 255 and
`= 211 to 243,
`the spectrum efficiency of the COFDM
`system lies from 0.4 to 1.5 useful bills/Hz and the
`necessary Eb/No ratio for a transmission without error
`varies from 3 to 12 dB respectively considering a coherent
`demodulation.
`
`l5
`
`(Intrusion rhlnnpl
`I: :
`R: Srlullvz “lylil‘ll
`rhlnnrl
`C : Cnhnnll dcmndulnllnn
`n : nm‘npnllnl drmndululnn
`
`
`
`
`
`t——————I——-~I—~——a——
`
`.4 ya
`.2
`/'
`_______.___
`
`SI lnHNON L [M l T
`R
`REb
`~~LOG<H~ —1
`H
`H “u
`2
`I—t—t-t—-¢——I-4
`
`7
`s
`5
`4
`3
`2
`srccmuu zrrrcrencv . am
`(m kuz)
`
`a
`
`-2
`
`Figure 5
`COFDM parlor mum!
`
`Although the concatenated codes previously described offer
`remarkable performance.
`their
`implementation in the
`context of the mass production domestic market is difficult
`because of the complexity of decoding the Reed-Solomon
`codes. An alternative solution consists in using CSRS
`codes (Cyclotomatically Shortened Reed-Solomon) [10]
`which are,
`like the Reed-Solomon codes. at a maximum
`
`
`
`
`
`_.,_-_._.__l__-..
`
`I
`i
`
`1
`1
`
`JPSK - COFDM
`.\
`.\
`l
`g
`;
`'l
`
`-\
`3
`using differential
`' \
`\
`.
`I
`AG '83 M \m ‘: demodulation
`i
`
`1
`l
`l
`i
`i
`13L—
`ZJASGTBEb/No
`(dB)
`
`Figure 6
`
`495K - COFDM performances
`
`6
`
`

`

`
`
`Le Floch, et 3.1.: Digital Sound Broadcasting to Mobile Receivers
`
`499
`
`In addition. by assuming that all angles of incident waves
`are uniformly distributed.
`the Doppler function can be
`given by:
`
`Pfdlfd) =
`
`1_
`7‘
`
`12
`rmax" fii
`
`for
`
`lfdl < lfmaxl
`
`where fmax is the maximum Doppler frequency and is given
`by fmax = fo.v/c (v is the speed of the vehicle).
`
`6.1 Frequency-selective
`
`fading
`
`The first characteristic of the multipath medium is the
`frequency variant aspect of the channel transfer function. It
`may be shown that the reciprocal of the delay spread Tm is a
`measure of the coherence bandwidth (ADC of the channel :
`
`(age :1 UT...
`
`is a
`The effect of the channel on the transmitted signal
`function of the choice of the signaling interval duration
`ts. If ts satisfies the condition ts >> Tm the channel
`introduces a negligible amount of intersymbol interference.
`This condition implies in the frequency domain that
`l/tS << (ADC and therefore that the channel is frequency-
`non—selective with respect to one modulated carrier of the
`COFDM signal.
`
`The effect of delay spread on the signal can be characterized
`by the parameter it = Tm/ts and the above condition
`becomes u. << 1.
`
`Computer simulations have been carried out to analyse the
`influence of it on the COFDM signal performance (fig. 8).
`with and without coding and taking into account a guard
`interval duration A equal to IS /4. In particular, the results
`point out the important gain provided by the coding.
`
`EFR
`
`6. TEMPORAL AND FREQUENTIAL COHERENCE
`OF THE CHANNEL
`
`This section intends to investigate the time and frequency
`coherence of the channel
`in relation to the transmitted
`signal. in order to determine the COFDM parameters l/tS
`and T5. These notions refer to the small surfaces of the
`channel representation (section 3) where the channel can be
`considered as locally invariant.
`
`The channel transfer response can be expressed in the tirne~
`frequency domain by :
`
`Him): 2 exp (j (pi) exp i-jZTlf‘til expijZitfdi t)
`i
`
`where (pi is the phase shift. Ti the propagation delay and fdi
`the Doppler shift for the im path.
`
`The channel is then characterized by the scattering function
`S (1. fd) which represents the power spectrum as a function
`of the time delay ‘1: and the Doppler frequency fd [11] [12].
`
`For simulation simplification. the random Variables fd and I
`are
`supposed to be statistically independent. This
`hypothesis leads to the scattering function of the form :
`
`s (x,fd]=K 15(1) Pfd[rd)
`
`where P1 ( ) and Pfd( ) are the probability density functions
`of I and fd. In most of the channel impulse measurements,
`the time delay function can be approximated by an
`1
`41TH,
`exponential distribution PrlTl =——e
`m
`
`where'I‘m is
`
`the standard deviation of the delays. namely the delay
`spread (fig. 7).
`
`Fifi)
`
`-30dB
`
`7.Tm
`
`t
`
`Figure 7
`
`Time delay density function
`
`ll-M
`
`ll-OI-
`
`Il-fl)
`
`
`
`runwalulls-II end:
`l-l/l [-1)
`‘(n
`
`Figure 3
`
`The performance of 4-PSK—COFDM in frequency-selective
`fading channels (differential demodulation and A = Isl-1)
`
`7
`
`

`

`
`
`500
`
`IEEE Transactions on Consumer Electronics, Vol. 35, No. 3, AUGUST 1989
`
`6 . 2 Tlme-selectlve
`
`fadlng
`
`The second characteristic of the multipath medium is the
`time-variant aspect of the channel transfer function.
`
`The coherence time of the channel can be given by
`(At)c~ l/fmax where fmax is the maximum Doppler
`frequency.
`
`The channel is called a slowly fading channel. or a time-
`non-selective channel. if the channel attenuation and phase
`shift are essentially fixed for the overall duration T5 of the
`signaling interval.
`
`This condition expressed by T5 << (At)c implies that B << 1
`where {3 is equal to fmax Ts and characterizes the effect of
`the rapidity of the fading on the signal.
`
`Simulation results. with and without coding, for different
`values of I?) are plotted in figure 9.
`
`BER
`
`|[»dle
`
`lE—OI-
`
`nut-nun «at
`-
`i
`I
`p
`)(I
`
`li—Olr
`
`
`
`
`I
`l
`‘I
`H
`I.
`II
`I.
`H
`ll
`I!
`i.
`[b/nu
`
`Figure 9
`The performance of 4-PSK-COFDM in lime-selective
`fading channels (differential demodulation)
`
`regards the
`The robustness of the COFDM signal as
`problem of time coherence ([3 = 0.04 corresponds to a
`vehicle speed of 600 k.p.h at 900 MHz for a symbol
`duration of 80 us). can be understood in the following
`manner. Considering the polar representation of a received
`pure frequency f0. as a function of the time (fig.10a). it is
`clear that the derivative of the phase of the received signal
`with respect to time (also called random FM). will highly
`tend to increase when the amplitude of the signal is closed
`to zero. This strong correlation between the modulus of the
`transfer function of the channel H(f°,t) and the random FM
`clearly appears in figure 10b and 10c. and it can be shown
`[7] that the value of the random FM is given by :
`
`211
`
`dt
`
`21:
`
`“H”:
`
`
`
`4.300:
`
`-l 300
`
`-Z.ODU
`
`~5.30fl
`
`3%:
`
`‘.300
`5.100
`
`Z.Jdfl
`
`(c)
`
`Figure 10
`Channel characteristics variation
`in the time domain at a given frequency
`
`1)
`
`a : Amplitude and phase variation at the given
`frequency Ifo (Polar representation)
`: Amplitude as a function of the time
`f0 = 900 MHz . V = 120 k.p.h
`e 2 Random FM as a function of the time
`same conditions as in b
`
`8
`
`

`

`Le Floch, et 3.1.: Digital Sound Broadcasting to Mobile Receivers
`
`501
`
`From this expression. one can see that. as the total energy
`appears in the denominator. large fluctuations of Fm(t) will
`occur near deep fadings.
`
`As it has been explained in section 4. the weightings of the
`Viterbi decoder are the real and the imaginary parts of
`
`Yj’k.Y*J-_1, k /(1-i) 61-32. This means that the soft decision
`channel decoder
`takes into account
`the effect of the
`
`attenuation at each time-frequency point (9131(2)‘ in such a
`way that the carriers affected by deep fadings. that are
`supposed to suffer
`important phase shifts. do not
`participate in a large extend to the maximum likelihood
`decoding process.
`
`6.3 COFDM parameters choice
`
`The conditions u << 1 and [3 << 1 imply that {mu 'l'm << 1.
`The spread factor defined as fmax Tm . or [(At)c (Am; 1".
`depends only on the channel. and the COFDM system in the
`described version is effective for channels with a spread
`factor much less than one.
`
`In fact. the parameters 1/ts and T5 of the COFDM system
`must be chosen to verify approximatively the two
`conditions (considering A = [5/4) :
`
`0.
`
` 1
`(or 1/t,<1mm)
`u< l
` l
`B < 0.02 (or Ts< 5mm)
`
`the spread factor of the channel must be less than
`Then.
`1.6 103. and leads to define the conditions of application
`of the adaptable CODFM system (fig. 11).
`
`C,(GHz)
`
`100
`
`10
`
`
`
`10
`
`100
`
`v(k.p.h)
`
`1000
`
`Figure 11
`
`Maximum carrier frequency as a function
`of the velocity of the receiver
`
`its
`let us suppose a delay spread of 1
`As an example.
`(typical case for suburban and urban areas) and a maximum
`Doppler shift of 100 Hz (which can be obtained with
`fo = 900 MHz and v = 120 k.p.h.). In this case. the spread
`factor
`is equal
`to 10" and the conditions become
`12.5 us < T; < 200 us.
`
`7. A DIGITAL SOUND BROADCASTING
`NETWORK
`
`The main difficulties that arise during the introduction of a
`new broadcasting service are caused by the lack of
`frequencies in the radio-electrical spectrum. This spectrum
`belongs jointly to all potential users and must be used to
`the full. This is why, in addition to efficiency in terms of
`power and spectrum usage. flexible frequency planning
`must be taken into account.
`
`and satellite
`research covers both terrestrial
`Our
`broadcasting and at present. we are looking into the
`following configurations:
`
`-
`
`~
`
`-
`
`a local urban broadcasting service in the UHF band
`using 4 to 7 MHz bandwidth. that can be shared with
`television signals. The COFDM system requires very
`low power and it is therefore possible to re-use TV
`channels operating in adjacent areas.
`
`a regional or national broadcasting service operating in
`the 60-200 MHz band based on the implementation of a
`single-frequency network using basic 4 MHz bands.
`Since the COFDM system is designed to function in the
`presence of multipaths.
`it
`is possible to generate
`«active echoes » from a network of
`transmitters
`spread throughout a given territory. They would be
`temporally synchronized and would all
`transmit
`the
`same
`signal. The
`spread of
`the
`« equivalent
`transmission channel» impulse response is. of course.
`dependent on the physical distance separating the
`various transmitters. It
`is therefore necessary to use
`very long COFDM symbols (=1 ms). with a guard
`interval capable of absorbing echoes from over 70
`kilometers away. Increasing the size of the symbols
`clearly results in an increased number of carriers
`transmitted in a given bandwidth. This apparent
`constraint in fact corresponds to a slight increase in
`receiver complexity.
`
`a national satellite broadcasting service operating on
`the 0.5 - 2 GHz frequency band based on the utilization
`of basic 4 MHz bands. This possibility. which requires
`the exclusive allocation of
`a
`frequency band
`nevertheless includes all the well-known advantages of
`satellite broadcasting [l3].
`
`8. SYSTEM DEVELOPMENT
`
`the modulation and coding
`8.1 Choice of
`parameters
`
`The first real system was developed with the aim of
`validating the principles of the COFDM system in the case
`
`9
`
`

`

`
`
`502
`
`IEEE Transactions on Consumer Electronics, Vol. 35, No. 3, AUGUST 1989
`
`It provides 16 stereophonic
`of UHF transmission [14].
`programmes. each with a rate of 336 kbitls.
`in a total
`bandwidth of 7 MHZ.
`
`The various programmes are multiplexed in such a way as to
`minimize receiver complexity. As
`the receiver has to
`process only one programme at a time.
`the coding and
`modulation procedures have to be applied separately to the
`various
`sources. The multiplexing of
`the different
`programmes cart be done in the frequency or in the time
`domain. In the case of frequency multiplexing. programme
`selection and demodulation can be performed jointly by
`decimating the FFI‘. Time multiplexing. which is more
`simple to implement, was retained for this first system.
`
`The basic symbol has a duration of 80 tts and includes a
`guard interval of 16 tts. Each symbol constitutes a
`multiplex of 448 carriers separated by 15625 Hz. modulated
`separately in 4—PSK. The guard interval absorbs the
`multipaths in most situations. Moreover.
`the modulation
`symbols are sufficiently short to provide temporal channel
`coherence as regards the incoming signal. oven at 200 kph
`and a carrier frequency of 1 GHz. This is a vital condition
`for demodulator operation, whether differential or coherent.
`
`is built up around a 24 ms frame structure
`The signal
`corresponding to the juxtaposition of 300 symbols. The
`first symbol. which is free of modulation.
`is used to
`synchronize th