Trim effect compensation using an artificial neural network

Trim effect is a skewing of the frequency versus temperature performance of a crystal oscillator as the frequency is pulled (trimmed) away from the oscillator´s nominal frequency. As TCXO (Temperature Compensated Crystal Oscillator) frequency versus temperature stabilities have improved to ppb (part per billion) levels trim effect has become more of a concern. Even though unwanted, the degradation of performance from trim effect is something generally accepted as a characteristic of TCXOs. This paper focuses on a method of compensating crystal oscillator trim effect. Through the use of an artificial neural network, trim effect compensation of AT cut crystal oscillators can be achieved with better than +/-15ppb stability over the industrial temperature range (-40 to +85 °C). This is more than a 10 fold improvement over the inherent trim effect found using state of the art polynomial function generator compensation. The theory of this compensation method with be discussed, and data showing the results of trim effect compensation on actual oscillators will be presented.