Circuit-aware system design techniques for wireless communication

When designing wireless communication systems, many hardware details are hidden from the algorithm designer, especially with analog hardware. While it is difficult for a designer to understand all aspects of a complex system, some knowledge of circuit constraints can improve system performance by relaxing design constraints. The specifications of a circuit design are generally not equally difficult to meet, allowing excess margin in one area to be used to relax more difficult design constraints. We first propose an uplink/downlink architecture for a network with a multiple antenna central server. This design takes advantage of the central server to allow the nodes to achieve multiplexing gain by forming virtual arrays without coordination, or diversity gain to decrease SNR requirements. Computation and memory are offloaded from the nodes to the server, allowing less complex, inexpensive nodes to be used. We can further use this SNR margin to reduce circuit area and power consumption, sacrificing system capacity for circuit optimization. Besides the more common transmit power reduction, large passive analog components can be removed to reduce chip area, and bias currents lowered to save power at the expense of noise figure. Given the inevitable crosstalk coupling of circuits, we determine the minimum required crosstalk isolation in terms of circuit gain and signal range. Viewing the crosstalk as a static fading channel, we derive a formula for the asymptotic SNR loss, and propose phase randomization to reduce the strong phase dependence of the crosstalk SNR loss. Because the high peak to average power (PAPR) that results from multicarrier systems is difficult for analog circuits to handle, the result is low power efficiencies. We propose two algorithms, both of which can decrease the PAPR by 4 dB or more, resulting in an overall power reduction by over a factor of three in the high and low SNR regimes, when combined with an outphasing linear amplifier. Thesis Supervisor: Gregory W. Wornell Title: Professor