FDD Channel Estimation Via Covariance Estimation in Wideband Massive MIMO Systems

Abstract

A method for channel estimation in wideband massive Multiple-Input Multiple-Output (MIMO) systems using covariance identification is developed. The method is useful for Frequency-Division Duplex (FDD) at either sub-6GHz or millimeter wave (mmWave) frequency bands and takes into account the beam squint effect caused by the large bandwidth of the signals. The method relies on the slow time variation of the channel covariance matrix and allows for the utilization of very short training sequences thanks to the exploitation of the channel structure, both in the covariance matrix identification and the channel estimation stages. As a consequence of significantly reducing the training overhead, the proposed channel estimator has a computational complexity lower than other existing approaches.

Structured channel covariance estimation from limited samples for large antenna arrays

“…However, as pointed out above, several schemes for DL multiuser precoding/beamforming in FDD systems make use of the user channel covariance matrix in the DL, which differs from the UL covariance since the frequency separation between the UL and the DL bands is large. The estimation of the DL covariance from UL channel samples has been considered in several works and it is generally another challenging task [7,[21][22][23][24][25]. We shall see that our scheme is able to accurately estimate the DL channel covariance by extrapolating (over frequency) the estimated parametric model in the UL.…”

“…Massive multiple-input multiple-output (MIMO) communication system, where the number of base station (BS) antennas M is much larger than the number of single antenna users, has been shown to achieve high spectral efficiency in wireless cellular networks and to enjoy various system level benefit, such as energy efficiency, inter-cell interference reduction, and dramatic simplification of user scheduling (e.g., see [2,3]). In a large number of papers on the subject, the knowledge of the uplink (UL) and downlink (DL) channel covariance matrix, i.e., of the correlation structure of the channel antenna coefficients at the BS array, is assumed and used for a variety of purposes, such as minimum mean square error (MMSE) UL channel estimation and pilot decontamination [4][5][6], efficient DL multiuser precoding/beamforming design, especially in the frequency Yang et al J Wireless Com Network (2023) 2023: 24 division duplexing (FDD) case [6][7][8][9][10], and multiuser DL precoding design based on statistical channel state information (CSI) [11][12][13].…”

Structured Channel Covariance Estimation from Limited Samples for Large Antenna Arrays

“…However, as pointed out above, several schemes for DL multiuser precoding/beamforming in FDD systems make use of the user channel covariance matrix in the DL, which differs from the UL covariance since the frequency separation between the UL and the DL bands is large. The estimation of the DL covariance from UL channel samples has been considered in several works and it is generally another challenging task [7,[22][23][24][25][26]. We shall see that our scheme is able to accurately estimate the DL channel covariance by extrapolating (over frequency) the estimated parametric model in the UL.…”

“…as we will see that this algorithm yields a computationally efficient update rule and excellent empirical results for the task of estimating the parametric ASF coefficients. Note that although the likelihood function in (25) is in a general form for any family of dictionary functions, the EM method can be applied only in the case where all the matrices S i have rank 1, which is the case when the dictionary functions ψ i (ξ) are Dirac delta functions. In contrast, the more general concave-convex procedure (e.g., see [1] for the application in this case) can deal with any type of dictionary, but yields significantly higher computational complexity, so that it is not suited for large M (massive MIMO case).…”

Structured Channel Covariance Estimation from Limited Samples for Large Antenna Arrays

“…However, as pointed out above, several schemes for DL multiuser precoding/beamforming in FDD systems make use of the user channel covariance matrix in the DL, which differs from the UL covariance since the frequency separation between the UL and the DL bands is large. The estimation of the DL covariance from UL channel samples has been considered in several works and it is generally another challenging task [6,[13][14][15][16][17]. We shall see that our scheme is able to accurately estimate the DL channel covariance by extrapolating (over frequency) the estimated parametric model in the UL.…”