Improved Models for DC-DC Converters

Abstract: To obtain high performance control of a dc-dc converter, a good model
of the converter is needed. It is suitable to consider the load to be included in
the converter model since it usually affects the dynamics. The load is often
the most variable part of this system. If the load current and the output
voltage are measured there are good possibilities to obtain a good model of
the load on-line. Adaptive control can then be applied to improve the control.
In peak current-mode control, the output voltage and the inductor
current are measured and utilized by the controller. This thesis analyses some
properties that can be obtained if the load current is also measured and
utilized for control. Accurate expressions for the control-to-output transfer
function, the output impedance, and the audio susceptibility are derived for
the buck, boost, and buck-boost converters operated in continuous
conduction mode in the case where the load is a linear resistor. If the
measured load current is utilized properly by the controller, the output
impedance becomes low and the control-to-output transfer function becomes
almost invariant for different loads. The use of load current acts as a
feedforward term if the load is a current source. However, if the load is a
resistor the load current is influenced by changes in the output voltage and
the stability is affected. Therefore, the use of load current is not a feedforward
action in this case. Instead it can be seen as gain scheduling, which can be
considered a special case of adaptive control.
In the thesis it is also shown that the two published models for currentmode
control, Ridley (1991) and Tan and Middlebrook (1995), give accurate
expressions for the control-to-output transfer function and the output
impedance but not for the audio susceptibility. A novel model for the audio
susceptibility is presented and it is used to improve the two published models.
Most of the results in the thesis are validated by comparing the frequency
responses predicted by the expressions and switched large-signal simulation
models.