Fifty Years of Progress in Quartz Crystal Frequency Standards
Proceedings of the 1996 IEEE International Frequency Control Symposium, pp. 33 – 46.
© 1996 IEEE. Personal use of this material is permitted. However, permission to reprint/republish this material for advertising or promotional purposes or for creating new collective works for resale or redistribution to servers or lists, or to reuse any copyrighted component of this work in other works must be obtained from the IEEE.
Marvin E. Frerking
Collins Avionics & Communications Division
350 Collins Road NE
Cedar Rapids, lowa 52498
The progression of developments in Crystal Frequency Standards is traced through the last half century. The paper emphasizes the underlying technical changes and innovations that have driven the remarkable progress, while keeping in mind the contribution of the great body of information that was already available at the beginning of the period.
The paper gives a treatment of the progress of ovenized crystal frequency standards from the early vacuum tube days through implementations using transistors, LSI, and hybrid construction.
The progression of TCXO developments is also covered including analog thermistor temperature compensation, digital compensation, and techniques using micro-computers, and self-sensed oscillators.
An examination of the past half century of progress in crystal frequency standards is indeed an exciting venture. The story reveals not only dramatic improvements in performance and size, but also a blend of previously known technology with innovations and improvements in implementation made possible by advancements in components and manufacturing techniques.
As we look at the resulting progress, the performance that was available near the beginning of the period in large rack mounted frequency standards is presently achievable in oven frequency standards with a volume of only 1-cubic inch (16 cm3). It is also nearly possible to achieve these results with the most advanced temperature compensated crystal oscillators in a space less than 1/2-cubic inch and with power consumption about two orders of magnitude below the early counterparts.
Oven controlled frequency standards have obviously seen very significant improvements in both performance and in size. The most dramatic improvements have been in size, however, made possible by the transistor, integrated circuits, and hybrid construction. The resulting size reduction has allowed precision units to be produced in about 1-cubic inch of volume. Many of the oven units in use today provide accuracies in the region of a few parts in 108 with parts in 109 being achievable using SC-cut crystals. A volume in the range of 1 – 4 cubic inches is typical at the time of this writing. Compared to units available in the late 1950s requiring over 100-cubic inches, the improvement is truly remarkable.
The list of innovations resulting in this progress is quite impressive. A partial list of some of the more significant changes is given below:
1. The use of transistors in frequency standards and oscillators.
2. Precision Overtone Resonators (energy trapping).
3. The use of varactors to pull the crystal frequency, particularly in TCXOs.
4. The use of varactor-thermistor temperature compensation in temperature compensated crystal oscillators.
5. The application of digital techniques to temperature compensation.
6. The use of externally temperature compensated crystal oscillators.
7. The discovery of the SC-cut crystal.
8. The use of microprocessors for temperature compensation and synthesis of the output signal.
9. The use of self-sensed crystal oscillators (dual mode).
10. The use of DC control circuits and transistor heaters in oven standards.
11. The application of hybrid and micro-electronic packaging techniques.
12. The use of a vacuum for thermal insulation (Although ironically standards during the early part of the half century often made use of the evacuated Dewar Flask).
13. The invention of SAW resonators and filters for UHF and microwave oscillators.
14. Etching techniques for VHF fundamental mode crystals.
Surprisingly, many of the techniques in use today were known at the beginning of the half century or shortly thereafter*. At the beginning of the period, in 1947, Professor Cady’s book  had been published and Raymond Heising had published his book  on advancements in theory, design, and production during World War II. Also, through such organizations as Galvin Mfg. Co. and the Western Electric Co., much of the manufacturing technology had been made available to the industry.
At this time, high frequency crystals had been developed that were stable under severe shock and vibration. A 100-kHz GT-cut crystal had been developed that had an aging rate after three months of 1 x 10-9 per day. VHF crystals and sealed metal crystal holders had been developed, but the holders leaked and high frequency crystal units drifted badly, in comparison to today’s standards.
By 1956, the 2.5-MHz and 5-MHz high frequency precision AT-cut fifth overtone designs were nearly complete. Other work in progress was the use of thermistors for temperature compensated crystal oscillators. This procedure was reported as early as 1955 on thermistor compensation by thermistor shunted capacitors .
The evolution of crystal oscillators is a fascinating one. One of the first quartz crystal oscillators used for frequency control was built by W.G. Cady in 1921 , , . It consisted of an amplifier with the output coupled to the input through a crystal. An X-cut crystal was used with two pairs of electrodes, one pair connected to the plate circuit of the output tube and the other pair to the grid circuit of the amplifier input.
The first Crystal Clock, is credited to Horton and Marrison in 1927 , , , . The crystal clock used a 50-kHz X-cut rectangular block of quartz supported by two heavy metal pole pieces. This was replaced by a quartz ring. The ring crystal was used until about 1937 at Bell Labs at which time it was replaced by the GT-cut crystal. The GT-cut crystal had a temperature coefficient that remained essentially zero over a large temperature range which resulted from the canceling effect between two modes.
* A significant amount of the material in this paper is drawn from the Proceedings of past Symposia, especially the article “A Quarter Century of Progress in the Theory and Development of Crystals for Frequency Control and Selection.” by E.A. Gerber and R.A. Sykes, presented at the 25th Annual Symposium of Frequency Control. As in that paper, no attempt will be made to establish credit, but references to individuals or institutions may be made when it clarifies the presentation.
Figure 1a  shows a photo of the 50-kHz crystal mounting used for the early Crystal Clock cited above. A photograph of the ring type crystal is shown in figure 1b. A photograph of a pressure point mounted GT-cut crystal is shown in figure 1c.
At the beginning of the period and for several years thereafter, oven frequency standards were implemented using vacuum tubes with the crystal in an oven, often using a Dewar flask to minimize thermal loss. The standards were large and often designed for mounting in a 19-inch relay rack. It was not uncommon to use a 1 MHz fundamental mode AT-cut crystal, although earlier standards used a 100-kHz GT cut. Aging rates as low as 2 x 109 per day were reported for the 100-kHz GT-cut crystal .
Bridge measurements were being made on 100-kHz GT quartz crystals as early as 1946 at the National Bureau of Standards (NBS) . High precision quartz crystal resonators were added in 1951 as a part of the primary standard of frequency. Temperature control at NBS was achieved by using a double oven with control accuracy to within 0.001 °C. Aging rates of 1 x 10-10 per day were achieved.
During these formative years, a frequency standard used in the industry was the General Radio, Type 1100-A rack mounted unit. This standard was produced from 1948 until about 1960. It used a 100-kHz second harmonic extensional mode quartz bar. A photograph of this resonator is shown in figure 2.
The accuracy of these frequency standards was reported to be a part in 107 for several months. The crystal temperature was controlled to 0.01 °C at normal ambient temperatures by a compensated thermostat cir-cuit. This standard also included a synchronous motor clock for evaluating its frequency directly in terms of time. The clock was driven by a 1000-Hz signal derived from the crystal. A photograph of this standard is shown in figure 3.
In 1955, results were reported on a Bridge Balancing Oscillator utilizing a 1-MHz GT-cut crystal . An AFC circuit was employed in connection with the bridge to reduce the effects of the vacuum tube oscillator that was not in the oven. Aging rates of a few parts in 1010 per day were achieved. A photograph of the bridge oscillator is shown in figure 4 .
In 1958, results were reported on work at Bell Laboratories on a cryogenically cooled crystal using a 10-MHz AT-cut overtone crystal, operated at liquid helium temperatures of 4.2 ° K . Two Dewar flasks were used, the outer flask contained liquid nitrogen at 77 ° K to minimize the helium loss from the inner flask. A crystal Q of 500,000 was achieved and the aging rate was less than 1 x 10-10 per month. A photograph of this standard is shown in figure 5 .
A vast amount of work was also done to design and produce frequency standards for the commercial and military markets. These standards were large compared with today’s, but they were generally more compact than those designed for use in the laboratory environment where everything possible was done to extract the ultimate in performance. Most of the crystal frequency standards described in the remainder of this paper address units produced in the industry for use in communications and navigation equipment.
At the beginning of the half century and for several years thereafter, transistors were not yet available. Since the active circuits using vacuum tubes were large, only the crystal was normally enclosed in the oven. It was necessary to use an AGC circuit to limit the crystal drive level because vacuum tubes limit at fairly high levels.
To avoid drift in the operating point of the control amplifier, AC oven control circuits were frequently used . A rather ingenious oven control circuit used two heater windings with different temperature coefficients in a bridge circuit. An audio frequency power amplifier was connected to the windings. When the oven was below the operating temperature, the oven windings were unbalanced to provide positive feedback to the amplifier input, and the amplifier oscillated. The resulting signal heated the oven. As the desired temperature was approached, the oven windings became nearly identical, balancing the bridge. This reduced the positive feedback so that the level of oscillation provided just enough power to stabilize the oven temperature. These and other clever techniques are a great credit to the ingenuity of the scientists, engineers, and technicians using the components available at that time to achieve rather impressive results.
In the 1950s, one of the standards developed was the 40K-1 oscillator at Collins Radio. (Refer to figure 6 for a photograph.) The oscillator measured 18 7/8 x 8 3/4 x 7 inches and used a 1-MHz fundamental mode crystal in an oven using a Dewar flask. The aging rate was specified to be better than one part in 108 per day after one month of continuous operation, though some units achieved aging rates approaching 1 x 10-11per day. Over 1000 of these units were produced for the U.S. defense system.
With the introduction of the transistor, a wealth of opportunities for improvements in the construction of oven frequency standards was made available. It suddenly became possible and practical to include the oscillator stage as well as isolation amplifiers and the AGC circuit in the oven. Ironically, the transistor stages drifted so badly with temperature, that it was almost mandatory to place them in the oven.
Precision oven standards of the early 1960s often used a 5-MHz fifth overtone crystal, developed by Warner at the Bell Telephone Laboratories and manufactured by Bliley Electric Company. The drive level was limited by AGC circuits to achieve the lowest possible aging rate. Aging specifications of about 3 x 10-10 per day were common and many crystals were better than 1 x 10-11 per day. The temperature stability of the standards was quite good and results of about 5 x 10-11 per degree centigrade were available. Power requirements were high by today’s standards; around 25 watts, using foam insulation, and a warm-up time of 30 minutes was considered reasonable. A typical unit used in the Naval Tactical Data System, measured 4 7/8 wide by 19 inches long by 8 3/4 inches high. (12 x 48 x 22 cm).
A significant amount of activity was taking place in the late 1950s to develop a ruggedized frequency standard for missile borne applications. One missile borne frequency standard  used a ruggedized 5-MHz crystal unit and held 1 x 10-9 during vibration from 10 Hz to 2000 Hz. Another missile borne frequency standard developed at Collins Radio used a 11-MHz fundamental crystal in an all solid state design. The acceleration sensitivity of AT-cut crystals was about 1 x 109 per G.
Also during the 1950s, some experimental work was done under sponsorship of the U.S. Army to develop a change of state oven using paradiethoxybenzene with a triple point at 70.47 ° C. Various problems with this technique prevented widescale commercialization.
A precision double oven frequency standard was developed at Motorola during the early 1960s. The unit used a 2.5-MHz fifth overtone AT-cut resonator with a Q of 4 million. A photograph of this unit, designated the 1011 Frequency Standard, is shown in figure 7. The unit had a temperature stability of ± 1.5 x 10-10 from 0 to 35 ° C. The aging rate was specified at 5 x 10-11 per day after four weeks; however, many of the crystals achieved rates better than 1 x 10-11 per day.
Another oven development of the 1960s was the use of DC-oven control circuits. As noted earlier, one of the disadvantages of the AC-oven control circuits was that the sidebands, related to the audio oscillation, appeared on the RF-output spectrum of the signal. The DC ovens eliminated the sidebands; however, the operating point of the solid state DC amplifiers drifted badly and necessitated placing at least the early stages of the amplifier in the oven. At this time most of the ovens still used heating elements formed by wrapping resistance wire around the oven, but thermistors replaced the resistance wire as the temperature sensor of choice.
During the 1970s, the use of resistance elements gave way to the use of power transistors for the heating elements. This had several advantages, such as ease of manufacture and power that was proportional to the current rather than the square of the current, thus linearizing the gain of the control loop. Also, nearly all of the power was dissipated in the heater rather than in the control amplifier, resulting in greater efficiency. Additionally, the use of transistor heaters virtually eliminated the need for a fast warm-up winding.
Power transistor heaters, unfortunately, resulted in a near point heat source that sometimes resulted in significant temperature gradients across the oven. Fortunately, the advent of the SC-cut crystal, discovered in 1974 by EerNisse , greatly reduced the susceptibility of the standards to temperature gradients.** Frequency stability levels of about ± 1 x 10-8 were achieved over a large temperature range with AT-cut crystals, with improvements to a few parts in 109 when SC-cut crystals were used. An oscillator incorporating a DC-oven control circuit with transistor heaters and an SC crystal was designed by Hewlett Packard in 1978. A photograph of this oscillator is shown in figure 8.
The paradigm of using an AGC circuit in moderate precision applications was gradually changed since transistor oscillator stages limit at a lower level than vacuum tubes, and since SC-cut crystals are less sensitive to drive level. Elimination of the AGC circuit significantly simplified some standards and eliminated the need to place all the amplifier stages in the oven. The availability of operational amplifiers resulted in further improvements in the oven control circuit and also made it possible to leave the control amplifier out of the oven entirely in some semi-precision applications.
It is interesting to note that some of the techniques have brought the process to a complete circle. For example, at the beginning of the half century, only the crystal was normally in the oven. Then as time progressed, both the oscillator/amplifier stages and the oven control stages were in the oven. Further advancements are again resulting in a reduced amount of circuitry in the oven.
The use of SC-cut crystals during the 1970s also allowed substantial reductions in the warm-up time. With great care, a warm-up time of 3 minutes from a -55 °C turn on (to about 3 x 10-8 of the final frequency) was achieved for the Global Positioning Systems (GPS). This is essentially the state of the art at the time of this writing for many production units.
The crystal of choice evolved from a 100-kHz GT cut to a 1-MHz fundamental AT cut. This resonator was soon displaced by the 5-MHz fifth overtone AT cut. The ATs were then displaced in many applications by the 10-MHz third overtone SC cuts. Other resonators were, of course, being used (e.g., a 5-MHz third overtone SC cut) but the cited units were and still are often employed. Power consumption was and is generally greater than about 1.5 watts at the cold temperature extremes of the military range. Surface mount resonator holders have also been developed and will likely become more important as the need for miniaturization continues.
The use of hybrid construction techniques allowed reductions in the size of oven standards during the latter part of the period and often hybrids were used in the oven assembly. A significant advancement was made in both power and size reduction in the tactical miniature crystal oscillator (TMXO) sponsored by the U.S. Army. Much of the work was accomplished from the mid-1970s to the 1980s . The TMXO involved using hybrid construction techniques in an oven assembly that was itself placed in an evacuated enclosure. The overall volume was about 1-cubic inch, and the power consumption was only 250 milliwatts. The temperature stability was ± 1 x 10-8 from -40 °C to +75 °C. A photograph, a later version of the TMXO, is shown in figure 9. The construction of the TMXO involved a number of rather exacting, if not exotic, techniques but because of economic issues many applications have used either conventional ovens, or precision TCXOs, that are approaching oven accuracy.
At the time of this writing, oven standards are widely used. They often use conventional insulation such as foam, with small chip parts and surface mounting. Both AT- and SC-cut crystals are used depending on the stability desired. Transistors are normally used for the oven heaters. The temperature stability varies from a few parts in 109 to parts in 107. Aging rates of about 5 x 10-11 per day to parts in 109 per day are common, and sizes as small-as-1/2-cubic inch can be purchased. A few OCXOs for high precision applications have achieved much higher performance. One standard, using a BVA resonator for a spacecraft application, showed an aging rate of a few parts in 1012 per day .
The acceleration sensitivity of quartz crystals has been difficult to improve for large quantity manufacturing. The typical acceleration sensitivity for AT cuts during the 1960s was around 1 x 10-9 per G, and for many crystals this is still the level of performance. Some applications have required better performance, and improvements have been achieved by different mounting techniques. Often, better performance is achieved by selection of low G sensitivity units. Some results in the order of 1 x 10-10 per G or better have been reported, for example, using the BVA resonator . Some work has been done to compensate for the G sensitivity of the crystal using electrical means in connection with accelerometers . To date, these results have been primarily applicable to the static acceleration and the low part of the vibration range.
The evolution and progress of the TCXO during the last fifty years, like that of oven standards, is truly an exciting story. Unfortunately, because of proprietary considerations, much of the work on TCXOs was not published. A considerable amount of information was available to the author on early TCXO progress at Collins Radio Company and this paper draws heavily from that source. Much work on TCXOs took place at other facilities as well, as noted above, but unfortunately not much of that work was published and may be only superficially discussed in this paper.
It was during the last years of the 1950s that two crucial elements of the TCXO came together; namely, the varactor diode and the thermistor. Prior to that time, a large pull range was usually achieved by the use of reactance modulators to pull crystal. These modulators drifted badly with temperature. The varactor diode on the other hand was relatively stable; it was available in convenient values of capacitance, and it was physically small. Additionally, the control voltage range was compatible with compensation networks working from the relatively low voltages required by transistor oscillators.
Some temperature compensation was accomplished prior to the availability of varactors by using thermistors in parallel with capacitors or inductors, but it was the introduction of the varactor that made compensation a reality when a large temperature range was involved. One of the results reported on temperature compensation used a thermistor in parallel with an inductor. The parallel circuit was placed in series with the crystal to obtain a temperature coefficient of ± 0.1 ppm per degree from 59 °F to 113 °F. A photograph of this oscillator is shown in figure 10 .
A variety of thermistors also became available with values from a few kilohms to several megohms. A considerable selection of beta values was available, particularly values of several thousand that were ultimately used in practical TCXOs covering a wide temperature range. Early attempts were also made to temperature compensate crystal oscillators using bimetal springs to apply mechanical force to the crystal and compensate its temperature coefficient. These techniques were far inferior to electrical compensation and were displaced by the varactor.
The most difficult part of the task of compensation during the early days of TCXO development was to find a thermistor network that would match the temperature coefficient of the crystal as transformed by the nonlinear capacitance vs. voltage curve of the varactor.
Some of the early work attempted to linearize the curve as much as possible. Thus a crystal with widely separated turning points was used so that it was fairly linear in the operating temperature range. Unfortunately, this also required that the crystal be pulled over a fairly wide frequency range. Multiple resistance decade boxes were used to simulate thermistors over the temperature range in attempts to devise a suitable linear thermistor network. This was a very difficult task because the thermistor were quite nonlinear.
The attempt to develop a linear wide temperature range thermistor-varactor TCXO network was basically unsuccessful. As a result of considerable ingenuity, however, a unique combination of three thermistors of grossly differing resistance and beta values was found that would match the cubic curve of the AT-cut crystal . Furthermore, the network had the property that only one thermistor and one resistor had a predominant effect on the curve in any given temperature range. Hence, three separate resistors were available to adjust the compensation in the cold temperature, room temperature, and the hot temperature regions with relatively lesser effects on the rest of the curve. An iterative test procedure was worked out so that after several temperature runs an accuracy of ± 0.5 ppm to ± 1 ppm could be obtained over the entire temperature range from
-55 °C to +75 °C. An additional adjustment for overall frequency pullability was made by placing a test select capacitor across the varactor. The resulting thermistor network is shown in figure 11. The temperature region where each element is most active is shown in parentheses. Using this network, a fixed set of thermistors could be used and only the resistors were adjusted for the particular crystal being used.
One of the first applications of TCXO’s covering a wide temperature range was the unit used in the Marine Corps. PRC-47 single sideband back pack radio which was developed in the very early 1960s. A photograph of this TCXO, with an accuracy of ± 0.5 ppm, is shown in figure 12. This unit, produced by Collins Radio Co., used the three thermistor compensation network shown in figure 11.
Other approaches and networks were also devised in which the thermistors were characterized individually, and selected for best match in a given TCXO. For example, a two thermistor network was used in some TCXOs where a lesser temperature range was required. The approach of using thermistors across capacitors has survived, and many TCXOs today are compensated using this technique over limited temperature ranges, particularly where low cost is required.
One of the limitations of the three thermistor network of figure 11 was that it did not track well above the upper turning point of the crystal, therefore, four and five thermistor networks were devised for applications requiring an upper temperature above about 80 ° C. A five thermistor network was capable of compensation up to 120 ° C. When more than three thermistors were required, computers were often used, in connection with initial temperature measurements, to custom design each TCXO network.
Another method of temperature compensation, pursued essentially in parallel with the nonlinear resistor thermistor network development, was to use “piecewise linear compensation” in which transistors were used to produce smaller isolated temperature regions where a single thermistor and resistor could be used to track the crystal curve. Results of this research were reported in 1968 , . Attempts were made to develop a servo system using motor driven potentiometers to accomplish the adjustments automatically. The segmented approach was fairly successful in providing independence of the segments but unfortunately required a large number of components.
Temperature compensation techniques were also developed in which the cubic curve of the crystal is generated electrically using a single temperature sensor. The power series generated from this voltage can be shaped by adjusting resistors that determine the coefficients of the expansion. Custom ICs were developed to minimize the number of components required for this approach .
During the mid to late 1970s, digital compensation also became practical where a temperature sensor was digitized to address a memory. The contents of the memory was preprogrammed with the required compensation voltage, and the output of the memory was connected to a D/A converter that fed the varactor. This approach realized a truly segmented method of compensation , . The economics, however, continued to favor analog compensation, and at the time of this writing, many TCXOs continue to use analog networks.
A refinement of the digital approach uses a microcomputer to implement the digital compensation. The idea appeared in publications as early as 1978 , but at the time microcomputer technology was still in its infancy and because of economic issues and size constraints microcomputers were not used extensively in production until the mid-1980s.
The use of the microcomputer offered several advantages over classical digital compensation. It became easy to interpolate between stored memory words reducing the size of the memory and the frequency jumps caused by limited memory. Another advantage was that the microcomputer could be made to be a part of the D/A and A/D converters. For the A/D converter, it allowed such temperature sensing schemes as a thermistor controlled multivibrator or a dual slope integrator to be used with the microcomputer determining the period. For the D/A converter, the microcomputer could be used to generate a variable duty cycle rectangular wave whose DC value was extracted by a simple low pass filter composed of a resistor and a capacitor. The photograph of an early microprocessor compensated crystal oscillator, built in 1984 at Rockwell International, is shown in figure 13.
Another temperature compensation scheme that was developed for very large scale production involved using multiple analog segments each covering a 20- to 30- degree temperature range. In this procedure, a linear temperature sensor is used; a digital memory holds the gain and offsets constants for each temperature segment. Nearly the entire TCXO was integrated in this way, resulting in a small inexpensive oscillator.
Surprisingly, during about the first three decades there was relatively little improvement in TCXO accuracy using direct temperature compensation methods in which the crystal is pulled to the correct nominal frequency. Many of the TCXOs produced in the early 1960s had an accuracy of ± 0.5 ppm from -40 °C to +75 °C. Over 50,000 of these units were produced at Collins Radio, and they were a significant factor in the worldwide acceptance of single sideband that required an accuracy of about ± 30 Hz in the high frequency radio bands up to 30 MHz.
Many more TCXOs were produced to an accuracy of ± 1 ppm. There seemed to be a dividing line at about ± 1 ppm, and accuracies beyond this became increasingly difficult. Part of the problem was that fine tuning of the analog network to fit the crystal curve exactly was very difficult. An even more serious limitation was due to hysteresis, and no matter how carefully the compensation network was tuned, it seemed impossible to compensate to better than a few parts in 107.
A comprehensive study of crystal hysteresis was carried out in connection with quartz thermometer research with the results reported at the Frequency Control Symposium in 1968. It was concluded that: “with inadequate vacuum baking or inadequate cleaning of the crystal mounts, header, or can the hysteresis effect can be aggravated. However, two experiments have shown rather conclusively that with due attention to processing of the crystal, the remaining hysteresis effects are not related to contamination” .
It could not be determined if the remaining hysteresis was an intrinsic property of quartz or attributed to a defect structure. Clearly, the hysteresis was affected by the angle of cut. At the time of this writing, almost 30 years later, the underlying cause(s) of hysteresis are still unknown.
An interesting approach was developed under U.S. Army sponsorship around 1970 in which the errors from an analog thermistor network were corrected by a fine digital compensation network. A few units were produced to a specification of ± 5 x 10-8 but the results were heavily dependent on selecting crystals with very low hysteresis , .
The advent of digital compensation allowed minimization of the curve fitting errors, but it was still difficult to reliably achieve compensation results better than 2 to 3 x 10-7 over a wide temperature range in large scale production. Fortunately, a significant breakthrough was reported in 1978 in which the output frequency was not obtained by pulling the oscillator to the nominal frequency. This allowed the oscillator to use an SC-cut crystal that was too stiff to pull to frequency. A second mode, the B-mode of the crystal, which had a large temperature coefficient, was used in a dual mode oscillator for temperature sensing . Accuracies in the 10-9 range were postulated for the technique and computer simulations indicated that a part in 10-8 could be expected for scan rates of 3 °C per minute.
Practical difficulties were encountered with both the dual mode oscillator and with the behavior of the B-mode over temperature. Research by the U.S. Army resulted in a variation that focused on exploiting the difference in temperature coefficients between the fundamental and third overtone C-modes of an SC-cut crystal to build a self-sensing temperature compensated crystal oscillator. Much of the work was classified and results were not generally published until the 1989 time frame , , , , . Circuits were developed that produced temperature compensated oscillators to a specification of ± 3 x 10-8 over the military temperature range, although some units achieved an accuracy of about ± 1 x 10-8.
The units that were being produced used a 10-MHz third overtone SC-cut resonator in an HC-40/U holder. The beat frequency between the third harmonic of the fundamental and the third overtone was used to provide a temperature sensor. The beat frequency had a temperature coefficient of around 100 ppm per degree centigrade. This microcomputer compensated crystal oscillator, (MCXO) has been produced in limited quantities at the time of this writing. A photograph of an early version of this oscillator is shown in figure 14.
The use of external compensation has been a very significant factor in breaking through the accuracy limits on TCXOs. Since it was not necessary to pull the crystal to frequency, the varactor circuit, a significant source of instability, could be eliminated. Much of the instability actually resulted from the high impedance point between the crystal and the varactor, making the TCXO sensitive to stray capacitance changes such as caused by moisture being absorbed into a printed circuit wiring board.
Even more significant, however, was that external temperature compensation allowed the use of crystal cuts other than ATs. Much of the published work revolved around the use of external temperature compensation in connection with a dual mode oscillator, and the results have already been cited in the MCXO. At the present time, a lesser published technique is also in use in which an independent temperature sensor is used rather than a dual mode oscillator. Using this technique with an SC-cut reference crystal, over 1000 units were manufactured to specifications of ± 2.5 x 10-8 from -55 °C to +95 °C for the MILSTAR program at Rockwell International during 1994 and 1995. Some engineering models achieved an accuracy of about 1 x 10-8.
The technique used with AT-cut reference crystals has demonstrated stabilities better than ± 1 x 10-7, using third overtone resonators and 1 to 3 x 10-7 with fundamental units.
The output signal for the ECXO has been synthesized in various ways ranging from pulse deletion to direct digital synthesizers (DDS) or DDSs used in combination with phase locked loops.
The accuracy of TCXOs has improved by almost two orders of magnitude during the half century from about 1 ppm to 2 x 10-8. The size of TCXOs has also been reduced dramatically with many units in the accuracy range of a few ppm to even parts in 107 being available in surface mount packages about 0.5 inches square and a few tenths of an inch thick. Units with accuracies in the low part in 108 region are currently being manufactured in substantial quantities in a volume of about 1-cubic inch with engineering models being produced in a volume of about one third this value. A photograph of a small ± 5 x 10-8 TCXO is shown in figure 15.
At this time the performance of crystal frequency standards is limited primarily by the crystal. Historically, this has often been the case during the past fifty years. As improvements are made in resonator technology and processing, we can expect to see continued progress in crystal frequency standards. The impact of integrated circuit technology and computers, both in the frequency standards themselves and in the manufacturing and test processes, will undoubtedly also play a large roll in advancing the state of the art of crystal frequency standards.
 Gerber, E.A. and Sykes, R.A., “A Quarter Century of Progress in the Theory and Development of Crystals for Frequency Control and Selection,” Proceedings of the 25th Annual Symposium of Frequency Control, 1971, pp 1-45.
 Hart, R.K., Hicklin, W.H., and Phillips, L.A., “Tactical Miniature Crystal Oscillator,” Final Report, ECOM-0301-F, 1973, Georgia Institute of Technology, Sponsored by U.S. Army Electronics Command.
 Schodowski, S.A., “A New Approach to High Stability Temperature Compensated Crystal Oscillators,” Proceedings of the 24th Annual Symposium of Frequency Control, sponsored by the U.S. Army Electronics Command, 1970, p. 200.
 Duckett, P.D., Peduto, R.J., and Chizak, G.V., “Temperature-Compensated Crystal Oscillators,” Proceedings of the 24th Annual Symposium of Frequency Control, sponsored by the U.S. Army Electronics Command, 1970, p. 191.
 Boor, S.B., Horton, W. H., Angove, R.B., “Passive Temperature Compensation of Quartz Crystals for Oscillator Applications,” Proceedings of the 19th Annual Symposium of Frequency Control, sponsored by the U.S. Army Electronics Command, 1965, p. 105.
 Fischer, M.C., Kusters, J.A. and Leach, J.G., “Dual Mode Operation of Temperature and Stress Compensated Crystals,” Proceedings of the 32nd Annual Symposium on Frequency Control, 1987, pp. 389-397.
 Messina, J., Bowman, D., Filler, R., Lindenmuth, Rosati, V., Schdowski, S., “Results of Long Term Testing of Tactical Miniature Crystal Oscillators,” Proceedings of the 43rd Annual Symposimun on Frequency Control, 1989, pp. 47-50.
 Kawashima, H., Sato, H., and Ochiai, O., “New Frequency Temperature Characteristics of Miniaturized GT Cut Quartz Resonators,” Proceedings of the 34th Annual Frequency Control Symposium, 1980, p. 131.