Alan J. Fougere
Falmouth Scientific, Inc.

Neil L. Brown
Woods Hole Oceanographic Institution

Edward Hobart
Woods Hole Oceanographic Institution

The Mark III conductivity-temperature-depth (CTD) profiler has been the mainstay of modern physical oceanographic research¹. The MKIIIB CTD provides high quality oceanographic data when used by skilled personnel and are frequently re-calibrated. The design objective of the Integrated CTD system was to attain the same high level of performance while reducing the necessity of frequent re-calibration through the enhancement of long term stability. This required re-consideration of both the electronic approach and the re-design of the physical sensors. The result of this work is a CTD system with improved measurement precision. All three primary sensors are newly designed to achieve long term measurement stability and to optimize system sampling performance without the limitations of existing technologies.

Advances in the state of the art in electronic and microcontroller technologies has enabled the development of improved methods of analog sensor signal processing. During the MKIII development, micro-processors and their software development tools were primitive and difficult to use. The revolution in high speed micro-controllers has allowed their use in the Integrated CTD allowing for real time numerical correction for drift in the analog signal processing circuitry.

The key component of this design is a phase shift oscillator as described by Brown 19682. Through the addition of a precision reference network and real-time numerical correction the performance of the oscillator circuit has been enhanced. This requires that the oscillator output frequency be stable for short periods of time over which it is recalibrated against the precision resistance reference network. The network accurately simulates the output of the sensors for known values of the measured parameters. The internal microprocessor is then used to mathematically correct for drift in the electronics.

Consideration was given to the use of a Wein Bridge Oscillator³. This technique can only be used with two terminal high resistance devices, such as Thermistor or high resistance two electrode conductivity cells (i.e., cells which have relatively small inside diameter and long length); and cannot be calibrated against reference resistors due to the error introduced by the unknown resistance of the required electronic switches. Additionally, it cannot be used with four terminal devices such as strain gage bridges, inductively-coupled conductivity sensors or low output sensors, such as platinum resistance thermometers. Since it cannot be calibrated in situ the electronic components must be selected to have extremely low temperature coefficients and extremely low long term drift. Even with all these steps the requisite long term stability in the electronics is almost impossible to achieve particularly in the frequency determining capacitors. For these reasons Wien Bridge technology was dismissed.

The design utilizes a unique phase shift oscillator that has the ability to convert the output of a 4-terminal network such as a temperature bridge or strain gage bridge to frequency with a high degree of linearity and sensitivity. It can also be used with a complex network of sensors and active electronics such as the conductivity circuit described below. For example it can be readily configured to produce a 50% change in frequency for a 1% change of one arm of an equal arm bridge thus making it ideal for use with low output devices such as platinum thermometers and strain gage pressure sensors.

Extreme accuracy is achieved by the use of a simple network of ultra stable precision resistors which simulate the output of the sensor network at 3 precisely known values of the sensed parameter, which are determined by calibration. This simulation network is periodically used (once per second) in conjunction with the microprocessor to calibrate the oscillator thus eliminating the effect of calibration drift due to temperature, aging or supply voltage variations on the electronics.

Figure la shows a simplified diagram of the oscillator. It consists of two amplifiers Al and A2, a simple two pole phase shifting network, a quadrature network, and a sensor bridge. The quadrature network output voltage is shifted 90 degrees from its input. The sensor output is either in phase or 180 degrees out of phase. The sum (Er) of quadrature voltage (Eq) and the sensor voltage (Es) is applied to the input amplifier (Al), thus forming a closed loop, which will oscillate at a frequency at which the sum of the phase shift between the oscillator output (Eo) and the combined quadrature and sensor output circuits (Er) plus the phase shift through the phase shifting network totals 180 degrees.

Figures lb and 1c show the relationship between the oscillator output voltage Eo, the sensor output voltage (Es), quadrature voltage (Eq) and the resultant voltage(Er). It can be seen that as the sensor output voltage (Es) magnitude varies, the phase angle (0) between the resultant (Er) and the oscillator varies from angles below 90 to angles above 180 degrees. This in turn will cause the oscillator frequency to change so as to maintain a total of 180 degrees phase shift.

The sensitivity of the oscillator, that is, its percentage change in frequency for a given change in sensor output is inversely proportional to the magnitude of Eq. The phase shift of the resultant voltage (Er) depends only on the ratio of the magnitude of the sensor voltage (Es) to the magnitude of the quadrature voltage (Eq). Consequently the oscillator can be readily designed to have a frequency shift of 50% for a change in sensor output voltage (Es) equal to 0.25% of the sensor input. This means that low output devices such as strain gage pressure transducers can be very effectively used. Amplifier Al has an automatic gain control circuit to compensate for the small changes in gain (i5%) that are required as the oscillator frequency changes with the sensor output.

Microprocessor controlled electronic switching is provided to connect any one of the calibration circuit outputs or the sensor output into the oscillator circuit. The input to the calibration or sensor circuit is permanently connected to the oscillator output. Since there is no current flowing in the quadrature and sensor output series circuit (Al has effectively infinite impedance), the "ON" resistance of these switches has no effect on the oscillator frequency.

The 3 basic requirements for accuracy in any sensor system are as follows:

1. Stable transducers including the effects of fouling.

2. Sensor electronics with stable or accurately known transfer function.

3. Calibration against known standards of conductivity, temperature and pressure.

Since frequent re-calibration of instrumentation is costly and degrades data taken between calibrations, achieving items 1 and 2 is mandatory. Item 3 is readily achieved with existing facilities or can be established with existing commercially available equipment.

It can be shown that if the electronics are linear, a 2-point calibration against a fixed reference network will completely determine the transfer function of the circuit and fully correct for any drift in the electronics. Figure 2 illustrates this concept for a temperature sensor utilizing a platinum thermometer⁴. The basic temperature bridge is made up of precision resistors Rl, R2, and Rf and the thermometer Rt. The self calibration circuit consists of precision resistors Ra, Rb and Rc which are mounted in close proximity to the thermometer and its reference resistor Rf and consequently can be assumed to be at about the same temperature. The overall procedure for the system is as follows.

First, a laboratory calibration is performed where several values of a temperature (T) and output data (X) are obtained and (assuming a linear relationship) a first order polynomial fit established such that

T = Ac + Bc * X (switch Swl in "Et" position in Figure 2)

At the same time when T is at the bottom of the temperature range Swl is set to "Elo" position and the value of X noted (Xlo). Similarly Swl is set to the "Ehi" position and the value X noted (Xhi). Using the values of Ac and Bc from above the values of Tb and Thi are calculated where Tb and Thi are the exact temperatures simulated by the reference network. Ideally the reference network Ra, Rb, Rc and Rf should have zero temperature coefficient and be absolutely stable. However if they are at approximately the same temperature as the sensor then the temperature coefficient of these resistors will only result in a modified calibration and will be completely repeatable and stable with temperature. In the case of conductivity and pressure it may be desirable to include a relatively crude temperature sensor network as well as the reference network to numerically correct for any residual temperature effects in the reference network.

This approach using a precision network to accurately simulate a sensor output at accurately known values of the measured parameter permits the use of electronics using standard components with essentially no requirement for stability, thus drastically reducing the cost, size and power consumption associated with traditional methods. The overall accuracy is thus dependent only on the transducer, its reference resistor and the simulation network, which in the case of a platinum thermometer, is four precision resistors.

The sequence of events for a single measurement would be as follows:

1. Switch Swl is set to the Ed position (see Figure 2) which is the nominal "zero calibrate" position and a reading of the system output (Xlo) taken.

2. Switch Swl is set to Ec2 position which is the nominal "full scale calibrate" position and a reading (Xhi) taken.

3. Switch Swl is set to Et position, (the sensor output) and a reading taken.

From the outputs Xlo and Xhi for accurately known simulated temperatures Tb and Thi (determined from the lab calibration) we calculate A and B from the two simultaneous equations below such that:

Tb = A + B * Xlo Thi = A + B * Xhi

The final step is to calculate the true temperature from the coefficients A and B and the system output X obtained above.

T=A+B*X (T is true temperature)

The above discussion has focused on temperature measurement. However, similar arguments can be made for the measurement of other parameters.

The above discussion has assumed a linear relationship between the measured parameter and the output of the electronics. However, the above arguments still apply in the case of moderately non-linear sensors providing the appropriate number of calibration and reference points are implemented. Platinum resistance thermometers have a resistance to temperature relationship in the oceanic range which can be very accurately described by a 2nd order polynomial. The same is true of many pressure transducers. Conductivity can be made to be perfectly linear. There are many signal processing techniques that are inherently linear or very close to it. Hence in the ICTD implementation a 2nd order polynomial fit with 3 calibration and reference points has been used to fully correct for drift in the electronics.

Over a 14 month period the repeatability for a platinum temperature sensor and phase shift oscillator combination was better than +/-.003 degree Celsius. Laboratory calibrations which use a 3rd order regression fit of temperature sensor data to the standard leave residual errors in the noise level (.0001 degree Celsius), reference calibration section and Figure 9. It is not possible to separate these small residual errors from the limitations of the calibration equipment and procedures used at Woods Hole Oceanographic Institution, (WHOI) CTD calibration laboratory.

An ongoing investigation by Robert Millard of WHOI has shown that the new titanium strain gage transducers made by Paine Instruments can be numerically corrected for static and dynamic temperature effects leaving residual errors of about 0.0l5%⁵. These residual errors are due to hysteresis, which is essentially the only source of non systematic error. The anticipated accuracy after numerical correction is about the same as the Digiquartz but the strain gauge transducers are extremely rugged and are about one forth the cost.

The low sensitivity of strain gage transducers (2.5 millivolts per volt at full scale) is not a problem for the phase shift oscillator described above. The circuit for pressure is essentially the same as for temperature with the strain gage bridge replacing the temperature bridge. The new transducers have an internally mounted platinum resistance thermometer to measure the strain gage temperature. The internal temperature measurement is used by the microprocessor to numerically correct for the effects of temperature on the transducer.

In the present design the sensor has been highly thermally isolated to eliminate temperature transient response of the transducer. Strain gauge pressure transducers exhibit transient response which is direct result of small temperature differences across the strain gauge bridge and its integral temperature compensating elements when the sensor is changing temperature rapidly. The current pressure transducer mounting extends the thermal time constant of the transducer to 40 minutes virtually eliminating any transient temperature response of the sensor. This thermal isolation is achieved by the mounting of the transducer in a large block (large thermal capacitance) which is connected to the bottom end cap via a short length of 60,000 PSIA tubing (large thermal resistance), reference Figure 3, thus generating the long thermal time constant.

The ICTD uses a newly developed platinum resistance thermometer. Platinum thermometers were selected as the measurement of temperature is defined by international temperature scale (ITS-90). ITS-90 defines temperature through the use of platinum resistance thermometry combined with fixed physical reference points, triple point of water, mercury, and gallium for the oceanographic range of temperatures. Additionally, high purity helical wound platinum elements when properly built do not exhibit calibration shifts exhibited by less stable Thermistor elements.

The new thermometer uses a .109" (2.7 mm) 316 5.5. sheath into which a four bore ceramic holder with four helical wound high purity one mil diameter wire platinum elements are installed. The construction is completed by the filling of the element using fine ceramic powder which ensures low thermal resistance of the element to the sheath without the voids typical of thermal grease filled assemblies. The resultant construction yields a laboratory grade platinum resistance thermometer with a time constant of 400 milliseconds.

The ICTD relies on Thermistor for high speed temperature measurements. The same electronic interface supports either an exposed (20 millisecond time constant) or a stainless steel sheathed .050" (1.2 mm) diameter (80 millisecond time constant) high stability probe for high speed temperature measurements. The derivative of the Thermistor can be found and added to platinum temperature measurement by the instrument or can be independently transmitted and combined via post processing by the user.

A serious limitation to the long term stability of CTD's is the use of electrode type conductivity sensors. Since the electrodes must be exposed directly to sea water they cannot be reliably protected from marine fouling. The small dimensions of these cells makes them particularly vulnerable to even minute amounts of fouling. Earlier applications of inductively-coupled conductivity sensors were not particularly successful for a number of reasons as follows:

Figure 4b details the present approach that eliminates the previous measurement error sources. This approach uses classic negative feed-back technique to reduce the effect of these error sources to negligible levels. The operation of this circuit is as follows. The input voltage Eg is applied to one side of drive winding Wl. Assume for the moment that the output of amplifier Al is zero the applied voltage across Wl will be Eg. The induced voltage in Wl will be equal to the applied voltage less the voltage drop across Rwl. Since the sense winding W2 is wound with exactly the same number of turns as Wl, and since essentially zero current flows through Rw2, the voltage across the sense winding will be exactly equal to the induced voltage in Wl and W2. Hence the input to the amplifier Al will be equal to the applied voltage minus the induced voltage. In other words if the output of Al is zero the input is equal to the error voltage. However, negative feed-back will reduce the error by a factor equal to the gain of Al by applying an additional voltage across Wl to reduce the error voltage so that the induced voltage is essentially equal to the applied voltage. In practice the circuit reduces the error voltage from about 0.05% to about 0.0005%.

A similar technique is used in the output toroid (W3 and W4) to essentially eliminate the effect of resistance of W4. Since there is no current flowing in W3 the input to the very high gain amplifier A2 will only be zero when the product of the current and the turns on W4 is exactly equal to the current in the sea water circuit regardless of the resistance of Rw4. The amplifier A2 is designed to extremely high gain, which reduces the output toroid current ratio error of less than 0.001%.

This now permits the use of toroidal transformers unprotected from pressure, thus dramatically reducing the thermal mass, complexity and cost, while essentially eliminating the errors caused by the instability in the voltage ratio. The relatively large diameter .96" (24 mm) and short length 1.73" (44 mm) of the hole forming the sea water path and the relatively low overall volume of the housing results in very low thermal inertia compared with the existing electrode types. These effects are discussed in detail by Lueck⁷. The present implementation of the sensor uses 99.9% pure alumina oxide housing to mount the two cores, this is the same material as used in the MKIIIB CTD conductivity sensor. The use of alumina oxide allows the use of temperature and pressure correction coefficients as defined by Millard & Fofonof f⁸. The ceramic sensor eliminates isolation problems typical of coatings used in previous implementations of inductive measurement sensors, and allows the sensor to be cleaned with a bottle brush without effecting the conductivity calibration.

The 0.96" (24 mm) inside diameter of the center hole of an inductively coupled sensor further improves the long term stability compared with the 3 to 4 mm inside diameter typical of existing electrode type conductivity cells. The stability of the inductive cell is limited only by the stability of the cell geometry. The sensor also exhibits significantly shorter flushing length than a MKIIIB CTD cell. Flushing length being directly proportional to the length to diameter ratio of the sensor MKIIIB = 15 mm I 3 mm = 5, ICTD = 44 mm I 24 mm = 1.8.

Figure 5 shows the detailed implementation of the conductivity circuit in the ICTD. The circuit consists of two basic sections. The first section is the phase shift oscillator described above which acts as the signal generator for the drive winding of the input toroid of the inductively coupled sensor. The second section is the current balancing circuit which provides a current through the balance winding of the output toroid that exactly balances the current induced in the sea water circuit. This balance current is passed through a precision resistor (Rs) thus generating a voltage (Es) that is exactly proportional to conductivity. This sensor voltage (Es) is added to the quadrature voltage (Eq) of the phase shift oscillator in the same way as was done for the temperature circuit. Alternatively the multiplexer (MUX in Figure 5) can connect calibrating voltages across R2 and R3 instead of the sensor voltage (Es) to calibrate the electronics in the same manner as was done for temperature and pressure. This technique does not calibrate the current balancing circuit. However, it will be shown below that no calibration drift could be caused by this circuit.

The current balancing circuit works in the following manner. Any unbalance in the circuit will result in a finite input to the very high gain amplifier A2. This error signal is amplified by A2 and detected by the phase sensitive detector. The result is a D.C. signal being applied to the input of the integrator consisting of A3, Ri and Ci. The D.C. output from the integrator controls the resistance within the current balancing circuit (Ibal, which is connected in series with the balance winding (W4) of the output toroid. The very high gain of A2 and the essentially infinite gain of the integrator at zero frequency will ensure that the circuit will exactly balance the current through the sea water circuit. At balance the ampere turns product in the sea water circuit is exactly equal to the ampere turns product in the balance winding W4. The balance current (Ibal) is proportional to conductivity. The balance current (Ibal) is passed through a precision resistor Rs resulting in a voltage (Es) at the input of the phase shift oscillator. This input voltage (Es) will control the frequency of oscillation by changing the phase of the resultant voltage (Er).

The use of internal calibration and 2nd order polynomial regression places a large burden on the internal math capability of the processor. In order to ensure the processor has sufficient capability to perform data collection, numerical computation, bookkeeping, communication and/or data storage a processor was selected which has augmented mathematical capability. The selected controller is a Siemens, which is a derivative of the Intel 8051 line of micro-controllers. The 8 bit architecture of the Siemens part also has a pipe-line 16 bit multiply and divide unit, enhancing floating point operation throughput. The central processor is supported with two internal and two external UART's. The part has an internal 10 bit resolution AID convertor used for internal temperature measurement and correction. Additionally, the system supports and Analog Devices AD7871 14 bit AID convertor for user input of Fluorometer, Oxygen, pH, light transmission or other add on sensors.

The two internal UART's support primary high speed binary data transmission to the surface along with the Bi-directional command control interface from the surface. The remaining UART's allow for communication with add on serial interface type sensors.

The system is programmed in a derivative of standard ANSI "C" and cross complied for the 80C51 using "Franklin"⁹. This compiler fully utilizes the additional capabilities of the 80C517. Program memory is supported in a 64K PAL. Random access memory is supported by both 256 bytes internal to the part and 2K bytes in the external PAL.

The mechanical configuration of the instrument is shown in Figures 6 and 7. The external dimensions of the instrument housing are 4.5" (114 mm) by 11.3" (287 mm). All housing components are fabricated from titanium. This results in an instrument with high strength, low corrosion and low weight. The design has a maximum working depth of 16,000 PSIA (11,000 meters) with a normal operating depth of 10,000 PSIA (7000 meters).

Internally the instrument is configured as a vertical set of 8 cards, 5 sensor interfaces and three boards which support the micro-controller, frequency counters and communication. The 8 cards are interconnected via a common backplane mounted across the top horizontally. The configuration allows for ease of service by removal of the digital section without the necessity for complete system disassembly. The entire electronics assembly is surrounded by a fiberglass sleeve which reduces thermal transients and ensures electrical isolation from the pressure housing.

The instrument transmits sensor data up the sea cable using a digitally synthesized f/2f frequency shifted keyed format at 5000 and 10000 hertz. The high speed data up-link utilizes the telephone band 1200 - 2400 hertz to support a Bi-directional Bell-212 command/control telemetry system similar to the system described by Fougere (1990)¹⁰. The command control interface allows sampling speed/resolution, channel selection, and other operating parameters to be changed during the cast. The command control unit also interfaces via the 2 external controller UART's to externally connected serial devices. Command/Control protocol conforms to a subset of the IEEE ASCII "SAIL" communication standard¹¹.

Shown in Figures 8, 9, and 10 are residual calibration error curves for the ICTD after a standard calibration for conductivity, temperature, and pressure respectively. This data was obtained in the Woods Hole Oceanographic CTD calibration laboratory. The WHOI lab uses triple points of mercury, water and gallium to certify its primary temperature standard, which is measured by an ATB-1250 precision resistance bridge¹². Pressure inter comparisons were versus a "Ruska" precision air-weight dead weight tester, certified to .02% of reading.

The new system was recently tested in Shizouka Bay, near Tokyo, Japan. Given in Figure 11 is a typical salinity versus temperature plot for this area, (1500 meter cast). The instrument was post cruise calibrated, the post cruise calibration of temperature compared to within .001 Celsius. Pressure offset was less than 0.5 dBar.

The new Integrated CTD system is designed to overcome the cost, size, and, stability limitations of existing CTD systems. The design utilizes a unique phase shift oscillator that has the ability to convert the output of virtually any four terminal network to frequency with a high degree of linearity and sensitivity. The network can be as simple as a temperature bridge or strain gage bridge or as complicated as a network of sensors and active electronics such as the conductivity sensor described.

Extreme accuracy is achieved by the use of a simple network of ultra stable precision resistors that simulate the output of the transducer at values of the sensed parameter which are determined at the time of the lab calibration. This simulation network is periodically used in conjunction with the microprocessor to calibrate the oscillator thus eliminating the effect of calibration drift due to temperature, aging or supply voltage variations on the electronics.

Figure 1

BLOCK DIAGRAM PHASE SHIFT OSCILLATOR

Figure 2

EXAMPLE CALIBRATION CIRCUIT: TEMPERATURE

Figure 3

THERMAL ISOLATION: PRESSURE SENSOR MOUNTING

Figure 4

BLOCK DIAGRAM SINGLE VERSUS DUAL WINDING

INDUCTIVE CONDUCTIVITY SENSOR

Figure 5

ICTD INDUCTIVE CONDUCTIVITY CIRCUIT DETAIL

Figure 6

ICTD MECHANICAL DETAIL (SIDE VIEW)

Figure 7

ICTD MECHANICAL DETAIL (BOTTOM VIEW)

Detailed components (Figure 6 and Figure 7): 1) Conductivity sensor, 2) Platinum Resistance Thermometer, 3) Fast Response Thermistor, 4) Redundant Platinum Thermometer, 5) Option Connector, Oxygen Sensor, 6) Option Connector, 7) Guard Cage, 8) Pressure Port, 9) Pressure Housing, 10) Sea Cable Connector, 11) Command/Control Communication Port, RS-232, 12) Lifting Bail.

Figure 8

ICTD CONDUCTIVITY VERSUS ATB-1250 @ 22.5 CELSIUS

(WHOI CTD CAL. LAB.)

Figure 9

ICTD TEMPERATURE VERSUS ATB-1250

(WHOI CTD CAL. LAB.)

Figure 10

ICTD PRESSURE VERSUS RUSKA DEADWEIGHT

(AT 0 AND 25 CELSIUS)

Figure 11

ICTD FIELD DATA 1500 M CAST SHIZOUKA BAY JAPAN

BIBLIOGRAPHY

1. Brown, N.L., "A precision CTD microprofiler," Proc., IEEE Oceans '74 Conference, vol. 2, pp. 270-278, 1974.
2. Brown, N.L., "The Paraloc - a precise telemetry subcarrier oscillator", ISA Marine Sciences Instrumentation, vol. 4. p. 106, 1968.
3. Pederson, A.M., "A Modular High Resolution CTD system with Computer-Controlled Sample Rate", Marine Technology Society, STD Conference and Workshop, San Diego, CA February 1984.
4. Fougere, A.J., Brown, N.L., Hobart, E., "Digital Output Sensing Module for Oceanographic and Atmospheric Measurements", Marine Instrumentation '90, MTS, Conference Proceedings, pp. 46-51, Feb. 1990.
5. Toole, J., Millard, B., Bond, G., "Implementation of a Titanium Strain Gauge Pressure Transducer, for CTD", Submitted for Publication Deep Sea Research, WHOI Cont. #7865, 1990.
6. Brown, N.L., "An in-situ Salinometer for use in the deep ocean", ISA Marine Sciences Instrumentation, volume 4, pp. 563, 1968.
7. Lueck R.G., "Thermal Inertia of Conductivity Cells: Theory", Journal of Atmospheric and Oceanic Technology, December 1989.
8. Fofonoff, N.P., Hayes, S.P., Millard, R.C., "WHOI/Brown Microprofiler, Methods of Calibration and Data Handling", WHOI Technical Report #7489, 1974.
9. Kiel/Franklin C51 Cross Compiler, Franklin Software, Inc., 888 Saratoga Ave., San Jose, California 95129.
10. Fougere, A.J., Smith, S., "Bi-Directional Data Telemetry System for Use with CTD Data Collection Systems", Marine Instrumentation '90, MTS, Conference Proceedings, Feb. 1990.
11. Mesecar, R., "Serial ASCII Instrumentation LOOP (SAIL)" IEEE Technology Assessment Committee (TAC), Instrumentation Sub-Committee, TAC-81-1, August 1981.
12. Brown, N.L., Fougere, A.J., Mcleod, J.W., and Robbins, R.J., "An Automatic Resistance Thermometer Bridge", Temperature: Its Measurement and Control in Science and Industry, American Institute of Physics, vol. 5, pp. 719-727, 1982.