148 IRE TRANSACTIONS ON INSTRUMENTATION December

sult of one good experiment rather than the results of [20] N. E. Dorsey, Trans. Am. Phil. Soc., vol. 34, pp. 1-110; 1944.[21] J. W. M. DuMond and E. R. Cohen, Rev. Mod. Phys., vol. 20,many poorer ones. This does not mean that the other p. 82; 1948.experiments are worthless in the evaluation. If the re- [22] B. Edlin, J. Opt. Soc. Am., vol. 43, p. 339; 1953.[23] L. Essen, J. IEE, vol. 93, no. 9, pt. 3A, p. 1413; 1946.sult from Froome's last experiment had differed greatly [24] , Nature, vol. 159, p. 611; 1947.from the central value given by the other methods the [25] Proc. Roy. Soc. (London), A, vol. 204, p. 260; 1950.

[26] ,Nature, vol. 165, p. 582; 1950.presence of some unknown error in it would be sus- [27] Proc. Roy. Soc. (London), B, vol. 66, p. 190; 1953.pected. This experiment demarcates the value of c [28] L. Essen and K. D. Froome, Nature, vol. 167, p. 512; 1951.[29] --, Proc. Phys. Soc. (London), B, vol. 64, p. 862; 1951.within closer limits than all the other experiments. [30] L. Essen and A. C. Gordon-Smith, Proc. Roy. Soc. (London), A,On the basis of this evaluation the author concludes vol. 194, p. 348; 1948.

[31] K. D. Froome, Proc. Roy. Soc. (London), A, vol. 213, p. 123;that the best value for the speed of light is around 1952.299,792.5 kms, that it is improbable that the true value [32] -, Nature, vol. 169, p. 107; 1952.

[33] Proc. Roy. Soc. (London), A, vol. 223, p. 195; 1954.differs from this figure by more than 0.1 km, that it iS [34] Proc. Roy. Soc. (London), B, vol. 68, p. 883; 1955.higlhly unlikely that the true value differs by as much [34] Proc. Roy. Soc. (London), B, vol. 68, p. 883; 1955.

[35] ,J. Brit. IRE, vol. 16, p. 497; 1956.as 0.3 kms, and that it is more probable that the true [36] Nature, vol. 181, p. 258; 1958.value lies above this figure than below it. But the author [37] R. A. Houstoni, Nature, vol. 164, p. 1004; 1949.

[38] Proc. Roy. Soc. Edin., vol. A63, p. 95; 1950.cannot express his confidence in this view by a meaning- [39] A. Huittel, Ann. Physik, vol. 37, p. 365; 1940.ful number. [40] I. C. C. Mackenzie RE, "The Geodimeter Measurement of theRidgeway and Caithness Bases 1953," Ordnance Survey Pro-

REFERENCES fessional Papers, New Series No. 19. (See also: Ordnance SurvevProfessional Papers, New Series No. 18 and "Empire Survey

[1] C. I. Aslakson, Trans. Am. Geophys. Un., vol. 30, p. 475, 1949. Review," vol. V, no. 34, pp. 211-225.)[2] Nature, vol. 164, p. 711; 1949. [41] D. WV. R. McKinley, J. R. Astr. Soc., Can., vol. 44, p. 89; 1950.[3] Nature, vol. 168, p. 505; 1951. [42] J. Mercier, Ann. Phys., vol. 19, p. 248; 1923; vol. 20, p. 5; 1923.[4] , Trans. Am. Geophys. U., vol. 32, p. 813; 1951. [43] 0. Mettelstaedt, Ann. Physik, vol. 2, p. 285; 1929.[5] WV. C. Anderson, Rev. Sci. Instr., vol. 8, p. 239; 1937. [44] A. A. Michelson, Astrophys. J., vol. 54, p. 1; 1927.[6] J. Opt. Soc. Am., vol. 31, p. 187; 1941. [45] A. A. Michelson, F. G. Pease, and F. Pearson, Astrophys. J.,[7] H. Barrell, Proc. Roy. Soc. (London), A, vol 209, p. 132; 1951. vol. 82, p. 26; 1935.[8] H. Barrell and M. J. Puttock, Brit. J. Appl. Phys., vol. 1, p. E7; [46] J. F. Mulligan and D. F. McDonald, Am. J. Physics, vol. 25,

1950. p. 180; 1957.[9] E. Bergstrand, Ark. Mat. Astr. Fys., vol. 36, p. 1; 1949. [47] NBS Tech. News Bull., vol. 39, p. 1; 1955.

[10] --, Nature, vol. 163, p. 338; 1949. [48] E. K. Plyler, L. R. Blaine, and W. S. Connor, J. Opt. Soc. Am.,[11] L. E. Bergstrand, Nature, vol. 165, p. 405; 1950. vol. 45, p. 102; 1955.[12] E. Bergstrand, Ark. Fys., vol. 2, p. 119; 1950. [49] D. H. Rank, R. P. Ruth, and K. L. VanderSluis, Phys. Rev.,[13] , Ark. Fys., vol. 3, p. 479; 1951. vol. 86, p. 799; 1952.[14] Ann. Franc. Chronom., vol. 11, p. 97; 1957. [50] , J. Opt. Soc. Am., vol. 42, p. 693; 1952.[15] R. T. Birge, Rep. Phys. Soc. Progr. Phys., vol. 8, p. 90; 1941. [51] D. H. Rank, J. N. Shearer, and T. A. Wiggins, Phys. Rev., vol.16] K. Bol, Phys. Rev., vol. 80, p. 298; 1950. 94, p. 575; 1954.[17] WV. Culshaw, Proc. Phys. Soc. (London), B, vol 204, p. 260; 1950. [52] E. B. Rosa and N. E. Dorsey, Bull. Bur. Stds., vol. 3, p. 433;[18] --, Proc. Phys. Soc. (London), B, vol. 66, p. 597; 1953. 1907.19] R. D. Cutkosky, J. Res. Standards, vol. 65A, p. 147; 1961. 53] XW. Schaffeld, F.I.A.T., Final Rept. No. 895; 1946.

Automatic DC Data Logging SystemWILLIAM ARNETTt, SENIOR MEMBER, IRE

Summary-A system is described which digitizes dc voltages INTRODUCTIONand records the values on punched cards for subsequent machinedata reduction. High accuracy is achieved by comparing the input T N THE PAST FEW YEARS we have seen a greatvoltage to a bank of standard cells and measuring the difference. An increase in the use of digital data logging systems.automatic calibration routine is provided. Resolution is one part per - A typical function of such a system is to scan pe-million, stability a few parts per million for a 30-day period witho--t riodically a number of electrical input terminals, to con-adjustment. Contributions of various parts of the system to over-allerror are analyzed. Automatic calibration data are anlalyzed to vert the voltage across these terminals to digital form,establish system performance. and to record this digital value for observation or auto-

matic processing. The over-all accuracy of such systems* Received August 14, 1962. Presented at the 1962 International is typically 0.1 per cent to 1.0 per cent. The system to

Conference on Precision Electromagnetic Measulrements as Paper be described is capable of measurement precision of a

t COHU Electronics, Inc., San Diego, Calif. few parts per million, with commensurate stability (re-

1962 Arnett: Automatic DC Data Logging System 149

peatability) of readings extending over periods of at the data reduces it to manageable form; in fact, it isleast several months. Thus, it extends automatic data only in this form that the full significance of the datalogging into a domain previously accessible only to pre- may be appreciated.cision standards laboratory equipment and technique. The system was planned to measure 100 independent

First, a brief history of the need which inspired the data channels once an hour, 24 hours a day, 7 days adevelopment of this equipment will be given. The sys- week. This amounts to 16,800 data points per week,tem itself will then be described, and the sources of or nearly a million a year. Our analysis indicates thaterror therein analyzed. Finally, the performance of the using conventional standards laboratory methods forsystem will be compared to predicted results. measurement and desk calculator data reduction, each

In this age of highly complex systems, there has de- data point and corollary statistical analysis will costveloped an enormous demand for better measurements about one dollar including overhead. Cost analyses for

better in every way. We require more measurements, the automatic system, including data reduction, resultmore precision, more accuracy, more speed, and all at a in a predicted cost of about 22 cents per data point. Bothlower cost. Advances in data handling and high-speed calculations include maintenance and amortization ofcomputers have given us the ability to digest enormous equipment. The figures for manual operation make noquantities of data. All this has provided incentive for allowance for loss due to operator fatigue, error andinstrument manufacturers to produce more precise and nervous breakdowns.more accurate equipment. Digital voltmeter and dc The system to be described may be expanded readilycalibrators are commonly offered with 0.01 per cent to 1000 input channels simply by addition of scannerspecified accuracy. Such specifications are a bit star- input capacity. This would result in continuous opera-tling; however, carefully designed equipment can and tion of the data logger, each point being scanned hourly.does attain this accuracy. The scanning rate presently used (100 points in 6 min-

utes) can be increased with standard equipment toabout 200 per minute, or 12,000 per hour, with no lossin accuracy or resolution; this increased speed can be

One of the basic requirements for such instruments used either to accommodate more input channels or tois a stable reference voltage. The unsaturated standard make measurements more frequently, at the option ofcell is frequently used for this purpose, and although its the user.voltage output is very stable under controlled condi-tions, it can and does change with temperature and load SYSTEM OPERATIONand with time. In spite of these limitations, it is highlysatisfactory over a limited temperature range and its The measuring system is basically quite simple. Thecharacteristics are well established. Zener diodes have voltage to be measured is connected in series oppositioncome into use as voltage references, and offer some char- to a known stable voltage of such value as to produce aacteristics which make them more suitable than the difference voltage of less than 0.05 volt; the differencestandard cell for certain applications. Our initial in- voltage is amplified, measured by a digital voltmeter,vestigation of the zener diode as a reference revealed and recorded on punched cards. The mechanization ofthat information on voltage stability over long periods the system requires some care in order to achieve itsof time was simply not available. In fact, even short- stability and accuracy objectives. Some detail on thisterm stability data was not available to the precision will be brought out in the following text.required of a satisfactory reference for an instrument Referring to the block diagram, Fig. 1, the systemwhich must have 0.01 per cent over-all accuracy. It was consists of a precision power supply for the zener diodes,for the purpose of obtaining stability data on zener a closely controlled oven for the diodes under test, adiodes that this data logger was designed. Its use has scanner to sequentially sample all the diodes, a bank ofsince been extended to long-term stability testing of standard cells in a temperature-controlled housing, anproduction dc standards and quality control applica- isolation amplifier, an amplifier with a gain of ten, ations. digital voltmeter, a parallel-to-serial converter, a cardThe basic reason for the automatic data logger ap- punch, a digital calendar-clock, and a programmer.

proach was economic. To satisfactorily evaluate stabil- The zener diode power supply delivers 30 volts withity of voltage reference diodes requires a very large 0.01 per cent stability. The diodes are connected to thisnumber of data points on a reasonably large sample of supply through individual precision wirewound resistors.each of the devices to be tested. This data must be Separate voltage sampling terminals are provided foraccurate to a few parts per million and, because of the each diode. These are sequentially connected by thevolume of data, must be recorded in a form suitable for scanner to the isolation amplifier and bucking voltage.mechanical data reduction, since such a volume of data The bucking voltage consists of from one to twelve un-15 difficult to digest in raw form. Statistical analysis of saturated standard cells and a highly stable one-volt

150 IRE TRANSACTIONS ON INSTRUMENTATION December

.CAsLFIBRATION 1VOLTAGES

ZENER~~ OVEINA DIR

PRODUCTION CLOCKVOLTAGE _ _______

CALIBRATORS PROGRAMMER

Fig. 1 Block diagram.

supply tapped at each one tenth volt. The number of ture, system zero, zener supply voltage, and comparisoncells and tenths of a volt are selected by auxiliary of 12 working standard cells against a reference bank ofscanner contacts and a wired-program patch board. All twelve cells. This data is all logged on punched cardsscanner contacts are gold-plated. The isolation amplifier and may be used for correction of data; however, this isinput impedance is in excess of 10,000 megohms at dc. not actually necessary in the present system since theThis is used mainly to protect the standard cells from corrections required are insignificant. Some of this datashorted or open zener diodes and limits the standard is used, however, to establish system stability and prob-cell current to two or three milli-microamperes in the able error. The system accuracy is determined by itsworst case. The gain-of-ten amplifier is standard except analog portions. Let us now consider these individually.for precision feedback resistors. The digital voltmeter The + 1 amplifier has a loop gain of 107. For any rea-is a standard commercial tracking type with 0.01 per sonable variation in components, the gain is unitycent accuracy. The parallel-to-serial converter is a within 1 ppm, except for zero shift which, for thismodified commercial unit feeding an IBM card punch. amplifier, is not more than 2 microvolts over periods ofThe programmer is controlled by the scanner and pro- 400 hours or more. The gain-of-ten amplifier has 107vides the switching necessary for the automatic calibra- forward gain. iiis gain stability is essentially that of thetion program. The scanner is a slightly modified com- feedback resistors employed. These are specially se-mercial unit, providing 100-point, five-pole switching. lected wirewound types with a temperature coefficientTwo levels are used for input selection, two for bucking of 2 ppm. Experience with these resistors in this andvoltage selection, and one for control. other products indicates that the ratio is stable to 10The system functions as follows: Once each hour the ppmn under the environmental conditions encountered

clock initiates a measuring cycle. At the start of each in this equipment. Since the maximum input to thecycle, a series of ten calibration and check points are amplifier is the difference between the input voltage andlogged. Following this, the various inputs are selected, the bucking voltage, which is never more than 0.05measured and logged. The punched cards are collected volt, the error referred to the input is less than 1 micro-daily and sent to the tabulating room. A daily listing is volt. Zero drift may amount to another 2 microvolts.made for "quick look" and statistical analysis is made The digital voltmeter accuracy is + 100 microvolts aton a weekly basis. All data reduction is done on stand- its input, which becomes + 10 microvolts referred to theard accounting machinery which is quite adequate for amplifier input.the volume of data involved. For larger systems, high- The zener supply is regulated to 0.01 per cent. Fig. 2speed computers may be used as the data is in proper shows a typical 6.3-volt zener at 7.5 ma nominal andform for entry into a computer. The reduced data con- with a dynamic impedance of 10 ohms. The error duesists of a printed summary of distribution (histogram) to 0.01 per cent changes in supply voltage is less thanof data for each input channel on a weekly basis [(24 X 7) 10 microvolts. Temperature coefficient of zener diodes=168 points], plus calculation of average value and used is 0.001 per cent/'C. Zener diode oven is regulatedstandard deviation. A weekly time separation of data to 0.015°C. Variation in output due to temperature is,was chosen, somewhat arbitrarily, as providing a mean- thus, 0.015°CX10 ppm,/°C =0.15 ppm, or approxi-ingful separation of short-term effects (standard devia- mately 1 microvolt. The 1.0-volt supply is regulated totion) and long-term effects (change in average value in 0.001 per cent and can contribute up to 10 microvoltssuccessive weeks). error, depending on how many tenths of a volt are re-

Calibration and check points are: Digital voltmeter quired for the bucking voltage. Excluding the standardzero, digital voltmeter calibration, amplifier gain, ampli- cells, the maximum error is, thus, 2+1 + 2±+10+10+1fier zero, 1.0-volt supply, standard cell oven tempera- +10=36 microvolts, which for the 6.3-volt zener diodeture, ambient temperature, zener diode oven tempera- of our example is approximately 6 ppm (Fig. 3).

1962 Arnett: Automatic DC Data Logging System 151- 30 VOLTS E -30 VOLTS

.01 °/ STABILITY

RI .19 R2 R4 2R3

75ma 3200A

-10 ~~~~~~~~DVM

20-E-

_ <, ~ ~ ~ ~~~~~~~~R22 ZERO VOTSe -IOL 6VOLTS OUTPUT VOLTAGE

Fig. 4-Amplifier gain check.

age applied to the amplifier input during measurementis 0.05 volt. Since the amplifier gain is measured as

IOX3OXIO-4 -6 VOLTS above to 0.002 per cent, the data may be corrected toE~ 3200 9.4X VO 0.002 per centX0.05 volt, or 1 microvolt referred toFig. 2-Effect of change in supply voltage on output voltage. the input.

ERROR SUMMARY PERFORMANCE

Isolation amplifier zero drif, 2 Av System performance may be evaluated by analysisAmplifier gain error, max 1 ,uV of the calibration data and by comparison of data takenAmplifier zero 2 ,vDVM error 10lv on selected inputs. It is convenient to compare the1-volt supply 10 /V weekly averages and mean deviations of the variousZener supply 10 ,uvZener oven 1 ,uv channels. Histograms are also used in some presenta-

Total 36,uv tions. Zero readings, gain readings, and oven tempera-Error in ppm at 6 3 vots tures are not plotted since the variations in these param-

36 x6 ppm eters have proved to be insignificant.

E = 6.3 - 6 ppm The most direct over-all indication of system stabilityis the record of the twelve independent reference cells

Fig. 3. bucked against the twelve working cells. Fig. 5 shows aplot of the weekly averages and the standard deviations.

The cells used are unsaturated cadmium standard Over the period shown of 20 weeks, the average changedcells. Temperature is maintained at 280C + 0.050C. The 40 microvolts, and the mean deviation varied from 3.9temperature coefficient of this cell is stated by the man- to 10.7 microvolts. Using a standard deviation of 10ufacturer to be less than 10 ppm/0C. Variations due to microvolts, the probable error of any reading becomestenmperature should, therefore, not exceed 0.5 ppm. Sta- about 7 microvolts, and 99.7 per cent of the readingsbility of this type of cell is very good short term. The should be within 30 microvolts of the true value, assum-manufacturer of these cells states that a slow decrease ing the standard cells have not changed or that the datain EMF amounting to about 0.004 per cent per year has been corrected for this change. Therefore, in meas-may be expected. This amounts to less than 1 ppm per uring 6-volt zeners, the probable error is about 1 ppm;week, and since the cells may be referenced to a bank of the standard deviation is about 1.6 ppm, and the 3-normal cells as often as desired, this stability appears to sigma value about 5 ppm. Fewer than two readings in abe quite satisfactory. Normal cells may be used instead million should depart as much as 8 ppm.of unsaturated cells. The original reason for selecting It is interesting to note that the expected maximumunsaturated cells was because of their lower temperature error calculated previously is about equal to four timescoefficient and the possibility of accidental damage due the standard deviation of the measurements. Fig. 6to shorts, etc., during experimental work. shows a histogram of the data taken during one week on

In the calibration program, all zeroes are obtained a particular voltage calibrator at ten volts nominal out-by shorting the appropriate inputs. Digital voltmeter put. Note that the plot shows normal distribution. Fig.calibration is checked by measuring a single unsaturated 7 shows a plot of the weekly averages and standard de-standard cell with 5-digit readout. Amplifier gain is viations of this same unit for a period of eight weeks.checked by switching in the circuit of Fig. 4. Resolution Total drift in average is less than 0.5 ppm per week. Theis 15 microvolt change at the input for one-digit change direction of drift indicates rising voltage which could asin output, and precision is dependent only on the sta- well be attributed to reduction in the bucking. voltage;bility of the ratio R1R3/R2R4 and the accuracy of the in fact, the expected change due to cell drift is aboutdigital voltmeter. Since the calibration voltage is 1.5 twice this amount, so the real change could be eithervolts, resolution of the gain measurement is 0.001 per way. Errors of this magnitude are less than the probablecent. Another 0.001 per cent may be added to allow for error of measurement, however. The long-term driftchanges in resistor ratio. The maximum difference volt- may be determined by periodic reference to a bank of

152 IRE TRANSACTIONS ON INSTRUMENTATION Dcme

AVERAGE DIFFERENCE VOLTAGE AVERAGE DIFFERENCE VOLTAGE

430 3

420

4300

42

20~~~~~~~~~~~~~

410~~~~~~~~~~~~~~~~~~~3

700

I0

___~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~I 0~ ~ ~ ~~__0 5 1 1 0 2 Y

0 10 20 30 40 so very low deviation values (about 10 microvolts)forthe~~~ig.7-Vltgealiraormeaurda 10vots utpt)

Fi.6-otaecairto mesre t10 vot upt. endvaio fterfrec2hne0i,i at

Fig.8 shows a plotgo dai-vly maximubnrmnmm maael.thswllalwonetr.eemnaino

never separated~~~~~~~~~6 bymretan2. mcovls.n.uull.ppas.oberasnbl;loe.vlagsofe.amronly tn micrvolts whichis onedigiton thevoltmeer, dfficul problm, aszero difts ad gai changs wilOn evra dys llredigs er te ame Tes ae paya or prmien rle n etrmnin oeral

exceptionaly stablereference. These dta illustate the erors in masurementremarkablestabilityof the measuring system. Although We believe that the performance of this systemestab-~~~~~~~~~~~~..

it s qit prbabe hata vrygoo zeerdioe mgh lihesth fesibliy o fulyautmatd easremnthave the same long-termdrift as the reference cells, it to a few parts per million in the voltage rangeofafew~~~~~~~~~~~~~~~~~~~...................

is4asrnmclyimrbbeta th hour-by-hour vot 3oa0e es fvls