1752
`
`IEEE TRANSACTIONS O N COMMUNICATIONS, VOL. COM-27, NO. 12, DECEMBER 1979
`
`A Comparison of Modulation Techniques for Digital Radio
`
`JOHN D. OETTING. MEMBER. IEEE
`
`Absmct-This paper describes and summarizes the characteristics of
`the modulation techniques most applicable to digital radio. The modulation
`techniques discussed are on-off-keying (OOK)
`with coherent and
`noncoherent detection.
`quadrature amplitude modulation
`(QAM),
`quadrature partial response (QPR), frequency-shift-keying (FsK) with
`noncoherent detection, continuous phase FSK (CP-FSK) with coherent and
`(MSK), binary and
`noncoherent detection, minimum-shift-keying
`quaternary phase-shift-keying
`(BPSK, QPSK) with coherent and
`differentially coherent detection, offset-keyed QpSK (OK-QpSK), M-ary
`(M = 8, 16), and 16-ary amplitude and
`PSK with coherent detection
`phase-shift-keying (APK). Functional descriptions of these schemes are
`provided and their performance
`is compared
`in a series of
`tables
`summarizing the results of
`the literature
`of the past 20
`years. The
`modulation schemes are compared with respect to ideal (white Gaussian
`noise) performance, spectral properties, signaling speed, complexity, and
`the effects on performance of interference, fading and delay distortion.
`
`T
`
`I. INTRODUCTION
`HE crowded conditions prevailing in many regions of the
`radio spectrum combined with the increased emphasis on
`digital transmission have created a need for improved spectrum
`utilization techniques. The intelligent application of efficient
`digital modulation techniques provides one means of achieving
`improved spectral efficiency at a reasonable cost. This survey
`identifies and describes some of the more important modula-
`tion schemes applicable
`to digital radio, including some that
`have been developed in recent
`years and are not discussed in
`the classic textbooks. This paper should be of interest to any-
`one involved in communications planning and communications
`system engineering.
`to
`Much previously published material has been devoted
`comparing the performance of digital modulation methods. In
`many cases, comparisons were limited to one performance cri-
`terion (e.g., performance in a restricted band [26], spectral
`occupancy [76] , effects of phase distortion [ 1161 , or cross-
`talk [118, 1151). In other cases [77,98, 165, 1921 ,relatively
`few modulation schemes were considered. This paper presents
`an up-to-date comparison of the modulation methods of par-
`ticular applicability
`to digital radio channels, and provides an
`indexed bibliography for readers interested in greater depth.
`Section I1 provides functional descriptions of the modula-
`tion schemes considered to be representative of the techniques
`appropriate for digital radio applications. Section 111 summar-
`izes the performance characteristics
`of
`the representative
`modulation schemes by means of a series of tables together
`with source references where
`complete detaiirmn be found.
`
`respect to ideal
`The modulation schemes are compared with
`performance, spectral properties, signaling speed, complexity
`and the effects on
`performance of interference, fading and
`delay distortion.
`
`11. DESCRIPTIONS OF THE REPRESENTATIVE
`MODULATION SCHEMES
`
`There are. three basic modulation techniques: amplitude
`modulation (AM), frequency modulation (FM), and phase
`modulation (Ph4). Each of these basic techniques has a large
`number of variants, the most relevant of which will be dis-
`cussed briefly in this section. In recent years hybrid schemes
`(e.g., amplitude-and-phase-shift-keying-APK) have received
`increased attention because of their inherent economical use
`of bandwidth. Therefore, 16-ary APK is included as being rep-
`resentative of this large class of modulation techniques [ 1771 .
`The primary components of a system for transmitting digital
`data over a radio channel are illustrated in Figure 1. All of the
`digital modulation schemes discussed in this paper can be con-
`ceptualized in terms of radio frequency sinusoids (carriers)
`modulated by
`low frequency (baseband modulation) signals
`that convey the digital information. These baseband modula-
`tion signals may be filtered, weighted, or otherwise shaped
`prior to modulating the carrier in order to achieve desirable
`results. At the receiver, the baseband information is recovered
`by a detection process. Coherent detection requires a sinusoidal
`reference signal perfectly matched
`in both frequency and
`phase to the received carrier. This phase reference may be ob-
`tained either from a transmitted pilot tone or from the modu-
`lated signal itself. Noncoherent detection, being based on
`waveform characteristics independent of phase (e.g., energy or
`frequency) does not require a phase reference:
`Usually, detection is followed by a decision process that
`converts the recovered baseband modulation signal into a se-
`quence of digital bits. This process requires bit synchroniza-
`tion, which is generally extracted from the received waveform.
`With most modulation schemes, decisions can be made on a
`bit-by-bit basis with no loss in performance, but with some
`schemes an advantage can be gained by examining the signal
`over several bit intervals prior to making each bit decision.
`The portion of the received waveform examined by the de-
`cision device in making a single bit decision is called the ob-
`servation interval.
`
`16, 1979; revised July 17, 1979. This
`Manuscript received January
`work was supported in part by the Naval Research Laboratory under
`Contract N00173-76-C-0253.
`The author is with the Communications and Information Technology
`Division. Booz.Allen & Hamilton, Inc., Bethesda, MD 20014.
`
`Amplitude Modulation (AM) Techniques
`
`The simplest digital AM technique is double sideband (DSB)
`AM [79, pp. 173 ff] modulated
`by a binary signal. The double
`
`0090-6778/79/1200-1752$00.75 0 1979 IEEE
`
`1
`
`TMO1016
`
`

`
`
`
`
`
` OETTING: MODULATION TECHNIQUES
`
`
`
`
`
`1753
`
`When m ~ ( t ) is the Hilbert transform of mI(t), QAM reduces to
`
`SSB. When mI(t) and mQ(t) are independent binary data sig-
`nals, QAM is as efficient in required power and bandwidth as
`ideal SSB without the stringent filtering requirements. If mI(t)
`and ms(t) are three-level duobinary signals (+l, -1 or 0 )
`coded in such a way as to minimize intersymbol interference
`caused by filtering, the result is quadrature partial response
`for use in the Canadian 8
`(QPR), which has been proposed
`GHz frequency band [ 1781 . All of these techniques require
`coherent detection. Any phase tracking errors that occur result
`in interference between theIand Q channels, thereby degrading
`performance.
`When m,(t) and mQ(t) take on the values *I, QAM is iden-
`tical to quaternary phase-shift-keying (QPSK) discussed in the
`section on phase modulation techniques. These two techniques
`will differ, hqwever, when mI(t) and mQ(t) are not rectangular
`pulses.2
`
`Frequency Modulation (FM) Techniques
`The simplest FM technique is frequency-shift-keying (FSK)
`involving binary signaling by the use of two frequencies separ-
`ated by Af Hz, where Af, the frequency deviation,
`is small
`compared to the carrier frequency, f , . With FSK schemes, it is
`common practice to specify the frequency spacing in terms of
`the modulation index, d , defined as:
`
`OEMOOULATOR
`
`L - - - _ - _ _ _ J
`Figure I : Primary Components of a Digital Radio System
`
`sideband waveform is represented as:
`fDsB(t) = - [ l + m(t)] COS o,t
`A
`2
`
`(1)
`
`where m(t) is the modulating signal and w, is the carrier fre-
`quency (in radians per second). For the case of 100 percent
`modulation by a non-return-to-zero (NRZ) binary data wave-
`form (m(t) = kl), we have on-off-keying (OOK) modulation.
`Such an OOK waveform can be detected either coherently or
`noncoherently, but the
`difference in performance is slight
`compared to the required increase in complexity so that co-
`herent detection of OOK is not employed over radio channels.
`The properties of the DSB waveform can be modified by re-
`placing the NRZ m(t) by some other type of binary baseband
`signall (e.g., the “partial response” signal [18, 20, 341).
`Since the carrier conveys no information, efficiency can be
`improved by the use of double sideband suppressed carrier
`(DSB-SC) AM. The general form of the DSB-SC signal is:
`
`fsc(t) = A m(t) COS a c t .
`
`(2)
`
`d = A f T
`
`(5)
`
`For the case where m(t) takes on the values 0 and 1, we have
`the OOK situation described in the previous paragraph. When
`m(t) takes on the values -1 and 1, we have the case of binary
`phase-shift-keying (PSK), which will be discussed under PM
`techniques.
`Both DSB techniques involve the transmission of a redun-
`dant sideband. For applications in which spectral efficiency
`is
`important, the occupied bandwidth can be reduced by a factor
`of two by the use of single sideband (SSB) modulation. The
`SSB signal can be written as:
`f S S B ( t ) = A[m(t) cos w,t + m( t ) sin oct]
`
`(3 )
`
`where i ( t ) is the Hilbert transform [79, p. 311 of m(t). In
`practice, SSB signals are usually generated by the use of a
`to suppress the upper or lower sideband. The
`bandpass filter
`sharp cutoff characteristic required for the bandpass filter pre-
`sents implementation problems. Thus, a bandpass filter with
`smooth roll-off is often used. This procedure results in a vestig-
`ial sideband (VSB) signal [79, p. 1921.
`(QAM) is yet another
`Quadrature Amplitude Modulation
`AM alternative. This technique involves summing two DSB-SC
`signals 90” apart in phase as follows:
`fQAM(t) = A [ m I ( t ) cos a c t + m&) sin a c t ] .
`
`(4)
`
`1 Radios designed for analog AM transmission are sometimes used to
`transmit digital signals. These radios distort the digital signal through in-
`efficient high frequency or low frequency signal response. Occasionally,
`other distortions are deliberately introduced to improve spectral per-
`formance at the expense of communications efficiency.
`
`where T is the symbol duration (equal to the inverse of the
`data rate for binary schemes).
`schemes, FSK can be detected
`As with other modulation
`either coherently or noncoherently. Noncoherent detection
`can be effected by two bandpass filters followed by envelope
`detectors and a decision device [79, p. 2971. With this ap-
`proach, the frequency spacing must be at least l / T ( d > 1)to
`prevent significant overlap of the passbands of the two filters.
`Alternately, a discriminator can be used to convert the fre-
`quency variations to amplitude variations, so that AM enve-
`lope detection can be employed [82]. This approach elimi-
`nates the above constraint on d.
`Recently, considerable interest has arisen in modified ver-
`sions of FSK, including some coherent schemes. These schemes
`are based on the idea of continuous phase FSK (CP-FSK), in
`which the abrupt phase changes at the bit transition instants
`characteristic of other FSK implementations are avoided. This
`implementation of FSK results in rapid spectral roll-off and
`improved efficiency. The improvement is attained by the use
`of observation intervals greater than one bit [ 1,401 . This fea-
`ture enables narrower filter bandwidths than would otherwise
`be feasible. With coherent detection, values of d in the neigh-
`borhood of 0.7 have been shown to provide optimal perform-
`ance for any observation interval [ 1231 .
`Another FM technique that has received considerable in-
`terest in recent years is minimum-shift-keying (MSK), also
`called fast frequency-shift-keying. MSK is a special case of
`
`2 Nonrectangular pulses can be employed to eliminate abrupt transi-
`tions in the modulated waveform, thereby improving spectral character-
`istics and easing transmitter implementation.
`
`2
`
`

`
`1754
`
`
`
`
`
`
`
`
`
`IEEE TRANSACTIONS ON COMMUNICATIONS.
`
`VOL. COM-27, NO. 12, DECEMBER 1979
`
`0
`
`0
`
`0
`
`0
`
`0
`
`0
`
`0
`
`0
`
`0
`
`0
`
`LOCATION OF I th SIGNAL
`
`0
`
`0
`
`0
`
`0
`
`CP-FSK for which d = 0.5 and coherent detection is used. This
`technique achieves performance identical to coherent PSK and
`exhibits the superior spectral properties of CP-FSK. MSK has
`the additional advantage of the possibility of a relatively sim-
`ple self-synchronizing implementation [59] , an advantage that
`coherent CP-FSK with d = 0.7 does not share.
`
`Phase Modulation (PM) Techniques
`Almost by definition, digital PM schemes require coherent
`detection. There are three basic variations of binary phase-shift-
`keying (BPSK). The most straightforward approach is coherent
`BPSK, in which the carrier phase is shifted by 0 or 180 degrees.
`Detection requires a precise phase reference, which is normally
`obtained by performing a nonlinear operation on the received
`waveform. Since some phase reference extraction techniques
`exhibit 180" phase ambigbities, a modified form of PSK called
`Differentially Encoded PSk (DE-PSK) is often used. With DE-
`PSK, information is conveyed via transitions in carrier phase
`(e.g., no transition may correspond to a
`space and a 180"
`transition may correspond to a mark). Since a bit decision er-
`ror on the current bit
`will induce another error on the sub-
`sequent bit, the performance of DE-PSK is slightly inferior to
`that of coherent PSK.
`The third version of binary PSK is Differential PSK (DPSK),
`in which, as with DE-PSK, the information is differentially en-
`coded. The difference between DPSK and DE-PSK lies in the
`detector. With DPSK, no attempt is made to extract a coherent
`phase reference. Rather, the signal from the previous bit inter-
`val is used as a phase reference for the current bit
`interval.
`Since the phase reference signal is not smoothed over many bit
`intervals, the performance of DPSK is somewhat worse than
`that of DE-PSK.
`Quaternary PSK (QPSK) schemes will also be considered.
`Coherent QPSK involves encoding two bits at a time into one
`of four possible carrier phases spaced 90" apart. As in the bi-
`nary case,
`the data can be differentially encoded and differ-
`entially detected with a concomitant loss in performance (this
`scheme will be denoted DQPSIC). In recent years, a modified
`version of QPSK,
`called offset-keyed QPSK (OK-QPSK) or
`staggered QPSK (SQPSK), has come into use. This scheme
`offers advantages over conventional QPSK with regard to spec-
`tral efficiency, sideband regeneration and synchronization [45,
`50,94, 1441.
`the QPSK sig-
`OK-QPSK can be visualized by considering
`nal to consist of in-phase and quadrature components (as with
`QAM). With normal QPSK, during each 2 bit time interval of T
`seconds, the I carrier is binary PSK modulated by one bit and
`resulting sig-
`the Q carrier is modulated by the other bit. The
`nal can
`take on any one
`of four possible phases, and abrupt
`phase transitions of 0", 90", or 180" can occur. With OK-
`QPSK, the Q channel is shifted by T/2 seconds with respect to
`the I channel. The transition rules are designed so that when
`the I and Q channels are added together, the resulting signal
`can shift abruptly by 90" at most (but shifts can occur every
`T/2 seconds, compared to every T seconds for standard QPSK).
`
`S i ( d = a i cos Iwt + E i l
`
`It = 1,
`
`. ., 161
`
`Figure 2: Location of 16-ary APK Waveforms in Phase-Amplitude Space
`
`amplitude and phase-shift-keying (APK) [ 1771 . The resulting
`information signals are best visualized by a representation in a
`phase-amplitude signal space. Figure 2 shows the signal space
`location of each of
`the 16 possible transmitted signals for 16-
`ary APK. Because this scheme conveys four bits of informa-
`tion during each signaling interval, it has been proposed
`for
`use over digital radio channels [ 1761 .
`111. COMPARISON OF REPRESENTATIVE MODULATION
`SCHEMES
`In this section, the modulation
`schemes described in Sec-
`tion I1 are compared with respect to their performance under
`a variety of conditions characteristic of digital radio channels.
`A performance measure utilized throughout this section is the
`baseband equivalent EB/No (defined as the ratio of average
`signal energy per bit to noise power spectral density, as meas-
`ured at the input to the receiver) required to achieve a bit er-
`ror rate of
`lop4. This error rate is adequate for most general
`purpose digital radio applications.
`
`Ideal Performance
`In order to establish a baseline for comparison, Table 1 pr.e-
`sents the ideal performance of the representative modulation
`techniques in the presence of additive white Gaussian noise.
`References are included with each entry to facilitate obtain-
`ing values of required EB/No for other error rates3 Of par-
`ticular interest is the fact that CP-FSK with coherent detec-
`tion over a 3-bit observation interval can outperform BPSK
`and other equivalent techniques (which are optimal only when
`the observation interval is confined to one bit). The identical
`ideal performance of QAM, MSK, and QPSK attests to their
`[ 17,
`underlying similarities, often discussed in the literature
`601. Indeed, offset-keyed QAM [45], MSK, and OK-QPSK
`differ only in the weighting functions applied to the I and Q
`channels.
`
`Hybrid AM/PM Techniques
`The ever increasing need for bandwidth conservation has
`led to the use of a class of hybrid AM/PM techniques called
`
`3 Care must he exercised to convert various SNR measures to E,/No
`and to convert symbol error rates to bit error
`rates. Also, since average
`EB/No is used throughout, the values shown for theOOK schemes are
`always 3 dB helow the peak values given in many references.
`
`3
`
`

`
`OETTING: MODULATION TECHNIQUES FOR DIGITAL RADIO
`
`1755
`
`TABLE 1
`IDEAL PERFORMANCE OF REPRESENTATIVE MODULATION
`SCHEMES
`
`TABLE 2
`RELATIVE SiGNALING SPEEDS OF REPRESENTATIVE
`MODULATION SCHEMES
`
`TYPE
`
`MODULATION SCHEME
`
`SPEED
`lb/r PER Hz)
`
`IdBI’
`
`REFERENCE E8INO
`
`AM
`
`FM
`
`PM
`
`~
`
`OOK
`COHERENT DETECTION
`OOK - ENVELOPE DETECTION
`
`OAM
`
`OPR
`
`~
`
`CP FSK
`-
`CP-FSK
`
`~
`
`I d = 11
`
`COHERENT DETECTION
`Id = .71
`
`NONCOHERENT DETECTION
`Id = .71
`
`MSK I d = .5l
`MSK - DIFFERENTIAL ENCODING I d = .51
`
`BPSK
`
`~
`
`COHERENT DETECTION
`
`DE-BPSK
`I DPSK
`QPSK
`
`DOPSK
`I OK-OPSK
`
`8-ary PSK
`
`~
`
`COHERENT DETECTION
`
`16-ary PSK -COHERENT DETECTION
`
`AMIPM
`
`l 6 ~ a r y APK
`
`. FOR BIT
`
`lo4
` ERROR RATE OF
`
`
`
`t CALCULATED FROM RESULTS FOR BPSK
`* * DISCRIMINATOR DETECTION
`
`,
`
`I
`I
`
`I
`
`0.8
`
`1.7
`
`2.25
`
`1.0
`
`1.9
`
`1.9
`
`0.8
`
`0.8
`
`1.9
`
`2.9
`
`3.1
`
`12.5
`
`9.5
`
`11.7
`
`26
`
`145
`
`178
`
`11.8” 0.8 FSK- NONCOHERENT DETECTION
`64
`
`10.7
`
`9.4
`
`10.4
`
`9.4
`
`.
`
`9.9 0.8
`
`10.6
`
`9.9
`
`I
`I
`
`11.8 1.6
`I
`
`I
`
`1
`I
`
`I
`
`26
`
`44
`
`44
`
`26
`
`26 t
`
`R7
`
`57
`
`165
`
`I
`
`I
`
`57 12.8 2.6
`
`17.2
`
`13.4
`
`
`d = FM MODULATION INDEX
`
`
`
`176
`
`176
`
`MiPM I 16-ary APK
`* FOR BITERROR R A T E O F l o 4
`ASSUMESTHREE-BITOBSERVATION INTERVAL
`
`I
`
`189
`
`12.4
`
`d = FM MODULATION INDEX
`
`Spectral Characteristics
`R is the data rate and W is the IF bandwidth) is an important
`figure of merit. In Table 2, we list the speeds for each technique,
`the modulation schemes can
`The spectral characteristics of
`EB/No required for a
`together with the
`be compared in many ways. Of particular interest is the extent
`BER when the
`to which a signal will interfere with signals in adjacent channels.
`signal is filtered at the indicated bandwidth (i.e., the degrading
`One measure of this
`quality is the attenuation of a
`signal’s
`effects of finite bandwidth are included). The results presented
`in Table 2 were derived from many different sources, so that
`power spectrum a specified distance from the center frequency.
`If, for example, we examhe the attenuation at an arbitrary
`slightly different filters5 were used
`in obtaining the results,
`distance of 8 / T Hz from the center frequency (Tis the symbol
`but these figures are indicative of the results to be expected. It
`duration), we find that with AM schemes the sidelobes are
`should also be noted that, in most cases, no rigorous attempt
`25 dB, with PM schemes the sidelobes are
`down by about
`was made to achieve the optimum combination of speed and
`down by about 33 dB, and with continuous phase FM the side-
`efficiency.
`lobes are down by 60 dB or more.4 In general, for frequencies
`far from the center frequency (large ( f - f,.)T),
`the spectrum
`of AM and PM signals falls off a s p 2 while that of CP-FM sig-
`nals falls off as f-4 ..
`While these numbers appear to indicate a significant advan-
`tage for FM schemes, they should be put in the proper per-
`spective. First, the figures quoted for PM systems assume that
`abrupt phase transitions occur.
`If phase transitions
`can be
`made to occur more smoothly, improved spectral characteris-
`tics can be achieved. Second, modifications can be made
`to
`AM schemes. For example, Shehadeh and
`Chiu [I051 show
`that a type of continuous phase AM exhibits a spectrum that
`falls off as f P 4 . Finally, sidelobes can always be reduced by
`suitable. post modulation filtering, although a penalty
`in per-
`formance is incurred. Thus, the spectral merits of
`the various
`schemes can only be judged on the basis of a detaiied study of
`the tradeoffs between cost and performance.
`Another spectral property of interest is the bandwidth re-
`so-
`quired to transmit at a
`specified information rate. The
`called “speed” of a modulation technique (equal to R/W, where
`
`Effects of Interference
`Another factor in evaluating potential modulation schemes
`for digital radio is the effects of co-channel and adjacent chan-
`nel interference. We have already discussed one aspect of adja-
`cent channel interference-the out-of-band
`attenuation of the
`various schemes. It was pointed out that MSK and the other
`CP-FSK scheines enjoy a large advantage over
`the AM and PM
`schemes, when
`no post-modulation filtering
`is employed.
`Another aspect of adjacent channel interference is the amount
`of performance degradation caused by a specified level of in-
`terference. Table 3 illustrates the effect of an in-band CW in-
`terferer (10 dB or 15 dB down in power from the desired sig-
`nal) on the EB/No required for a
`lop4 BER. This situation
`can be a model either for interfering sidelobes from adjacent
`channels or for the main lobe from a co-channel interferer. Of
`the schemes for which data are available, noncoherent FSK
`and BPSK show the least degradation from ideal performance,
`while the 8-ary and 16-ary schemes exhibit the most degrada-
`tion (compare
`to Table 1). Unfortunately, no analytical or
`
`4 A modified version of MSK known as sinusoidal FSK (SFSK)
`achieves an attenuation of 95 dB I15 I .
`
`5 In most cases, a simple three-pole filter
`used.
`
`or a Gaussian filter was
`
`4
`
`

`
`1756
`
`
`
`
`
`
`
`
`
`IEEE TRANSACTIONS ON COMMUNICATIONS,
`
`VOL. COM-27, NO. 12, DECEMBER 1979
`
`TABLE 3 ,
`PERFORMANCE OF REPRESENTATIVE MODULATION SCHEMES
`IN THE PRESENCE OF cw INTERFERENCE
`
`TYPE
`
`MODULATION SCHEME
`
` DETECTION
`
`AM
`
`
`
`ENVELOPE DETECTION
`
`~
`
`I CP-FSK - NONCOHERENT DETECTION
`
`Id = .71
`
`FM
`
`I
`
`I
`
`.
`
`
`
`I
`
`FM
`
`-
`
`10.5
`
`9.2
`
`I FSK
`
`OPR
`
`~
`
`NONCOHERENT DETECTION
`id =11
`CP-FSK - COHERENT DETECTION
`r d = .71
`-
`CP-FSK- NONCOHERENTDETECTION
`Id = .71
`
`14
`
`1
`
`2
`
`0
`
`1
`
`f
`
`f
`
`1
`
`1
`
`
`
`13"'
`
`18".
`
`174
`
`174
`
`TABLE 4
`PERFORMANCE O E REPRESENTATIVE MODULATION SCHEMES
`ON A RAYLEIGH FADING CHANNEL
`I
`I OOK - COHERENT
`I OOK
`OA M
`
`AVERAGE
`
` IdBI'
`
`I EnINn
`I
`I
`I
`
`17'
`
`19'
`
`I REFERENCE I
`I
`1
`I
`I
`I
`
`79
`
`79
`
`MSK Id = .5l
`MSK - DIFFERENTIAL ENCODING Id = .5)
`BPSK - COHERENT DETECTION
`DE-BPSK
`
`PM
`
`DPSK
`
`OPSK
`
`DOPSK
`
`OK-OPSK
`
`8-ary PSK - COHERENT DETECTION
`l6ary PSK - COHERENT DETECTION
`
`AMIPM
`
`l 6 a r y APK
`
`FOR B I T E R R O R RATE OF
`
`11.0
`
`12.0
`
`12.2
`> 20
`
`- 20
`
`9.7
`10.3
`
`9.8
`
`14 0
`
`15.8
`> 24
`
`147, 191
`
`147
`
`147
`
`147. 191
`
`147
`
`191
`
`191
`
`
`
`
`
`d = FM MODULATION INDEX
`
`simulation results are available for any of the CP-FSK schemes
`or for OK-QPSK.
`
`Effects o f Fading
`on digital
`Fading is another problem often encountered
`radio links.
`If the fading is caused by two resolvable multi-
`path components, then the
`results of .Table 3 can be utilized
`(the CW interferer can represent the signal from the secondary
`path). If the fading is caused by a large, number .of equal ampli-
`tude components, the Rayleigh fading model [79, p. 3481 is
`more appropriate.6 Table 4 presents the performance results
`for a Rayleigh fading channel. Because of the severe effects of
`Rayleigh fading, a required bit error
`is assumed
`rate of
`in this table (although this error rate is rather high for digital
`radio applications, error
`control coding could be used to
`achieve the desired ;ate). Because thk values of Table 4 repre-
`sent simply a weighted average of the ideal performance curves,
`the 'relative
`performance of the schemes does not differ
`markedly from that
`indicated by Table 1. It should be em-
`phasized that for the AM and AM/PM schemes, the fading is
`assumed to be slow. enough that tde decision threshold can be
`continuously adjusted to the optimai value.
`
`Effects of Delay Distortion
`in selecting a
`Yet another factor that should be considered
`modulation scheme for digital radio applications is the effect
`of delay d i s t ~ r t i o n . ~ Most of the delay distortion observed on
`
`of the
`6 Neither of these models takes into account the fade rate
`signal. A discussion of the effects of fade rate
`on the probability of
`error is beyond the scope of this paper, but these effects are discussed
`in some of the references [ 14, 28, 35; 37,47,65, 79, 1611.
`7 If the distortion observed on the received signal can be modeled
`by passage of the transmitted
`signal through a linear filter, the delay
`distortion is the derivative of the phase-frequency characteristic of the
`filter [77, p. 851.
`
`PM
`
`-
`
`MIPM A
`
`-
`
`~
`
`~
`
`* FOR BIT ERROR RATE OF
`1 ASSUMES OPTIMUM VARIABLE THRESHOLD
`* ' ASSUMESTHREE BITOBSERVATION INTERVAL
`
`d = FM MODULATION INDEX
`
`TABLE 5
`PERFORMANCE O F REPRESENTATIVE MODULATION SCHEMES
`IN THE PRESENCE O F DELAY DISTORTION (d/T = 1)
`
`OK-OPSK
`8-ary PSK - COHERENT DETECTION
`16-ary PSK - COHERENT DETECTION
`
`1
`
`9 8
`
`15 8
`
`-25 <25
`
`116
`
`175
`
`l 6 a r y APK
`
`\MlPM
`i
`* FOR BIT E R R O R RATE OF 104
`
`.. ASSUMES THREE-BIT OBSERVATION INTERVAL
`
`d = FM MODULATION INDEX
`
`line-of-sight radio links is introduced by the radios and not the
`channel. Table 5 presents some results derived primarily from
`Sunde's comprehensive treatment of' the effects of delay dis-
`tortion on pulse transmission [ 1 161 . The performance of the
`representative systems is shown for quadratic and linear delay
`distortion for the case in which the maximum differential de-
`lay (relative to the mid-band delay)
`is equal to the symbol
`duration. Note that some modulation schemes are severely af-
`fected by one type of distortion or the other. Notably, DQPSK
`
`5
`
`

`
`OETTING: MODULATION TECHNIQUES FOR DIGITAL RADIO
`
`1757
`
`LOW
`
`t
`
`t
`
`
`
`LL
`
`t
`t
`t
`LCPFSK.
`L D P S I
`DlSCRlMlNATOR DETECTION
`FEK - NONCOHERENTOETlCTlON
`OOK - ENVELOPE OLTtCTlON
`
`1 1 1 1 1
`
`1
`
`COMPLEXITY
`
`DOPSK
`
`ACKNOWLEDGMENT
`The author wishes to acknowledge Mr. D. I. Himes of the
`Naval Research Laboratory, Washington, D.C., for suggesting
`the study that eventually led to this paper.
`
`HIGH
`
`APPENDIX A: SUBJECT INDEX
`
`Amplitude Modulation
`AM-FSK:30
`Binary Coherent
`general: 32,70,77,80, 119,192
`spectrum: 75,105
`filtering: 26
`synchronization: 90
`delay distortion: 1 16
`fading: 109
`Binary Noncoherent: 70,192
`M-ary: 52, 192
`OK-QASK: 45,62
`OOK: 35,70
`
`QAM
`
`general: 67, 80
`spectrum: 98, 145
`filtering: 98
`experimental: 98, 145
`synchronization: 89
`delay distortion: 1 16
`interference: 98
`QASK: See Amplitude and Phase Modulation,
`QPR: 188
`Partial Response: 18,20,34,90
`Radios (digital transmission over): 67
`
`Frequency Modulation
`Chirp modulation: 55
`CP-FSK
`general: 1,40, 71 157
`spectrum: 21,76
`filtering: 71,72, 165
`experimental: 66,71, 165
`demodulation: 123, 157, 164
`delay distortion: 165
`interference: 73, 118, 155, 165
`frequency stability: 165
`M-ary: 1, 118, 165
`
`Digital FM
`spectrum: 21,75
`filtering: 25,74,82
`experimental: 63,113,122
`demodulation: 126, 133
`interference: 104, 113
`
`Fisure 3: Relative Complexity ot'Representative Modulation Schemes
`
`delay distortion,
`suffers severe degradation from quadratic
`while the coherent biorthogonal schemes (QAM, MSK, and the
`variations of QPSK) are degraded significantly by linear delay
`distortion. Thus, delay distortion can be an important criterion
`in the selection of a modulation scheme for digital radio.
`
`Cost and Complexity
`Finally, some mention should be made of the cost and
`complexity of the modulation schemes discussed in this paper.
`In general, it is difficult to evaluate the cost of a particular
`scheme without conducting a
`full-scale investigation of the
`cost and complexity tradeoffs involved with the many alterna-
`tive implementation options. Nevertheless, the modulation
`methods can be ranked according to their inherent complexity,
`and the results of such a ranking are presented in Figure 3.
`
`Application of the Results
`The results presented in this section in conjunction with the
`referenced source material
`can be used to solve systems engi-
`neering problems. In many applications, most of the alterna-
`tives can be eliminated simply by determining and applying
`the physical and economic constraints.
`Suppose, for example, that it
`is desired to determine the
`best modulation scheme for transmitting 90 Mbits/s over a
`40 MHz radio channel. A quick glance at Table 2 shows that
`only QPR, 8-ary or 16-ary PSK, and 16-ary APK can achieve
`the necessary signaling speed (2.25 bits/s/Hz). Ofthese schemes,
`QPR is the most efficient (smallest required E,/N,,)
`and the
`least complex. These considerations have prompted at
`least
`one proposal to employ QPR in the Canadian 8 GHz frequency
`band [ 1781 .
`
`IV. CONCLUSIONS
`This paper has treated some of the more important consid-
`erations relevant to the selection of modulation schemes for
`digital radio applications. In the process, we have summarized
`a portion of the vast body of digital modulation literature of
`the past 20 years. Hopefully this work will prove useful to
`communication engineers both as a summary of modulation
`characteristics and as a guide to the available literature.
`Clearly, this paper does not cover every aspect of digital
`modulation. Many factors(e.g., the effects of frequency offset,
`imprecise carrler phase recovery and limiting) are
`omitted
`from the discussion. Many of these topics are covered in detail.
`in the references at the end'of
`this paper. As an aid to using
`these references, Appendix A presents a complete
`subject
`index.
`
`6
`
`

`
`1758
`
`IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. COM-27, NO. 1 2 , DECEMBER 1979
`
`fading: 65
`biphase FM: 148
`capture effect: 9 1
`threshold performance: 43
`FSK (coherent)
`general: 86, 192
`spectrum: 141
`experimental: 124
`demodulation: 102
`delay distortion: 116
`phase reference: 1 10
`fading: 39
`M-ary: 124
`FSK (noncoherent)
`general: 24,30,80, 192
`filtering: 26, 27, 30
`demodulation: 82, 121, 164
`delay distortion: 1 16
`frequency uncertainty: 178
`interference: 41
`fading: 28, 39, 158
`nonlinearities: 85
`M-ary: 23, 165, 192
`MSK (See also SFSK, TFM)
`general: 17,42, 59, 151, 160
`spectrum: 15,76,84, 166
`filtering: 44, 71,72,84,97, 180
`experimental: 68,71,96, 144
`demodulation: 59, 156, 171
`delay distortion: 96
`p