Time-domain Channel Prediction Algorithm For OFDM Systems

Recently, OFDM (orthogonal frequency division multiplexing) and adaptivetransmission have been identified as two key technologies to improve communications overa wireless channel. For wideband channels, OFDM systems have a superior ability tomitigate inter-symbol-interference (ISI) without the need for complex equalizationcompared with the single-carrier (SC) systems. Meanwhile, adaptive transmission is knownto provide a high throughput performance for OFDM systems, provided that the accuratechannel-state-information (CSI) can be used for the transmission parameters adjusting.Therefore, the combination of OFDM and adaptive transmission has been considered as apromising technology for the wireless communication systems. In adaptive transmission,accurate CSI for the upcoming transmission frame is needed. In fast variation channels, theestimated CSI will be soon outdated and channel prediction is the sole way to provide thefuture CSI. This thesis studies the channel prediction technology in OFDM systems. Themain contributions and novelties of this thesis are:1. The merits of the time-domain channel prediction method for OFDM systemsover frequency-domain method are analyzed, and a novel prediction method whichimproves the prediction performance utilizing the sparse characteristics of channel isproposed. By analyzing the structures of the time-domain and the frequency-domainchannels, it is found that the channel tap in the time-domain is more predictable than thesubcarrier in the frequency-domain, and the reason is that the structure of time-domainchannel is simpler than that of frequency domain channel. Considering the sparse featuresof realistic radio channel in the time-domain, an enhanced prediction method exploiting thesparse features is proposed to eliminate the noise and improve the prediction performance.The performance of the proposed method is compared with existing methods, and thetime-domain channel prediction technique is one of the most important OFDM channelprediction methods in the future.2. In the case that the channel changes abruptly in OFDM systems, it is approvedthat the remaining information in the prediction error of representative LMSpredictor can be utilized, and a new method utilizing the remaining information isproposed to improve the prediction performance. When the radio channels change abruptly (the abrupt change denotes the abrupt appearance or disappearance of scatteredcomponents), it usually takes a long time to converge to an accurate prediction fortraditional LMS (least-mean-square) prediction method. In the case that some scattercomponents appear abruptly, the evolution of the prediction coefficients as well as theprediction error of LMS prediction method are analyzed, and it is found that the remaininginformation in the prediction error can also be exploited to improve the predictionperformance. Thus, a new double prediction method which uses another LMS predictor toexploit the remaining information is proposed, and this method can speed the convergenceprocess and maintain the accuracy of prediction. In the case that some scatter componentsdisappear abruptly, it is analyzed that the effect on prediction accuracy can be neglected.The applicability of the new double LMS prediction method for smooth time-varyingchannel is also verified.3. Based on the discussed channel prediction methods, a novel coherent detectionmethod with no processing delay for delay-sensitive applications employing OFDMsystem is presented, and an enhanced solution which makes the detected data as pilotsis proposed. In data-aided detection, pilot symbols known to the receiver are transmittedperiodically to enable channel estimation at each pilot position. This method always suffersfrom a large processing delay, which thus restricts its application in delay-sensitive services.The proposed method periodically inserts the pilots at a low rate, and employs channelestimation followed by linear prediction and interpolation to obtain the CSI for symboldetection. A improved method, which takes the detected data as the ‘pilot’ to enhance theprediction accuracy, is also proposed. Compared with traditional prediction techniques, thenew method can compensate for the processing delay without losing much detectionaccuracy, and it is very appropriate for delay-sensitive services.