Equations (6) and (7) show that T1 and T2 are calculated using complex trigonometric functions; which can not be evaluated precisely with micro-controllers. These equations were implemented using neural networks. Architecture of neuron used in hidden layer is shown in Figure 4. First these equations were implemented in Matlab Simulink. Input output samples were derived through simulations as explained in [11]–[12] and then used for training of ANN. The designed neural network consists of three layers. First layer is the input layer which takes angle B and sector number Formula as the inputs which is provided to second layer consisting of ten neurons with tan-sigmoidal activation function. Structure of second layer with five neurons is shown in Figure 5. Third layer consists of two neurons, as there are two outputs T 1 and T 2, with linear activation functions. Tan-sigmoidal activation function was used in hidden layer being the best activation function for modeling any mathematical expression [2]. Learning rate was selected to be 0.05 and the momentum factor was kept at 0.5. Number of epochs given for training was 15000 but network trained with only 5781 epochs as shown by performance cure of the ANN training in Figure 6. Each training session took a long time of about 30 minutes over P-IV 1.7 GHz computer as the allowable error was set to 1x10. Number of training sessions was carried out to adjust the weights and biases to minimize the error in T1, T2 and to reduce the resulting harmonic losses. The network was trained for different conditions i.e. for each sector data. Levenberg-Marquardt training algorithm was used as it provides the fast training response with respect to gradient method of training. Parameters of ANN model are given in Table 1. The weights and biases of trained network were taken for implementation with micro-controller.or different voltage requirements T_1 and T_2 are changed while Ts remains the same. To change the frequency rate of change of angle β is controlled. Time in which reference vector completes one revolution over the circle decides the output frequency of the inverter. Thus both voltage and frequency control are achieved by controlling T_1 , T_2 and rate of change of angle β .