8, AUGUST i 9 8 4
935
AbstractThe automatic frequency control (AFC) loopis beginning to 11. AFC CONFIGURATIONS play an important role in digital data links. There is very little published The AFC loop configurations discussed below employ literature which deals with AFC loop implementations and performance in three basic frequency discriminator techniques. They may be noise (in contrast to the large body of literature dealing with phaselocked categorized as loops). This paper presents a brief discussion of eight differentAFC loop impledifferentiate and multiply discriminator; mentations, including the block diagram, discriminator characteristic, and crossproductdiscriminator loop tracking error in the presence of gaussian noise. Tracking performdiscrete Fourier transform discriminator. ance of the various loops is compared in the presence of noise for CW, These'discriminator techniques are suitable for CW, and in DPSK, and MFSK signaling.
some cases DPSK and FSK. signals. Several variations on these
1. ~NTRODUCTION
IGITALdatatransmissionoversatellitelinkshasbeen dominated coherent by biphaseshiftkeyed (BPSK) and quadraturephaseshiftkeyed (QPSK) systems. The "heart" s ( t ) z= A COS [mot @(t)] n ( t ) (21 1 ofthecoherentdemodulator is someform ofphaselocked loop (PLL) and, as a result, a wealth of published informafion where n ( t ) is white Gaussian noise with twosided spectral is available on this subject. density The coherent PSK systems have excellent performance in the presence of Gaussian noise; however, they are not very ' S n ( f )=No/2. (22) tolerant of other link disturbances such as high Doppler rates, specular multipath, and ionospheric anomalies. These effects The noise process n ( t ) may be represented as have bekome more important in recent years, particularly withtheproliferationofairborneusers. As aconsequence,, n(t) = n l ( t ) cos w 0 t n 2 ( t j sin wet. (23) there is motivation for moving toward differentially coherent detection of PSK signals (DPSK) and multiple frequencyTheprocesses /zl ( t ) and 122 ( t ) areindependent,Gaussian shiftkeyed (MFSK) systems. MFSK signaling is also used processes with spectral densities frequently in frequencyhopped (FH) systems. These systems generallyrequireautomaticfrequencytracking(AFC)loops s,*,(O = sn,(t> = N o . (24) whicharemorerobustinthepresenceoflinkdisturbances TheAFCloopstobediscussedhavethebasicconfiguration than the PLL. AFC loops are also being used in conjunction of Fig. 1 , The VCO output is assumed to be of the form with PLL's toperforminitialfrequencyacquisitionofthe receivedsignal.Furthermore,burstmodeQPSKmodemsalv,(t) = 2 cos 0 1 t (25) mostuniversallyemployanAFCloop of some form (often operating on a CW o r BPSK preample so that the AFC loop which results in baseband outputs of does not operate during QPSK data transmission). Incontrasttothesituationwith PLL's, therehasbeen relativelylittlepublished on the performance of AFC loops. I(t) = A cos [ A u t t @(t)] Section I1 of this paper gives a brief discussion of eight difn l ( t ) cos A u t n 2 ( t ) sin A u t ferentAFCloopconfigurations(sevenofwhicharedigital implementations) which have been analyzed at ST1 over the and pastfewyears.Thenoiseperformanceresultsarepresented without derivations due to space limitations. Several, but not Q(t)= A sin [ A u t I@(t)] all,, of the results have been compared with both simulation n l ( t ) sin A u t n 2 ( t ) cos A u t and experimental data for verification. Data are presented where available and show good agreement with analytical where results. Section 111 containsasummaryofresultsandcompares Am =  ~ 1 . the performance of different configurations, where appropriate, for CW, DPSK, and MFSK signaling.
basic configurations which are particularly well suited to DPSK and MFSK signaling are also presented. A total of eight configurations are discussed.' The received signal is assumed to be of the form
Paper approvedby the Editorfor Satellite and Space Communicationof the IEEE Communications Society for publication after presentation at the Internst i O d Conference on Communications, Boston, MA, June 1983. Manuscript received June27, 1983; revised November22, 1983 Theauthor is with StanfordTelecommunications, Inc., Sunnyvale, CA
94086.
The differentiator AFC (DIFAFC) configuration in Fig. 2 . The discriminator is that described by Park
is shown [3]
' The originator of each ofthe algorithms is, in somecases, not knownby the author. Credit is given when possible.
>
'
936
IDISCRIMINATOR
vco
LOOP FILTER
~ LOW ~ PASS =
Y
I F iNPUT
@++
Fig 2 . DIFAFCconfiguration.
FILTER LOOP
An AFC configuration which employs a "cross product discriminator" is showninFig. 4 . TheCPAFC is similar t o the DIFAFC with the.differentiators replaced by delays. This is very convenient for a digital implementation as shown. Integrateanddump (I/D) filters are used as presampling filters t o limit the input noise. The filter noise bandwidth is (2.21)
asthe"idealdetectorwithinverselimiting."TheDIFAFC is particularly amenable to analog implementation in contrast totheotherconfigurationstobediscussed,whicharebest implemented digitally. Results pertaining to other analog AFC loop implementations are given in [ 1 1 and [ 2 ] . The discriminator output, which is the loop error control voltage, can be shown to be
The discriminator output is represented by (2.22) which is the cross product of the vectors p s ( k )and p s ( k  I ) , where
vs(k)= Zk + jQk
(2.23)
is the baseband representation of the input signal. The discriminator characteristic for a sinusoidal inputis This characteristic is plotted in Fig. 3 for two different lowpass filters. By analogy with phaselocked loops, a reasonable estimate of threshold is given by
D,(w) = A sinc2 ~
where
A : T 1 ) sin A U T ,
(2.24)
(2.12) where ,AF(peak) is the frequency error corresponding to the peak of the discriminator Scurve. T h eD I F A F C is useful for tracking CW, DPSK, or t h e averagefrequency of a nF S K signal.Theinputbandwidth, B L p l must be wide enough to pass the input signal spectrum, without serious distortion, for proper operation. The noise performance of the DIFAFC has been shown to be [ 5 ]
5 . Loop threshold is
U A ~ T= , 0.07.
The performance of the CPAFC in the presence of Gaussian noise can be shown to be [21
(2.25)
for
f o r BL
< BI.
(2.13)
93 7
( .
oo
NCO
FILTER
C
tktkl=T1
I .o
t"
I I
tk
t TnI
O A f T , = 0.07.
.I.OL
(2.210)
Fig. 5. CPAFCdiscriminatorcurve. The CPAFC is useful for tricking a CW carrier (or the average value of an FSK signal). ADPSK signal of the f o r m
TheloopperformanceinGaussiannoisecanbeshownto be [61
0.05 1 (BL/Bz)Bz2
Oaf2 =
S ( t ) = A d ( t ) cos ( w o T
+ 4)
(2.27)
SNRz
where d i t ) = ? I is the differentially encoded data, generates a loop error voltage (2.28) thus causing the error voltage t o have the wrong sign when d k ' # d k  1 , Severalvariations of theCPAFCsuitablefor use with DPSK'signals are given below. 1 ) Cvoss ,Product AFC with Sample Excising fov DPSK (SEAFC): TheCPAFCqanbemadetooperatewithDPSK signalsbyexcisingsampleswhichoccurarounddatatransitions. Let us consider the case for
+
[ y] 1)
1

___
2 SNR]
(2.21 1)
N
N>2 and
NT1 == Td
BZ
Note thatthe above operation implies that bit sync is available. In the absence of bit sync, as during signal acquisition. the loopcanbeoperatedwithoutsampleexcising,asdiscussed above, as long as the number of samples per bit, N , is relatively large ( N 8). The large N will ensure that performance 1.5 dB from that given will not be degraded more than about by (2.26). N should be decreased to N = 2 and sample excisingbegunassoonasbitsync is achievedinordertotake advantage of a reduced input bandwidth ( E 1 = N J T d ) . A comparison of performance predicted by (2.21 1) and digitalsimulationresults is shown in Fig. 8 . Excellentagreement is obtained. 2 ) Angle Doubling AFC (ADAFC) .for DPSK Signals:
N T , = Td
(2.29)
>
6. In this case, where Td is the data period as shown in Fig. if we excise those values of v,(n) [see(2.28)lwhichare derivedaroundadatatransition,then d , = d,*1 , and the loop error voltage has the proper sign. Thus, the CPAFC can be made to work with DPSK data by excising the sample formed at the data transition and using the N  1 remaining samples as shown in Fig. 7.
93 8
J
tktkl'1
.
BI
100 Hz
= lKHz
00)
N=2
3 
2
1
10 100
. A
1
1000
SNRI
Fig. 8. Comparison of computed and digit simulation results for an SEAFC loop with DPSK signaling. The effect of AGC on the loop bandwidth is included. (Data provided by G. Spcci, STI.)
!
Another modification of the CPAFC for operation with DPSK is shown in. Fig. 9. Basically, the instantaneous phase of the carrier, relative to the receiver reference, is doubled, thus removingtheDPSKdata.Toshowthis,consider(26)and (27 j for the noisefree case.
This technique can be readily extended to a configuration suitable for QPSK signals by introducing second a angle doubling operation, su'ch thatanglequadrupling is achieved. Alternatively,ananalogfrequencyquadruplerfollowedby the CPAFC will accomplish the same thing. The ADAFC can be made independent of data synchronizaI(fn) = A d , COS [O(t,>] (2.212) using by tion a time invariant opposed (as integrateandto dump) input filter and selecting an appropriate sample rate. Q(t,) = A d , sin[O(t,>l (2.213) discriminator The summing bandwidth reduced by be may the angle doubled output samples as shown in Fig.'9. where d, = +1 is the differentially encoded data. Then, referThe discriminator characteristic can be shown to be ring t o Fig. 8, M
Xn = In 2
where
D , ( A f ) = A 4 sinc4 ( n A f T , )
(2.215)
m=l
 sin (m411AfT1)
+ (M  m ) sin
[(m
+ M ) 4IIAjT,].
(2.216)
This result is plotted in Fig. 10 with M as a parameter. As ill 1/M. approximately decreases as range trackingthe increases, Note that for higher values of M ,the discriminator characteru, = 0 withpositiveslopeandsufficient Itshouldbenotedthatanalogfrequencydoublingfollowedisticpassesthrough b y .t h eC P A F C is mathematicallyequivalenttotheADAFCamplitudesuchthatfalselockcouldoccur. Loop threshold (when sync bit is available). estimated is occur when to AfT, = 0.04 for M = 1.
d,? = 1.
939
r
r
 7
k)
V,(t)
B I T SYNC
t.
X1
I
I"k
I
I
I I
I L .
OISCRIMINATOR
I I
B I TS I N C
_____ANGLE DOUBLING
I J
______J
BL <l BI
3 2
'I I
1,N> I)=
SNRI ( 1
. S".ETTJf
Fig. 10. ADAFC discriminator characteristic for some values of M . The loop performance in Gaussian noise is [ 7 ]
0.05 1 (BL/BI)BIZ
Oaf2(M=

1

+ 2.5
SNRI' SNRI
++I
2
7)
0.5
SNRI~
(2.217)
0af2(M> 1, N
= 1) =
0.05 1 (BL/BI)B12
SNR, ( 1

7)
3 ) CrossDot Product AFC (CDPAFC) for DPSK Signals: A third variation of the CPAFC configuration, which is suitableforusewithDPSKsignals, is shown in Fig. 11. Inthis scheme,boththecrossanddotproductsofadjacentsignal samples are computed and multiplied together to remove the data modulation from the loop error signal. The (CDPAFC) is mathematically equivalent to the ADAFC with M = 1 . Thediscriminatorcharacteristicandnoiseper(M = 1 ) . formancearethesameasshownfortheADAFC Note that, as in the case of the ADAFC, the CDPAFC can be madeindependentofdatasynchronizationbyusingatime invariant input filter and selecting an appropriate sample rate. 4 ) CrossProduct A F C w i t h Decision Feedback (CPAFCDF) for DPSK Sigmling: The CPAFCDF is a variation of the crossdotproductAFCand is ofconsiderableinterestforDPSK signals.AblockdiagramofthisAFCconfiguration is shown in Fig. 1 2 . The cross and dot products of the input samples are formed in the same manner as for the CDPAFC; however, datadecisionsareperformedonthedotproduct,andthe estimated data are used to resolve the sign ambiguity of the loop error voltage. The discriminator characteristic is given by DE(A.f) = A,2 sinc2 (nAfT1) sin ( ~ I I A ~ T , ) (2.219)
sgn [cos ( 2 n A f T l ) ] which is showninFig. when
1 3 . Threshold is estimatedtooccur
M 2OS SNRI~
o A f T 1 = 0.07.
(2.220)
940
a, AUGUST I 984
Fig. 12.
1.d
94 1
tion, is considerably more tolerant to frequency fluctuations thantheADAFC or CDPAFCconfigurationsbefore losing lock. TheloopperformanceinGaussiannoise is approximated by [41 0.05 1 (BL/BI)B12
oaf
(1
2 P,)2 SNRI (1
2 NBL/BI)
+ 20 P,(BL/BI)(
(1

1  1.6 P,)
(2.221)
2 P,)2
where
oaf Q Td
. 6 .5
A comparison of computed and measured performance is showninFig.14.ThesedataweretakenwithaPN/DPSK receiver with code tracking loop and AGC loop operating. A secondorder CPAFCDF was employed since the loop was required to track a high Doppler rate in the intended application. The computed curve includes 0.8 dB correlation loss as The computed indicated by data error rate measurements. curvealsotakesintoaccountthechangeinloopbandwidth duetoAGCsignalsuppression,atlowSNRI.This is given, for a secondorder loop, by
.4
.3
.2
I
2
I
1
0.1
I
0
I
1
I
2
I
5
I
6
I
7
I
8
I
9
I
10 11
3
3 . 4
SNRI(dB)
Fig. 14. Comparison of computed and measured CPAFCDF performance for a PNlDPSK system.Computed curve assumes 0.8 dB of correlation loss as evidenced by data error rate measurements. Change in bandwidth due to AGC signal suppression due to low SNR, is also included. (Data provided by M. Paff, STI.)
1 1
+,
we have
4i0 where
and
M1
(2.33)
(7) ( y )]

I , sin
(2.44)
( D F TA F C )
ak = R K + j M k Im and Q ,
aresamplesof ~ ( t and )
Q(f)
(2.35) givenby(26)
Theconfiguration of theDFTAFC is shown inFig.15.and The DFT can be viewed as a spectrum analysis algorithm and (27). whichprovidestheequivalent of afilterbank.TheDFTThe computes the coefficients. 1 M1 ak = X(mTs)eikmOTs12 M m=0 where
p0wer7) of thekthoutputattime
T~ is
(2.36)
Pk = (ak I2 = R k 2 + M k 2
(2.31)
(2.37)
where (2.32)
X(rnT,) = I , t j Q ,
A f = frequency error.
942
LPF
AID
Rk
+
P k = Rk2 t Mk2
S(t)
LPF
AID
Qll
O0
NCO
f 
LOOP FILTER
B I T SYNC M
Fig. 15. DFT AFC configuration. The discriminator characteristic is plotted in Fig. 16. Threshold is estimated to occur when
= 0.17. UA~TO
(2.38)
is given b y [ 8 I
where
BI=.
I
T O
Note that when the DFT analysis interval is synchronized with the bit timing interval, and the DFT interval is equal to or less than a bit interval, then the DFTAFC can be used with DPSK andFSKsignaling(tracksaverage.frequencyintheFSK case). Degraded operation can be achieved before bit synby employing an analysis interval chronization is available that is considerably less than the data bit interval (To T d ) . 1 ) DiscreteFourierTransformAFCwithDecisionFeedback (DFTAFCDF) for MFSK Signaling: One of the principal applications of the DFTAFC configuration is for MFSK signaling. The MFSK signal consists of M tones with spacing 1/T, (Hz) and frequencies Fig. 16. Discriminator curve of DFT AFC algorithm. where 2rI wo = 
TO
To = DFT analysis interval
k = 0, f 1, f 2 , ....
Thus, the filters corresponding to k = f ( 2 m  1) [designated Pk(Dj1 are matched to the M data tones. The filter responses are shown in Fig. 17 for M = 8 . The filter noise bandwidth is ] / T O The . filters corresponding to
fo
. for
m = 1 , 2,
.... M / 2 .
Pk (AFC), are spaced in between the FSK (2,311) and designated tone frequencies as shown by the dotted linesof Fig. 1 7 . In the case of the DFTAFCDF, the kth received tone is detected by the DFT filter P , , . and the discriminator output
943
NCO
LOOP FILTER
4
3
2
1
Af T
a7 6 
f k 
1(AFC)
(2.313)
as shown in Fig. 18 [9]. That is, once the decision is made as to what data tone was sent, the discriminator error number is formed from the adjacent DFT filter bins. This constitutes decision feedback. The DFT analysis interval is assumed to be synchronized with the data tone period. This scheme results in considerably improved performance over configurations which track average frequency since it can employ a narrower discriminator bandwidth and the result does not depend on the data tone distribution. The discriminator characteristic has the same form as that given by (2.37), centered on the transmitted tone. The threshold i s estimated to occur when u a f T o = 0.17. TheMFSKdetectormust,ofcourse,haveaninitialfrequency uncertainty of less than onehalf the tone spacing for proper operation. In applications where the initial uncertainty is greaterthanthisamount,afrequencyacquisitionphase employing the DFTAFC discussed above may be employed. The loop noise performance in Gaussian noise is given by [ 81
50
40
30
20
uA
(HZ)
10
8 6 
4 
aAj2 =
0.31 BL ( P , / N o ) ~ o ~ (l ~ e ) ~ (1 ~NBLT,)
2 
1 16
C/No
(dB)
Fig. 19. SecondorderAFCperformance in noise.Measureddatainclude effects of hoptimingloop jitter, quantizationnoise,andAGC.(Data provided by B. Campeau, STI.) (2.314) where The computed and measured results show excellent agreement. There is somedivergenceathigh C/No where the A/D quantization noise becomes evident in the measured data.
exp (P,TO/~NO)
BI=.
To
is comparedwithmeasuredperThe computed performance formance in Fig. 19. The experimental data were taken with an FH/8FSK receiver employing a secondorder DFTAFCDF with AGC. The computed result includes the effect of loop bandwidth change due to AGC signal suppressions at low SNRI. Thisisgivenforasecondorderloopby(2.222).
The compatibility of the various AFC algorithms with MFSK signaling formats is summarized in CW, DPSK, and Table I. The loop performance equations are summarized in Table 11. Detailed comparisons are given below for the different signal formats.
944
TABLE I
CONFIGURATION
AFC
cw
MODULATION
FORMAT
LOOP
BANDWIDTH MUST
BE CONSIDERABLY LESS
ON T H E STATISTICS
TONE OCCURENCES.
NBL/BI<<l
UIF4FC
? .38 B I
CPAFC
i .21 B I
UFTAFC
t .51 8,
IUFIAFCUF
.51 B I
SEAFC
k .21 RI
AUAFC ( N = l
,051 ( B L / b I ) d l Z
M 5NK.
0.1126
(BL/BI)
BI2
SNRI4
CPAFCDF
0.026 (BL/Bl)
(12 Pe)'
B12
t .21 s1
C11.6 Pe)
SNRI2
20 Pe (BL/BI)
'Terms o f t h e fnrm 1
B & a r e general'ly s m a l l I
945
1.0
0.8 0.6
A . CW Signals
In some application, a CW carrier or residual carrier component is transmitted for synchronization or Doppler tracking purposes. The DIFAFC, CPAFC, and DFTAFC are potential choices for this application. In order to get a meaningful performance comparison, we must first select parameters such that the loops have equivalent tracking ranges, i.e., from Table 11. 0.38 BI (DIFAFC) = 0.2 1 BI (CPAFC)

0.4
BLTO.Ol
0.5 1
0.2
Y
To (DFTAFC)
which leads t o 2.43
(3.11)
g.5 fT B
u
w
(3.12)
: :
w
4
0.1
z
ar
.Of
.O
0
Y
N w
i
(3.13)
TO
where To is the DFT analysis interval. Comparing performance for SNRI < 1 with the above values for B I , we find
UA~'
6 z
.04
(DIFAFC) =
0.40 BL
.92
(PSI& 1' T ,
forSNRI
< 1(3.
4)
Oaf2
f o r SNR,
< 1,
(3. 5
.01
4
I
2
I
0
SNRI ( d B )
I
2
I
4
I
6
NBL/BI< 1
width BI and possibly loop bandwidth BL thanusedduring tracking.Lack of bit sync during acquisition can be'accommodated, for those configurations'that use it, by taking several samples per bit such that the percentage of samples affected NBL/BI < 1. (3.16) bythedatatransitions is small.This is consistentwiththe widerinputbandwidthandpullinrangeoftenrequiredto The conclusion is that these configurations provide essentially BI = N / T d (Secidentical performance (within the accuracy of the approximate handleinitialfrequencyuncertaintiessince tion 11B1). A wide 'BL is oftenemployedforrapidloop performancesolutions)atverylowSNRI.Normalizedpersetting. The acquisition mode is not 'discussed here in detail; formance is plotted in Fig. 20 versus SNRI. The DIFAFC exhibitsbetterperformanceforthe higher' values of SNRI however,performancecanbeinferredfromresultsalready presented. than the other considerations. The DFTAFC and CPAFC The AFC loop generally operates with the minimum input ( N = 1 ) have essentially identical performance for the chosen bandwidth during tracking, in order to achieve the best perparameter set. The tracking 'mode parameters for the various Theseresultsindicatethattrackingperformance is very formance. loopsaresummarizedinTable 111. Aperformanceco.mparisimilar for all the loops as they approach threshold. The analogDIFAFC gives improvedperformanceforSNRI 0 . son is shown in Fig. 21. ThaCPAFCDF appears to achieve the best noise performance of any configuration for DPSK signalThe performance of the CPAFC can be improved at SNRI ing(assumingperfectbitsync).TheCDPAFCandADAFC 0 dB by increasing N ; however, this improvement is'limited also provide excellent noise performance but threshold almost by the restriction that NBL T 1 . 3 dB higher (this situation may change if there is significant B. DPSK Signals bittimingjitter,sinceitwillprobablyhaveagreatereffect on the CPAFCDF). Several of the AFC configurations discussed above are The DFTAFC and DIFAFC are not as efficient in noise as suitableforusewithDPSKmodulatedsignals.Referringto thecrossproductdetectorloops;howtrver,theyexhibitas Table I the DIFAFC, SEAFC, CDPAFC, ADAFC, CPAFCDF, threshold about 2 dB lower than the CPAFCDF. and DFTAFC are recommended candidates. The SEAFC exhibits a threshold approximately equivalent The DIFAFC is the only configuration which does not t o t h a t o f theCPAFCDFbuthassignificantlypoorernoise require bit sync for full performance. The CDPAFC and performance. ADAFC can, be made independent of bit sync by subsituting lowpass a filter (LPF) for the input integrateanddump C. MFSK Signals (I&D) filter. However, the LPF must be at least twice as wide The DFTAFCDF (Section 11C1) wasderivedspecifically as the I&D filter to prevent degradation due to intersymbol forusewithMFSKsignalsandprovidesveryefficientperinterference, and poorer loop performance will result. The AFC loop generally has an acquisition mode to accom formance in noise as given by (2.314). An "average frequency" tracking loop such as the DIFAFC modatethelackof,bitsyncandlargeinitialfrequencyunrequiresthatthediscriminatorcurvespantheentiresignal certainties. This mode is characterized by a wider input band
>
>
<
946
AFC
TRACKING
RAllGE. AF/Ru
THRESHOLO
OAf"U
LOOP ERROR
2
"Af
i O N F I tiURA1 ION
2
/RU
DlFAFC
t 0.75
0.25
SEAFC ( N = 2 )
0.12
0. I 4
CDPAFC d
AORFC ( M = 1 )
t 0.12
0.04
0.051 ( E L / R o )
E5'"O
r1mL'8;J
['
exp (  [$/No)
'
2.5
' (ES/No)
(ES/No) 0.5
'I
CPAFC
OF
! 0.2
0.07
where P
_I_
2
OFTAFC
t 0.51
0. I7
*Es/No
__
Ne RC
and
symbol
'ate
Fig. 21. Loop tracking performance comparison for DPSKsignalmodulation. (parameters selected for best performanceafter bit sync has been acquired.)
947 ACKNOWLEDGMENT
This paper presents previously unpublished results obtained by the author and his associates at Stanford Telecommunications,Inc.TheDIFAFCnoiseperformance,(2.13), representsajointeffortby Q. Huaandtheauthor(1978). . . TheresultfortheSEAFC,(2.211),wasobtainedbyA. J. VanDierendonck (1 98 1 ). Theremainingnoiseperformance results were obtained by the author. Preliminary work on the ADAFC and CDPAFC algorithms was performed by B . Currivanand S. Chu. The DFTAFCDF J. Ohlson. The algorithm was proposed by M. Miller and CPAFCDFalgorithmwasfirstsuggested, to thebest of t h e authors knowledge, by the Mission Research Corporation.
0.6
0.5 0.4
PI
0
0.3
PI PI w
z
w
3 G 
0.2
T
0
E c+Af
CL
0
0
0
y
4
5
0
REFERENCES
D. Richman, Colorcarrier reference phase synchronization accuracJ in NTSC color television, Proc. IRE,vol. 42, Jan. 1954. L. H. Enloe, Decreasing the threshold in FM by frequency feedback. Proc. IRE, vol. 50, Jan. 1962. J. H. Park, AM FM detector for low SIN, IEEE Trans. Comrnun. Technol., vol. COM18, Apr. 1970. F. D.Natali, Noise performance of a crossproduct AFC with decision feedback for DPSK signals. in Conf. Rec. N a f . Telecommun. Conf., Nov.1983. Q : Hau and F.Natali, unpublished, 1978. A. J. Van Dierendonck, unpublished, 1981. F. Natali, unpublished, 1981. ,unpublished,1982. M. Miller andJ. Ohlson, private communications. 1982.
.06
.05
.04
.03
.02
1
6
4 0 2
P s To/N,
.O I
2
(dB)
8
*
Francis D. Natali received the B.S. degree in electrical engineering from Rensselaer Polytechnic Institute, Troy, NY, in 1960, and the M.S. and Ph.D. degreesinelectricalengineeringfromthe State University of New York at Buffalo in 1964 and 1967, respectively. He joined Stanford Telecommunication, Inc., Sunnyvale, CA. shortly after it was formed in 1973. He has been a technical contributor in the areas of spreadspectrum signaling, demodulators, and ranging receivers, including applications to the Global Positioning System. Previous to joining STI, he was employed by PhilcoFord in the Communications Sciences Department, and Cornell Aeronautical Laboratories in the Systems Research Department.
bandwidth. For example, a n 8FSK signalwithtonespacing fo o.5T0, f o 1.STo,fo l / T o uses tone frequencies of l.STo, fo 3.5T0 about the carrier frequency fo. Let us set the discriminator peaks of the DIFAFC to coincide with the tonesfo k 3.5To. Then from Table 11,
Bz (DIFAFC)=
3.5To
0.38
= 9.21 T o .
(3.31)
A comparison of the performance of the DFTAFCDF and the DIFAFC is showninFig.22 for an8FSKsystem.Theadvantage of theDFTAFCDF is clearlyshownforthissignal format.