Research On Receiver Design Under Non-ideal Information Transmission Scenarios

With the rapid evolution of wireless communication,the communication system becomes more and more complicated owning to the diversification of service scenarios,the considerably complexity of the network structures and the rapid growth of traffic volume.Under the influence of the practical engineering,there are various undesir-able phenomena during information transmission,e.g.,systems with strongly nonlinear distortion that comes mainly with the engineering of transceivers,the channel uncer-tainty for situations without a complete knowledge of the channel,high-dimensional input-output relationship in massive MIMO systems and so on.Thus information may no longer be transmitted over linear Gaussian channels.In the presence of such non-ideal issues,the nearest neighbor decoder,which is the optimal decoding rule for linear Gaussian channels,loses its optimality.It is of great significance to seek more effective receiver schemes for information transmission in non-ideal scenarios.Nearest neighbor decoding remains a convenient and robustness solution for gen-eral channels.The key message of this paper is that the performance of the nearest neighbor decoding can be improved by generalizing its decoding metric to incorpo-rate channel state dependent output processing function and codeword scaling function.The validity of generalized nearest neighbor decoding(GNND)in dealing with various non-ideal problems is verified by applying it to actual non-ideal scenarios.Further,we adopt the generalized nearest neighber decoding into the complex scenes without us-ing the statistical channel model,and propose a data-driven information transmission scheme.The main contributions of our work are summarized as follows.1)Recognizing that the generalized nearest neighber decoding is generally mis-matched with respect to the actual channel law,we use a known lower bound of the mismatched capacity,namely the generalized mutual information(GMI),as the per-formance metric.By maximizing the generalized mutual information,we establish the optimal generalized nearest neighbor decoding rule.We do not make any special as-sumption to the channel in soving the maximization problem.Thus the solution is hold for the general channels.However,considered the computational limitations in practical implementation process,we also examine several types of reduced complexity gener-alized nearest neighbor decoding rules by restricting the processing function and the scaling function in different ways.2)We apply the generalized nearest neighber decoding to several non-ideal scenar-ios,including channels with nonlinear effects,and fading channels with receiver chan-nel state information(CSI)or with pilot-assisted training.Two notable consequences are in order.First,compared with the conventional approach of decomposing the non-linear channel output into the linear superposition of a scaled channel input and an un-correlated residual term,the GNND using the nonlinear processing function can better align the channel input to match the transformed channel output,in such a way that the resulting GMI performance measure is maximized.Second,compared with the con-ventional approach where the channel state is first estimated and then treated as if it is perfect in decoding,it is beneficial for the receiver to directly estimate the channel input and perform the GNND rule.These shed new insights into the architecture of receiver design.3)In complex scenarios,the underlying physical mechanism of channel is not suffi-ciently understood by us to build a dependable channel model,or is known but yet too complicated to prefer a model-driven design.When the encoder and the decoder know the statistical channel model,it is found that the optimal channel output processing func-tion is the MMSE estimator,thus hinting a potential role of regression,a classical topic in machine learning,for this model.Without utilizing the statistical channel model,a problem formulation inspired by machine learning principles is established,with suit-able performance metrics introduced.A data-driven inference algorithm is proposed to solve the problem,and the effectiveness of the algorithm is validated via numerical experiments.