Approximate Sphere Decoding Based Model Predictive Control of Cascaded H-Bridge Inverters

The sphere decoding algorithm (SDA) offers a computationally efficient technique for direct model predictive control in cascaded H-Bridge (CHB) inverters. However, the variable computational complexity offered by the algorithm for each time step makes its real-time implementation quite challenging due to higher values of prediction horizon. We propose to overcome these limitations by using the partial graph processing approximate computing techniques. The proposed algorithm offers constant computational complexity for each time step by retaining K number of best candidates at each layer of the tree. The value of K is then used to decide the reduction in computational complexity and degree of approximation in the identified solution. The simulation results reveal that the proposed approximate SDA provides a sub-optimal solution with a significant improvement in energy and computational efficiency.