IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. ... Abstract-A new version of a noise thermometer has been re- ... trical leakages at high temperatures. ... exceeding fO.l K at the freezing point of copper (2.58-.
IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 42, NO. 2, APRIL 1993
39 1
A High Temperature Noise Thermometer for Accurate Thermodynamic Temperature Measurements Luigi Crovini, Antonio Actis, and Roberto Galleano
Abstract-A new version of a noise thermometer has been realized at the Istituto di Metrologia Gustavo Colonnetti (IMGC) to determine the thermodynamic temperature of the melting and freezing points of copper. Low-noise, low-input-capacitance amplifiers, cross-correlating amplifiers and specially shielded resistive sensors were designed, realized, and tested. They are described in this work and their relevant features are discussed as well as the problems associated with EM1 and electrical leakages at high temperatures.
I. INTRODUCTION noise thermometry program is underway at the Istituto di Metrologia Gustavo Colonnetti (IMGC) to determine the thermodynamic temperature between 1234 K and 1360 K. On the grounds of a previous experiment that produced results for temperature between 900 K and 1235 K [ l ] a new instrument was developed in view of (a) extending the measurements up to the melting point of copper (ca. 1357 K); (b) reducing, with respect to the previous version, the errors due to the lead resistances and to the electrical leakages; and (c) providing an enhanced immunity from EM1 and other sources of extra noise. The experiment was designed to achieve an uncertainty not exceeding fO.l K at the freezing point of copper (2.58U estimate).
f m c e control
A
METHOD 11. NOISE THERMOMETRY The noise thermometry method in use at IMGC [2] is based on the direct comparison of the Johnson noise from two resistors ( R I and R 2 ) , the former kept at a reference temperature, TI close to 298 K, and the latter at the unknown temperature, T2. The working conditions are such that (i) the noise levels are very closely the same with both resistors at the input, (ii) the transmitted frequency band is the same for the two resistors and (iii) the noise from the leads and the effect of insulation leakages can be measured and accounted for. In the ideal condition of no electrical resistance in the leads and no insulation leakage, the following equation Manuscript received June 12, 1992. The authors are with the Istituto di Metrologia “G. Colonnetti,” 10135 Torino, Italy. IEEE Log Number 9207027.
conputer
Fig. 1. Schematic diagram of the IMGC noise thermometer. LNA, lownoise amplifier; F, filter. The EM1 detector causes the DVM (HP3458 A) to overload when interference greater than five times the thermal noise occurs. The overload condition causes the sample to be rejected and the 10-s measurement to be repeated.
holds
where A2 is the squared ratio of the noise voltage from the two resistors and R,, is the noise-equivalent resistance of the amplifier, as determined in the same frequency band. A random variable x , of zero mean value, accounts for the random error in measuring T2, its standard deviation being given as follows
The frequencies fi and f2 limit the transmitted band. A significant correction to T2,as given ( l ) , is due to the lead noise as generated in the high-temperature source, ( T T ) ~ ~ , and in the low temperature source, ( r T ) L l .As a consequence, a deduction equal to (rT)L2/R2 - A ~ ( ~ T ) L ~ / R ~ must be applied to T2 as determined in (1). In our case, such a deduction amounts to approximately 0.8 % . Another additive correction is due to electrical leakages: It more details on this correction being amounts to 2 X provided in the following. The schematic diagram of the IMGC noise thermometer is shown in Fig. 1.
0018-9456/93$03.00
0 1993 IEEE
IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 4 2 . NO. 2 , APRIL 1993
392
10
v
TABLE 1 CHARACTERISTIC OF THE LOW-NOISEAMPLIFIER
I
-10
v
Fig. 2 . Electrical circuit of the low-noise amplifier. T I : 4 X BF246 A; T2: ECG433; R1 = 470 Q ; R2 = 4 . 7 Q ; R3 = 10 kQ; R4 = 1 kfl; C1 = 470 pF.
111. AMPLIFIER Several low-noise, low-input-capacitance amplifiers have been designed, realized and tested. The electrical scheme of the amplifier chosen for the noise thermometer is shown in Fig. 2 and its properties are summarized in Table I. It embodies a cascode FET-input stage employing four high-transconductance, low-input-capacitance FET in parallel in the first stage (BF 246 A) and a lownoise bipolar transistor in the second stage (ECG 433). Transistor T2 forces the drain of T I to follow its source signal, thus reducing the amplifier input capacitance. Furthermore, a feedback network ( R I ,R2)neutralizes TI gateto-drain and gate-to-source capacitances and stabilizes both the amplifier noise and gain. The amplifier stability was further enhanced by means of stabilized supply voltage, low temperature-coefficient resistors (50 ppm /K) and by operating the amplifier in a cabinet where the temperature is stable to within k 0 . 3 K in one hour. Experimental determinations of R,, and of its stability showed a drift less than 0.06% in 19 h, i.e., less than 0.003%/h. The lead resistance of the high-temperature source is 0.6 8, approximately, and ( T T )is~about ~ 700 8 K. This being the largest correcting term, it has to be determined by means of direct measurement. A cross-correlating amplifier, shown in Fig. 3, is used for this purpose. It uses for both A , and A2 four commercially-available low-noise operational amplifiers connected in parallel (LT1028) and an analog multiplier (AD534) for MUX. The band-pass filters, F, transmit the noise between 10 kHz and 50 kHz, the higher frequency being limited by the multiplier dynamic performance, The repeatability of the cross-correlating amplifier was tested in a shielded room with a 3 8 resistance at 273.15 K. The result was better than 0.04% (la estimate) over 3600 s of continuous integration. The cross-correlating amplifier was calibrated to within f 0.1 %with several resistances at the ice point providing a range of noise values including that produced by the high temperature leads. The required accuracy in measuring the lead noise is +0.2%. With the cross-correlating amplifier described above, the noise in the leads can be measured directly. However, the very low level of noise at the amplifier input requires special arrangements such as operation in a shielded room and decoupling of computer and data acquisition systems. Only in such conditions is the level of interferences negligible. Alternatively, the lead-noise measurement can be
Input stage: four BF246A FET in parallel, cascode circuit Total transconductance: 44 mS Cascode 2nd stage: bipolar, ECG 433 Theoretical noise resistance: 16 fl Open-to-closed-loop ratio: 100 Total gain: lo4 Input current: 5 f 1 nA Input capacitance (excluding connections): 4 pF Frequency limits to within dB: kHz, 2oo kHz Effectiveequivalent noise resistance (To = 296 K , f = 10-100 kHz: 4 2 . 6 fl) Input noise voltage (10-100 kHz): 0.78 n V / & Input noise current (10-100 kHz): 0.04 P A / & 10 V rechargeable batteries for more than 24 h of Power supply: .. . continuous operation
l
l
I
-
Fig. 3. Cross-correlating amplifier. A , and A*, low-noise amplifiers; F , filters; MUX, analog multiplier. The input noise is that produced by a lead loop.
carried out on a dummy sensor geometrically identical to that used for measuring T2, equipped with Pt leads as the other, and submitted to the same temperature profile. The leads in the dummy sensor, however, are much thinner, i.e., 0.1 mm instead of 0.5 mm, thus producing a resistance greater than that of the sensor by a factor of 25. In such a condition, the lead noise level is of the order of 18 x lo3 0 K and thus it can be measured with either the cross-correlating amplifier or with the noise thermometer of Fig. 1 without excessive precautions. The correcting ~ obtained by dividing the measured term ( T T ) is~ then noise by the ratio between the lead resistances of the dummy sensor and the true sensor.
-
IV. SOLUTION FOR INTERFERENCE PROBLEMS The R.M.S. value of the noise voltage produced by R , or R2 is approximately 0.8 pV and their ratio must be deterinined to within a few parts in lo5. Such accuracy is attained only when special care is exercised in order to maximize the shielding effectiveness and the rejection of common-mode interferences. Special low-noise cables and several different input-circuit arrangements were made and tested. The final arrangement is shown in Fig. 4.It is the only one which produces a “high-quality noise” when a resistor at the end of a long-stem is placed into a high-temperature fur-
CROVINI, ACTIS,
AND
393
GALLEANO: A HIGH TEMPERATURE NOISE THERMOMETER
*
tor to reveal small interference peaks by means of a sensitive spectrum analyzer. A further level of analysis was provided by monitoring the variance of the detected noise for numerous successive 100 s periods of time. During a temperature measurement a threshold circuit set to five times the noise RMS value (EM1 detector in Fig. 1) and a computer program for the on-line evaluation of the variance of the measured voltage enabled the rejection of part of a measurement if it is affected by interference.
-
cu
Thermostat
Fig. 4. Input arrangement of the noise thermometer showing the connections of shields to the ground point. The low-temperature resistor is shown in an enclosure with circulation of a temperature controlled liquid.
nace. A noise is of “high quality” when it exhibits a white spectrum in the measurement frequency band (5-100 kHz) arid when it does not contain, within the required level of accuracy, components due to interferences, external noise and unstable products of the amplifier internal noise (e.g., parasitic oscillations). A careful inspection of the noise voltage in the time domain with the apparatus in its final configuration must not detect any correlated component due to either sine oscillations or peaks of pulses. If these conditions are not achieved, grounding, shielding, and input arrangement must be modified in order to achieve the high-quality condition. Some basic rules hold: Use star grounds connection as far as practicable within each amplifier stage with the shields and the chassis directly grounded to a common point. When a ground bus has to be used, as for instance, inside the low-noise amplifier, this should be as short as possible and of low resistance (e.g., < 10 ma) and the connections of the various branches to it follow a given sequence in order to avoid unwanted feedback to the input. Shielding continuity must exist everywhere and a particular shielding symmetry be realized at the input; and, last but not least, to ensure that the connections to computers provide full isolation with ground circuit separation. There are, however, other precautions to be taken that may vary from case to case. For instance, the hightemperature resistor with its cable acts like an antenna more than any other part of the noise thermometer, this behavior being enhanced with the single-ended configuration. Thus both electric and magnetic shielding is required for it. EM1 may be broadcast by currents in deflecting yokes of computer monitors, switching power supplies and other devices with switching power. They were identified and their noise eliminated by means of a lengthy trial-and-error procedure. A sufficiently high sensitivity in detecting interferences was achieved when operating the noise thermometer with the high-temperature source set at room temperature. Thus the input noise power is approximately 5% of that produced at 1084”C, making it possible for a skillful opera-
1
V. EQUIVALENT CIRCUITOF A HIGH-TEMPERATURE NOISE SOURCE The high-temperature sensing resistance is a coiled thin wire of Pt, wound onto a silica support, as shown in Fig. 5. It is placed at the bottom of a closed-end silica tube. Four Pt wires are attached to the coil to transmit the noise voltage and to accurately measure the resistance of the coil. At a temperature near 1084°C (1357 K), the electrical leakage through the silica spacing discs cannot be neglected. It depends on both temperature and frequency; the frequency dependence is shown in Fig. 6. The hightemperature resistor was designed in such a way as to minimize leakages everywhere, particularly where the temperature begins to be lower than that of the sensing coil. With reference to Fig. 5, the leakages can be represented by three frequency-dependent conductances corresponding to, respectively, the coil support and the nearest spacing disc and to the two other discs immediately above the latter. Discs at temperatures below 900°C (1 173 K) do not, in fact, introduce appreciable leakages. The noise output of the equivalent circuit is:
Go ti2 = 4kBT
[l
T‘ +GI - + G2T T
+ R2(Go+ G, +
R2
(4)
- With Go, G I , G2 the leakage conductances as averaged in all transmitted frequencies. The values of R2 and of the conductances are calculated from the measured sensor resistance and from the G-values as determined with a conductance/capacitance bridge. Typical values are shown in Fig. 5. The actual values for a particular resistor must be determined at the end of the experiment. The sensor resistance is measured at 400 Hz, and then corrected to account for the frequency-dependent conductances of the insulators in order to obtain a mean resistance in the frequency range 5 kHz-100 kHz. VI. EXPERIMENTAL RESULTS The repeatability of the noise thermometer was determined at 1233 K and at the melting point of copper using a measurement sequence which enabled us to observe the noise of both resistors for 3000 s and reduce any residual instability in the amplifier noise and gain. At 1233 K the measured square-voltage ratio, A 2 , exhibited a standard equivalent to 0.15 K. A step in deviation of 1 X temperature of 1 K produced an increase in A2 of 1.4 X
IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 42, NO. 2, APRIL 1993
394
t- 10
c m l
1.0010
1
1
0
1357.8 K
1357.8 K
1300 K
850 K
1.ooc0
-
1.oW7
-
*
