In wireless communications, the frequency spectrum is a valuable commodity as the
ever increasing number of wireless users demands more efficient usage of the already scarce frequency resources. Communication transceivers rely heavily on frequency conversion using local oscillators (LOs) and therefore the spectral purity of the oscillators in both the receiver and the transmitter is one of the factors limiting the maximum number of available channels and users. For that reason, a deeper understanding of the fundamental issues limiting the performance of oscillators, and development of design guidelines to improve them, are necessary.
During the last fifteen years, there has been tremendous growth in wireless mobile
systems [l]-[3]. These systems have been made possible by technological advances in the field of integrated circuits (ICs) allowing a high level of integration at low cost and low power dissipation. There is also great interest in integrating complete communication transceivers on a single chip. This single chip implementation of the systems results in a new environment for oscillators which has not been investigated before.
In digital applications, the timing accuracy of the clock signal determines the maximum clock rate and hence the maximum number of operations per unit time. In
microprocessors and other synchronous very large scale digital circuits, the clock signal is generated by on-chip oscillators locked to an external oscillator. Ring oscillators are commonly used for on-chip clock generation due to their large tuning range and ease of integration. In the IC environment, there are additional sources affecting the frequency stability of the oscillators, namely, substrate and supply noise arising from switching in the digital circuitry and output drivers. This new environment and the delay-based nature of ring oscillators demand new approaches to the modeling and analysis of the frequency stability of the oscillators.
It will be shown that all oscillators are periodically time-varying systems, and that their time varying nature must, therefore, be taken into account to permit accurate modeling of phase noise. Also due to the periodically changing operation points of the active devices in the oscillator, many of the noise sources have a periodically timevarying power spectrum; they are cyclostationary. In this work, a time-variant model which is capable of properly assessing the effects of both stationary and cyclostationary noise sources is presented.
The approach presented here explains the exact mechanism by which spurious
sources, random or deterministic, are converted into phase and amplitude variations.
This time variant model makes explicit predictions about the relationship between
waveform shape and 1/f noise upconversion. It also shows that the upconversion can
be reduced by exploiting the symmetry properties of the waveform. This result is particularly important in CMOS oscillators because it shows that the effect of inferior 1/f device noise can be reduced by proper design.