WO2017108075A1 - Method and apparatus for enhancing user selection in a mu-mimo system - Google Patents

Abstract

The present invention provides a user selection enhancing unit 300, an enhanced scheduling unit 501 ', a multi-user-multi-input-multi-output, MU-MIMO communication system 500 and a method 100 for enhanced user selection in a multi-user-multi-input- multi-output, MU-MIMO communication system 500, the method comprises the following steps: determining 101 at least one system operating value 302 that indicates the actual systems performance; calculating 102 a number K* of user's equipments, UE, to serve simultaneously in the system on the basis of a signal-to-interference-plus-noise- ratio, SINR, value, wherein the SINR value is determined based on the at least one system operating value 302; and providing 103 the calculated number K* of UEs to serve to a scheduling unit 501, 501 ' for selecting at least one user S.

Description

METHOD AND APPARATUS FOR ENHANCING USER SELECTION IN A MU-

MIMO SYSTEM TECHNICAL FIELD

The present invention relates to the field of wireless communication techniques. In particular, the present invention describes a method and an apparatus for enhancing user selection in a multi-user-multi-input-multi-output, MU-MIMO communication system. A method and/or a system is provided for enhancing the user selection procedure in MU-MIMO downlink transmission by determining the best number of users to serve based on system operating values. The invention can be used in both frequency-division duplex (FDD) and time-division duplex (TDD) systems. Aspects of the present disclosure relate generally to wireless communication systems and, more specifically, to multiple antenna transmission technologies, such as those configured for MIMO precoding/beamforming. Certain embodiments of the invention relate to user group selection for MU-MIMO downlink transmission.

BACKGROUND

In radio, MIMO communication is a method for multiplying the capacity of a radio link using multiple transmit and receive antennas to exploit multipath propagation. A MU-MIMO is a set of multiple -input and multiple-output technologies for wireless communication, in which a set of users or wireless terminals, each with one or more antennas, communicate with each other. A MU-MIMO system, in which a multi-antenna base station (BS) communicates with multiple users simultaneously, is a key enabling technology to provide substantially increased spectral efficiency and enhanced link reliability in broadband wireless communication systems. In MU-MIMO downlink transmission, the spatial degrees of freedom offered by multiple antennas can be advantageously exploited to enhance the system sum-rate through scheduling multiple users simultaneously by means of space-division multiple access (SDMA). Such a multiple access protocol requires more complex scheduling strategies and transceiver methodologies, but does not involve any bandwidth expansion. In spatial multiple access, the resulting inter-user interference is handled by the multiple antennas which, in addition to providing per-link diversity, also give the degrees of freedom required to separate users in the spatial domain.

MU-MIMO transmission has been enabled in state-of-the-art wireless network standards, including LTE cellular networks, IEEE 802.1 lac WLAN, and WiMAX. In real-world MU-MIMO systems, the BS is usually equipped with more antennas than the user's equipment (UE) due to various practical limitations and factors, including equipment size, cost, power consumption, and computational capabilities. In a recent development of MU-MIMO technology, known as massive MIMO or large-scale MIMO, few hundreds of antennas are employed at the BS in order to send simultaneously different data streams to tens of users.

A major fundamental challenge arising in MU-MIMO transmission is how the BS should choose the set of users to serve in order to optimize a certain performance metric of interest, e.g., sum-rate, weighted sum-rate, QoS-based metric, etc. The choice of the best user subset depends on the precoding method adopted, as well as on the user channel state information (CSI) available at the BS. Although dirty paper coding (DPC) is the theoretically optimal non-linear precoding scheme, since it achieves the capacity of a MU-MIMO downlink channel, it is highly complex to implement in practice. For that, linear precoding schemes, such as zero-forcing beamforming (ZFBF) and block diagonalization (BD), are often employed in both industry and academia.

In a MU-MIMO system with linear beamforming, a BS with Ντ transmit antennas can only serve up to K < Ντ users out of the total number of active users (i.e., users having traffic and requesting access to the medium). The Ab users' rates are inter-coupled and depend on their channel orthogonality as well as on their channel strength. Finding the optimal set of users (of cardinality K) requires a brute-force exhaustive search over all possible user subsets, whose complexity is prohibitively high when the number of users is large. Several suboptimal, mostly greedy, user selection algorithms have been proposed in order to determine the selected user subset. When MU precoding and user selection are performed using accurate CSI, the transmission scheme can remove the inter-user interference, usually through channel matrix inversions and projections. Nevertheless, in practical system settings where user selection and precoding rely on imperfect CSI, e.g. imperfect channel estimation, quantized/delayed CSI, limited-capacity feedback link (in FDD mode), and calibration errors (in TDD mode), the inter-user/stream interference cannot be completely eliminated, resulting in system performance deterioration. Specifically, the system becomes interference-limited and the throughput saturates (ceiling effect) for increasing signal-to-noise ratio (SNR).

State-of-the-art user selection solutions cannot optimize the system performance, even when providing a user subset with excellent orthogonality properties, as they usually fail to identify the optimal number of UEs to serve for each operating regime.

The present invention builds on the observation that serving the right number of users is vital in interference-limited MU-MIMO systems and proposes a very low- complexity method for determining this number in order to enhance the user selection process and significantly improve the system performance.

As previously discussed, in MU-MIMO systems with imperfect CSI, the performance is dominated by the uncanceled inter-user interference, whose effect is more evident and detrimental at high SNR operating regimes. Existing user grouping methods aiming only at selecting a set of users with very good channel orthogonality properties and strong channels would be very inefficient in future-generation communications, especially in dense cellular networks with universal frequency reuse, where a) the inter- user (and inter-cell) interference will be increased, b) the channel dimensions to estimate will be larger, e.g., as more antennas are deployed at the BS (massive MIMO), and c) the number of active users/nodes is expected to increase massively.

As discussed above, determining the best user subset in MU-MIMO downlink transmission is a combinatorial optimization problem, which requires a prohibitively complex brute-force exhaustive search over all possible user subsets. Several suboptimal user selection algorithms have been proposed, mostly following a greedy selection approach.

In G. Dimic and N. D. Sidiropoulos, "On downlink beamforming with greedy user selection: performance analysis and a simple new algorithm ", IEEE Trans. Signal Process., vol. 53, no. 10, pp. 3857-3868, Oct. 2005, a greedy user selection (GUS) based on sum-rate metrics was derived. The user with the highest rate is first selected and the next user(s) is selected as the one(s) that gives the highest sum-rate among the remaining unselected users.

This method was further improved in J. Wang, D. J. Love, and M. D. Zoltowski, "User selection with zero-forcing beamforming achieves the asymptotically optimal sum rate ", IEEE Trans. Signal Process., vol. 56, no. 8, pp. 3713-3726, Aug. 2008 using sequential water-filling to eliminate users with zero transmit power after selection and power allocation. Both algorithms are of high complexity since calculation of the precoders (matrix inversions) is required each time a new user is added in the set (at each step). Furthermore, in the case of imperfect CSI, the above algorithms - in spite of their high complexity - fail to terminate when the optimal number of users is selected.

In T. Yoo and A. Goldsmith, "On the optimality of multiantenna broadcast scheduling using zero-forcing beamforming", IEEE J. Sel. Areas Commun., vol. 24, no. 3, pp. 528-541, Mar. 2006, a technique coined as semi-orthogonal user selection (SUS) is proposed. The first user is selected as the one having the highest channel quality indicator (CQI) metric (either channel norm or effective channel), and the next user is selected within a user set of ε-orthogonal users as the one having the largest CQI value. Using this method, the BS chooses users that have high CQI and are mutually semi-orthogonal in terms of their channel directions.

Despite the fact that SUS reduces the complexity as compared to GUS, such methods also exhibit the flaw of not being able to identify the best number of users to serve. As such, there are redundant users in the selected user subset, i.e., users that can be deleted from the selected user subset to yield a performance increase. Note that the fact that the aforementioned techniques do not find the optimal number of users is an inherent flaw of any greedy incremental algorithm due to the non-iterative cumulative user selection procedure.

In S. Huang, H. Yin, J. Wu, and V. C. M. Leung, "User selection for multiuser

MIMO downlink with zero-forcing beamforming" , IEEE Trans. Veh. TechnoL, vol. 62, no. 7, pp. 3084-3097, Sept. 2013 it is proposed to add the "delete" and "swap" operations as means to tackle the redundant user issue. However, this method increases the complexity, even as compared to GUS, as it involves matrix inversions, projections, and an iterative procedure. Furthermore, its performance is sensitive to CSI inaccuracies due to the increased error in the projection operation using imperfect CSI and does not necessarily find the best user subset in the imperfect CSI case.

In US7630337 B2 the main idea is to select sequentially a certain number of users in a greedy fashion based on quantized channel direction and channel gains. The user with the highest quantized channel gain is first selected, and the second user is selected based on the value of the product of channel gain and orthogonality. Multiple rounds of feedback reporting are required and only two users are selected. Actually, it is a simplification of the semi-orthogonal user selection (SUS) proposed by Yoo and Goldsmith. Furthermore, only 2 users are selected and there is no method or way to find the best number of users to be served. Even a modification or generalization by selecting more than 2 users through repeating the procedure will fail to identify the best number of users.

Both US8934432 B2 and US8150330 B2 describe range reduction algorithms. This range reduction scheme reduces the number of users or user pairs to consider for searching in order to specify the best user group. Firstly this range reduction algorithm is used to reduce the dimensionality of the search space by excluding the weakest users from the selection procedure. In other words, if this range reduction algorithm is pessimistic, the system will serve fewer users than the optimal and the performance will be decreased due to reduced multiplexing gain. If the range reduction algorithm is optimistic, more users will be selected, resulting again in throughput loss due to inter-user interference. Secondly, and most important, the range reduction method is basically a threshold operation. The method consists in determining the desired user group by searching among the strongest users. Furthermore, the threshold operation is performed using the channel gains of the users and therefore it should be updated every coherence block, i.e., every time the users' channel gains change, which happens in the order of milliseconds in real systems. Finally, users with weak channels are penalized and are excluded from the user selection process. This is detrimental from a fairness point of view and will result in rate/service starvation for the weak users.

SUMMARY

In view of these disadvantages, there is a need to improve the throughput of future wireless communication systems and to avoid the sum-rate ceiling effect due to remaining interference, which can be achieved by inferring the optimal number of users to serve at each operating regime and scenario. The proposed invention turns out to be very efficient, fair and improves remarkably upon existing state-of-the-art solutions.

The present invention has particularly the object to provide a user selection scheme that is less complex, less unfair and performs well in high-complex communication scenarios.

The above mentioned object of the present invention is achieved by the solution provided in the enclosed independent claims. Advantageous implementations of the present invention are further defined in the dependent claims. In particular this invention generally targets scenarios in which a set of users is selected to transmit to; more specifically, it focuses on MU-MIMO downlink transmission performing precoding and user selection using imperfect channel knowledge. The invention can be applied to MU-MIMO systems operating in either TDD or FDD mode. More importantly, the FDD implementation of the idea involves using a signal to report necessary scalar information to the BS, enabling thus the detectability and strengthening patent protection.

The invention is most likely applied to MU-MIMO downlink transmission. However, if replacing the space domain with other domains, e.g., OFDMA techniques and joint spatial division and multiplexing (JSDM), the invention can be applied to other transmission methods, too. Similarly, the invention applies to a multiple access (uplink) scenario. Finally, the method determines the best number of users to serve, but it can also be used to determine the optimal number of streams to send to each user according to a certain metric.

A first aspect of the present invention provides a method for enhanced user selection in a MU-MIMO communication system, the method comprises the following steps: determining at least one system operating value that indicates the actual systems performance; calculating a number of user's equipments, UE, to serve simultaneously in the system on the basis of a signal-to-interference-plus-noise-ratio, SINR, value, wherein the SINR value is determined based on the at least one system operating value; and providing the calculated number of UEs to serve to a scheduling unit for selecting at least one user.

A system operating value is provided as an input value for the inventive method.

It is basically an operating value that defines a structural part of the system itself, such as a number of operable antennas and/or a requirement to operate the system, such as a needed number of streams at each UE, statistical, long-term or large-scale channel information, or an operating SNR that needs to be fulfilled.

Thus, the method finds in a simple and efficient manner the best balance between multiplexing gain i.e., serving more users, and per-user rate affected by the uncanceled inter-user interference based on an estimated and/or approximate expression for the expected SINR. The number of users that maximize the expected/approximate sum-rate of the system is then found. Once this operating point is determined, i.e., once the optimal number of users K* is found, the Scheduler can select the K* users based on different criteria and performance metrics. The main advantage of the proposed invention is its enhancement of the performance of MU-MIMO wireless communication systems employing a very low- complexity method for determining in advance the cardinality of the best selected user subset based on system operating values. The proposed invention improves the state-of- the-art solutions both in terms of complexity and performance. Specifically, the proposed solution improves the sum-rate performance of a MU-MIMO downlink transmission as compared to SUS in all SNR ranges and having lower complexity. Remarkably, it even outperforms the high-complexity GUS at moderate to high SNR values. Without the invention, the MU-MIMO system will operate in a suboptimal point, especially not in the right point that balances multiplexing gain and per-user link rate, resulting in lower system performance, especially in terms of throughput.

The invention is related to a user selection technique as such. It is enhanced user selection unit, whose output can be provided to any user scheduling method adopted in the system. In other words, it is a method and/or a system adding intelligence to user selection techniques by modifying their operation through determining the cardinality of the selected user set for different system configurations.

Additionally, the invention can be employed as a simple yet efficient user selection method. Finally, the method is a general process and does not only work for maximizing the sum rate.

In a first implementation form of the method of the first aspect, the at least one system operating value includes the number of transmitting antennas and/or the receive antennas and/or the number of streams to send to each UE. Thus, the proposed approach is fully scalable, in terms of computational complexity, since nearly fixed system parameters are used, and only requires the computation of a one-dimensional variable that represents an approximate SINR as a function of users asking for service, which can be derived in closed-form in most cases or can be numerically evaluated or approximated. To optimize the expected performance is a simple integer optimization problem. In this sense, the advantages brought over state-of-the-art solutions, especially those involving matrix inversions and recursive operations are evident.

In a second implementation form of the method of the first aspect as such or according to any of the previous implementation forms, the at least one system operating value further includes the operating signal-to-noise ratio, SNR, and/or a channel estimation error and/or a channel state information, CSI, CSI error. The SNR may be defined as the ratio of the transmit power P to the background noise variance. Based on the above system-level parameters, an effective or approximate received SINR can be calculated, which is a function of the number of users to serve K. This SINR expression can be given in closed-form in most cases and its implementation is very easy.

In a third implementation form of the method of the first aspect as such or according to any of the previous implementation forms, the at least one system operating value is a real scalar value. In the most general case, if the knowledge of correlation among antennas, users, large-scale channel information, etc., a matrix with complex values may have to be included. However, since the number of UE to serve is a scalar value, a sticking to scalar values allows a simple computing method that can be implemented easily.

In a fourth implementation form of the method of the first aspect as such or according to any of the previous implementation forms, a system performance metric of interest is calculated, wherein the number of UE to serve is further calculated on the basis of a system performance metrics of interest function to satisfy the performance requirements defined by the MU-MIMO system, e.g. a scheduling unit or a resource allocation block. Thus, once this operating point is determined, users can be selected according to the desired performance criterion, e.g. QoS, largest queue length, service priority, fairness, etc. The system performance metrics of interest function may take on different forms depending on the performance target and criterion. As an example, commonly used performance metrics in wireless systems may include coverage, capacity (sum-rate), weighted sum-rate, energy/power consumption, delay and other for standard utility functions of users-applications.

In a fifth implementation form of the method of the first aspect as such or according to any of the previous implementation forms, the number of UEs is calculated by following formula

K* = argmaxt_f PM( SINR_k, K) , wherein K is the best number of UE's to serve simultaneously, K is the maximum possible number of UE to serve and PM is a calculated performance metric of interest function.

In a sixth implementation form of the method of the first aspect according to the fifth implementation form, the expression PM(SINR_k,K) is calculated by following formula PM{SlNR_k, K) = ^ w_k - R_k (SINRk

k=l

wherein Wk is a k-t UE real-valued weight and ¾ is the rate of a k-t UE. In a homogeneous systems with specific channel characteristics, the Shannon rate might be applied, that would lead to a more specific formula for the expression PM (SINR_k, K):

PM{SlNR_kk, K)

l g₂ (l + SlNR_k {K)) .

In a seventh implementation form of the method of the first aspect according to the fifth implementation form, the expression PM(SINR_k,K) is retrieved from a look-up- table indicating a specific set of modulation and coding. Thus, the system performance metric of interest is pre-stored in the system based on previously evaluated communication schemes that are already approved.

In an eighth implementation form of the method of the first aspect according to the fifth implementation form, the SINR_k(K) value is an effective or approximate received SINR value of a k-t UE as a function of K.

In a ninth implementation form of the method of the first aspect according to the any one of the fifth to eighth implementation forms, K is the maximum possible number of UEs to serve.

The expression PM(SINR_k,K) may be calculated for users with homogeneous channel statistics by the following formula:

SNR (N_T - K + 1) - Q - T⁴

SINR

K + SNR (K - 1)^■ Vl - τ² + _N _ ^τ _χ ^■ (1 - τ²)^2■ SNR* wherein Νχ is the number of transmit antennas, τ is the quality of channel estimation operation, Q is the number of streams and SNR is the ratio of a transmitted power to a background noise variance.

In a second aspect of the present invention a user selection enhancing unit in a MU-MIMO communication system is provided. The user selection enhancing unit comprises an input means configured to receive at least one system operating value that indicates the actual systems performance; a calculation means configured to calculate a number of user's equipments, UE, to serve simultaneously in the system on the basis of a SINR, value, wherein the SINR value is determined based on the at least one system operating value; and an output means configured to provide the calculated number of UEs to serve to a scheduling unit of the MU-MIMO system for selecting at least one user.

In a first implementation form of the user selection enhancing unit of the second aspect, the calculation means is further configured to calculate a system performance metric of interest, wherein the number of UE's to serve is further calculated on the basis of the system performance metrics of interest function.

The user selection enhancing unit according to the second aspect and its first implementation form is adapted to perform the calculations already described with reference to the implementation forms of the method of the first aspect. Similarly, it must be clear that the at least one operating value received by the enhancing unit include those described with reference to the first to third implementation forms of the method of the first aspect.

The user selection enhancing unit according to the second aspect, and its implementation forms, achieves the same advantages and technical effects than the waveguide structure of the first aspect and its respective implementation forms.

In a third aspect of the present invention an enhanced scheduling unit in a MU- MIMO communication system is provided. The enhanced scheduling unit comprises a user selection enhancing unit according to the second aspect of the present invention or its implementation forms that is configured to provide the calculated number of UE's to serve and an input means configured to receive at least a selection value that indicates the actual systems performance.

The enhanced scheduling unit according to the third aspect, and its implementation forms, achieves the same advantages and technical effects than the waveguide structure of the first aspect and its respective implementation forms.

In a fourth aspect of the present invention a MU-MIMO communication system is provided. The MU-MIMO communication system at least comprises a user selection enhancing unit according to the second aspect of the present invention or its implementation forms and a scheduling unit for selecting at least one user.

The MIMO system according to the fourth aspect, and its implementation forms, achieves the same advantages and technical effects than the waveguide structure of the first aspect and its respective implementation forms.

In a fifth aspect of the present invention a computer program product for implementing, when carried out on a computing device, a method according to the first aspect. The computer program product according to the fifth aspect, and its implementation forms, achieves the same advantages and technical effects than the waveguide structure of the first aspect and its respective implementation forms.

It has to be noted that all devices, elements, units and means described in the present application could be implemented in the software or hardware elements or any kind of combination thereof. All steps which are performed by the various entities described in the present application as well as the functionalities described to be performed by the various entities are intended to mean that the respective entity is adapted to or configured to perform the respective steps and functionalities. Even if, in the following description of specific embodiments, a specific functionality or step to be full formed by esternal entities is not reflected in the description of a specific detailed element of that entity which performs that specific step or functionality, it should be clear for a skilled person that these methods and functionalities can be implemented in respective software or hardware elements, or any kind of combination thereof.

BRIEF DESCRIPTION OF DRAWINGS

The above described aspects and implementation forms of the present invention will be explained in the following description of specific embodiments in relation to the enclosed drawings, in which

Fig. 1 shows a flow chart of a method according to a first embodiment of the present invention.

Fig. 2 shows a flow chart of a method according to a second embodiment of the present invention.

Fig. 3 shows user selection enhancing unit according to a first embodiment of the present invention.

Fig. 4 shows user selection enhancing unit according to a second embodiment of the present invention.

Fig. 5 shows a system of a user selection enhancing unit and a scheduling unit according to an embodiment of the present invention.

Fig. 6 shows an enhanced scheduling unit according to an embodiment of the present invention.

Fig. 7 shows a MU-MIMO system according to an embodiment of the present invention. Fig. 8a-8b show a first throughput performance comparison of different user selection techniques with and without user selection enhancement according to the present invention.

Fig. 9a-9b shows a second throughput performance comparison of different user selection techniques with and without user selection enhancement according to the present invention.

Fig. 10a- 10b shows a third throughput performance comparison of different user selection techniques with and without user selection enhancement according to the present invention.

Fig. 1 la-1 lb shows a fourth throughput performance comparison of different user selection techniques with and without user selection enhancement according to the present invention.

DETAILED DESCRIPTION OF EMBODIMENTS

In Fig. 1 a flow chart for a method 100 for enhanced user selection in a multiuser-multi-input-multi-output, MU-MIMO communication system 500 is shown. The method 100 comprises a determining step 101 in which at least one system operating value 302 that indicates the actual system performance is determined. Subsequently, a calculating step 102 is performed, in which a number K* of user's equipments, UE, to serve simultaneously in the system on the basis of a signal-to-interference-plus-noise- ratio, SINR, value is calculated, wherein the SINR value is determined based on the at least one system operating value 302. Subsequently, a providing step 103 is performed, in which the calculated number K* of UEs to serve is provided to a scheduling unit 501, 501 ' for selecting at least one user.

In Fig. 2 a flow chart for a method 200 for enhanced user selection in a multiuser-multi-input-multi-output, MU-MIMO communication system 500 is shown. The method 200 comprises a determining step 201 in which at least one system operating value 302 that indicates the actual systems performance is determined. Subsequently, a calculating step 202 is performed, in which a number K* of user's equipments, UE, to serve simultaneously in the system on the basis of a signal-to-interference-plus-noise- ratio, SINR, value is calculated, wherein the SINR value is determined based on the at least one system operating value 302. Subsequently, a providing step 203 is performed, in which the calculated number K* of UEs to serve is provided to a scheduling unit 501, 501 ' for selecting at least one user. Subsequently, another calculation step 204 is performed, in which a system performance metric of interest function PM is calculated.

The system performance metric of interest function PM can be any function, which takes on the SINR value as input; the SINR may be an exact SINR value or an estimated value or an approximated value. Additionally, the PM function can also depend from the number of users K, either explicitly or implicitly (since the SINR value in multiuser communication systems is in turn a function or depends on the number of users K). The function PM is supposed to indicate a standard and widely-used performance metric of the physical layer and network performance, including capacity, sum-rate or aggregate throughput, weighted sum-rate, delay constrained sum-rate, energy efficiency or consumption. The PM function can be any meaningful utility function of user applications containing at least the SINR value (exact or estimated/approximated). The form of the PM function can be designed from case to case depending on the characteristics of the specific implementation of the invention and the traffic application.

In the following paragraphs examples of the function PM will be given with respect to specific contexts and system configurations. In the example for which simulation results are provided, the PM function is the sum-rate of the system serving K users. However, it is clear that the invention is not limited to these forms of the function PM.

In the following, a detailed description of the inventive methods 100, 200 is provided. It is leveraged on the observation that, in interference-limited MU-MIMO downlink transmission with imperfect CSI, it is more important to know the optimal number K of users to serve rather than which users exactly to select based on their compatibility properties.

The invention originates from the need to find an efficient and low-complexity way to estimate or infer this optimal number K of users using system parameters that do not change very frequently. The objective is to boost the system performance and outperform existing prior art. Furthermore, differently from the state-of-the-art strategies, the invention is a very low-complexity method that does not require the implementation of complex iterative procedures and does not involve any computationally heavy matrix inversions. Evidently, this results in a complexity gain that is increasingly appealing as the number of active users grows. Moreover, identifying the optimal number K of users to serve provides significant performance gains for SNR increasing in MU-MIMO systems 500 with imperfect CSI. This is due to the fact that the inter-user interference increases and becomes the limiting factor for SNR increasing.

Below, a system model is used for assessing the performance of the invention. It should be noted that the invention does not work only for this system model structure. The below structure is given for application and evaluation purposes mainly. A MU- MIMO downlink system 500 is considered, in which a BS with Ντ transmit antennas sends signals to a maximum of M_s users, each equipped with N_R antennas. Omitting the time and frequency index, the received A¾-dimensional signal vector at the k-th UE can be expressed as

y_k = H_kx + n_k (1) where y_k e £^{NR X1}, H_k e £^NR ^XNT i_s the channel matrix between the BS and the k-th UE x e £^Ντ^χ1 includes the transmitted symbols, and n_k e £^Ντ^χ1 represents the additive white Gaussian noise (AWGN) for which n_k ~CN (0_NR, I_NR) . Using linear precoding, the transmit vector x takes on the form x = P_s , where Pe £^NT^XMS denotes the precoding matrix (transmit linear filter) utilized by the BS and s = [s], S2, SMS] contains the complex-valued unit-power symbols intended for the Ms served UEs. We assume that Q≤ N_R streams are sent to each UE.

Expanding equation (1) yields

M_S M_S (2)

Vk = ^HkPs + n_k = H_k ^ Pi Si + n_k = H_kp_ks_k + H_k ^ p_f s_f + n_k

i=l i=l,i≠k where p_k e £^Ντ^χ1 denotes the k-th column of P, i.e., P = [pi, p2, PMS]■

The wireless channel between k-th UE and BS, described by H_k and that both BS and each k-th UE have imperfect knowledge of H_k. The estimation of H_k is modeled as

where E e £^NR^XNT represents the channel estimation error matrix consisting of standard complex normal random variables, i.e., E_mn ~CN (0,1) . This widely used model can capture various scenarios of imperfect CSI. In one embodiment, the system operates in TDD mode and this setting can model the result of channel estimation errors using pilot signals. It can also be used to model calibration errors between the BS and the UE. In another embodiment, the system operates in FDD mode and it can model the error/quality of the channel estimation process at the UE side. The parameter τ e [0,1] captures the CSI imperfections and describes the quality of the channel knowledge (or the quality of the channel estimation technique). Perfect CSI corresponds to the case of r = 1, whereas τ = 0 implies that H_k is independent of H_k, i.e., there is no correlation between the actual and the estimated channel (worst case). Note that the invention is not tailored to the specific CSI model described above and can be easily adapted to work under various conventional receiver structures. For instance, the following model can be also used

with SNR_e/f being the effective SNR during the training phase. Evidently, the effective SNR can be easily mapped to the parameter r.

The system parameter τ is one of the inputs of the proposed inventive block and is considered to be known at the BS. It represents the channel estimation (CE) and/or CSI error. Note that in one embodiment (FDD mode), the value of r, which is a scalar and does not change frequently, can be reported to the BS through signaling. There are various methods for estimating the variance (or sample variance) of the CE/CSI error, ranging from kernel-based to data-aided ones. These are methods for pilot-aided channel estimation, as well as for semi-blind and blind channel estimation. Without assuming any particular estimation method (e.g. ML, MMSE, LS), the error variance can be calculated using the background noise variance, the variance of the symbols, and the covariance of channel samples. Moreover, there are many methods based on channel measurements that allow us to have an accurate estimation of the error variance.

In TDD systems, τ may represent the calibration error and no additional feedback is required. In FDD systems with imperfect CE at the receiver side, the value of τ may be reported back to the BS through feedback in the uplink channel. This may increase the signaling overhead, but this increase is very low or negligible. First, as τ represents the variance of the CE error (second-order statistical information), it does not varies rapidly. It can be reported every few thousands of samples, resulting in negligible overhead. Considering the most challenging situation of periodic reporting of τ e [0,1] every 100ms, additional 2 or 3 bits may be needed with scalar quantization, so the uplink feedback rate will be 20 to 30 bits/s at each sub-band. Even if it should be reported on every band (which is very inefficient in practice), the signaling overhead is at worst 10% of the total throughput. In realistic implementations, the signaling overhead is expected to be lower (optimizing the reporting period and exploiting sub-band correlation) and is estimated to be around 3 to 5% of the achievable throughput.

In the considered downlink MIMO system 500, BS employs linear precoding. For exposition purposes, it is considered that the block diagonalization (BD) precoder of the concatenated broadcast channel [H₁, H₂ ... H_Ms ] is adopted as a means to remove the inter-user interference at the BS. Evidently, the invention can operate with any precoding technique used in the system. Using BD, for each H_k it holds that

H_k ^~p_k = 0 V/c≠ i (5)

Upon signal reception, each k-th UE utilizes a linear minimum mean-squared error (MMSE) filter w_k e C^w«^xl given by w_fc = (R_k + HkPkPk Hk ^HkPk (6) where R_k denotes that the interference-plus-noise covariance matrix, which is obtained as

Once again, it is emphasized that the invention is not tailored to a specific receiver and can be easily adapted to work under various conventional receiver structures.

The actual signal-to-interference-plus-noise ratio (SINR) at the -th UE is given by

For the sake of clarity, using a state-of-the-art scheduling unit, like the scheduling unit 501 described in the following with reference to Fig. 5, all operations are performed within the same block. The scheduling unit receives CSI and CQI information for all users as well as information on the users' traffic (queue length, minimum rate, etc.) and the performance requirements (QoS, priorities, etc.). The scheduling unit employs a certain (usually complex) user selection algorithm which, based on the users' CQI and taking into account all constraints and requirements, provides the set of users to serve, denoted by S.

According to the invention a new user selection enhancing unit 300 is proposed, whose structural block diagram is depicted in detail in Fig. 3. Therein an input means 301 retrieves at least one system operating value 302. In a calculating means 303, the SINR value is calculated and the number K of UE to serve is retrieved. This number K is provided to a subsequent scheduling unit (not shown in Fig. 3).

According to Fig. 4, a further embodiment of a user selection enhancing unit 300 is shown. The user selection enhancing unit of Fig. 4 describes further elements and details and can be used in addition or in alternative to the unit of Fig. 3. Therein the calculating means 303 are provided in greater details. The elements of the user selection enhancing unit of Fig. 4, which did not change and were already described with reference to Fig. 3 will not be described again. The calculating means 303 at least comprise an SINR-calculation block 3031 at which the SINR-expression can be retrieved. Subsequently, a PM-function calculation block 3032 is provided to apply a system performance metrics of interest function PM. The PM function might be obtained from a LUT that is stored in a PM-storage 306. Finally, an optimizer 305 determines the integer value of K that optimizes the objective function PM ((SINR), K^ and its output is denoted by K .

For instance, at least one of the following system operating values 302 or a function of them might be provided to the input means 301 of the user selection enhancing unit 300 of Fig. 3 or Fig. 4:

• number of antennas in the system 500, such as BS-antennas Ντ and UE- antennas NR);

• number of streams Q to send at each UE;

• operating SNR (usually defined as the ratio of the transmit power P to the background noise variance);

• the CE/CSI error τ (quality of CE operation);

• statistical CSI (e.g. channel correlation, long-term statistics, large-scale channel attenuation). It should be noted that the CE/CSI error τ can take on different values for different users. In that case, τ may either represent a vector containing each user's CE/CSI error (in the most general case), or it can be a real-valued scalar value of a function of each user's individual CE/CSI errors, e.g. a weighted mean, median, etc.

Based on these operating system parameters 302, preferably provided as real- valued scalars to reduce computing efforts, the user selection enhancing unit 300 calculates first an effective or approximate received SINR at the calculating means 303 (e.g. a SINR-calculation means 3031), which is a function of the number of users to serve K and is denoted as SINR .

The SINR is the ratio of the signal power divided by the interference power and the noise power. In its most general representation the SINR is given by SINR = S/ (/ + N), where S is the (useful) signal power, / represents the interference power and N is the noise power. Depending on the system employed, at least one among S, I and N may take different forms, so that the expression of SINR may be different for different systems and different transmission technologies. This SINR expression can be given in closed- form and its implementation is very easy. The SINR calculating unit 3031 calculates an effective or approximate received SINR of each user k or the average peruser received SINR, which we denote by SINR. This effective SINR is, for example, a function of the number K of users. In one embodiment of the invention, the channel may be approximated by independent Rayleigh fading and each user k has independent and identically distributed channel statistics. In that case using block diagonalization precoding, the effective average received SINR is approximately given by

SNR (N_T - K + 1) - Q - T⁴ (9) K + SNR {K - 1)^■ Vl - T² + _n ^N_^t ₁ ^■ (1 - T²)^{2 ■} SNR²

wherein Ν is the number of transmit antennas, τ is the quality of channel estimation operation, Q is the number of streams and SNR is the ratio of a transmitted power to a background noise variance.

Then, SINR is entered to the PM Calculator block 3032 and is used to calculate the system performance metric of interest, which is denoted by PM ((SINR), K) . The PM metric is determined by the scheduling/resource allocation policy adopted by the system 500 and can take on the form of sum-rate, weighted sum-rate, fairness-based throughput, QoS-based throughput, weighted queue-aware throughput, etc.). Note that the PM metric is usually a system performance metric rather than an individual/per-user metric. Thus, an output means 304 of the inventive user selection enhancer 301 outputs the optimal number K of users to serve. In one possible embodiment, it would take on the form of the sum-rate, or of a weighted sum-rate:

wherein Wk is a k-t UE real- valued weight and ¾ is the rate of a k-t UE. One option for the rate is the Shannon rate, i.e.

R_k = \og₂ (l + SINR_k) (11)

One other option is the rate achieved by using a specific set of modulation and coding schemes. It would be obtained by indexing a LUT in the PM-storage 306. In one possible embodiment, it would be a fairness-based throughput, or a QoS-based throughput. In one possible embodiment, it would be a weighted, queue-aware throughput.

Results for the following two cases are provided: a) using (Shannon) rate formula under ideal link adaptation and Gaussian signalling according to equation (11) and b) quantizing the Shannon rate using a certain modulation and coding scheme. For evaluation and demonstration purposes, we assess the performance of one embodiment of this invention, in which the optimal number K of users to serve is the one that maximizes the expected sum-rate

K^* = org max PM ((SINR), tf) (¹²) wherein the PM metric PM((SINR_k,K)) is determined on the following form of SINR, i.e.

Once the number T is determined, there are at least two different implementation forms of the inventive system 500. The above method of obtaining K is one example on how to use the method in a particular system and does neither limit nor constrain the applicability and generality of the method.

According to Fig. 5 a possible configuration of the inventive system 500 is shown.

Therein, the optimal number K of users to serve retrieved from the user selection enhancing unit 300 is reported as an additional input to a scheduling unit 501 as a means to indicate how many users K should be scheduled. The user selection enhancing unit 300 operates as a new controller linked to the scheduling unit 501 to boost the performance of the system 500.

The optimal number K of users to serve retrieved from the user selection enhancing unit 300 provided by the proposed invention can be directly incorporated in all user selection algorithms and can work with any user selection algorithm and for any performance metric (objective function). For instance, if GUS or SUS techniques are employed in the system, the information provided by the proposed block enhances (i.e. it adds "intelligence" to) these methods by indicating when the user selection procedure should be terminated. In fact, the proposed invention not only succeeds to improve the performance of existing state-of-the-art user selection methods, but also may accelerate the user selection process by terminating it when the best number of users is attained.

As said above, the inventive method 100, 200 is a general process and is not tailored to only work for maximizing the sum rate. The scheduling unit 501 is now capable of not only maximizing the weighted sum-rate but is also capable of scheduling the users according to their priorities or to the status of their queues/buffers, e.g. the users with the largest queue length or largest delay, etc.. Scheduling of the users by the scheduling unit 501 may be done by selecting a selection policy provided by at least one of selection value inputs 503. A more detailed description of the selection policies used in the system 500 will be given in the following with reference to Fig. 6. The invention finds in a simple and efficient manner the best balance between multiplexing gain, i.e. serving more users and per-user rate, affected by the uncanceled inter-user interference. Once this operating point is determined, i.e. once the optimal number K of users is found, the scheduling unit 501 can select the K users that would satisfy the performance metric, e.g. the K users with highest delay, or largest queue length, etc. Serving more or less than T is a simple design choice of the scheduling unit 501, however the system 500 will be dominated by inter-user interference and the user rates will be low affecting thus the queue length, delay, etc. as most probably the packets will not be decoded and retransmissions will be required. As in current mobile communication standards (e.g. Evolved Packet System in Long-Term Evolution (LTE) and in 4G systems), the Scheduler is tightly interacting with link adaptation and the HARQ process, the system 500 further includes an HARQ unit 502. The HARQ Unit 502 is responsible of handling the HARQ functionalities of all UEs, i.e. handles transmission errors and retransmissions. The HARQ unit 502 processes received status messages (e.g. acknowledgement from the UE whether the transmission was successful or not) and provides status reports from HARQ processes as input to the Scheduling Unit, so that the latter can determine if either a new transmission or a retransmission should be made.

In Fig. 6 another embodiment of the inventive system 600 is shown. Here the user selection procedure can be incorporated into an enhanced scheduling unit 501 ' that combines the user selection enhancing unit 300 and the scheduling unit 501 according to Fig. 5. The system 600 also includes a HARQ Unit 502. The HARQ Unit 502 is identical to that described in relation to the system 500 and will not be described further. Thus, the enhanced scheduling unit 501 ' is composed by two main units. First, a user selection enhancing unit 300 provides the cardinality K of the optimum user subset. Second, a user selector 504 outputs the set of users S of the K users to serve according to some selection policy. The selection policy might be provided by selection value input 503.

In one embodiment of this invention, the policy is a random selection K of K users to serve. No additional information is provided as input to the proposed block.

In another embodiment of this invention, the policy is the selection K of K users with the highest CQI metric. The CQI information for the users has to be provided as input 503 to the user selector 504.

In another embodiment of this invention, the policy is the selection K of K users with the highest priorities. Priority-based information for the users has to be provided as input 503 to the user selector 504.

In another embodiment of this invention, the policy is the selection K of K users with the most/least demanding QoS requirement. The QoS requirement information for the users has to be provided as input 503 to the user selector 504. In one embodiment of this invention, the policy is the selection K of K users according to a suitable metric to satisfy the performance requirements defined by the Scheduler/Resource Allocation block. The suitable metric can be different for different systems and applications and includes, but is not limited to a sum-rate, aggregate throughput weighted sum-rate, and fairness-based utility function. In this context fairness indicates whether users or applications are allocated a fair share of system resources (e.g. data rate or access to the medium).

Finally, it is highlighted that the proposed invention also constitutes a search space reduction algorithm for high-complexity and/or combinatorial schedulers.

One of the major strengths of the proposed invention is the extremely low complexity. In fact, computing the optimum number K requires in most cases a single computation involving at worst an N-dimensional vector (or even with lower dimension if some previous knowledge of the range of optimal numbers of served users is available). Neither iterative algorithms or exhaustive search are required, nor matrix inversions to calculate the precoders of selected users and decide on the optimal number K . The proposed idea aims at deriving first an approximate yet efficient closed-form expression on the effective/average SINR based on system operating values 302 and then finding the number K of users that maximizes a performance metric PM, which is usually a function of this SINR and can also be expressed in closed-form.

Specifically, to find K , the real-valued scalar SINR may be calculated, for instance on the basis of equation (12), using basic arithmetic operations (additions and multiplications) of Νχ, τ, SNR and Q. Then, the objective function PM is calculated, which is a simple function of the scalar SINR or of the scalars SINR and K, and the value of K maximizing this real-valued scalar objective function is obtained using a simple discrete optimization method.

Considering the greedy user selection GUS, the first user is selected as the one having the highest channel gain or rate. Then, from the remaining unselected users, GUS finds the user that provides the highest total throughput together with the previously selected users. The precoders in the precoding unit 701 are calculated at each step for each potential user group. The algorithm terminates when K users are selected or the total throughput does not increase by adding more users. This GUS algorithm needs to search over no more than NK user sets, i.e. the complexity is reduced compared to the exhaustive search method. Nevertheless, computationally demanding matrix operations are required. The proposed invention enjoys very low complexity for two major reasons. First, only basic arithmetic operations (additions and multiplications) on real-valued scalar quantities are required. Finding the number K that maximizes the real-valued scalar objective function is also standard and of low complexity. No vector operations (projections/multiplications, norm calculations) are required as in the case of GUS, SUS, exhaustive search, etc. Note also that the complexity does not result in performance degradation. Our method can even outperform GUS in certain operating regimes.

Second, in prior solutions, several operations (often complex, such as matrix inversion) are required at each step of the algorithm. In the inventive method 100, 200, the best number K of users will be the same and will not be recalculated, as long as the slowly-varying system operating values 302 remain the same.

In Fig. 7 an exemplary MU-MIMO system 700 is shown in which the inventive solution is embedded. Therein, K numbers of users have UE with a number N_R of receiving antennas. A BS is shown that has a number Νχ of transmitting antennas. The enhanced scheduling unit 501 ' provides data streams to selected users S to a precoding unit 701. While figure 7 shows the enhanced scheduling unit 501 ' embedded in the MU- MIMO system 700, it will be clear to the skilled person that also of the user selection enhancing unit 300 in combination with the scheduling unit 501 as described in Fig. 5 can be used as well.

In the following, a performance evaluation is discussed in which the simulation results according to Fig. 8 to Fig. 11 are described. For evaluation purposes, we consider the throughput performance of the invention considering a downlink MU-MIMO system 500 operating in TDD mode with bandwidth of B = 20 MHz, N_T = 64, N_R = 2 and a total number N = 40 of available UEs. Block diagonalization precoding is employed as downlink transmission scheme and Q = 2 streams are sent to each UE. The wireless channel between the k-th UE and the BS, described by H_K is comprised by independent and identically distributed zero-mean and unit- variance complex random variables. In addition, channels between UEs and BS are assumed to be independent. Lastly, the channel estimation error of Ηκ is modelled as described in equation (3). The optimal number K of users in the system 500 is calculated under the assumptions of equations (9) to (13).

In Fig. 8 to Fig. 11 the performance of the proposed invention is compared with the high-complexity greedy user selection (GUS) and the main SUS method which serves as a baseline. The effect of the proposed user selection enhancement in standard, low- complexity user selection techniques, namely norm-based user selection (NUS) and random user selection (RUS), is also shown. The enhanced NUS and RUS, enhanced according to the proposed invention are indicated as eNUS and eRUS respectively. The main purpose showing the performance of eNUS and eRUS is to demonstrate that the proposed enhancement technique still shows significant gains even if combined with very simple user selection methods such as NUS and RUS.

According to Fig. 8a and Fig. 8b a first throughput performance comparison of different user selection techniques with and without user selection enhancement having τ = 0.6 is shown. The performance of the enhanced algorithms according to the invention is denoted by eSUS, eNUS, and eRUS, respectively. The Shannon rate is plotted in Fig. 8a, wherein the quantized rate with MCS is plotted in Fig. 8b.

According to Fig. 9a and Fig. 9b a second throughput performance comparison of the invention, combined with norm-based selection, against state-of-the-art solutions for τ = 0.6 is shown. The Shannon rate is plotted in Fig. 9a, wherein the quantized rate with MCS is plotted in Fig. 9b. The inventive concept substantially improves performance with respect to the SUS method for most of the SNR values. Remarkably, it performs better than GUS at high SNR, and close to it in the rest. It should be noted that this is achieved with much less computational complexity.

According to Fig. 10a and Fig. 10b a third throughput performance comparison of the invention, combined with norm-based selection, against state-of-the-art solutions for τ = 0.6 is shown. The Shannon rate is plotted in Fig. 10a, wherein the quantized rate with MCS is plotted in Fig. 10b. To highlight the importance of scheduling the right number of users, in Fig. 10a and Fig. 10b the invention is combined with random user selection in the User Selection block and is denoted as enhanced RUS (eRUS). The obtained throughput is still notably higher than with SUS, and beats GUS again at high SNR. Note that RUS can represent the performance of the system when the PM metric does not take into account the user/system throughput. Interestingly, selecting users based on other metrics than sum-rate but using the right number of users K may provide higher throughput than complex scheduling schemes (SUS and GUS) that aims at maximizing the sum-rate. This showcases the cardinal importance of serving the right number of users for a given system configuration.

According to Fig. 11a and Fig. l ib a fourth throughput performance comparison of the invention, combined with norm-based selection, against state-of-the-art solutions for T = 0.8 is shown. The Shannon rate is plotted in Fig. 11a, wherein the quantized rate with MCS is plotted in Fig. 1 lb. Even though gains are now smaller, improvements are still achieved, especially at high SNR. Note that the higher the CE/CSI error (lower τ), the more pronounced the gains of the invention are.

The main advantage of the proposed invention is its enhancement of the performance of MU-MIMO wireless communication systems 500, 600, 700 employing a very low-complexity method for determining in advance the cardinality of the best selected number K of users K based on system operating values 302. The proposed invention improves the state-of-the-art solutions both in terms of complexity and performance. As shown, the invention improves the sum-rate performance of a MU- MIMO downlink transmission as compared to SUS in all SNR ranges and having lower complexity. Remarkably, it even outperforms the high complexity GUS at moderate to high SNR values. Without the inventive concept, the system 500, 600, 700 will operate in a suboptimal point, meaning not in the right point balancing multiplexing gain and peruser link rate that would result in lower system performance, especially the throughput performance.

Furthermore, the inventive approach is fully scalable, in terms of computational complexity, with the number of antennas at both BS and UE, the number of active users K and the number of streams Q sent to each UE, since it only requires the computation of a one-dimensional variable SINR, which can be derived in closed-form. The optimizer 305 of the expected performance is a simple integer optimization problem. In this sense, the advantages brought over state-of-the-art solutions, especially those involving matrix inversions and recursive operations, are evident.

Finally, the output K of the user selection enhancing unit 300 is provided as input value to the scheduling unit 501, 501 ' and can work in conjunction with existing user selection algorithms, e.g., GUS and SUS. In that case, incorporating the value of K in the latter block can accelerate its operation as it will terminate the complex sequential algorithms when the optimal number of users is attained. Note also that serving less or more users than K will normally result in lower performance.

The present invention has been described in conjunction with various embodiments as examples as well as implementations. However, other variations can be understood and effected by those persons skilled in the art and practicing the claimed invention, from the studies of the drawings, this disclosure and the independent claims. In the claims as well as in the description the word "comprising" does not exclude other elements or steps and the indefinite article "a" or "an" does not exclude a plurality. A single element or other unit may fulfill the functions of several entities or items recited in the claims. The mere fact that certain measures are recited in the mutual different dependent claims does not indicate that a combination of these measures cannot be used in an advantageous implementation.

Reference signs

100, 200 Method for enhanced user selection

101 Determining step

102 Calculating step

103 Providing step

204 Calculating step

300 User selection enhancing unit

301 Input means

302 System operating value

303 Calculation means

3031 SINR Calculator

3032 PM Calculator

304 Output means

305 Optimizer

306 Look-Up-Table

500, 600, 700 MU-MIMO communication system

501 Scheduling unit

501 ' Enhanced Scheduling Unit

502 HARQ Unit

503 Selection value input

504 User Selector, input means

701 Precoding unit

* Optimal number of UE's to serve simultaneously

K Maximum number of UE's to serve simultaneously

UE User's equipment

s Selected User

NT Number of transmitting antennas

NR Number of receiving antennas

Q Number of streams

τ Channel estimation error

PM System Performance Metric of Interest function w Real-valued weight of UE

R Rate of UE

Claims

Claims What we claim is:

1. A method (100) for enhanced user selection in a multi-user-multi-input-multi- output, MU-MIMO communication system (500), the method comprises the following steps: determining (101) at least one system operating value (302) that indicates the actual systems performance; calculating (102) a number (K*) of user's equipments, UE, to serve simultaneously in the system on the basis of a signal-to-interference-plus-noise-ratio, SINR, value, wherein the SINR value is determined based on the at least one system operating value (302); and providing (103) the calculated number (K*) of UEs to serve to a scheduling unit (501 , 501 ') for selecting at least one user (S).

2. The method (100) according to claim 1 , wherein the at least one system operating value (302) includes the number (Νχ) of transmitting antennas and/or the number (NR) receive antennas and/or the number of streams (Q) to send to each UE.

3. The method (100) according to claim 1 or 2, wherein the at least one system operating value (302) further includes the operating signal- to-noise ratio, SNR, and/or a channel estimation error (τ) and/or a channel state information, CSI, and/or a CSI error.

4. The method (100) according to one of the previous claims, wherein the at least one system operating value (302) is a real scalar value.

5. The method (100) according to one of the previous claims, further comprising: calculating (204) a system performance metric of interest (PM), wherein the number (K*) of UE to serve is further calculated on the basis of the system performance metrics of interest function (PM).

6. The method (100) according to one of the previous claims, wherein the number (K*) of UEs is calculated by following formula:

K^* = argmaxt_f PM( SINR_k, K), wherein K is the number of UE's to serve simultaneously, K is the maximum possible number of UE to serve and PM is a calculated performance metric of interest function (PM).

7. The method (100) according to claim 6, wherein the expression PM(SINR_k,K) is calculated by following formula: PM(SINR_k, K) =∑£₌₁ w_k ^■ R_k ^■ _SINR_k) wherein Wk is a k-t UE real-valued weight and ¾ is the rate of a k-t UE.

8. The method (100) according to claim 6, wherein the expression PM(SINR_k,K) is retrieved from a look-up-table (306) indicating specific set of modulation and coding.

9. The method (100) according to one of the previous claims 6 to 8, wherein the SINR_k(K) value is an effective or approximate received SINR value of a k-t UE as a function of K.

10. The method (100) according to one of the previous claims 6 to 9, wherein K is the maximum possible number of UEs to serve.

1 1. A user selection enhancing unit (300) in a multi-user-multi-input-multi-output, MU-MIMO, communication system (500), the user selection enhancing unit (300) comprises: an input means (301) configured to receive at least one system operating value (302) that indicates the actual systems performance; a calculation means (303) configured to calculate a number (K*) of user's equipments, UE, to serve simultaneously in the system (500) on the basis of a signal-to- interference-plus-noise-ratio, SINR, value, wherein the SINR value is determined based on the at least one system operating value (302); and an output means (304) configured to provide the calculated number (K*) of UEs to serve to a scheduling unit (501 , 501 ') of the MU-MIMO system (500) for selecting at least one user (S).

12. The user selection enhancing unit (300) according to claim 1 1 , wherein the calculation means (303) is further configured to calculate (204) a system performance metric of interest, wherein the number (K ) of UE's to serve is further calculated on the basis of the system performance metrics of interest function (PM).

13. An enhanced scheduling unit (501 ') in a multi-user-multi-input-multi-output, MU-MIMO, communication system (500), the enhanced scheduling unit (501 ') comprises: a user selection enhancing unit (300) according to one of the claims 10 to 12 configured to provide the calculated number (K ) of UE's to serve; an input means (504) configured to receive at least a selection value (503) that indicates the actual systems performance.

14. A multi-user-multi-input-multi-output, MU-MIMO, communication system (500) at least comprising: a user selection enhancing unit (300) according to one of the claims 1 1 to 12; and a scheduling unit (501) for selecting at least one user (S).

15. Computer program product for implementing, when carried out on a computing device, a method (100) for enhanced user selection according to one of the claims 1 to 10.