Abstract:In this paper, we investigate the recently introduced new learning technique-rrelevant vector machine (RVM) to construct a Multi-user detection (MUD) scheme through perfect sampling, which forms a new a posteriori parameter distribution estimation scheme based on Markov chain Monte Carlo (MCMC) algorithm. First set a prior on the parameter to be estimated, then have the parameter as a state variable. As the parameter in MUD is concerned, the number of w is prefixed, then make use of the given prior and perfect sampling to acquire a posteriori estimation of w. Once Bit error rate (BER) representing a kernel function output increases, a timely updating of mean and variance of prior distribution is implemented to acquire the 'needed sample'. Thus the evaluation of w forms an adaptive close loop process, which generates predictive distribution rather than makes point prediction, meanwhile the sparseness is acquired through a Jeffreys prior. Compared with other Bayesian detection scheme, this algorithm is especially efficient in solving high dimension and global optimization problems. Digital result shows that the detection performance is better than that of conventional Support vector machine (SVM) as well as Minimum mean square error (MMSE) scheme especially under massive user environment.

Online Training of SVM for Multiuser Detection

A Fast Online SVM Algorithm for Multi-user Detection

On-line support vector regression with multiple samples

SVM multiuser detection based on heuristic kernel

Study of Multiuser Detection: the Support Vector Machine Approach.

A novel blind multiuser detection algorithm based on support vector machines

Online background subtraction based on sufficient statistics

Improved Partial Parallel Decision Feedback Multiuser Detection for Multiuser-MIMO System

On-line Support Vector Machine Training Algorithm and Its Application

Real-time intrusion detection system based on one-class SVM

Multi-User Detection Based on Improved Cheetah Optimization Algorithm

Blind Multiuser Detection Based on Kernel Approximation

Probabilistic tangent subspace method for multiuser detection

Online Kernel-Based Structured Output SVM for Early Expression Detection

Pedestrian detection based on modified dynamic background using gaussian mixture models and hog-svm detection

The Study Of An Improved Relevant Vector Machine Mud Based On Perfect Sampling

Fast Pedestrian Detection And Tracking Based On Vibe Combined Hog-Svm Scheme

Multi-sphere Classification Algorithm Based on 1-SVM

An Uplink MC-CDMA Multi-user Detector Based on Incremental Support Vector Machine

Multiuser Detection Algorithm Based on Belief Propagation in Multiple-Input Multiple-Output Systems

Detection and Segmentation of Moving Objects Based on Support Vector Machine