Technical Field

The present invention relates to the field of communication and particularly to the channel estimation methods in a multiple-input multiple-output (MIMO) Orthogonal Frequency Division Multiplexing (OFDM) communication system.

Background

In modern communication systems, achieving high data rates, low latency, high reliability and connectivity have become the main performance targets with the recent advances. Satisfying these requirements in a wide range of use-cases and deployment scenarios is a challenging task, and this makes the accurate channel knowledge at the receiver even more critical. Therefore, channel estimation is one of the most crucial blocks of a communication system and it can be performed based on pre-defined reference signals also known as pilots.

The global standardization bodies such as 3GPP conduct necessary studies to define modern communications standards such as Long Term Evolution (LTE), Long Term Evolution-Advanced (LTE-A) and lately 5G New Radio (NR). 5G waveform is based on Cyclic Prefix OFDM (CP- OFDM) (as in LTE and LTE-A) for both sub-6 GHz and millimetre-wave frequencies and supports single and multi-user MIMO as a key enabler to achieve high data rates. Both LTE and NR define reference signals for wide range of physical channels to enable channel estimation at the receiver. The reference signals which are defined for coherent demodulation of the control and user data in both uplink and downlink are known as demodulation reference signals (DMRS). The design of DMRS to allow multiple layers of data transmission through a MIMO system needs to consider many aspects concurrently. Some of the factors, which have been considered in DMRS design of 5G, include the pilot density, the power variation in frequency, number of layers with orthogonal pilot symbols, configuration flexibility, the location of pilots to support low-latency demodulation, and allowing a common receiver structure for different configurations.

Both 5G and LTE utilize code-division multiplexing (CDM) for orthogonal transmission of pilot signals in different MIMO layers. To support pilot transmission in more than one layer in the same resources, orthogonal cover codes (OCC) or cyclic shift (CS) operations are used to achieve orthogonality among pilots. For example, in 5G, every DMRS configuration (except single layer transmission) includes an OCC based pilot allocation in the frequency domain, and it is possible to define up to 12 orthogonal layers by combining CDM and frequency domain multiplexing (FDM). Usage of CDM based pilots also have an inherent advantage over other orthogonal methods such as time or frequency domain multiplexing due to the processing gain. It is noted that Channel State Information Reference Signal (CSI-RS) and Sounding Reference Signal (SRS) defined in NR also utilize CDM structure to extend single port operation to multiple ports. Therefore, CDM based pilot allocations are widely used in the state-of-the-art communications systems.

Even though CDM based designs have certain advantages and attractive properties, they rely on the assumption that the channel does not change over the resource elements where the CDM is defined. For example, when the CDM group location consists of resource elements in frequency, and the channel is frequency selective or there is time synchronization error between transmitter and receiver, then the orthogonality in frequency domain is lost at the receiver. Similarly, when the CDM group location consists of resource elements in time, and when the channel is fastfading or there is a frequency synchronization error between transmitter and receiver, then the orthogonality in time domain is lost at the receiver. This is detrimental for channel estimation performance especially for high spectral efficiency scenarios, as the performance is limited by the channel estimation error as a result of inter-layer interference at pilot symbols. Therefore, it is important to evaluate and deal with such cases to achieve high data rate targets in wide range of channel scenarios.

It is essential for the channel estimation algorithm in a MIMO-OFDM system to correctly recover CDM pilots at the receiver and interpolate estimated channel values at the pilot locations to cover all the resource grid. Recovering the channel at pilots of each MIMO layer belonging to the same CDM group from the received signal and obtaining an initial estimation at pilot locations is referred as pilot depatterning operation. Pilot depatterning and interpolation of the pilots can be performed jointly in an optimal MMSE estimator. However, such an optimal depatterning is not very practical due to excessively large complexity. In particular, it requires inversion of a large matrix, which consists of the pilots values at every different pilot occasion in real time. Another problem with optimal depatterning is that when the DMRS pilots are used for multiuser interference measurement, and the users utilize different resource grid sizes to be interpolated, the performance will be degraded due to mismatch among users. Hence, a lower complexity channel estimation method is implemented in realistic receivers.

Conventional channel estimation procedure for MIMO systems involves two separate stages: a pilot depatterning stage and an MMSE estimation stage which interpolates the estimated channel values in pilot locations after pilot depatterning. The pilot depatterning is performed based on least-squares (LS) method with the assumption that radio channel stays flat in the pilot depatterning occasions, and it has linear complexity. The interpolation is performed using the estimated pilot values after depatterning operation. This method has much lower complexity compared to optimal depatterning. The optimal depatterning method reduces to conventional method, if the channel is exactly same in the CDM resources. However, due to multipath and mobility, the wireless channel has a certain delay and Doppler spread, which might cause small or large changes in the channel in time and/or frequency. Also, when there are non-idealities such as time synchronization error or carrier frequency offset (CFO) error, the channel can not stay flat, which causes serious performance loss in channel estimation for conventional receivers. Especially, considering the fact that modern communication systems need to support for high date rate and spectral efficiency via MIMO technology, the channel estimation errors can limit the performance of such systems in previously said conditions.

In a MIMO-OFDM system, the received signal at the fcth subcarrier of nth OFDM symbol at the mth receive antenna, can be expressed as: for m - 1, ..., N _R , n — 1, ...,N _slot , k - 1, K, i — 1, ...,N _L where H^ ^l) is the channel coefficient observed at the fcth subcarrier of the nth OFDM symbol between ith layer and znth receive antenna, X^ _k is the complex pilot symbol carried at the fcth subcarrier of the nth OFDM symbol at the ith layer and is the complex Gaussian noise component effective at fcth subcarrier of the nth OFDM symbol at the mth receive antenna. Note that N _L is the number of data layers, N _R is the number of antennas at the receiver, N _siot is the total number of OFDM symbols in the channel estimation window/slot and K is the total number of OFDM subcarriers in the transmission band. It is noted that N _slot K resource elements define a resource grid for a given layer. In the channel estimation problem, the aim is to obtain an estimate of the effective channel coefficient given the pilot (reference signal) symbols and the received signal at the location of pilot symbols.

In practical systems, the pilots should be allocated orthogonally between N _L MIMO layers to avoid inter-layer interference, which can considerably degrade the overall performance. There are different multiplexing options for pilot allocation to satisfy this goal. In Code Domain Multiplexing (CDM), the pilot symbols in different layers belonging to the same CDM group use the same resource elements, i.e. (n,fc) in equation (1 ), while the separation is achieved via various codes such as orthogonal cover codes (OCC). In Time Domain Multiplexing (TDM), the pilot symbols in different layers are transmitted in different OFDM symbols in a slot. In Frequency Domain Multiplexing (FDM), the pilot symbols in different layers are transmitted in different subcarriers within an OFDM symbol. One or more of these schemes can be utilized to ensure orthogonality of pilot symbols. In CDM, a certain list of MIMO layers shares the same resource elements in time and frequency, therefore belong to the same CDM group. Because of that, they have to be orthogonalized in the code domain. Each CDM group can be characterized by its base location indices set in terms of subcarrier and OFDM symbol indices. For example, CDM group j can be specified with base sets (Kj,Lj , which means that all layers in CDM group j have pilot signals located in the subcarrier indices generated according to the set ₇ and OFDM symbol indices generated according to the set Lj. Also, each CDM group can support at most |C ₇ | = |% ₇ ||£ ₇ | number of orthogonal layers, where Cj is the set of indices of layers belonging to the CDM group j and 1. 1 denotes the cardinality of the set.

Different CDM groups should be transmitted in different resource grid locations, and they should not overlap. Therefore, no symbol should be transmitted in other layers in the resource elements corresponding to the those of given CDM group. For example, given CDM group i and another CDM group j with ( ₇ ,£ ₇ ), then and i Cj. Different CDM groups can be multiplexed using FDM and/or TDM. Let F and T denote the number of CDM groups separated via FDM and TDM, respectively; then J = FT and, where J is the number of different CDM groups in the resource grid.

In practical systems, the pilot allocation pattern specified with CDM base sets, i.e. (Kj,Lj) for CDM group j, needs to be repeated regularly in frequency and time to increase pilot density. This is because of the fact that the sufficiently dense pilot allocation is required to capture the channel effects and changes in time and frequency domain in the given grid. This means that the pilots of the CDM group j are not only located in the pilot locations given in base sets ( ₇ ,£ ₇ ), but they are also located in the repeated occasions generated according the base sets. One example of pilot allocation is given in Figure 1 . In this example, ₇ = {1,2} and Lj = {3,4}, and the resource grid contains K = 120 subcarriers and N _siot = 14 OFDM symbols. Also, for CDM group j, the pilots are repeated at every 6 subcarriers in frequency for entire grid and repeated once in the time domain such that there are 6 OFDM symbols between the repeated pilot symbols. Hence, the pilot symbols of CDM group j are located in subcarrier indices, i.e., {1 , 2, 7, 8, ..., 109, 1 10, 1 15, 1 16} and at OFDM symbol indices given as {3, 4, 1 1 , 12}. However, while applying the orthogonal cover code (length-2 in frequency and length-2 in time), each repeated pilot occasion is treated separately, which implies that pilot depatterning is applied separately to the repeated pilot occasions. In the present disclosure, this is referred as pilot depatterning group. For example, the pilots located at the subcarriers {1 ,2} of the OFDM symbols {3, 4} form a pilot depatterning group, and the pilots located at the subcarriers {7,8} of the OFDM symbols {3,4} form another pilot depatterning group. First, the conventional channel estimation in MIMO-OFDM systems involving CDM pilot allocation in the prior art is described. In all channel estimation methods, the common goal is to obtain the estimates n = l, ...,N _slot , k = l, ...,K, 1 = 1, ...,N _L using the received symbols at pilot locations and pilot values according to the system model given in equation (1 ) in this embodiment. For a given layer i and receive antenna m, can be estimated separately for each (m, i) pair using the same procedures in a practical communications system. The main operations in a conventional channel estimation procedure are illustrated in Figure 2. In the first step, the signal is received in a receive antenna, e.g. receive antenna m, and OFDM demodulation is performed to obtain received symbols in the resource grid of receive antenna m. In the second step, the conventional pilot depatterning is performed at all pilot depatterning groups using the corresponding OCC and received symbols at the resource grids of all layers in a CDM group, e.g. CDM group j. As a result of this step, for a given specific pilot depatterning group, a single channel estimation value is assigned as the estimate for all locations in the group; however, channel estimations at different pilot depatterning groups are most likely to be different. Pilot depatterning operation needs to be performed for all CDM groups, i.e.j = 1, ... ,J to complete this step. In the third step, the estimated values in pilot locations are used to obtain the channel estimation for all resource elements in the resource grid using an interpolation method. Interpolation procedure can be performed using following different exemplary methods: MMSE estimation in both time and frequency (2D-MMSE), MMSE estimation first in time and then in frequency or first in frequency then in time (MMSE 1 D-1 D), linear interpolation, nearest point interpolation, sliding window averaging. In the final step, all operations given in the first three steps of the procedure are performed for all remaining receiver antenna signals, i.e. m = 1, ... , N _R .

In one example, the conventional pilot depatterning operation in the second step of conventional channel estimation procedure given in Figure 2 is based on Least-Squares (LS) with the assumption that radio channel stays flat in the pilot depatterning occasions. An example of this procedure is provided. For this purpose, how OCC is applied in a pilot depatterning group at the transmitter, and how the estimation can be obtained for each layer by depatterning operation are explained. Suppose that a pilot depatterning group in the resource grid of a layer belonging to the CDM group 1 is located at the subcarriers {1 ,2} of the OFDM symbols {1 , 2}. This means that there are four pilots in the pilot depatterning group. The layer indices in the CDM group 1 are given as = {1,2, 3, 4} in this example, and all layers in have the same locations for all pilots in their respective resource grids. When length-2 OCC in frequency and length-2 OCC in time are applied to achieve orthogonalization, the transmitted pilot symbols can be written as: where X ⁽ⁱ⁾ = with each X^ _k is as defined in the equation (1 ) and a _{11 :} a ₁₂ ,a ₂₁ , “22 ^are complex-valued pilot symbols with unit amplitude. For mth receive antenna, the received and OFDM demodulated signals at the considered pilot depatterning group are Y^, Y^\ Y^ and

Then, based on equations (1 ) and (2),

The orthogonality of the pilots using CDM at the receiver is based on flat channel assumption, i.e. for any (m, i) pair. Under this assumption, and based on equation

(2) the following equation is obtained - h _L for brevity) form, this can be expressed as Bh + w = y _p . The least squares (LS) solution for estimate h =

According to equation (4), the elements of h becomes:

Herein, is the common channel estimate value assigned to the pilot locations in the considered pilot depatterning group between ith layer in the CDM group 1 and receiver antenna m, that is h-i= H^' ¹ ^ = = H^’ ^l \ This example procedure is repeated for all pilot depatterning groups in the resource grids of all layers in all CDM group for a given receive antenna signal. As a result of the operations of conventional pilot depatterning step, initial estimates at all pilot locations between ith layer and mth receive antenna are obtained and they can be stored in a vector,

In the conventional pilot depatterning procedure, the main assumption is that the radio channel stays flat in a pilot depatterning group. If this assumption fails, then the orthogonality of the pilots at the receiver is lost. Such an assumption also needs to hold for all layers due to cross terms as exemplified in the equation (5). For example, can be expressed in terms of actual channel values based on equations (3) and (5) as: which implies that even if the flatness assumption holds in the pilot depatterning group for layer 1 in the CDM group, the estimate will be affected by the channel values in layers {2,3,4}, introducing extra interference for the estimate. This implies that the estimation performance can be improved if the terms due to other layers are cancelled. For example, the interference term coming from layer 2 is given it j _s noted that is the difference in the channel at the first two subcarriers of the first time symbol, and is the difference in the channel at the first two subcarriers of the second time symbol.

In practical systems, the channel flatness assumption rarely holds perfectly due to effects such as large delay spread in the wireless environment, Doppler spread due to mobility, carrier frequency offset (CFO) and possible time synchronization errors. Such a loss in orthogonality can result in serious performance loss especially in transmission scenarios requiring high spectral efficiency to ensure high data rates and throughput (via large number of MIMO layers, higher modulation order and code rates). Also, if the channels of other layers in the CDM group are not flat, then the performance is degraded, even if the channel is entirely flat for the layer whose channel is to be estimated

The patent numbered US101 16478B2 is related with scattered pilot pattern and channel estimation method for MIMO-OFDM systems. The method and an apparatus are provided for reducing the number of pilot symbols within a MIMO-OFDM communication system, and for improving channel estimation within such a system. However, this document does not disclose a channel estimation method for MIMO-OFDM systems based on iterative pilot depatterning which filters the estimated channel values at the pilot locations to obtain the channel changes in the pilot depatterning group and then cancels the inter-layer interference iteratively to achieve performance improvement. For MIMO systems using CDM, no known method with near-optimal performance with practical complexity by compensating the loss due to the change of channel at the pilot locations has been observed. Therefore, there is a need for a computer implemented method and a data processing apparatus of channel estimation with practical computational complexity but have improved and near-optimal performance in MIMO-OFDM systems utilizing CDM in pilot allocations.

Summary of the Invention

The present invention relates to a channel estimation method and apparatus for MIMO-OFDM communications systems with an improved performance compared to conventional receivers. In accordance with a particular embodiment of the present invention, a computer implemented method and a data processing apparatus for channel estimation in MIMO-OFDM systems utilizing CDM in pilot allocations is provided, which is based on cancelling the inter-layer interference in pilot depatterning procedure in an iterative manner by using the filtered channel estimates which capture the changes in the channel at the pilot locations. The disclosed invention has a practical computational complexity with near-optimal performance and can be used in any MIMO-OFDM system utilizing CDM in pilot allocations.

In accordance with an embodiment of the present invention, a computer implemented channel estimation method is provided for MIMO-OFDM communications systems utilizing CDM in pilot allocations. The method is based on iterative pilot depatterning which cancels the inter-layer interference effectively by using the channel estimates of other layers, wherein the estimates are obtained via filtering operation to capture the channel changes at the pilot locations accurately. The received signal is OFDM demodulated to obtain received symbols in the resource grid and an initial pilot depatterning is performed to the received symbols. The iterative pilot depatterning is applied. At each iteration, the channel estimates at pilot locations are filtered, and pilot depatterning with inter-layer interference cancellation is performed using the received signal and filtered channel estimates at the pilot locations of other layers which lead to interference. The channel values at pilot locations are updated and if the predetermined number of iterations is reached iterative process is terminated, if it is not reached, then iteration count is increased and the next iteration starts. The estimated values in pilot locations as a result of iterative pilot depatterning are interpolated to obtain the channel estimation values for all resource elements.

In accordance with an embodiment of the present invention, a data processing apparatus for channel estimation is provided for MIMO-OFDM communications systems utilizing CDM in pilot allocations. The apparatus includes a pre-processing module for receiving and performing OFDM demodulation to obtain received symbols in the resource grid and performing an initial pilot depatterning to the received symbols; an iterative pilot depatterning performer module for filtering the channel estimates at the pilot locations, performing pilot depatterning with inter-layer interference cancellation using the received signal and filtered channel estimates, updating the estimated channel values at all pilot locations, and determining if the pre-determined number of iterations is reached and if it is, then terminating the iterative process; if it is not, then increasing the iteration count and starting the next iteration; and a channel estimator module interpolating the estimated values in pilot locations to obtain the channel estimation values for all resource elements.

Brief Description of the Figures

Figure 1 illustrates an exemplary pilot allocation in a resource grid of a layer belonging to a CDM group.

Figure 2 illustrates the flowchart of operations for conventional channel estimation in a MIMO- OFDM system involving CDM pilot allocation.

Figure 3 illustrates the flowchart of the procedures for the disclosed invention for channel estimation in MIMO-OFDM systems based on iterative pilot depatterning.

Figure 4 illustrates the pilot allocations for two exemplary 5G Physical Downlink Shared Channel (PDSCH) Demodulation Reference Signals (DMRS) configurations.

Figure 5 illustrates the mean-squared error (MSE) versus signal-to-noise ratio (SNR) performances for DMRS Type 1 and Type 2 configurations on Tapped Delay Line-C (TDL-C) channel with 1 ps delay spread for disclosed iterative method, conventional method and optimal method.

Figure 6 illustrates the MSE versus channel delay spread performances on TDL-C channel for DMRS Type 1 and Type 2 configurations when SNR is 15 dB for disclosed iterative method, conventional method and optimal method.

Figure 7 illustrates block error rate (BLER) versus SNR performances for 64-QAM with code rate 3/4 for DMRS Type 1 and Type 2 on TDL-C channel with delay spread 1 ps for disclosed iterative method, conventional method and optimal method.

Figure 8 illustrates the MSE versus SNR performances for DMRS Type 1 and Type 2 configurations on TDL-C channel with 100ns delay spread for disclosed iterative method, conventional method and optimal method.

Figure 9 illustrates the MSE versus SNR performances for DMRS Type 1 and Type 2 configurations on TDL-C channel with 300ns delay spread for disclosed iterative method, conventional method and optimal method.

Figure 10 illustrates the MSE versus SNR performances for DMRS Type 1 and Type 2 configurations on TDL-C channel with 700ns delay spread for disclosed iterative method, conventional method and optimal method.

Figure 1 1 illustrates the MSE versus channel delay spread performances on TDL-C channel for DMRS Type 1 and Type 2 configurations when SNR is 25 dB for disclosed iterative method, conventional method and optimal method. Figure 12 illustrates the BLER versus SNR performances for 256-QAM with rate 2/3 for DMRS Type 1 and Type 2 on TDL-C channel with delay spread 1 ps spread for disclosed iterative method, conventional method and optimal method.

Reference List

100. Computer implemented method

101. Receiving and performing OFDM demodulation to each receive antenna signal

102. Performing an initial pilot depatterning in all pilot depatterning groups for all CDM groups

103. Performing iterative pilot depatterning

1031. Filtering the channel estimates at the pilot locations at each iteration of iterative pilot depatterning operation

1032. Performing pilot depatterning with inter-layer interference cancellation at each iteration of iterative pilot depatterning operation

1033. determining if the pre-determined number of iterations is reached or not at each iteration of iterative pilot depatterning operation

104. interpolating the estimated values in pilot locations

Detailed Description

Hereinafter, the detailed descriptions of the embodiments of the present invention will be given with accompanying drawings. The present invention relates to a computer implemented channel estimation method and a data processing device configured to implement this method for MIMO- OFDM communications systems with near-optimal performance and low-complexity. The disclosed computer implemented method is based on the idea that the inter-layer interference in pilot depatterning procedure can be cancelled in an iterative manner very effectively by using the channel estimates of other layers obtained in previous iteration, wherein the estimates are obtained via filtering operation to capture the channel changes at the pilot locations accurately. The most important observation, which lays the foundations behind the innovative steps of the present disclosure is that in order to cancel the inter-layer interference effectively, the differences in the channel at the pilot locations need to be estimated and learned accurately. This implies that the estimating the difference in the channel values is more important than the actual channel values themselves for the interference cancellation.

The disclosed computer implemented method (100) which enables the channel estimation for multiple input multiple output (MIMO) orthogonal frequency division multiplexing (OFDM) systems comprises the steps of

• receiving and performing OFDM demodulation to each receive antenna signal to obtain received symbols in the resource grid of each receive antenna (101 ), • performing an initial pilot depatterning in all pilot depatterning groups using the corresponding OCC code and received symbols for all CDM groups and each receive antenna signal (102),

• performing iterative pilot depatterning at all pilot depatterning groups (103), wherein each iterative step comprises: o filtering the channel estimates at the pilot locations to obtain channel changes at the pilot depatterning groups in the resource grid of all layers for each receive antenna signal (1031 ), o performing pilot depatterning with inter-layer interference cancellation using the received signal and filtered channel estimates at the pilot locations of other layers causing interference in the CDM group for each receive antenna signal (1032), o determining if the pre-determined number of iterations is reached (1033); if it is reached, then terminating the iterative process, and if it is not reached, then increasing the iteration count and starting the next iteration of iterative process,

• interpolating the estimated values in pilot locations after iterative pilot depatterning (103) to obtain the channel estimation values for all resource elements at each layer for each receive antenna signal (104).

In the present invention (100), as a first step, the signal is received and OFDM demodulation is performed (101 ) to obtain received symbols in the resource grid of each receive antenna m, for m = AS a second step, an initial pilot depatterning is performed (102) at all pilot depatterning groups using the corresponding OCC and received symbols for all CDM groups and each receive antenna signal, i.e. CDM group j, for j = 1, and receive antenna m, for m = 1, ...,N _R . In one example, the initial pilot depatterning operation (102) is based on Least-Squares (LS) estimation method with the assumption that radio channel stays flat in the pilot depatterning occasions. An exemplary procedure for LS-based method is already provided for the second step of conventional channel estimation procedure above.

In the present invention (100), as a third step, iterative pilot depatterning operation (103) is performed. Herein, each iteration starts with filtering the channel estimates at the pilot locations in the resource grid of all layers for each receive antenna signal (1031 ). After that, for each layer belonging to a CDM group, pilot depatterning with interlayer interference cancellation (1032) is performed, using the received signal and filtered channel estimates at the pilot locations of other layers causing interference in the CDM group for each receive antenna signal. As a result of this step (1032), the estimated channel values at pilot locations are updated. Finally, it is checked if the pre-determined number of iterations, denoted by Qmax, is reached (1033) or not, and if it is reached, then the iterative pilot depatterning operation is terminated, finishing the iterative pilot depatterning step (103) of the present invention (100). If the pre-determined number of iterations is not reached yet, then the iteration count is increased by one and the next iteration starts again with filtering the estimates at the pilot locations (1031 ), which are obtained as a result of the pilot depatterning with interference cancellation operation (103).

In the first iteration of the initial pilot depatterning step (103) of the present invention (100), the filtering operation (1031 ) is applied to the estimated channel values at pilot locations that are obtained as a result of initial pilot depatterning (102). As noted previously, channel estimate after initial pilot depatterning (102) in a pilot depatterning group is a common value assigned to all pilot locations in the considered pilot depatterning group for a given layer. This implies that the initial pilot depatterning (102) assigns a single estimate for all pilot locations in the pilot depatterning group. However, in the iterative pilot depatterning step (103) of the present invention (100), it is essential to capture and estimate the changes in the channel in the pilot depatterning group for each layer. This is due to the fact that the interference in the channel estimates after initial pilot depatterning operation (102) consists of terms that are differences in the channel in pilot locations. In order to cancel the interference effectively, learning how the channel changes is required. By using a filter based on the channel correlations in the pilot locations, the changes in the channel in a pilot depatterning group can be estimated. Without the filtering step, the estimated channel values would be same for all pilot locations, and such an information is not useful for the interference cancellation. That is, interference cancellation based on the output of initial pilot depatterning (102) without any filtering operation (1031 ) to obtain channel changes would not improve the performance of conventional channel estimation, and even worse, it would actually increase the noise power in the estimates. Another advantage of filtering operation (1031 ) is to improve the quality of channel estimates at pilot locations after initial pilot depatterning (102) operation by noise reduction achieved with filtering. At the second iteration and after the second iteration of the iterative pilot depatterning step (103) of the present invention (100), the filtering operation (1031 ) is applied to the estimated channel values at pilot locations that are obtained as a result of pilot depatterning with interlayer interference cancellation (1032). In this case, even though each pilot location has already a unique estimate for a given layer, the filtering operation (1031 ) is still useful as it improves the quality of the estimates. It is also noted that the filtering operation (1031 ) considers only the estimates at the pilot locations; that is, it is applied to the channel estimates at the pilot locations and only produces new estimates for the pilot locations. Therefore, it is different than interpolation operation (104), which produces estimates for all data and pilot locations.

The filter used at the iterative pilot depatterning operation (103) while filtering the channel estimates at the pilot locations (1031 ) needs to be able to estimate the channel changes at the pilot locations accurately. MMSE based filters are good candidates for this step (1031 ), as they use the second order statistics such as channel correlation between the pilot locations to obtain the estimates. In one example, the filter used at the iterative pilot depatterning operation (1031 ) is based on MMSE estimation in both time and frequency (2D-MMSE). In another example, the filter used at the iterative pilot depatterning operation (1031 ) is based on MMSE estimation in time dimension first, and then in frequency dimension. In another example, the filter used at the iterative pilot depatterning operation (1031 ) is based on MMSE estimation in frequency dimension first, and then in time dimension. In another example, the filter used at the iterative pilot depatterning operation (1031 ) is based on MMSE estimation in frequency dimension first, and then one of the following methods in time dimension: linear interpolation, nearest point interpolation, sliding window averaging. In yet another example, the filter used at the iterative pilot depatterning operation (1031 ) is based on MMSE estimation in time dimension first, and then one of the following methods in frequency dimension: linear interpolation, nearest point interpolation, sliding window averaging.

In one example, 2D-MMSE based filtering used at the iterative pilot depatterning operation (103) while filtering the channel estimates at the pilot locations (1031 ) is using the following expression: wherein denotes filtered channel estimation values at pilot locations of the resource grid between ith layer and mth receive antenna. For all iterations of iterative pilot depatterning operation (103) except the first one, h ™' ¹ - ¹ is the estimated channel values at pilot locations between ith layer and mth receive antenna obtained as a result of pilot depatterning with interlayer interference cancellation (1032) at the previous iteration. For the first iteration, h.^ is the estimated channel values at pilot locations between ith layer and mth receive antenna obtained as a result of initial pilot depatterning (102). R _h ^l p _hV in equation (6) is the correlation matrix for wireless channel among pilot locations for ith layer, and a ² is the noise variance. This operation given in equation (6) is repeated for each receive antenna signal m = 1, ...,N _R and all layers i = 1,2, by using corresponding correlations matrices for each layer. It is noted that the subscript (or superscript) p appearing in the vectors in (6) indicates that those vectors carry values only for pilot locations and not all the resource grid.

In one example, the correlation matrix R _h ^l _VhV given in equations (6) can be calculated based on robust channel estimation method introduced in Robust MMSE channel estimation in OFDM systems with practical timing synchronization, by Vineet Srivastava et al., IEEE Wireless Commun. and Networking Conf., Atlanta, GA, USA, 2004, the entire contents of which are incorporated herein in its entirety. It is noted that the layer index i is dropped from the variables for the brevity of the description in the following. To calculate the correlation matrix, the correlations between resource elements in time and in frequency are calculated independently, and they are merged through a Kronecker product. The correlation matrix in frequency domain is denoted by R^ _hP , and it is denoted by R^ _hP in time domain. First, define the matrix R^ , which can be calculated for a given number of channel taps L by wherein Nff _t is the FFT size used in OFDM modulation and R^(a,b') is the channel correlation value between ath and 6th subcarriers in the resource grid, and it is the element of the matrix R^ at ath row and Z?th column. Next, R^ _hP can be obtained from R^ by taking only the rows and columns corresponding to the pilot locations, i.e subcarriers carrying a pilot value. Similarly, define the matrix R^, which utilizes the Jakes model by wherein D is the maximum Doppler spread, T is duration of the OFDM symbol, R^ (a, 6) is the channel correlation value between ath and 6th OFDM symbols in the resource grid, and it is the element of the matrix R at ath row and 6th column. Also, / ₀ C) is zeroth order Bessel function of the first kind. R^ _hP can be obtained from R by taking only the rows and columns corresponding to the pilot locations in time domain, i.e OFDM symbols carrying a pilot value. Then, the combined correlation matrix can be calculated as R _hPhP - R^lv® R^lv, wherein ® indicates the Kronecker product of the matrices. It is noted that, to be able to use the equations in (6)-(8), the channel values in the resource grid h or in the pilot locations in h ^p are ordered as first in frequency then in time. This means that, for a given layer, where h is a column vector carrying the channel values at the subcarriers of the jth OFDM symbol in the slot, for j - 1,2, ...,N _slot and h ^p is a column vector carrying the channel values at the subcarriers of the jth pilot OFDM symbol in the slot, for j = 1,2, ...,N _P , wherein ^denotes the total number of pilot OFDM symbols in the slot. It is also noted that R _hPhP matrix is common for each layer in the same CDM group.

In one example, pilot depatterning with interlayer interference cancellation (1032) procedure applied at the iterative pilot depatterning step (103) of the present invention (100) involves the following steps. After obtaining the filtered channel estimates at pilot locations for all layers, then for a given layer, inter-layer interference terms are subtracted from the received signal to obtain an intermediate signal, then an estimate is obtained at the pilot location by dividing intermediate signal to the pilot symbol sent at the corresponding pilot location. For instance, an example pilot allocation scenario is considered, which is also used as an example to the pilot depatterning operation in the second step of conventional estimation previously, wherein the received and demodulated signals at the considered pilot depatterning group satisfy the equations given in (2). According to that, for mth receive antenna signal and the pilot located at the first subcarrier of the first symbol, the following equation holds

If the channel estimates at the pilot locations after the latest filtering operation (1031 ) in an iteration are denoted by for the four layers, then, in order to find the channel estimate for the first layer, inter-layer interference terms are subtracted from the received signal to obtain an intermediate signal,

Based on (10), the channel estimate for the first layer becomes:

For other layers, the channel estimates at the first subcarrier of the first symbol can be expressed similarly as:

The channel estimates at the other pilot locations in the considered pilot depatterning group can be obtained by following similar steps. If the channel estimates at the pilots located at the second subcarrier of the first symbol after the latest filtering operation (1031 ) in an iteration are denoted by for the four layers, then the channel estimates can be expressed as:

If the channel estimates at the pilots located at the first subcarrier of the second symbol after the latest filtering operation (1031 ) in an iteration are denoted for the four layers, then the channel estimates can be expressed as:

If the channel estimates at the pilots located at the second subcarrier of the second symbol after the latest filtering operation (1031 ) in an iteration are denoted H^' ² \ H^’ ³ \ H^' ⁴ ^ for the four layers, then the channel estimates can be expressed as:

The example for pilot depatterning with interlayer interference cancellation (1032) procedure given above needs to be applied for pilot depatterning groups in the layers at all CDM groups. Updated estimates at pilot locations between ith layer and mth receive antenna are stored in the vector h p and its previous value is overwritten.

The present disclosure has improved and near-optimal performance for channel estimaton in MIMO-OFDM systems as it will be shown in the numerical examples. This is a result of applying filtering to the channel estimates to obtain channel changes at pilot locations in a pilot depatterning group, and effectively canceling the interference terms by using filtered estimates, wherein such a “filter and cancel” approach does not exist in the prior art. The performance improvement is based on the fact that the interference terms include the channel changes in pilot locations at other layers. Failure to performing of these steps (not obtaining the channel changes or not cancelling the interference terms) or performing them without the specific order as indicated in the disclosure will prevent obtaining the performance benefits of the present disclosure.

In one example, iterative pilot depatterning step (103) of the present invention (100) is applied to a pre-determined subset of pilot locations in the resource grid. In another example, said predetermined subset of pilot locations can be specified as all pilot locations except the ones which are located at the first k^ ^d0e and last k^ ⁹⁶ pilots of the transmission band, wherein k^ ^d0e and k^ ^d s ^e _{are non-ne} g _a tive integers. In this case, the pilots at the edge of the transmission band specified by k ^d0e and k^ ^dae are not included in the iterative pilot depatterning operations (103), and for those pilots, the estimated values after initial pilot depatterning (102) are kept without any updates through iterative pilot depatterning (103) until all iterations are completed. In the present invention (100), as a fourth step, the estimated values at pilot locations after iterative pilot depatterning (103) operation are interpolated (104) to obtain the channel estimation for all resource elements in the resource grid at each layer for all receive antennas. Interpolation procedure (104) can be performed using following exemplary methods: MMSE estimation in both time and frequency (2D-MMSE), MMSE estimation first in time and then in frequency or first in frequency then in time (MMSE 1 D-1 D), linear interpolation, nearest point interpolation, sliding window averaging.

In one example, at the step of interpolation (104) of the estimated values at pilot locations after iterative pilot depatterning (103), 2D-MMSE interpolation is implemented using the following expression: wherein h ^2 _e stores the final channel estimation values at entire resource grid between ith layer and mth receive antenna, h is the outcome of the iterative pilot depatterning (103) step of the present invention (100) and carries the estimated channel values at pilot locations between ith layer and mth receive antenna. in equation (16) is the correlation matrix for wireless channel between at all resource grid positions and at pilot locations for ith layer. R _h ^l p _h p in equation (16) is the correlation matrix for wireless channel among pilot locations for ith layer, and a ² is the noise variance. This operation given in equation (16) is repeated for each receive antenna signal m = 1, N _R and all layers i = 1,2, ... , N _L by using corresponding correlations matrices for each layer.

In another example, the correlation matrices R _h ^l _h p and R _h ^l p _h p given in equations (16) can be calculated using robust channel estimation method in two dimensions. It is noted that the layer index i is dropped from the variables for the brevity of the description in the following. To calculate a correlation matrix, the correlations between resource elements in time and in frequency are calculated independently, and they are merged through a Kronecker product. The correlation matrices in the frequency domain are denoted by R^ _p and R^p and they are denoted by R^p and R PfiP in the time domain. By using the matrix R^ defined in (7), R^, can be obtained by taking only the columns of R _Rh corresponding to the pilot locations in the frequency domain, and R P P can be obtained by taking only the rows of R^ _p corresponding to the pilot locations in the frequency domain. Similarly, by using the matrix R^ defined in (8), R^ _p can be obtained by taking only the columns of R _RR corresponding to the pilot locations in the time domain, and R^ _h p can be obtained by taking only the rows of R^ _p corresponding to the pilot locations in the time domain. Then, the combined correlation matrices can be calculated as R _hhP = R^p® R^P ^ar| d

⁼ hPhP® PtiP’ wherein 0 indicates the Kronecker product of the matrices. It is noted that, herein, the channel values in the resource grid h or in the pilot locations in h ^p are ordered as first in frequency then in time. This means that, for a given layer, vector carrying the channel values at the subcarriers of the jth OFDM symbol in the slot, for j = 1,2, ...,N _siot and h ^p, i is a vector carrying the channel values at the subcarriers of the jth pilot OFDM symbol in the slot, for j = 1,2, ...,N _p , wherein N _p denotes the total number of pilot OFDM symbols in the slot. It is also noted that R _h p _h p matrix is common for each layer in the same CDM group.

The disclosed invention (100) has a practical computational complexity as in conventional receivers. The iterative pilot depatterning step (103) of the present invention (100) is based on filtering operation (1031 ) at the pilot locations and pilot depatterning with interference cancellation (1032). The filtering of the channel estimation values at the pilot locations (1031 ) in the iterative pilot depatterning step (103) and interpolation of the estimated channel values (104) can be implemented by using pre-stored matrices for a set of channel and SNR parameters, that is, taking the inverse matrix offline, and calling the necessary matrices for given channel conditions from memory in real time. Therefore, both filtering (1031 ) and interpolation (104) operations require only matrix multiplication to perform estimation, which can be implemented very efficiently in the hardware, and the matrix inversion is not performed in real-time operation. Also, pilot depatterning with interference cancellation (1032) is a low-cost operation and can be performed in linear-time complexity with simple arithmetic operations. Also, the disclosed invention can achieve a good performance with very small number of iterations. Typically, Qmax=1 or Qmax=2 is already enough to achieve the best performance that can be achieved with the disclosed method (100), which means that filtering (1031 ) and pilot depatterning with interference cancellation (1032) operations do not need to be performed many times. Therefore, the complexity of the disclosed invention (100) is only slightly higher than conventional method.

The disclosed invention has an improved and near-optimal performance. In order to show this explicitly, the optimal pilot depatterning procedure is also described. In that case, pilot depatterning and interpolation of the pilots are performed jointly in an optimal MMSE estimator. The procedure is given for a single receive antenna case, as the channel estimation for multi- receive antenna case can be performed by repeating the described process for each antenna separately. The first step is to derive the MMSE estimator for a single layer OFDM system. The received signal at the pilot locations in an OFDM system can be expressed as: (17) where y ^p is the received signal at the pilot location in the OFDM resource grid, X is a diagonal matrix carrying pilot values at the corresponding pilot locations, h ^p carries the channel values at the pilot locations and n is zero-mean complex Gaussian noise vector independent of h ^p , where each component is independent and identically distributed with variance a ² . Therefore, MMSE channel estimator for all resource elements based on observations y ^p can be expressed as: where h carries the channel values at all resource elements. Note that the first correlation matrix is obtained as Rh _y p = R _hh pX ^H , and second correlation matrix is obtained as R _y p _y p = XRftPhpX ^H + a ² I, due to independence of n and h ^p . Therefore, the channel estimate becomes

In a MIMO-OFDM system utilizing CDM in pilot allocations, h (and h ^p at pilot locations) should include channel values for all layers belonging to the same CDM group. For example, CDM group j contains |C _; -| number of layers, therefore, the corresponding vectors can be defined as: where X _t is a diagonal matrix of pilot values for ith layer, h _t is the channel values at all grid of ith layer and h is the channel values at the pilot locations of ith layer. The channel estimate equation given in (19) for a single layer can be generalized to the multiple layers. For this purpose, it is noted that the correlation matrices specific to a layer are same in the same CDM group, as the pilots are located in the same resource elements in all layers. Therefore, if the channel correlation matrices for a given layer is denoted by R^P and fiftpftp, then combined correlation matrices for all layers in the CDM group j can be obtained as: wherein 0 indicates the Kronecker product of the matrices. Therefore, the equation given in (19) can be rewritten for multiple layers case as: where h _mmse is the optimal joint depatterning and MMSE channel estimate for the channel values at all resource grid of all layers in CDM group j. By using the equations in (20) and (22), h _mmse can also be expressed as: wherein

BftftP and RhPhP need to be calculated to obtain h _mmse according to equation in (22). When 2D- MMSE channel estimation is considered for optimal performance, the channel correlation matrices can be obtained by finding the correlations between resource elements in time and in frequency independently, and they are merged through a Kronecker product. R _hh p and RhPhP can be calculated using robust channel estimation method in two dimensions. The correlation matrices in the frequency domain are denoted by R^p and R^p and they are denoted by R^p and R^p _h p in the time domain. By using the matrix defined as R^ in (7), R^> can be obtained by taking only the columns of R^ _h corresponding to the pilot locations in the frequency domain, and R^ptf can be obtained by taking only the rows of R^p corresponding to the pilot locations in the frequency domain. Similarly, by using the matrix R^ defined in (8), R^p can be obtained by taking only the columns of R^ corresponding to the pilot locations in the time domain, and c ^an b ^e obtained by taking only the rows of R^ _p corresponding to the pilot locations in the time domain. Then, the combined correlation matrices can be calculated as R^ = R^p® R^hP and RhPhP is noted that, herein, the channel values in the resource grid channel values in the pilot locations in h and pilot values X _t used in equation (20) are ordered as first in frequency then in time for each layer, i.e. i = 1,2, ..., N _L . This means that, for ith layer, where h ^] _t is a column vector carrying the channel values at the subcarriers of the /th OFDM symbol in the slot at ith layer, for / = 1,2, ... , N _siot and h ’ ^} is a column vector carrying the channel values at the subcarriers of the /th pilot OFDM symbol in the slot at ith layer, for / = 1,2, ... , N _p , wherein N _p denotes the total number of pilot OFDM symbols in the slot. X ^J _t is a diagonal matrix carrying the pilot values at the jth pilot OFDM symbol in the slot at ith layer. The channel estimates for other layers in different CDM groups can be obtained by following the same steps described above. The optimal method has a large computation complexity and is not very practical to implement, as it requires the calculation of inverse of large matrices at every pilot occasion on real-time on the hardware as the matrix to be inverted depends on the actual pilot values as well. However, its performance is provided in the simulation results as a performance target for the disclosed invention.

The performance of the computer implemented method (100) for channel estimation in MIMO- OFDM systems based on iterative pilot depatterning is provided using the pilot structure for 5G introduced by 3GPP in Release 15 standards. In 5G, the pilot symbols for data demodulation is called Demodulation Reference Signals (DMRS), downlink and uplink data channels are called as Physical Downlink Shared Channel (PDSCH) and Physical Uplink Shared Channel (PUSCH). The DMRS defined for both PDSCH and PUSCH have the same structure. Considering the wide range of scenarios that needs to be supported by 5G, the DMRS structure is very flexible and can be configured via relevant configuration parameters. PDSCH mapping type defines if the slot is conventional downlink slot (Type A) or a special slot structure defined in 5G called mini-slot (Type B), dmrs-TypeA-Position defines the starting symbol of first DMRS in the slot (3 or 4). dmrs- AdditionalPosition indicates if there are additional OFDM symbols in the slot which carries DMRS (0,1 ,2 or 3), dmrs-Type specifies the frequency domain pattern of DMRS in a given symbol (Type 1 or Type 2), and maxLength indicates if the CDM group is defined in 1 (single) or 2 (double) symbols, i.e. |£y| = 1 or 2.

In Figure 4, two examples for DMRS configurations are provided. The top two resource grids show the DMRS patterns for the first example. In that case, there are N _L = 4 orthogonal layers and 1 resource block (K = 12 subcarriers). For larger resource block sizes, the given pattern is repeated in the frequency domain, therefore the entire transmission band is collection of such resource blocks and N _siot = 14 OFDM symbols. For the first example, DMRS parameters are configured as PDSCH mapping type = Type A, dmrs-TypeA-Position= 3, dmrs-AdditionalPosition = 3, dmrs-Type = Type 1 and maxLength =1 . Herein, there are two CDM groups with CDM group 1 having layers C = {1,2} with = {1,3} and £ _t = {3}, and CDM group 2 having layers C ₂ = {3,4} with K ₂ ⁼ {2,4} and £ ₂ = {3}. The CDM groups are multiplexed in frequency domain with each other, therefore F = 2, T = 1 and J = 2. For Type 1 , each pilot depatterning group is utilized three times in the frequency domain inside each resource block. For example, in Figure 4, the subcarrier starting positions of pilot depatterning groups for first example are {1,5, 9} for each symbol. The regular extension of the pilot depatterning groups in frequency domain is automatically carried out throughout the transmission band of the data by taking as the base reference. However, time domain allocation is configurable via dmrs-AdditionalPosition parameter. For example, there are pilot symbols at 4 OFDM symbols (at locations {3,6,9,12} in the slot) in Figure 4. For Type 1 and maxLength =1 , there can be maximum 2 different CDM groups, and each CDM group can carry maximum two layers implying the maximum number of orthogonal layers that can be supported is 4 for that configuration.

In Figure 4, the bottom two resource grids show the DMRS patterns for the second example. In that case, the main difference is that dmrs-Type = Type 2 is used instead of Type 1 compared to the first example. For this case, there are again two CDM groups with CDM group 1 having layers group 2 having layers C ₂ = {3,4} with ₂ ⁼ {3,4} and £ ₂ = {3}. For Type 2, each pilot depatterning group is utilized two times in the frequency domain inside each resource block. For example, in Figure 4, the subcarrier starting positions of pilot depatterning groups for second example are {1,7} for each symbol. This shows that Type 2 has lower density in the frequency domain, however it can support larger number of layers in general. This is because of the fact that there can be maximum 3 different CDM groups multiplexed in the frequency domain for Type 2, and each CDM group can carry maximum two layers with maxLength =1 implying the maximum number of orthogonal layers that can be supported is 6 for that configuration. Only N _L = 4 of them is utilized in this example. If maxLength =2, the maximum number of layers with orthogonal pilots doubles compared to single symbol case and becomes 8 and 12 for Type 1 and Type 2, respectively. This is due to the fact that each CDM group can carry maximum four layers instead of two this time, and the maximum number of CDM groups does not change.

Even though the examples are provided using 5G PDSCH DMRS pilot signals, the disclosed method can be applied as a channel estimation method in any pilot allocation scheme involving CDM groups. Some examples include but not limited to channel estimation with multiport CSI- RS, SRS, PUSCH DMRS in 5G, or UE specific DMRS in LTE.

In simulations, the DMRS parameters are chosen as in the examples given in Figure 4 and the 5G waveform is utilized. In particular, TDL-C channel model is used with delay spread 1 or 2 ps with no user mobility. The modulation type is 64 or 256-QAM. The channel coding is NR LDPC with base graph 1 and the code rates are either 2/3 or 3/4. The channel decoder is min-sum algorithm with 20 iterations. The transmission band consists of K = 600 subcarriers, and the subcarrier spacing is 15 kHz. The number of orthogonal layers is N _L = 4 and number of receiver antennas is N _R = 8, no MIMO precoding is used, carrier frequency 3.5 GHz. The synchronization is assumed to be perfect and Soft MMSE equalizer is used as MIMO detector. For channel estimation, 2D-MMSE is utilized as the interpolation method for all considered methods, that is, conventional method, optimal receiver, and disclosed iterative method. 2D-MMSE is also utilized as the filtering method in the disclosed iterative method. The maximum number of iterations Qmax = 1 for all results in the disclosed method.

In Figure 5 the MSE versus SNR performances are provided for DMRS Type 1 and Type 2 configurations on TDL-C channel with 1 ps delay spread for disclosed iterative method, conventional method and optimal method which performs MMSE estimation with pilot depatterning jointly. It is observed that conventional receiver has a significant performance loss compared to optimal receiver. This is because of the fact that the orthogonality assumption does not hold in a frequency selective channel and this causes inter-layer interference at pilot symbols in a CDM group. The disclosed invention significantly improves the performance of the conventional method and has the same error performance as the optimal method when SNR is less than 15 dB. As SNR increases (i.e. SNR > 20) the performance of the disclosed iterative method is slightly worse than that of the optimal receiver, however at that point, achieved MSE values are already less than 0.003, this implies the link level performance of the disclosed iterative method and optimal method will still be very close to each other even for transmission schemes requiring high spectral efficiency. Another important observation is that for optimal and disclosed iterative method, Type 1 configuration is always better than Type 2. This is because of the fact that Type 1 has larger pilot density and inter-layer interference in frequency selective channels is handled much better in both optimal and disclosed iterative method. For conventional method, Type 1 provides better performance, when the performance is limited by noise (low SNR). However, as SNR increases, Type 2 has better error performance compared to Type 1 . This is because of the fact that the pilots in a CDM group is closer in Type 2 as it can be seen in Figure 4. Hence, the change in channel coefficients in a CDM group in Type 2 (which causes the loss of orthogonality) is less than that of Type 1 . Because of that, Type 1 is affected in a more dramatic manner from frequency selectivity, when the conventional receiver is used.

In Figure 6, the MSE versus channel delay spread performances on TDL-C channel are provided for DMRS Type 1 and Type 2 configurations when SNR is 15 dB for disclosed iterative method, conventional method and optimal method. It is observed that disclosed iterative method achieves the performance of optimal method quickly, as the delay spread starts to increase. It is important to note that when delay spread is very low, almost 0 ns, (flat fading channel), the conventional receiver achieves the performance of the optimal receiver as expected due to the fact that orthogonality at the receiver is preserved. In that case, the iterative approach is not necessary and conventional receiver can directly be used for optimal performance. However, as delay spread increases, then the performance of the conventional receiver rapidly deteriorates. For example, when delay spread is 100 ns, disclosed method already improves the performance of the conventional receiver, and the performance difference increases as delay spread increases. In Figure 7, block error rate (BLER) versus SNR performances are provided for 64-QAM with code rate 3/4 for DMRS Type 1 and 2 on TDL-C channel with delay spread 1 ps for disclosed iterative method, conventional method and optimal method. The link level performances of the methods are consistent with MSE error performances given in Figure 5. For Type 1 , the iterative method has the same performance with the optimal receiver, and the performance gap between optimal receiver and iterative solution is 0.1 dB for Type 2 at 0.01 target error rate. Note that the performance of the conventional receiver is improved by 4.5 dB and 0.7 dB by iterative method for Type 1 and Type 2, respectively. The performance under perfect channel information is also provided, and the performance loss due to channel estimation error compared to ideal channel information is 1 and 1 .6 dB for disclosed method for Type 1 and Type 2, respectively.

In Figure 8, the MSE versus SNR performances are provided for DMRS Type 1 and Type 2 configurations on TDL-C channel with 100 ns delay spread for disclosed iterative method, conventional method and optimal method. It is observed that iterative method performs better than conventional receiver even with a relatively short delay spread.

In Figure 9, the MSE versus SNR performances are provided for DMRS Type 1 and Type 2 configurations on TDL-C channel with 300 ns delay spread for disclosed iterative method, conventional method and optimal method. Performance improvement via iterative method over conventional receiver starts to become more significant.

In Figure 10, the MSE versus SNR performances are provided for DMRS Type 1 and Type 2 configurations on TDL-C channel with 700 ns delay spread for disclosed iterative method, conventional method and optimal method. The performance improvements with disclosed invention are significant. It is noted that the optimal receiver also starts to have error floor due to large delay spread because of the fact that CP is not long enough to fully cover channel length for 700 ns delay spread.

In Figure 1 1 , the MSE versus channel delay spread performances on TDL-C channel are provided for DMRS Type 1 and Type 2 configurations when SNR is 25 dB for disclosed iterative method, conventional method and optimal method. The performance of iterative method is near-optimal especially as delay spread increases. The conventional receiver achieves optimal performance in flat channel (almost 0 ns, delay spread). Using iterative method is not necessary when delay spread is less than 100 ns for this example, however the MSE values are around 0.0001 , therefore it is possible to use Qmax=1 even in flat channels.

In Figure 12, the BLER versus SNR performances are provided for 256-QAM with rate 2/3 for DMRS Type 1 and 2 on TDL-C channel with delay spread 1 ps for disclosed iterative method, conventional method and optimal method. It is observed that the conventional receiver can not operate in Type 1 configuration, whereas iterative receiver has almost the same performance with optimal receiver. Therefore, disclosed iterative method in channel estimation procedure can ensure almost optimal performance for Type 1 . For Type 2, the performance gap between optimal and iterative receiver is 0.25 dB for 0.01 target BLER, whereas iterative receiver improves the performance of the conventional receiver by 1 .8 dB. The performance gap between iterative receiver and perfect channel information case is 1 .2 and 1 .8 dB for Type 1 and Type 2, respectively. The benefit of the disclosed method is clearly observed in this high spectral- efficiency transmission scenario.

The simulation results show that disclosed iterative computer implemented method (100) has almost the same performance with the optimal receiver in many cases, and mitigates the detrimental effect of the orthogonality loss in a CDM group to the channel estimation performance. The cost of each iteration involves filtering at the pilot locations (1031 ) i.e. the multiplication of the estimates at the pilot locations by an filtering/smoothing matrix, which can be implemented in the hardware very efficiently and pilot depatterning with interference cancellation (1032) which only involves simple addition/subtraction at pilot locations. Note that filtering matrix can be stored in the memory, as its elements do not need to be calculated in real-time due to independence of the matrix from actual pilot values. Such an approach is not possible with optimal receiver, and optimal receiver strategy is not likely to be implemented in practical systems due to large complexity. Therefore, disclosed computer implemented method (100) provides almost optimal performance with a reasonable complexity increase compared to conventional receiver.

The present invention also relates to a data processing apparatus to carry out the steps of disclosed computer implemented method (100) for channel estimation in MIMO-OFDM wireless communication system utilizing CDM groups in pilot allocations. In accordance with an embodiment of the present invention, the data processing apparatus comprises:

• a pre-processing module which receives and performs OFDM demodulation to each receive antenna signal to obtain received symbols in the resource grid of each receive antenna, and performs an initial pilot depatterning in all pilot depatterning groups using the corresponding OCC code and received symbols in the resource grid of all layers at all CDM groups for each receive antenna signal;

• an iterative pilot depatterning performer module, which, at each iterative step, filters the channel estimates at the pilot locations to obtain channel changes at the pilot depatterning groups in the resource grid of all layers for each receive antenna signal, performs pilot depatterning with inter-layer interference cancellation using the received signal and filtered channel estimates at the pilot locations of other layers causing interference in the CDM group for each receive antenna signal, updates the estimated channel values at all pilot locations, determines if the pre-determined number of iterations is reached and if it is, then terminates the iterative process; if it is not, then increases the iteration count and starts the next iteration of iterative process;

• a channel estimator module configured to interpolate the estimated values in pilot locations to obtain the channel estimation values for all resource elements at each layer for each receive antenna signal.

The present invention also relates to a computer program comprising instructions which, when the program is executed by a computer, cause the computer to carry out the steps of disclosed computer implemented method (100).

The present invention also relates to a computer readable data carrier having stored thereon the computer program of disclosed computer implemented method (100).

The various modificiations to the present invention (100) may be suggested to one skilled in the art, and the exemplary embodiments used in the description of the disclosed invention shall not limit the scope of the appended claims.