Lucent Technologies

Power Supply Networks

Jim Landwehr, Clive Loader and Scott Vander Wiel.

A large number of network interface units (NIU) are connected to a single power supply. When the system is powered up, each NIU requires an initial power surge. If all NIU's surged simultaneously, this would cause the system to overload and fail. Thus random delay times are built into the system. The problem is to estimate the probability of system failure, based on a number of design parameters:

  1. The capacity of the power supply in amps.
  2. The number of NIU's connected to the power supply.
  3. The current drawn by each NIU during the initial surge, and how long it is drawn for.
  4. The standard deviation, or more generally the distribution of, the delay times.
Mathematically, this can be formulated as a special case of the following problem: Let X1...Xn be indepentdent random variables with a density f(x). Fix a window width h. As a window of width h is moved accross the random variables, what is the maximum number of observations that fall within the window? The complexity of this problem makes exact analysis impossible, but since interest is mainly in rare events (i.e. we want the probability of system failure to be small), large deviation theory can be applied to obtain excellent asymptotic approximations to the probabilities of interest. These probabilities are largely characterized by two properties of the underlying density f(x): the maximum, and the second derivative of the density at the maximum.

[Statistics Home] [Statistics Projects]