Patent for Application Method and System for Adaptive Control of Real-Time Computer Graphics Rendering - Real-Time Rendering: Computer Graphics with Control Engineering - page 156

the resources on the computer. The effect on the rendering process attributed by
these external processes is defined as the disturbance 104 to the system.
Figure B.2 illustrates schematically the fundamental control system in a closed-loop
configuration. The controller module 201 works on the error between the output 204
of the plant 202 and the reference 203 (performance objective). Depending on the
design, the computed output of the controller module 201 will be fed into the plant
202 such that the plant's output may be regulated to the reference 203. This process
is iterative until the error between the plant's output and the user-defined reference
diminishes to a negligible value.
Figure B.3 illustrates the control system in componentised form localised within
a computer device. The controller 301 and the plant 302 share the resources from
this computer device, such as memory, data bus, and main processor's computation
bandwidth. The controller and plant are connected for data exchange via the main
memory using the shared memory 301 within the same execution space.
Figure B.4 illustrates the framework by which the control system is deployed in
a distributed computing environment. The controller 401 resides in a different com-
puter device from the plant 402 (the rendering process). The controller and the plant
are linked via an external network 403. The control action and the plant's output are
routed via bidirectional digital channel data 404 over this network.
CONTROL DESIGN AND MECHANISM
Due to the complexity in modern computer graphics hardware, rendering processes
may not exhibit linear properties over certain operating ranges. The present invention
describes a design technique that yields a controller which is capable of handling such
non-linearity during the system's operation. The approach consists of two strategies:
I. PID gain scheduling
II. Fuzzy control
The design process commences with collection of a qualified set of input-output
data pairs. The qualifications of the input and output variables are contingent
upon whether the quantities are both measureable and controllable. The data gen-
eration process involves selecting a range of inputs that are sufficient to drive the
dynamics of the rendering system. The derivation of the system model is based on
the system identification methodology where the model may be represented in a
linear auto-regressive (ARX) model or its corresponding state space representation.
i. Pid g ain s cheduling
After collecting the steady-state values of the input-output data, they are plotted
against each other as shown in Figure B.5. The example shows the output (frame rate)
is plotted against the input (vertex count). Empirically, the input-output relation-
ship is typically non-linear. The gain scheduling technique proposed in this inven-
tion requires piece-wise approximation of non-linear curves using straight line
segments. Each segment represents a linear region of operation by which linear