A Pareto Frontier based IIR Optimization for Industrial Reverberant Channel Emulation

The traditional channel emulation methods with FIR architecture are less capable of emulating the reverberation fading channel in industrial applications, which may require hundreds of taps to provide a satisfactory emulation accuracy. A possible solution is to employ an IIR filter to approximate the high-order FIR filter. However, classical FIR-to-IIR transformation methods can only provide optimized results when the filter order is treated as a fixed hyperparameter. As a result, researchers and engineers face the challenge of selecting an appropriate filter order through a trial-and-error process over a vast solution space. To ease this challenge, this paper proposes an agent-based Pareto frontier method to model this problem, which decreases the high-dimensional solution space into a one-dimensional curve, where either the best accuracy with given order or the minimized order with given accuracy can be guaranteed. The researchers and engineers can then simply search for a satisfactory solution on this frontier curve instead the whole solution space. To accelerate the construction of Pareto frontier, this paper further proposes agent function to approximate the error response, which can be simply obtained with a closed-form solution and satisfactory accuracy. The results of extensive experiments validate the efficiency and accuracy of the proposed method.