CSI Prediction based on Past Sample Clauses

CSI Prediction based on Past and Present Channel Estimates In Xxxxxxxx X0-0 we present and evaluate OFDM channel prediction algorithms that enable JT CoMP to work reliably for terminals with high pedestrian velocities at 2.66 GHz, or for vehicular velocities at 700 MHz. We mainly discuss FDD downlink prediction, but the method is relevant also for TDD and for uplinks. For FDD systems, predictions of downlinks must be based on downlink reference signals. We use a Kalman state-space channel prediction algorithm, making extensive use of recent results from [Aro11] and [ASA12]. Kalman predictors provide two advantages: First, they provide optimal linear predictions of multiple channels that are based on measurements over an unlimited past time window, and a selectable frequency window. Second, at no extra cost they also provide the second order statistics of the prediction accuracy. This statistical information can then be included into a robust precoder design in different ways (see Subsection 5.2.3 and Appendix A2-2). The result is less intra-cluster interferences when robust precoders are applied on the true, unknown, channels. The Kalman predictor is based on a linear state space model. This model can be based on Auto-regressive (AR) models for each estimated channel component. The state space model represents the second order statistics of the variability over time and frequency of all OFDM channel components of interest. A vector of state estimates is updated every time a Reference Signal (RS) is received, using channel measurements at the RS time-frequency locations. From the state estimate, the predicted CSI can then be calculated for a given prediction horizon. Kalman prediction performance: Theoretical observations. 6 dB 12 dB 18 dB Figure 5.16: Normalized mean square prediction error (NMSE) of the complex channel gain, in dB, versus the prediction horizon λ expressed in carrier wavelengths, for non-line-of-sight propagation with a flat Doppler spectrum (upper) for an 4th order AR approximation of a Jakes/Clakes spectrum (lower left) and for a Doppler spectrum with one main peak (lower right). The curves correspond, in pairs, to SNR levels of 6, 12 and 18 dB, respectively. In each pair, the upper and lower curves show the NMSE performance for frequency selective and flat fading channels, respectively. (Figures 6.8, 6.9 and 6.10 in [Aro11].) The figures above are theoretically calculated stationary Kalman prediction accuracies from [Aro11]. They refer to the simultaneous estimati...