Design Of A Novel Stochastic Nonlinear Operation Unit And Its Application In MIMO Signal Detection

Nonlinear function computation is widely used in many scientific and technological fields.Stochastic computing is a novel and efficient value representation and computation scheme for information and signal processing.The main challenges of stochastic computing-based nonlinear function implementation are poor generalization,low accuracy,and high latency.In this paper,we propose a multi-drive and multi-dimensional finite state machine(MM-FSM)to implement the main univariate nonlinear functions used in the fields of information and signal processing with low complexity and latency on a general-purpose platform.We provide a synthesis of the corresponding MM-FSM activation parameters and conditional parameters.To improve computational accuracy,we further propose an adaptive scaling algorithm for MM-FSM.The most remarkable feature of MM-FSM is that we can configure different types of nonlinear functions with the same MM-FSM structure.Therefore,MM-FSM can be used for a wide range of stochasticbased applications.Compared with the traditional stochastic computing and coordinate rotation digital computer(CORDIC)algorithm,the simulation results show that the proposed MM-FSM nonlinear function calculation unit has a significantly reduced complexity while ensuring the calculation accuracy.This paper first introduces the classical nonlinear computational structure,and then proposes two kinds of random sampling gates.We describe the structure,function,and advantages of these two gates compared to traditional logic truth tables.Then,we use these two kinds of logic gates to carry out the experiment of Bayesian inference of Rain model,verify the applicability of these two kinds of sampling gates,and give the error between these two kinds of logic gates and the theoretical value.We also give the structure and logic block diagram of the Rain model computing circuit.Secondly,we propose a multi-input and multi-dimensional finite state machine structure,with the input value of the structure as the independent variable of the function,and the output value as the dependent variable of the function.We added the ?-gate and CPT gate mentioned in Chapter 2 to convert the specific value into a random bit stream,which greatly reduces the complexity of our operation and reduces the operation cycle.The experimental results show that our MM-FSM nonlinear function operation module can calculate many simple one-dimensional nonlinear functions,such as EXP,SQRT,Log functions,etc.Our algorithm can make the whole system configurable The performance is enhanced,so that more nonlinear functions can be formulated.In addition,we propose a full-stochastic MCMC-MIMO detection operator based on MM-FSM based on the traditional MCMC-MIMO detection algorithm.Under the framework of the MCMC detection algorithm,we use the random sampling gate to convert the input probability value into a stochastic bit stream,then use the MM-FSM nonlinear operation module method to calculate the LLR,and update our LLR in the iterative process,and finally calculate The location of the constellation points.We use the proposed MM-FSM nonlinear calculation module to complete the calculation of the value of the nonlinear function in the LLR process,thereby approximating our LLR.The MCMCMIMO detection algorithm based on the MM-FSM computing unit designed by us greatly reduces the complexity on the basis of the traditional MCMC detection algorithm and the MMSE detection algorithm.