W. J. WILLIAMS, MEMBER, IEEE Summary-The phase-locked loop has been suggested as a possible replacement for the banks of contiguous filters used to process Doppler signals. Two basic systems are considered, differing only in the type of phase sensitive detector used. The two types of phase sensitive detectors are the sinusoidal type and the sawtooth type. The sawtooth-type phase sensitive detector is shown to be superior in several respects. The sawtooth-type detector is supenor with respect to noise rejection, acquisition time and system design considerations. A phase-locked loop system using the sawtooth-type phase sensitive detector is shown to meet the requirements for a typical situation. INTRODUCTION W[HE PURPOSE of this paper is to select a phase j comparison technique which exhibits the most desirable characteristics for space sensor applica- tions. Contiguous filters are usually necessary for processing pulse Doppler radar signals. However, for most applica- tions, these filters are undesirable because a typical bank of filters may weigh up to 50 pounds and occupy 1000 cubic inches. For this reason, the phase-locked loop has generated considerable interest. The phase-locked loop is aphase sensitive device which has the property of behaving like a tunable adaptive filter. It tracks a given signal, but admits noise only in a small bandwidth. Once the signal has been acquired, the phase-locked loop has a very good tracking capability even in strong noise. However, acquisition over the entire range of Doppler uncertainty is difficult for the phase-locked loop unless some type of frequency searching scheme is employed. Even with the searching problem, the phase-locked loop looks attractive in situations where lightweight and small dimensions are at a premium such as in space sensor applications. PHASE-LOCKED LooP OPERATION The phase-locked loop diagram is presented in Fig. 1. Phases of the input signal and the output signal are compared in the phase sensitive detector. At the output of the phase sensitive detector is a voltage that is some function of the phase difference between the two signals. Manuscript received by June 11, 1964. This paper was presented at the IEEE Region Six Conference, Salt Lake CitY, Utah, April 30, 1964. The author is with the Electronics and Space Division, Emerson Electric, St. Louis, Mo. This output is filtered by the low-pass filter H(s) and applied to the voltage controlled oscillator (VCO). The VCO frequency is a linear function of control voltage. Any change in signal frequency will cause a correspond- ing change in VCO frequency, thereby tending to keep the input and output in synchronism. EQUIVALENT SYSTEM In this case, the Fig. 1 system can be replaced by the equivalent system in Fig. 2. It has been shown that the equivalent noise may be injected just before the low- pass filter. In the no-signal case, A is zero, so that the frequency of the VCO is moved about in some random manner by the noise. If the signal is of sufficient ampli- tude, the system will function as a feedback control system and will be able to lock on to the phase of the signal. The equivalent system is linear except for the f(e) box, and resembles a type II servo. sin tisSt+ 0 )+n(l ) PHASE f (e)+n*( t) l;72s+l ) SENSITIVSE nk H(s} T Is+I DETECTORf cos ut VOLTAGE CONTROLLEU OSCI LLATOR h O+Kvo e = ir-0O=(z5-<R t +9 n*(t) is n(t) olded about u = Fig. 1-Basic phase-locked loop system. Fig. 2-Equivalent system. PHASE COMPARATORS This paper is concerned with the selection of a phase comparator which has the most desirable characteristic for acquisition and tracking of a space target. The sinusoidal characteristic is quite well known be- cause it is the natural result of multiplying or mixing 230
Williams: Phase Sensitive Detectors for Space Radar two signals. The two signals might be A sin (w,t + 0) (1) B cos wt. The difference term resulting by multiplying the two signals is AR 2 sin [(co-)t +O] (2) where e = [(co.-w)t + 0]. The quantity is the instantaneous phase difference between the two signals. However, this type of phase comparator presents two difficulties: 1) the output is proportional to signal strength as well as being sensitive to signal phase and 2) the gain is dependent on phase offset for any system using this type of comparison. This gain dependence is undesirable because system design is optimum for a set gain. However, to reduce gain dependence, an AGC control is required, which in turn adds complexity. The sawtooth characteristic is not common but may be obtained in several ways. The signal and reference may have a sawtooth shape, thus giving a sawtooth characteristic to their phase difference. The gain in this case is not phase sensitive, but remains sensitive to amplitude variations. An alternate method of generating the suggested sawtooth characteristic is to use the zero- crossing information obtained from two signals, to switch a flip-flop in such a manner that the integral of its output is a quasi-linear or sawtooth function of phase difference. This type of comparator is not gain sensitive to phase or signal amplitude. The system gain is not dependent on phase offset or signal amplitude in this case. The zero-crossing information may be obtained in a forward part of the loop, but the model remains the same for the linear portion of the comparator char- acteristic. correlation intervals, M is 7/2 \M PBL = 1 - ( Pn(4)d4) PBL = 1 - P.(O)d+ -X (sinusoidal) (3) (sawtooth) (4) where Pn(4) is the distribution of the phase noise. Re- sults may be obtained if Pn(s) is assumed to be Gaussian in distribution with a zero mean and a variance which is dependent on the input noise. By the approximation for large X,1 rx ~~~~1 1 -f(x) = g(y)dy < e-1/2x2 xz xV27r (5) and the binominal expansion, the expressions for break- lock probabilities become. 4oM 4PBL < -/2 (7r/2of)2 PBL .-- -- r-\/2r 2(JM PBL < e-1/2(r/a)2 rV2ir (sinusoidal case) (sawtooth case). (6) (7) The correlation period may be approximated as the reciprocal of the noise bandwidth. Table I presents the results for an M of one and some selected noise powers. As noted, the probability of break-lock is much smaller for the sawtooth comparator than for the sinusoidal comparator for equivalent noise power. The phase noise variance may be found as a function of input noise. 2= B, 2SINo (8) where B,, is the system noise bandwidth S/No is the input signal power to noise density ratio. Rey2 shows that for the case of the sinusoidal compara- tor the system noise bandwidth is THE BREAK-LOCK PROBLEM The system is assumed to be locked to the phase of the return signal from the target. The target can be lost in two ways: 1) the phase change of the return signal can exceed the tracking rate of the phase-locked loop, or 2) noise can cause a break-lock condition by exceeding the positive slope range of the phase com- parator. The loop will follow any signal whose band- width is less than the loop bandwidth. Widening the loop bandwidth admits more noise which increases the probability of break-lock due to noise. Fig. 3 illustrates that break-lock will occur if the magnitude of e exceeds 7r/2 in the case of the sinusoidal comparator, or 7r in the case of the sawtooth comparator. Break-lock prob- ability for the two cases, for a particular number of Bn=Kt Sec .1 + cos (e)KtT22IT, cps 2 1 + KtT2 cos (e) where K,=KA/2 in this case. Bryne3 shows that for the sawtooth comparator Ko 1 + KoT22/T cps Bn = 2 1 + KoT2 (9) (10) I E. Parzen, "Modern Probability Theory and Its Applications," John Wiley and Sons, Inc., New York, N. Y; 1960. 2 T. J. Rey, "Automatic phase control: theory and design," PROC. IRE, vol. 48, pp. 1760-1771; October, 1960. 3 C. J. Byrne, "Properties and design of the phase-controlled oscillator with a sawtooth comparator," Bell Sys. Tech. J., vol. 41, pp. 559-602; March, 1962. 231
232 IEEE TRANSACTIONS ON AEROSPACE AND NAVIGATIONAL ELECTRONICS December OUTPUT [. UNLOCK I7 at iF At (a) fseize = flock T To insure a good possibility of lock, the search must place the initial conditions within a positive lock area which is constrained by the seize frequency and the discontinuities. The phase plane is a good vehicle for discussion of this problem. Since frequency difference is the derivative of phase difference, the use of the phase plane is convenient for this application. Fig. 4 illustrates the situation for sinusoidal and saw- tooth phase-locked loops. In order for locking to occur, a search trajectory must pass through one of the areas proscribed by the seize frequency and the discontinuity limits. If the search is a ramp of frequency, the search trajectories will be parabolas. The phase difference in such a case will be _ UNLOCK e = at2 + eO so that (b) Fig. 3-Sinusoidal and sawtooth phase comparator characteristics. (a) Sinusoidal phase comparator. (b) Sawtooth phase comparator. e'= 2at. This is a parametric representation of TABLE I BREAK-LOCK PROBABILITY FOR ONE CORRELATION PERIOD Noise Power (a') Probability of Break-Lock Volts Squared Sinusoidal Comp. Sawtooth Comp. 1.OXt10 5. X 10- 7 5.OX 10-1281 1.OX10-2 3.7X10-57 3.5X10-228 1 .OX10-l 1.8X10-7 8.0X10-27 1.0 1.4X10-1 3.5X 10-7 SEARCH AND ACQUISITION Some type of search is required to obtain initial lock. Initial lock is necessary to assure positive locking in a reasonable length of time. It is possible that in the noise-free case a phase-locked loop may lock for a rather large frequency offset. In the locking process, many dis- continuities may be passed, and therefore the circuit may flicker for a long while before locking. Locking is very unlikely when the frequency offset is large and the signal-to-noise ratio is small. If the offset frequency is less than the seize frequency, locking will be quite probable even in noise. The seize frequency is the maximum mistuning or offset that cannot cause a dis- continuity to be crossed in the locking action. The seize frequency has been found to have the following relationship to the maximum lock frequency for the systems considered here3' 4 4 D. Richman, "Color-carrier reference phase synchronization accuracy in NTSC color television," PROC. IRE, vol. 42, pp. 106-133; January, 1954. (13) - + eO = e. 4a The frequency may be discretely searched in a linear fashion, in which case the search trajectories appear as in Fig. 5. The search trajectory should pass through a lock point as it passes through the locking area. Since the initial phase difference is unknown, the search trajec- tory must accommodate all possible initial phase differ- ences so that the trajectory passes through a locking area for any value of phase difference. The rapid search trajectories may easily miss the scattered positive lock areas of the sinusoidal case so that slow search is essential. The sawtooth comparator produces a con- tinuous corridor, however, which cannot be missed. The discrete search is more attractive than the con- tinuous search since it can be digital in nature. The voltage controlled oscillator is simply advanced in fre- quency in discrete steps with the locus of frequency be- ing a ramp function. If the system locks onto the signal during one of the steps, the stepping sequence is inter- rupted and the system then is free to follow the signal over the entire range of frequency and phase. The time required to search the frequency range of interest may be found from the search trajectory. If in (13), eo is considered to be zero, e/2 = e. 4a (13a) 77; 2 UNLOCK (11) / UNLOCK (12)
Williams: Phase Sensitive Detectors for Space Radar The constant may be evaluated for the desired trajec- tory by choosing a typical point on the trajectory. Thus, FAST SEARCH 2 (HIT) Ws__ FST SEARC (AISS) a = - . (14) 0SLO SEARCH -27T ir -7ff0'; 2ff/2 57 27r L The time for the trajectory to pass through the area of LOC| LOCK AREA interest is 8w W T =-zL de'. (15) (a) The time required to search the entire frequency range is 8T, T's ~~~~~~~(16) . FAST SEARCH T2W s(17) (H T) _ FAST C ASEARCH / / e~~~~~~~~~HIT) since LOCK AREA -7r | (A s( = _max (17) '4 ~~~~~~and \ maX= 7r BW (18) 8 T~2 (b) T(8= (- ) (19) Fig. 4-Four search trajectories and locking areas with continuous BW 2 ramp searching. (a) Sinusoidal case. (b) Sawtooth case. The phase-lock loop system must lock during the time that the search trajectory is in the lock area. There must then be sufficient time to determine if locking has occurred so that searching may be stopped. The slow search necessary for positive lock in the case of the sinusoidal comparator is not practical for typical situations. The search time for a typical set of parame- ters will be in the order of several seconds, when it must l /2 2 l l | | | tS _ be fractions of a second. A rapid search has a probability Ws T<_ _*of slightly greater than 0.5 of achieving lock on a single SLOW SEARCH pass. This probability improves for additional passes but this again increases total search time. FAST SEARCH (hIISS) | HBecause of its continuous acceptance corridor, the FAST SEARCH (HIT) sawtooth comparator offers considerably more promise. (a) The probability of lock in a single pass approaches one if the system is designed to respond rapidly enough. e The system speed of response is dependent on the damp- ing ratio and the natural resonance frequency. The term (e - Vw,t) represents the damping term of the re- sponse of a second-order under-damped system in the linear region. ___SLLW SAIRH +1 + K]'2 K TA - i eL T, lI e + T-e= T . (20) FAST SEARCH (HIT) Thus, (b) 1 + KT2 Fig. 5 Search trajectories and locking areas with discrete jump 2= * (21) searching. (a) Sinusoidal case. (b) Sawtooth case. 2T1 The settling time for the damping term to decay to 0.05 is 1964 233
234 IEEE TRANSACTIONS ON AEROSPACE AND NAVIGATIONAL ELECTRONICS December 6T1 tsK= *(22) 1 + KT2 To clarify the situation let us consider an example. Suppose that it is necessary to acquire in 0.2 sec, in a bandwidth of 20 kc. Further, the signal-to-noise density is 1.8 X 103. To detect satisfactorily, a filter bank of 100 filters (each with a 200-cycle bandwidth) would be necessary with 8 db taken as the minimum signal-to- noise ratio for detection at a filter. The following selection of parameters is presented as being satisfactory for the situation using the sawtooth comparator system. T2/ T1=0.04 K=2rX104 T2=2X10-4 T, 5X10-3 T search 0.2 sec Frequency step rate 130 steps/sec. These values produce with the equations previously derived, Bn= 1256 cps PBL < 1.26 X 10-7/correlation period PBL < 1.58 X 10 4/sec Time to lock<1.9X10-sec Time detect lock > 1.9 X10-' sec. These values are well within the realm of possibility. It is interesting to consider that the search time required for the sinusoidal system for positive lock will be 8 /T1 2 Ts B ( ) = 1.0 sec. BW KT2J CONCLUSION It appears from the results that the phase-lock loop system using the sawtooth comparator based on zero- crossing information is considerably superior to the sinusoidal comparator system both in break-lock proba- bility and acquisition time. A suggested system for a phase-locked system with search is shown in Fig. 6. Fig. 6 Phase-locked loop system with search. A Unified Approach to the Error Analysis of Augmented Dynamically Exact Inertial Navigation Systems SAMUEL L. FAGIN, SENIOR MEMBER, IEEE Summary-A model for pure inertial navigation systems is de- fined which describes a wide class of dynamically exact systems for terrestrial and space applications. Based on this model a standard error block diagram is described and it is shown how the conventional simplifying assumptions lead to the specific version used for slow moving terrestrial applications. To efficiently treat augmented inertial systems, a generalized approach to error analysis of estimation systems is introduced. This approach, discussed in an earlier paper, is amplified as it pertains to augmented inertial systems. It is shown how the standard error Manuscript received August 14, 1964. This paper was presented at the 1964 \Western Electronic Show and Convention, Los Angeles, Calif. The author is with the Inertial Division, Sperry Gyroscope Com- pany Division of Sperry Rand Corporation, Great Neck, N. Y. propagation block diagram can be used to place the augmented in- ertial navigation system error analysis in the canonical form neces- sary for a generalized error analysis. Augmentations considered are position, velocity and attitude fixes. Specific examples are given. I. INTRODUCTION I N A RECENT paper,1 it was pointed out that, assuming optimal use of redundant data, a wide class of "estimation systems" could be conveniently studied employing a generalized digital computer im- 1 S. L. Fagin, "Recursive Linear Regression Theory, Optimal Filter Theory, and Error Analyses of Optimal Systems," presented at the 1964 IEEE International Conv., New York, N. Y., March 23-26.