1942 IEEE TRANSACTIONS ON COMMUNICATIONS, DECEMBER 1974 Correspondence Lkear Amplkcation with Nonlinear Components , D. C.COX Sa(t) Abstract-A technique for producing bandpass linear amplification with nonlinear components (LINC) is described. The bandpass signal first is separated into two constant envelope component signals. All of theamplitudeandphase information of the original bandpass signal is contained in phase modulation on the component signals. These constant envelope signals can be amplified or translated in frequency by amplifiers or mixers'which have nonlinear input-output amplitude transfer characteristics. Passive linear combining of the amplified and/ortranslated componentsignals produces an amplified and/or translated replica of the original signal. I. INTRODUCTION {COMPONENT1 E ( t ) cos w o t TFig. LINC amplifier. .I. and Conventionalsolid-statelinear power amplifiers are difficult to build a t low microwave frequencies and impossible to build a t high microwave and millimeter wave frequencies. In many transmitter applications,such as the amplification of single-sideband (SSB) signals or the amplification of a multiplexed set of several separately modulated low-level carriers, a linear overall input-output relationship is required in the transmitter power amplifiers. Nonlinear solid-state power amplifiers are readilyavailable a t low microwave frequencies and constant-amplitude phase-lockable signalsources (GunnandImpatt diodes andmagnetrons)are available in the high microwave and millimeter wave region. Also, for high-power applicationsin the microwave region, nonlinear electron tube amplifiers and power oscillators aremore readily available than are linear amplifiers. A technique that makes use of these available nonlinear amplifiers or phase-lockable oscillators to produce bandpass linear amplification with nonlinearcomponents (LINC) is described in this correspondence. The overall input-tooutput transfer function of these LINC amplifiers is linear over a wide range of input signal levels but the internal R F amplifying devices can be highly nonlinear or, in fact, even constant-amplitude phase-locked oscillators. + 4(t) I ~ 1 , ( t )= i ~ , / 2 ) sin ~wot SzG(t)= (E,/2) sin Coot - + ( t )1. I (4) Note that SI,( t ) and Sza( t ) can be represented by constant-amplitude vectors rotating in opposite directions -with increasing values of + ( t ) as in Fig. 2. With E (1) 2 0 the vectors occupy onlyquadrants 1 and 2. Stability requirements discussed later furtherrequire b ( t ) 5 ~ / 2 Since . SI, and SZa can be separated, they can be amplified separat,ely in nonlinear amplifiers. In fact, the two amplifiers illustrated in Fig. 1 can be two power oscillators which are phase locked or injection locked to &,(t) and S,,(t). Subtracting GSz,(t) from GSla(t), where G is the identical amplifier gain, yields the amplified output GXl, ( 2 ) - GSZ,( t ) = GE, sin 4 ( t ) cos wot = GS, ( t ). (5) A similar argument can be carried through for the most general bandpass signal [I] 11. T H E BASIC PRINCIPLE The basicprinciple for thisLINC is separatingthebandpass input signal which may have either or both amplitude and phase (frequency)variations (i.e., AM and/or angle modulation) into two component signals S1and 8 2 which are constant amplitudewith variations inphase only. Thesetwo constant-amplitude anglemodulated signals can be amplified separately by anyamplifier with sufficient bandwidth regardless of itsamplitude linearity. The amplified component signals are passively combined to produce an amplified replica of the input signal. Consider first a constant-phase bandpass signal S,(t) = E ( t ) cos oot (1) with a real envelope E ( t ) 2 0 as the input to a LINC amplifier component separator (see Fig. 1 ) . The component separator produces the two constant-amplitude signals, &,(t) and Sza( t ) , related , t o S,(t) as follows. Substitute E ( t ) = E, sin 4 (1) , (2) where the constant E, equals the maximum value of E ( t ) and the equation defines 4 ( t ) , into (1) to yield 111. AN IMPLEMENTATION OF A SIGNAL COMPONENT SEPARATOR The first step required in this LINC amplifier signal component separator is to obtain the envelope E ( t ) and a constant-amplitude angle-modulated term p ( t ) = K cos Coot e ( t ) 1. As illustrated in Fig. 3, p ( t ) is obtained by limiting S ( t ) and E ( t ) is obtained either from a linear envelope detector or froma synchronous detector. The next step, illustratedin Fig. 4, is to separate SI ( t ) and SZ( t ) . Consider first the feedback path around the amplifier with a very large voltage gain, Gl = - Vo/V,. A phase-shifted p ( t ) is modulat.ed by Vo(t) to produce + Ol(t) = K sin Coot + e ( t ) + klVo(t)]. The mixer output Paperapprovedby the Associate Editor for Communication, Electronics of t h e I E E E Communications Society for publicationwlthout oral presentation. Manuscript received December 1 1 , 1973: revised July 26, 1974. The author is with Bell Laboratories. Holmdel, N.J. 07733. p ( t ) o 1 ( t ) = K2sin Coot is filtered to yield + e t t ) + klVo(t)]cos Coot+ e ( t ) ] (9) 1943 CORRESPONDENCE The largest value, which occurs when VOis small and sin klVo(t) = klVo(t),iskl.Thus,ifG~>>[(R1 R z ) / ( K z R 2 ) ] ( ~ / k(12) 1 ) , becomes + -KzRz . E ( t ) = -Rl sin klV01t) which is the same form as ( 2 ) . The approximation can be made as good as required by making GZsufficiently large. The size of GZwill be dictated by the distortion limits (see Section V) placed on the overall LINC amplifier. If kl and K are sufficiently large, then GI could be less than unity and the baseband amplifier could be replaced by a passive summingnetwork. This should increase the realizable bandwidth of such a component separator. If K , RI, and Rz are chosen such that (KZ/2)(Rz/R1)= E,, then % P Fig. 2. Signal component vector diagram. LIMITER (a) klVO(t) = -+w. (14) The outputs from the component separator of Fig. 4 [see (9) ] then become the desired signal components 4-b ENVELOPE + e(t) - +(t)] K sin [wet + 0 ( t ) + + (1) ] & ( t ) = Ksin OZ( t ) = Coot = (2K/Em)Sz(t) = ( 2 K / E m SI ) ( t ). Of course, the feedback loop must be designed to satisfy phase shift and gain conditions required for stability [a]. Note that if the phase modulator in the feedback loop does not produce a linear phase change as a function of modulating voltage Vo (i.e., if kl is a function of Vo), then the high-gain feedback loop will compensatefor this imperfection by distorting VO(t) so that sin [ k l ( V o ) V o ( = t ) ]V l ( t ) and k l ( V o ) V o (=t )+ ( t ) . The only requirement is that the two phase modulators shown in Fig. 4 have the same modulation characteristic kl(Vo). (If a phase modulator with a sin-’ characteristic could be realized then the feedback loop could be eliminated.) The matched modulator requirement can be removed by providing a second similar feedback loop with its own high gain amplifier, phasemodulator,etc.,forproducing Oz(t) directly from E ( t ) . The second loop would be identical to the one shown but driven by -E(t) to produce a phase modulator output of 0 2 ( t ). (C) Fig. 3. Detectors and limiters. IV. FREQUENCYTRANSLATIONWITHIN LINC AMPLIFIER LOOP LOW PASS FILTER (15) A MODULATOR In many microwave or millimeter wave transmitter applications it is desirable to translate a signal in the 10’s or 100’s of MHz to a higher frequency and linearly amplify it. This is not possible within the current state of the millimeter wave solid-state device art. It can be done, however, by separating the low-frequency signal into components O1( t ) and 02 ( t ) , translating thecomponents in frequency with the samemixing oscillator and amplifying as indicated in Fig. 5 to produce 01m(t) = (2K/Em) sin [(WO + + e o ) - +(t)l WI)~ and PHASE MODULATOR TKsin o,(t) ’ The mixers and amplifiers can be nonlinear. The outputis the ampliw. fied input translated to W O + kot+8(t)I Fig. 4. Signalcomponent separator. (4KG/Em)Sm(t)= (4KG/Em)E(t)COS V,(t) = (K2/2) sin [k,Vo(t)] (10) which has the proper slope for closed loop stability as long as I k,VO(t) I i T/2. Assume that the amplifier input impedance is large compared to Rl and RZso that Combining V o ( t )= -GlVi with (10) and (11) yields vo(t) + Gz (K2/2) Rz sin k1Vo ( t ) (R, Rz)VO( t ) + . (12) Stability requires that E (1) be restricted so I k’Vo(1) I 5 ~ / 2 .With ( t ) ) is 2kl/x. this restriction the smallest value of [sin k1V0( t )]/ ( VO [(WO + + e(t)l. W I ) ~ (17) Frequencytranslationwithina LINC amplifier may find application in point-to-point and satellite microwave and millimeter wave repeaters and for amplifying the multiplexed sets of modulated carriers used in future high-capacity mobile radio base stations. V. SOMECOMMENTS ON DISTORTIONANDNOISE Noise and distortion ( N and D ) affect the output differently depending on where they areintroduced into the LINC loop. A detailed analysis has not been made but several aspects can be seen from the equations in the previous sections. First, since a LINC amplifier is linear, the signal-to-noise ratio at the input will be preserved in the output. N or D added within the LINC can be considered in two classes. Symmetrically added N or D affects both & ( t ) and Sz(t) identically and usually occurs before the component separator. Assymetrically 1944 IEEE TRANSACTIONS ON COMMUNICATIONS, DECEMBER 1974 BAND PASS O m ( t ) FILTER cos W, t OSCILLATOR BAND PASS Fig. 5 . added N or D affects only one component component separator or amplifier. Frequency translation within a LINC amplifier. signal and occurs in a Symmetrically Added Noise or Distortion Symmetrically added N or D may occur in the limiter or envelope detector (Fig. 3). Assume that there exist amplitude fluctuations on the limiter output of K = K , ( l 6,) where 6, << 1. Then from (13) with + ‘C Level of input E ( t ) . Again, the LING amplifier does not ,enhanct noise which enters a component signal path. Effects of Finite Gain in the Signal Component Separator Feedbacl Loop The effect of finite gain in the feedback loop is obtained by using (13) in (12), and (9) to yield the LING output. K,2Rz/2R1 (21: and higher powers of 6a neglected, the LING output is Then, consider the distortion produced on a two-tone test Signal 0 the form S2(t) = Thus, 6, will appear on the output as an amplitude variation of 26,. Phase fluctuations on the limiter output will appear on the LING amplifier output unchanged. Forthe envelope detector,with E(1) = E, ( t ) E n d ( t ) , the N or D , End ( t ) , appears. in the LING amplifier output in the same proportion as at the detector output. + = + sin wtt - E z ( t ) COS sin wzt’ [wet + e z ( t )1 (22: + where wo = (w1 WZ) /2, and we = (a1 - 02) /2. Substituting (22) into (21), writing theoutput as a Fourie series, and solving for the signal power S and distortion power L with ( Rl Rz)/ (klGIR1) < 1 yields + Asymmetrically Added Noise or Distortion Phase fluctuation in a single component path is one type of asymmetrically added N or D . Consider the output as S a ( t ) = (GEm/2) {sin [mot + + ( t )+ & ( t ’ ) ] - sin Cwot - + ( t ) where end + e,(t) + e n d l l (19) If R1 = Rz then klGl of 100 reduces distortion so that S/D = 50 dB. Thus, the loop gain requirements on this implementatio~ of a LING amplifier are not toosevere. Even though analysis does not indicate any requirements whicl will prevent LING from being successful, it has shown the inheren sensitivity to balance (end and &). Thus, realization of an amplifie with low distortion and wide dynamic range requires careful design (20) VI. LABORATORY DEMONSTRATION OF THE FEASIBILITY OF COMPONENT SEPARATION AND RECOMBINATION is the N or D. The output can be rewritten as ‘COS + 1 @ ( t ) cos [Wet + & ( t ) f ‘@,dl. The sign of (1 cos end)1’2is always positive but the sign of (1 cos Ond)i’2 depends on the quadrant of end. One effect of endis a phase fluctuation in the output of f e n d . Also, for very small values o f . E ( t ) , the second term in the brackets defines a minimum LING amplifier output level with envelope = &a. A small modulation of the desired envelope E ( t ) is also produced by end. As an example consider end uncorrelated with E ( t ) and with a peak amplitude of 0.02 radians. This produces a peak fluctuation in the outputenvelope of about E, x 10-2 which 1s of the sameorder as the noise in the component signal path. Thus, the LINGamplifier does not enhance phase noise which enters a component signal path. A similar consideration of an amplitude fluctuahion, 6, << 1, in a component signal path shows that for large signal levels, 6, produces a small output phase perturbation of the order of 6,. It also modulates the envelope by a factor of approximately 1 $6,. When the peak value of E ( t ) / E , becomes of the same order as the peak value of 6,, extraneousphase andamplitude fluctuationson the output become quite large andthe amplitudefluctuations and phase fluctuations due to 6, become overpowering at about the same + for a two-tone test signal with maximum envelope swing. The key problems in realizing LING are producing the constan envelope signals and maintaining thephase and amplitude balancc A nonoptimum realization Of two component separators (similar t Fig. 4) and a combiner were assembled from available laborator, components to demonstrate the feasibility of the LING techniqut Envelope detectors and limiters were not included. The operatin frequencies, bandwidths, etc., were determined by available corn ponents. No attempt was made to optimize any component sinc the only objective was to demonstrate that the twoconstant-ampli tude phase-modulated signals could be generated and combines to produce a specified envelope varying narrow-band signal. A two-tone test envelope [fe = 10 kHz = we/27r in (22)] wa applied tothe two narrow-band component separators (fo = 7 MHz = w 0 / 2 7 r ) . The “two tones” a t f~ = fo fe and fz = fo out of the combiner [see (22) ] were examined with a spectrur analyzer. At maximum output for this nonoptimized assembly th strongest spurious signal was 22 dB below the level of the two equr amplitude “tones,” fl and fz. At a 9 dB lower output the stronges spurious signal dropped to 32 dB below the level of the tones. Thi + 1945 CORRESPONDENCE simple experiment demonstrated the feasibility of the LINC tech- above and below threshold operation, demonstrate the accuracy of nique, but did not give any indication of the minimumspurious the approximateexpressions. signal level possible from a well-designed amplifier. REFERENCES I. INTRODUCTION [ I ] M. Schwartz, W. R . Bennett,and S. Stein, CommunicationSystem New York:McGraw-Hill. 1966, sec. 1-6, 1-7. 1-8, andTechniques. Simulationresults are presented in Snyder [l, pp. 98-1041 for the F M demodulator described by and ch. 4. [ 2 ] J. G. Linvill and J: F. Gibbons, Transistors and Active Circuits. New York: McGraw-H111. 1961. J. J. D’Azzo and C.H. Houpis, Control System Analysis and Synthcsis. New York: McGraw-Hill, 1960. Comments on “Survivability Analysis of Command and Control Communications Networks-Part I” where F = [ O0 ROBERT E. LAWRENCE In the above paper,’ Dr. Frank points out many problems commonly encountered in the analysis of command and control comof the munication (C3) systems. In particular,hemakesmention erroneous results which arise from having “ignored the network’s structure once communication probabilities have been calculated.” An example of the difficulties which arise due to network structure is given by considering a simple system (see Fig. 1 ) whose link availability is assumed to be 90 percent. As Dr. Frank points out, it is incorrect to say that theprobability that neither A nor B receives the message is given by the calculation (1.0 - 0.81) (1.0 - 0.73) + (0.9) (1.0 - 0.81) (1.0 - 0.9) - 0.9) 1, G = [ (2k)1/2 and n = -kC.ZN x^(t) represents an estimate of z ( t ) and (1) is called the estimator equation. r ( t ) is the observable form of r ( t ) = C sin Coot + Mzl ( t ) ] + n ( t ) (3) where n ( t ) is a white Gaussian noise process with spectral densityN . The desired message is the second component of the process generated by ~ ( t= ) Fz(t) + Gw(t) where w ( t ) is a white Gaussian process, independent of n ( t ) ,with unit spectral density. The spectrumof the desired message is 0.051. However, the suggested correct solution is also incorrect. The proper solution is given by (1.0 ‘1, -k = 0.1171. Missile Complex A The demodulator described by (1) and ( 2 ) is based on small error approximations tothe conditional densityfunction derived by Kushner [a]. The demodulator is a special case of the extended Kalman filter [ 3 , p. 3381. Equation (2) is called the gain or variance equation. The steadystate solution of ( 2 ) is Commnd Port Missile Complex B Fig. 1. Exampleillustratingthe effect of overlappingcommunication paths. Paper approved by the Associate Editor for.Communication, Systems Disciplines of the IEEE Communlcatlons Society for pubhcatlon wlthout oral presentation. Manuscript received June 5, 1974; revised July 2 , 1974. The author is with Automation Industries, Inc., Silver Spring, Md. 1 H. Frank, I E E E Trans. Commun., vol. COM-22, pp. 589-595, May 1974. where y = (1 p = + 2pW2) 112 nL/k. Foroperationabovethe threshold (m2Vll< $), (4)and (6) accurately approximate the steady-state phase and message estimate variances,respectively. Unfortunately, the approximations do not show the nonlinearthresholdingphenomenon that characterizes nonlinear demodulators. Threshold Behavior of a Quasi-Optimum FM This correspondence presents a technique to obtain approximate Demodulator expressions, valid both above and below the operatingthreshold, for boththe phase and message estimate variances. The technique extends the work on the phase estimate variance by Snyder [1, pp. F. S. NAKAMOTO AND N. J. BER.SHAD 97-1021 based on the Fokker-Planck technique discussed by Abstract-An analytical technique is presented here forobtaining Viterbi [4]. The analytical techniquepresented here consists of two parts. approximate expressions for both the phase and message estimate variances for anFM demodulator. Numerical computations, for both The first part begins with the steady-state differential equations for the density functions of the phase and message errors. The projection is applied to obtain approxitheorem for conditional expectations mate solutions to the differential equations. The net result is the Paper approved by the Associate Editor for Communication Theory of theIEEE Communications Society for publication .without oral functional form for the phase and message errorvariances. The presentatlon. Manuscript received December 5, 1973; revised June 14, is present in both second part dealswith a jointstatisticthat 1974. This work was supported in part by the Offlce of Naval Research under Contract N00014-69-A-02000-9004. variance expressions. A Gaussian assumption is made to relate the The authors are with theSchool of Engineering, University of Califorjoint statistic to the individual variances. nia, Irvme, Calif. 92664.