Most modern digital and analog electronic devices require a time base to perform their intended function. Found in everything from cell phones to smart munitions, quartz crystal oscillators are widely used in many embedded applications. Quartz resonators’ high Q, excellent temperature performance, and superior long-term aging makes them the clear resonator of choice for many applications. The frequency versus temperature performance of a discrete LC oscillator will be on the order of several hundred parts per million (ppm) per °C, where a crystal oscillator (XO) will have roughly ±30 ppm over the entire industrial temperature range (–40 to +85°C). While being superior to a discrete oscillator, this temperature stability is not nearly sufficient for many modern applications.
The temperature-compensated crystal oscillator (TCXO) employs the use of an open loop compensation circuit to create a correction voltage to reduce the inherent frequency versus temperature characteristic of the crystal. The crystals used in TCXOs have frequency versus temperature characteristics that approximate a third-order polynomial, as seen in the nearby figure.
The early designs for TCXOs employed a network of thermistors and resistors to create a correction voltage. By using thermistors with different slopes and properly selecting the fixed value resistors, the correction voltage can be made to have a shape factor matched to the crystal’s frequency versus temperature performance. The correction voltage is applied to a varactor in the feedback path of the TCXO. This change in capacitance in the feedback path alters the tuning of the oscillator, thus changing the output frequency and compensating it for temperature effects. Thermistor/Resistor network TCXOs can achieve frequency versus temperature stabilities of around ±1 ppm over the industrial temperature range; however, they are limited in their curve-fitting capabilities because of the nature of using discrete thermistors and resistors.
Thermistor/resistor network TCXOs are still found in specialized environments including satellite and other space applications where modern solid-state devices do not have the radiation hardness to survive. Most TCXOs manufactured today utilize an ASIC which contains the oscillator circuit and a third- or fifth-order polynomial voltage generator. The polynomial generator is an analog output voltage but also has digital registers for setting the coefficients of the polynomial. The newest generations of TCXO ASICs can provide temperature performances of ±0.1 ppm over the industrial temperature range. This is a 10-fold improvement over what is obtainable with a traditional thermistor/resistor network TCXOs and also has the advantage of a much smaller footprint (5 mm × 3.2 mm).
Some high-precision applications require frequency versus temperature stabilities better than ±0.1 ppm. To meet these challenging specifications a different methodology is implemented. An oven-controlled crystal oscillator (OCXO) uses a heater circuit and thermal insulation to keep the crystal at an elevated temperature (≈15°C above the upper operating temperature limit). By controlling the crystal’s temperature and keeping it nearly constant, the frequency deviation due to ambient temperature changes is vastly reduced. OCXOs can achieve frequency versus temperature stabilities of ±0.005 ppm. This improved performance comes at the cost of a larger footprint and increased power consumption. The TCXO’s performance limit of ±0.1 ppm is due to several factors. First, the resonators are not perfect. Their frequency versus temperature stability approximates a third-order polynomial; however, higher order effects are present. Secondly, the polynomial generator is nonideal and induces some higher order artifacts, leaving the user with residuals of ±0.1 ppm. A new methodology which uses an artificial neural network (ANN) to create the correction voltage has recently been demonstrated. The ANN is superior in that the neural network is not inherently shape limited like a third-order polynomial. If enough data is presented to the ANN, it can “learn” the crystal’s temperature performance shape and correct for it. This new methodology has been shown to provide ±0.01 ppm frequency versus temperature stability over the industrial range. The ANN algorithm can achieve OCXO temperature performance in a much smaller footprint, and without the need for the power-hungry oven.
The evolution of quartz crystal time bases over the last 70 years has seen the frequency versus temperature stability improve by a factor of several thousand. As our need for more stable oscillators in smaller packages with less power consumption grows, the development of better compensation schemes is paramount. The ANN demonstrates a technology that has much potential. Its ability to adapt and change its shape factor makes it ideal for complex compensation problems.
Circuit Cellar 291 (October 2014) is now available.