动态频谱接入中基于最小贝叶斯风险的稳健频谱预测

Minimum Bayesian Risk Based Robust Spectrum Prediction in Dynamic Spectrum Access

针对频谱感知错误累积造成频谱预测性能恶化问题,该文提出一种基于最小贝叶斯风险的稳健频谱预测策略。分布拟合检验表明频谱预测输出服从正态分布,定义频谱预测输出的贝叶斯风险函数,证明使贝叶斯风险函数最小的频谱预测输出判决门限将使频谱预测的均方误差最小,求得了使贝叶斯风险最小的最优判决门限,构建稳健频谱预测策略。仿真结果表明,与固定判决门限的神经网络频谱预测相比,稳健频谱预测策略改进了频谱感知错误下的频谱预测性能,改善了非授权用户的动态频谱接入性能。

The accumulation of miss detection and false alarm in spectrum sensing leads to the persistently decreasing of prediction accuracy in spectrum prediction. This paper takes neural network based spectrum prediction for example, and presents a minimum Bayesian Risk based spectrum prediction to solve this problem. The distribution fitting shows that the prediction output follows the normal distribution. The expectation of prediction mean square error is defined as the Bayesian Risk, and the optimal detection threshold of the prediction output is derived through minimizing the Bayesian Risk. Through this method, the prediction accuracy is insensitive to the spectrum sensing errors. Compared with the traditional spectrum prediction with fixed detection thresholds, simulation results demonstrate the robust spectrum prediction keeps the prediction accuracy stable, and improve the performance in dynamic spectrum access.