Fast Simulation of Ultra-Reliable Coded Communication System via Adaptive Shaping of Noise Histogram

To estimate the probability of an event, conventional Monte Carlo (MC) needs $100/P_{\mathrm {e}}$ simulation runs to attain a 10% precision, where $P_{\mathrm {e}}$ is the probability of the event. It therefore encounters difficulty in simulation-based evaluation of packet error rates for ultra-reliable communication under its stringent requirement. Many fast simulation techniques for evaluating the probability of rare events have been proposed. However, a more efficient method for coded communication systems that can adaptively exploit the code structure and concentrate the generated noise vectors to the error-prone regions is desirable. We propose a method which seeks to adaptively learn a certain optimal histogram of the noise vectors and generate the noise vectors accordingly. The said histogram is a one-dimensional function and hence is easy to work with. The adaptation mechanism is code-agnostic. Simulation with cyclic redundancy check-aided polar coding in additive white Gaussian noise shows an approximately 10-100 times speed-up compared to conventional MC.