Solutions to Parallel and Distributed Computing Problems: Lessons from Biological Sciences

10.5 NEURAL NETWORKS APPLIED TO THE CHANNEL-ASSIGNMENT PROBLEM

Neural network algorithms to solve the channel-assignment problem have been proposed by several researchers [6, 17]. Kunz [17] used Hopfield and Tank's model for each neuron in the network, creating one neuron for each channel i at each base station j (i.e., a cell). An energy or cost function representing frequency-separation constraints and channel demand is formulated, and is then minimized by the neural network. Kunz uses the following energy function:

f(x) = (1+tan gh(λ, x))/2

where λ is a constant.

Funabaki and Takefuji [6] proposed a neural-network parallel algorithm for the channel-assignment problem. All input values are sequentially updated, while all output values are fixed. Then all output values are sequentially updated, while all input values are fixed. Their network model is based upon the hysteresis McCulloch-Pitts neuron model [19], where the input/output is given by

Note that U_i and V_i are the input and output, respectively, of the ith processing element. The changes in the input U_i are given by partial derivatives of a computational energy function E(V₁,…, V_n), which is also called a motion equation. In [6], the authors proposed an energy function that considers all interference constraints in the channel-assignment problem. They also fixed the channel in one or more cells in order to accelerate ...