CESIUM AND RUBIDIUM FREQUENCY STANDARDS STATUS AND PERFORMANCE ON THE GPS PROGRAM M. J. V a n M e l l e GPS N A V S T A R O p e r a t i o n s Rockwell Spare and O p e r a t i o n Center Falcon A i r Force B a s e C o l o r a d o Springs, CO
This paper is an update of the on-orbit operational performan~eoJ the frequency standards on the last Block I NAVSTAR satellite (GPS-lo), the complete BlocL I1 NAT'STAR satellites (GPS-13 to 21) and the Block JM NAVSTAR (GPS-22 to 40) satellites. Since the status of the GPS constellation is now at FuU Operational Capability (FOC), a minimum of twenty-four satellites are in position with all the necessary tests successfully completed. The evolution of frequency standards on board the GPS vehicles will be presented with corresponding results. Various methods and techniques will be presented to show on-orbit life time, down time, state of health telemetry, on-orbit trending and characterization of all the frequency standards. Other topics such as reliability, stability, clock quirks and idiosyncrasies of each vehicle will be covered.
INTRODUCTION The evaluation of the space-rated freqnency standards on the GPS program started with the Block I concept validation program and the full-scale development vehicles of which only one is still functional: GPS-10 (PRN-12). The production vehicles are divided into two groups, Block I1 (GPS-13 through 21) and Block IIA (GPS-22 through 40j. Each vehicle include!: two Rubidium Frequency Standards (made by Rockwell) and two cesium Frequmcy Standarc's (made by Frequency and Time Systems as the prima~ysourcc and, Kernco and Frequency Electronics Inc. as secondary sonrces on selected vehicles). The cesium clocks are considered primary because of their degree of radiation hardness, their extremely low frequency drift, or aging, which does not require any Kdlman filter modeling, and the shorter modeling time between turn-on and activation for GPS users. The actual on-orbit GPS Frequency Standard operating history (shown in Figure I for the last Block I and all Block I1 satellites, and in Figure 2 for the Block IIA satellites minus the four
vehicles in the Eastern Launch Site awaiting launch) illustrates the results of these hardware implementations. The operating life history of the production models of both cesium and rubidium frequency standards will be briefly discussed. This will be reviewed in order to calm the doubting Thomas's or Henny Pennies, that the sky is not falling in regard to (1) the amount of disabled clocks that have recently been occurring, Dec 94 - July 95, (2) the reliability of the clocks and (3) the combined projected lifetime of the four clocks (7.5 years) on each of the vehicles. A brief history of the rubidium clocks on the Block I vehicles is given in Table 1. The major problems were corrected via modifications (the final modified clock for Block IIflIA is Modification number 12). The non-generic problems were never repaired. From the sample of 30 Block I n~bidiumslaunched, the average age was 1.5 years with a maximum of 12.5 years and a minimum of one day. The minimum acceptable hardware reliability requirement for a five-year life rubidium clock was 0.765, which equates to a 3.8 year projection life.
The history of the rubidium clocks on the Block IVIIA vehicles is given in Table 2. Of the six disabled clocks, four may be retried with possible degraded performance. The final production model #12 RFS's have not acquired much on-orbit operating time, since the cesium clocks have traditionally been preferred over the rubidiums. This is because of the advantage of cesium over rubidiums in terms of radiation hardness, lower drift rate by a factor of 100, no Gfield tuning or frequency biasing needed, and a shorter warm-up time before the vehicle can be set healthy (2.6 days versus 6.4 days on the average). There have been eleven turn-on: with six powered down, for a total time of 120 months or 87,380 operational hours, as of 110 November 1995. Since the hours of operation (sample size) are so small, a point-in-time fail~lrerate estimate must be used. If the two failures are used, then the calculated failure or an Mean Time Between Failure (MTBF) of 4.3 years. If the six disabled rate is 26.5 x and an MTBF of clocks are considered complete failures, then the failure rate is 79.6 x 1.4 years. The operating history of the cesium clocks on the Block I vehicles is as follows: a total of six clocks (three preproduction models and three Model 1 production clocks) with an average life time of 5.9 years (Maximum of 9.3 years and a minimum of 3.3 years). Since five of the failures were caused by cesium depletion, the final production model (Model 2) had an increase of cesium fill (1.0 grams to 1.5 grams). The history of the cesium clocks on Block IIJIIA vehicles is given in Table 3. Of the thirty-one CFS's powered-up, nineteen are still operating, with an age range of 6.5 years to seven months. Of the twelve clocks which have been disabled, six have been labeled failures and six may be given a second chance with possible degraded performance. If the five failures, excluding one GFE clock, are used, then the calculated failure rate (via the point-in-time failure rate estimate) is 7.2 x and the MTBF is 15.8 years. If the ten disabled clocks (excluding the two GFE clocks) are considered failures, then the failure rate is 1 4 x and the MTBF is equal to 7.9 years. The manufacturer signed up for a minimum acceptable hardware reliabilitv requirement for a 7.5 year life of 0.663, or 4.3 years per clock. Taking the reliability numbers uf both rubidium and cesium clocks, plus having to meet the navigation payload reliability number of 0.934 for 7.5 year life, the number of clocks per vehicle came out to be two rubidioms and
two cesinms. Another figure to remember is the mean mission duration value of six years, a specification which five vehicles have already surpassed. In summary, the complete GPS Block IUIIA clock status is included in Table 4.
ON-ORBIT PERFORMANCE In order to acquire the exact performance characteristics of the operating on-orbit frequency standard, the L-Band signal must be evaluated. This signal is affected by the (Frequency Synthesizer Distributor Unit) FSDU (which is commanded by the NDU), atmospheric effects, ephemeris uncertainties, monitor station variations, spacecraft effects and other factors. All of these factors are fed into a Kalman filter, which is a computer algorithm for processing discrete measurement data in an optimal fashion. There are several parameters which are instrumental in evaluating the operational performance of the frequency standards. The first two parameters are in the navigation message. One is al, which is the frequency offset (sec/sec). This is the filter's estimate of the frequency difference, or offset between the satellite's frequency standard and the GPS composite clock (a nominal frequency). This is a continuous absolute value. One can also take the daily average of the difference between the minimum and maximum values of a1 as a possibli: trending signature. Another parameter is the frequency drift in sec/sec2, a 2 term. This is the rate of change of the drift term. Another parameter that is used daily to evaluate the clock's performance is the Estimated Range Deviation (ERD). An ERD is the difference between a range determined from the aposteriori state estimates during a Kalman interval and the range determined from the navigation trpload data that is valid for the same time. These ERD's compare the current filter estimates each 15 minute period to the prediction made from previous filter estimates (considered to be a minimum range error either induced primarily, by clock movement or satellite positional change). Examples of these ERD's are in Figures 3 and 4. Plots of these estimates provides us one more clue of evaluating the performance of each spacecraft's clock. Continuing the investigation of a potential clock problem, a correlation of these ERD plots to the telemetry monitor values must be examined. Along with these clock monitor values, th: a1 and a2 terms must be observed for movement. One important aspect of Kalman filter operation is to provide accurate continued measurement updates, every fifteen minutes. Unfortunately, there are periods when the spacecraft is not in view of a monitor station, and the filter must estimate aging through the a2 term, with no real measurement verification. Also, different monitor stations (with obvious different clock errors) contribute errors into the filter estimation and subseqoently the prediction process. An operating limit is set on the ERD value in terms of meters (eg. 10 meters or 8 meters). When the limit is exceeded, a new irpload must be sent to the vehicle to correct the new terms in the navigation message. A new experimental limit has recently been defined as 5 meters, resulting in a fifteen percent improvement in the URE. This also has increased the work load on the Master Control Station (MCS) crew, which now performs approximately twelve more uploads per day on the 25 vehicles. As a rule, the MCS contacts each vehicle twice a day,
once to update the navigation message. This equates to a total of 60 t~ 70 supports per day for the MCS crew. Another set of parameters which appear to effect the clock's performance are environmental effects, such as prolonged radiation effects (i.e. passing through the Van Allen belts every twelve hours) and irregular solar activities. Thermal variations, either induced by delta-v maneuvers or eclipse seasons, appear to be causing the older cesium clocks (> 5 years) the most problems in terms of ERD's. Maybe aging of the electronics, resulting in a degraded temperature coefficient causes frequency changes. The eclipse season causes the cesium clock's temperature to decrease by three degrees centigrade with a f l o C variation. The rubidium does not have this problem since a heater (ABTCU) keeps the nlbidium clock at a stable temperature, 0.1" C. When the satellite enters eclipse season especially during the first eclipse season, an ephemerischange could also occur, which is corrected by manual intervention from the operators.
ON-ORBIT TRENDING The most important objective of trending analysis is to determine when a particular frequency standard is no longer useful for providing a navigation signal. Particularly elusive is the time frame - whether it be in days or months - when a clock will expire. Note that this is different one d r , than not meeting specifications. The stability specification of the clock, 2 x 10-'"t for the cesium clock, is so tight, that if the clock is performing at 3 x at one day, thr URE of 4.8 meters (1 s) can still be met with extra maintenance by the MCS, for the space segment. There are several vehicles now that do not meet the stability specification, but the Air Force is reluctant to switch to another clock. The Air Force will determine when a clock will be disabled by many factors., These factors include: 1. how burdensome to the MCS crew are extra daily uploads and/or Kalman maintenance in order to correct the a1 and'a;? terms? 2. how old is the clock (> 5 years)? 3, how old is the vehicle (> 6 years)? 4. how many clocks are left to be tried?
5. what is the world situation (conilicts/trouble spots)?
6. what is the condition of the vehicle in terms of performance operation and other subsystems? and 7. what is the condition of the entire constellation?
The navigation signal must be made available 98% of the time with 21 spacecraft. In predicting the useful operating lifetime of the clock, the most important performance parameter is the stability of the clock. This is what most effects the user, and is the most sensitive parameter.
The next set of parameters are the a1 and a2 terms and their deterioration andlor fluctuations, and the ERD's. The last set of parameters which effect the performance of the clock and that of the navigational signal is internal to the clock. The cesium clock has 18 monitors (combination of analog and digital) and the rubidium clock has 11 monitors (combination analog and digital). Of all the telemetry monitors on the cesium clock, there are only a few that could vary and not effect the performance of the clock. The rest of the monitors will cause an upset of the performance by any detectable movement (minimum step size). One of the monitors having particular character or individuality is the cesium beam current monitor. This trending parameter is hard to interpret in the sense that each of the 19 operating clocks has a slightly different signature as seen in Figure 5. SVN-17 has the normal stair-stepping decline in heam current. Since each clock starts off at a different absolute value, each has a different rate of decline (the higher it starts, the faster it drops) and each has a different final plateau. So each drop in beam current may or may not effect the stability, a , or a2 terms, or ERD's. Furthermore, each clock will degrade or age at a different rate. Even though each clock is huilt to the same specification and from the same set of drawings, when one compares stability performance in terms of parts in l0I4, there will be variations in their outputs. The other parameter that might change without detrimental effects on the performance is the loop-control voltage, which normally will move slightly one way or another, depending on what electronic changes or aging occur in the loop in order to keep the same 10.23 MHz frequency output to the FSDU. The parameters which are catastrophic to clock performance if any movements are observed are the cesium oven temperature, R F level (power shift and or spectrum change), electron multiplier gain changes, ionizer voltage, and any input current changes to the total clock or to individual units such as the quartz oven. The rubidium clocks have the same type of monitors and the same type of loop control voltages. The lamp voltage monitor which detects pressure changes and photo cell degradations is somewhat similar to the cesium beam current in terms of end-of-life predictions. To predict the exact ( one week) end-of-life of either type of clock is extremely hard. This was tried on SVN-20 with Cesium No. 3. After 4.5 years of operation, the stability was > 2 x lo-'" one-day with four to five extra uploads needed per week. Maybe two to three months of less than useful life could have been squeezed out of the clock. Other parameters that are incorporated into the trend analysis include on-orhit temperature of the spacecraft, any FSDU - NDU infltrence, or L-Band effects.
SCHEDULES The last of the Block I satellites, NAVSTAR 10 (PRN 12), launched in 1984, is scheduled to be disposed of in the June 1996 time period. The main problem is that the solar arrays have lost their efficiency (design life of five years) and can no longer support the navigation payload. On November 18, 1995, the payload was set unhealthy. Kalman filter tests, frequency standard tests, sun sensor test, etc., will be performed in February and March 1996. The two rubidium frequency standards, yet to he powered up after 12 years of on-orbit storage, will be tested for stability, temperature coefficient, VCXO and turn-on characteristics and any other tests the
clock community would like to have performed. The last Block IIA satellite launched, GPS-37, reached orbit in March 1994. This completed the 24 satellite constellation. There are four vehicles in the Eastern Launch Site in storage waiting to be launched. The next launch is planned to be positioned in "Plane C" in March 1996. There are available launch slots for summer 1996 time frame. The total on-orbit times for both rubidium and cesiums are staggering for the first operational satellite system ever to utilize both types of production frequency standards. The on-orbit times for all rubidiums exceed 60 years of operation, while the cesium on-orbit times are more impressive with over 125 years of operation. The GPS clock utilization times in their operational sequence are shown in Figure 6.
CONCLUSION As verified by on-orbit performance data, most of the major generic problems, especially with the rubidiums, have been corrected. I will admit that the rubidium short life times, the phase jumps that occur within the first 3 to 4 months of operation and the changing drift rate within the first 6 months are on-going problems. There have been 31 out of 48 cesium clocks activated with 19 currently operating. Of the 12 disabled clocks, half may he reactivated with possibly degraded performance. There have been eleven rubidium clocks activated with 37 remaininpl to be turned on. Of these eleven clocks, five are still operating and four to be reactivated f c ~ future use. The average age of all the disabled clocks is 1.65 years. The average age of the currently operating clocks is 2.9 years. The average age of the space vehicles is 4.5 years, which equates to 60% of the design life (7.5 years). The total number of clocks turned on is 42, which equates to using only 44% of the available clocks. The usage and performance to date indicates that the number of clocks (four), originally determined in the 1982 proposal, will support both the spacecraft design Life of 7.5 years and the mean mission duration of 6.0 years. For the more quick-look-managerial type, a user-friendly smiley face chart, Figure 7, has been concocted in order to eliminate reviewing all the Kalman drift rate residuals, Allan variance stability curves, and ERD's figures. Each little qnirk and idiosyncrasies of the vehicles combined with clock performance in terms of ERD's are for your (management) eyes only. ACKNOWLEDGMENTS The author would like to acknowledge the following companies and services who were involved and contributed to the GPS frequency standard program over the past twenty years: Air Force Space Division (JPO) and ZSOPS (FAFB) w
Naval Research Laboratory (NRL)
National Institute of Standards & Technology (NIST)
Rockwell (RFS Manufacturer) w
Frequency and Time System (CFS Manufacturer)
Kernco & Frequency Electronics Inc. (Second Source CFS Manufacturers)
. TABLE 1
RUBIDIUM FREQUENCY STANDARDS
- BLOCK I
TYPES OF PROBLEUS
VCXO DRIFT R A E ATOMIC LOOP *
OPERATIONAL TO END
NOMlNAL TURN-ON (TESTS)
AGE (YR) Hi Low
0.53 10-3 27 12.5-0.4 1.9 5 1 - 0.03 28 05 01 05 l DAY 1 7 7. ... 125- M I 12 3.5 . 0 1 l MONTH
.. ,.... .. .. .. I5 12.5 - M 3 'ONE STILL AVAILABLE
HOW FIXED MOD. PiN 3 MOD. P M 4 MOD. PM I I NONGENERIC NON GENERIC NON-GENERIC
NIA NIA NIA
R U B I D I U M F R E Q U E N C Y S T A N D A R D S B L O C K I1
DISABLED 6 C L O C K S
SYMPTOMS ERRACTIC MONITORS HEATER CIRCUITRY; L A M P VOLTAGE ERRACTIC DRIFT RATE LARGE:
FREQUENCY M O V E M E N T S INSIDE T E M P E R A T U R E COWROLLER FREQUENCY M O V E M E N T S DRIFT R A T E L A R G E
02 A V E = IOYR OPERATIONAL5C W C K S O L D E S T = 1.3YRS. A V E R A G E = 0.8YRS.
CFS BLOCK I1 VEHICLES
No. of CFS - 19 Operating
' 6 YRS. > 5 YRS. > 4 YRS. > 1 YRS. > 2 YRS~ i YR. c l YK.
12 -.DISABLEDA G E
4.3 YRS. 0 > 2 E-13 ;U I day 3.5 YRS. VCXO OR SERVO 2.8 YRS. FUSE 2.5 YRS. ' VCXO: AF 2.3 YRS. *SECOND SOURCE RF 2.2 YRS. 35 DAY CYCLIC PATTERN 2.1 YRS. CBILOW. HIGH BACKGROUND NOLSE 1.8 YRS. SECOND SOURCE - RF 1.1 YRS. EMULT: Af .7YRS. VCXO, DAC, SERVO or EMULT 0.1 YRS. ' SOLAR COEFFICENT 0. I YRS. D E S E R T STORM: Al ' AVAILABLE FOR POSSIBLE DEGRADED OPERATION
- NOV. 24.1995
- GPS B-TJI~
lkh JCA 1C1 ?i
IYh ?Yh JCs 4C>
28 kith 1Yb
SPARE SL'SI'IXT orb:funx(i SPAUB
IYh ZYb JCI
4u 29 0-
rzr IrUu1 d s . a . ~ 7 .~a."o10d.p
.SP.\YS SPAYC: OI'ER\llNl: SP,\RE
U l M l MAR %
A i m . 0.8 s [email protected]"id~, a n .Mi
cr.l.4 la." @ L d s.0-13 % l o " id llrJ'I. h z u n iJa = la " h h v n n .Ws
5 0 4C1
32 o m l l ~ l U Y O 1 d r r i a M hF m c h m c m 1 1 / Z W l d 1-rarure 8~rnUu k ~ b hkr l i b P-t VCS L.d U4.) fmq - f l y 1 q d k p l t m 4 t h h q e
A U t i rn
Smnd Sawrt C-m
r r l FW;SYD -p
A d w U x 10'"kia~hrpr W H A Y 94 NONE
' s .ad drlh nu d w w I= URb I l m l l h p m Y 1 ~ a . - 1.2 x I T " @ Id.yIwZfRb
I)*rurbnct ~bs=LW+ A I l T CU .U=nytC,o r Ob x IC" @ Idry. b p w - 1 0 0 s tOm'hlalun 1IYh. 8 . 1.91 L f " Q 1dS. 1.b~. EHI) *mu?: CHI c 1.b II,~ ua J(C& R-arth. l n u x t
IYb IYb K-d
:nns: o = u.:
1 5 x l ~ " 0 IU'S. 1 2 xlO."Q 1 0 d . y ~ d n r . IJ r 11."16~!. nn .VC&
Illh ?Yb >Us
AV I n 9lW. I'mhkna rr n~
H w. V I W . Cwqww mnm*nrrun lrl(b a = 0-4 I I U " B 1 dry. L'.aUd -,lor Aylnl: .SJ s 1 ~ " l d a y .
h k h nn%
r 2: BLOCK llA FREQUENCY STANMRD CONFGURATION
ICK IIA FREOUENCY STANOARO GCNFIGURATIGN
c ~ a w m mnsism ~v I , " . " , ,
1 rspr I
I . . ? , ,
tSllt!AlEO RANGF D E V I A I ION FOR SVll-20
~ L . I
m UNCLASS~FIEU rn RMS-ERD FOR SV: 2 0 m-rn ----
EIQ : *.st C I Q - 4.3. ( 0 1 ..*I
, , - ,-MIX
m . 3
n = o
DRIFT STATE RESIOUAL FOR SV: 20
tr m n
E C s.ooma¶ P
;e.oom,m D Y
m?aun omarm (ul UU
- a w u w s am-2
FSTINATFD RANGE D N I ATION FOR SVN-79
- nLI?. .
na.8 1 1 ~ RO s : 9.
8 . 8
FOR SV: 29
YU E m : I.U 0..I
r . r
8.1 ru.s ma vmma4e.m I . . I I . U ~
OWFT STATE RESlDUU FOR SV: 29
FIGURE 5 C e s i u m Beam Current for Three Different Clocks Data from 11/1/94 to 11/1/95
..................................................... .........,.............................. .....
.......-.SVN ... 17.....;-...-...( I . . &
Rockwell GPS Operations, Folcon AFB, Colorodo
GPS CLOCK UTIL !ZATl ON I N OPERATIONAL
I 1 I I
IN OPERATIONAL SEQUENCE
1 1 t30195
AVERAGI AGE 24 OPERATING CLOCKS 2 885 YRS
WERAGE AGE 18 DISABLED CLOCKS 1 Irc5 YRS
Rb Clock 2 Horlz
Cs Clock 3 Solid
ROMAN NUMBERAL = PARTITION NUMBER ARABIC m E R = NUMBER OF CLOCKS TURNED ON
Questions and Answers DAVID ALLAN (ALLAN'S TIME): Given our opening talk by Captain Foster and work that Dr. Winkler did some years ago on measuring lifetimes of clocks, I wonder, given the importance of the lifetime of this system, why we're still using MTBF rather than half-life, as was recommended by Dr. Winkler. That was a very excellent piece of work, and it's a much better measure of lifetime than MTBF.
M.J. VAN MELLE (ROCKWELL SPACE AND OPERATION CENTER): Right, I still don't - this is a reliability person that I gave all the data to who said that all I can figure out was that it seemed to be was a little high. Plus, I think most of the problems that we had may be workmanship. I don't know how hard it is. Remember, this is the first production vehicle we've ever had with production rubidium and cesium clocks on board. To me, it's still in the infant stages. But to answer your question, I don't know. It probably would be a little better the way you're suggesting.
DAVID ALLAN (ALLAN'S TIME): Well it's the work that Dr. Winkler did on those clocks some years ago, and it's a very excellent measure. Perhaps, in terms of this being a PTTI planning meeting, it's something we should think about for some future representation. I don't know whether we can do anything here, but it certainly is an issue. M.J. VAN MELLE (ROCKWELL SPACE AND OPERATION CENTER): Sounds good to me. MARTIN BLOCH (FEI): Van, we've discussed this many times. You say workmanship, but something bothers me. If you take a look at the performance of similar clocks that were made by the same manufacturers on the ground, they outperform the space segments significantly, where life is over 10 years on the cesium and the same on the rubidium. I'm wondering if there's some other reason that we're overlooking on what is happening in space - it just sounds that the number of failures, with all the care that space hardware is supposed to take, that workmanship doesn't sound to me is a good excuse; unless what you're implying is that we do worse for space hardware than we do for military or commercial hardware. M.J. VAN MELLE (ROCKWELL SPACE AND OPERATION CENTER): Well with the rubidium, you know, Rockwell made it, Rockwell is not a clock manufacturer. They took Efratom's physics package and they took your oscillator, and we just packaged it up and put it together; and tested it only like three or four months on the ground. Then we launched it. Maybe during the six-month period, it would have a failure on the ground. Maybe the launches affected it, but we still do a lot of fault testing with vibrat~Jn and so forth. We only vibrated it once; you know, maybe the second vibration killed it. But, the FI?)S's are production clocks made by the cesium manufacturers, and they seem to have a little longer lifetime. I can't explain why the Block-I's are lasting more than the Block-II's. But then, we only had a sample of six; and here we had a sample of 19; maybe the sample size was too small compared to the complexity of the standards themselves.
MARTIN BLOCH (FEI): I wonder if anybody else has some thoughts as to the other effects that influence the life performance. M.J. VAN MELLE (ROCKWELL SPACE AND OPERATION CENTER): And we're talking about if it goes to 5 x lo-'" , i t's no good.