| Mobile communication is undergoing booming prosperity, along with the soaring number of users and the diversification of cliental need, current mobile commumcation systems dominated by voice service can no longer satisfy the rapidly changing cliental expectation, and will finally be replaced by new multi-medium services such as the transmission of video, images, text and high rate date etc., which have strict requirement to network bandwidth. Future high-speed communication calls for a new transmission technology with high spectrum frequency, enough system capacity and higher transmission speed. Then the proposal of Code-Division Multiple Access (CDMA) obtained this objective.However, due to the no-ideal orthogonality between the signature codes and different transmission delays since time-varying character of wireless channel, which leads to Multi-access interference (MAI). MAI will further deteriorate along with growing number of enabled users. Traditional CDMA signal detection technologies named correlation detection, Traditional signal detection technologies involving address code correlation calculation based on the direct spread-spectrum theory, named correlation detection, which lacks the anti-interference capacity.Multi-user detection (MUD) is one of the efficient anti-inference technical solutions newly developed on the base of relation detection tchnology, it is a crucial technology in CDMA communication system. It take use of all available information of the multi-access interference to implement joint detection among aimed users, therefore suppresses the multi-access inference, efficiently utilizes spectrum resources, enhances substantially the system capacity and lowers the requirement to the power control. This paper studies this basic point in direct spread-spectrum code-division multi-access (DS-CDMA) and multi-carrier code-division multi-access (MC-CDMA) communication systems.Real-time Character is the key of the multi-user detection's practical performance. This paper thoroughly analyses the parameters which determine the degree of precision and the velocity of convergence in this algorithm, pursuing an optimized algorithm which enables real-time detection, and finally concludes that the blind adaptive multi-user detection algorithms applied to base stations and mobile terminals must be separately discussed according to their specific characters, making full use of available information.Based on all the known spreading sequences in subregions for users, it is theoretically feasible to eliminate multi-access inference using the statistical and structural information of these pseudorandom spreading sequences. After repeated researches and comparisons, this paper proposes three high-speed parallel Inference Cancellation intelligent algorithms:(1) Parallel Inference Cancellation (PIC) algorithm realized plausible inference counteraction and amelioration of system's performance with relatively low calculation complexity and short processing delay, and it is also one of relevant feasible practical technologies. However, inaccurate decision values from the former stage will result in false interference cancellations at the latter stage, which will affect sequent decision. This paper, throughout a serial of theoretical analysis and simulation, explored and revised the Hebb learning rule, and can be applied to the adjustment of the anti-inference factor, and further included considerations of the variations of channels and variety of inferences to individual client, then proposed the Partial Parallel Inference Counteraction (Hebe-PPIC) algorithm. This algorithm avoid false counteraction resulted from inaccurate judgment, decreased the bit error ratioand number of algorithm stages.(2) To further optimize the real-time character of anti-inference factor, the SIR-FPIC algorithm was proposed to apply the estimated SIR of the former stage to construct a membership function as the ICF of the latter stage, consequently realized real-time adjustment to anti-inference factor according to variation in tracked channels, improved the accuracy of judgments and further lowered BER.(3) Based on the high speed and powerful parallel processing function of neural networks and MMSE criteria, using Hopfield neural network's PIC structure and its inherent rapid decline character in its Energy function, this paper present a Hopfield-based parallel inference-counteracting multi-user detection algorithm (GMHNN), and proved this algorithm can converge to global minimum.In mobile terminals, besides the informations of object user, to obtain all spreading sequences of enabled users is impractical. For this reason, this paper also presents two blind multi-user detection algorithms which are adaptive to mobile terminals.(1) Based on the Principal Component Analysis, this paper replaced Eigenvalue Decomposition with neural network's learning mechanism, and presented a new blind adaptive multi-user detection algorithm (NPCA) based on bilinear neural network tracking signal subspace. In this algorithm, instead of calculations of high order covariance matrix, characteristic vectors and characteristic values were determined by the sampling value directly from input vectors, consequently decreased the complexity of calculation and enhanced the real-time character.(2) As for the too low velocity of convergence in Least Mean Squares (LMS) blind multi-user detection algorithm and the too high calculation complexity in RLS and Kalman algorithm, this paper proposed the List Mean Square algorithm (LMS) based on the a new changeable step length momentum factor. In this algorithm, a larger step length is adopted to speed up convergence at starting stage, then dynamically adjusts step smaller in neighborhood of optimum convergent point in order to lower maladjustment Error, consequently solved the conflict between varying convergence rate and steady-state error and suppressed multi-access inference.Theoretical analysis and Monte Carlo simulations confirmed that the proposed five algorithms improve the performance of detection.
|
PDF Full Text Request