Network Assisted Inter-cell Codeword Cancellation for Interference-limited LTE-A and Beyond

Transcription

1 Zhou, G.; Xu, W.; Bauch, G. : Network assisted inter-cell codeword cancellation for interference-limited LTE-A and beyond. IEEE Wireless Communications and Networking Conference (WCNC), Istanbul, Turkey, April 6-9, IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting / republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to server or lists, or reuse of any copyrighted component of this work in other works.

2 Network Assisted Inter-cell Codeword Cancellation for Interference-limited LTE-A and Beyond Guangxia Zhou Technische Universität Hamburg-Harburg D-1073, Hamburg, Germany Wen Xu Intel Mobile Communications D-85579, Neubiberg, Germany Gerhard Bauch Technische Universität Hamburg-Harburg D-1073, Hamburg, Germany Abstract Network assisted interference cancellation and suppression (NAICS) is being studied for 3GPP LTE-A to mitigate interference in multi-cell networks. To accomplish this goal, various interference mitigation techniques have been proposed. Contrary to conventional techniques, a rate splitting technique splits the interference in a way that part of the interference can be canceled at users through base station coordination. Rate splitting has been shown to have potential to achieve the channel capacity. This paper studies the benefits of the rate splitting scheme and proposes a new inter-cell codeword cancellation scheme for 3GPP LTE-A and beyond. Simulation results for interference-limited LTE-A scenarios show that the inter-cell codeword cancellation can be a very attractive interference coordination scheme due to its low complexity, low overhead and superior performance. Index Terms MIMO-OFDM systems, interference mitigation, NAICS, SIC/PIC, rate splitting, 3GPP LTE/LTE-Advanced. I. INTRODUCTION Current state-of-the-art wireless systems, e.g., the Third Generation Partnership Project (3GPP) Long Term Evolution Advanced (LTE-A) [1], are expected to utilize frequency spectrum efficiently. To meet this demand, 3GPP LTE-A aims at providing more flexible frequency spectrum management, e.g., carrier aggregation, heterogeneous networks. The fundamental of this flexibility is to adopt a tight frequency reuse to increase spectral efficiency in a multi-cell network. However, this leads to strong inter-cell interference on cell edges and severe performance degradation. Hence, network assisted interference cancellation and suppression (NAICS) has become a hot topic for 3GPP LTE-A release 1 [] [4] to improve the performance of cell-edge user equipments (UEs). Han-Kobayashi (HK) rate splitting [5] has been shown to potentially achieve the interference channel capacity in the information theory community. In [6], a simple HK rate splitting strategy is shown to achieve the capacity region of a two-user Gaussian interference channel to within one bit. In this strategy, each BS splits its information into two substreams: common and private streams. The common stream is decoded at both users, while the private stream is decoded only at the intended user. In light of decoding and subtracting the common stream from an interfering base station (BS), the user cancel part of the interference and accordingly obtains significantly capacity gains. Inspired by HK rate splitting strategy, in which decoding part of interference can improve the performance, this paper proposes an inter-cell codeword (CW) cancellation scheme for 3GPP LTE-A and beyond. This inter-cell CW cancellation is based on an assumption that the spatially multiplexed streams are separately coded and modulated before the spatial multiplexing, which is often called multiple-codeword (multi-cw) multiple-input multiple-output (MIMO) transmission. Each coded stream is referred to as a CW. In this way, a part of CWs from an interfering BS can be decoded and cancelled in the cancellation process. LTE-A spatial multiplexing supports a maximum of two CWs and eight-layer MIMO [1]. We can treat one CW as a common CW and the other as a private CW. Then, with the aid of the inter-cell CW cancellation scheme, the cell-edge user s performance can potentially be improved in interference-limited LTE-A networks. This paper studies the benefits of the inter-cell CW cancellation scheme and makes three novel contributions. First, an effective and practical inter-cell CW cancellation scheme is proposed for LTE-A and beyond. Second, a CW-tolayer mapping procedure is designed for the inter-cell CW cancellation. The current LTE-A spatial multiplexing CW-tolayer mapping is suitable for single-user MIMO, but it needs to be extended to support our proposed scheme. In this paper, a new CW-to-layer mapping and the corresponding selection method are investigated. Third, we identify the requirements of control signaling for the inter-cell CW cancellation. The current LTE-A does not support this type of inter-cell CW decoding. Thus, we propose a signaling procedure to deal with this situation. In addition, four potential solutions are proposed to deal with the security/privacy issues caused by inter-cell CW cancellation. II. SYSTEM MODEL Without loss of generality, a universal frequency reuse cellular network is considered here. Two UEs located at cell edge areas, such as UE 1 and UE, receive interference from nearby BSs. This scenario can be modeled as a two-user interference channel. Namely, BS 1 and BS serve UE 1 and UE, respectively. At the same time, BS 1 causes interference to UE, and vice versa. Both BSs use LTE physical downlink shared channels (PDSCHs) for data transmission. Each BS and each UE are equipped with N t transmit antennas and N r receive antennas, respectively. Both BSs support two CWs transmission and the corresponding baseband transmission chains are illustrated in Fig. 1(a). Let CW q i denote the q-th CW transmitted from BS i. For each CW q i, a stream of i.i.d information bits are coded using turbo coding (not necessarily with turbo codes) and scrambled. The block of scrambled bits is then modulated into a block d q i Q [1 N q symb,i ] of /14/$ IEEE 5

3 modulation symbols, where N q symb,i is the number of modulation symbols of CW q i and Q denotes an M-ary QAM set. Afterwards, the modulation symbols are partitioned into N layer symb,i blocks x q i (k) = [x1 i (k),..., xvq i i (k)] T of v q i modulation symbols, where v q i is the number of layers reserved for CW q layer i and Nsymb,i is the number of modulation symbols per layer. For simplicity, an identity matrix is taken as a precoding matrix and thus each layer is directly mapped onto an antenna port. Accordingly, we have vi 1 + v i = N t. Finally, both blocks x 1 i (k) and x i (k) are mapped onto a transmit vector x i (k) which corresponds to all available transmission layers. The index k = 1,..., N layer symb,i specifies the resource element (RE) used for downlink transmission. (a) Transmitter structure. (b) Receiver structure (take UE 1 as an example). Figure 1: Transmitter and receiver structures for inter-cell codeword cancellation scheme in an interference channel For simplicity, we assume that the BSs are timesynchronized, such as in the case of the 3GPP time division duplexing LTE-A (TD-LTE-A) or synchronous frequency division duplexing LTE-A (FDD LTE-A). In this case, the signals at UEs become synchronous and quasi-synchronous such that the received signals at a given RE k can be expressed as r 1 (k) = H 11 (k)x 1 (k) + H 1 (k)x (k) + n 1 (k), r (k) = H 1 (k)x 1 (k) + H (k)x (k) + n (k), where H ij (k) denotes an N r N t Rayleigh fading matrix from BS j to UE i, whose entries are i.i.d. complex Gaussian distribution with CN (0, σij ). The average energy per transmit symbol x i (k) is E s and the transmit power is uniformly distributed over the transmit antennas (Φ xx = E s /N t I Nt ), where I m denotes an m m identify matrix. For CW q i, the average energy per transmit symbol x q i (k) is vq i /N t E s. (1) n i is the vector of additive complex white Gaussian noise of zero mean and covariance matrix Φ nn = σni Nr. In addition, the signal-to-noise ratio (SNR) of UE i is given by SNR i = (σii E s)/(n t σn), the interference-to-noise ratio (INR) of UE i is given by INR i = (σij E s)/(n t σn) for j i and the corresponding signal-to-interference ratio (SIR) is given by SIR i = σii /σ ij. The RE index k will be ignored in the following for ease of notation. III. INTER-CELL CODEWORD CANCELLATION SCHEME For inter-cell CW cancellation, the two CWs transmitted from a BS can be classified into two types based on a decoding strategy. If a CW is decoded at both UEs, then the CW is labelled as a common CW; otherwise it is labelled as a private CW. In general, this leads to three decoding strategies: 1) If both CWs are labelled as common CWs, then they should be decoded at both UEs. ) If one CW is common and the other is private, then only the common CW should be decoded at both UEs. 3) If both CWs are labelled as private CWs, then they should be decoded only at the intended UE. Fig. 1(b) illustrates an iterative multiple CWs detectiondecoding structure in UE 1. A MIMO detector is used to decouple the transmitted layers within the same CW l i and to calculate log likelihood ratios (LLRs) L(CW l i), where i denotes a BS index and l denotes a layer index. Then the LLRs are mapped to the corresponding CWs. Each CW is descrambled and fed to an a posteriori probability (APP) decoder, e.g., turbo decoder, which produces LLRs of information and coded bits. Afterwards, the LLRs of coded bits are fed back to refine the LLRs calculation of the MIMO detector in the next iteration. The MIMO detector iteratively exchanges information with the APP decoder in a turbo-type manner. After several iterations, the LLRs of information bits of the intended CWs are good enough and the APP decoder outputs hard decisions of the information bits in the final iteration. For simplicity, we assume that both BSs have two transmit antennas (N t = ). In each BS, the two CWs are labelled as common CWs and tied to two transmit antennas, respectively. Different from the other type of receiver, which treats the interference as noise, this receiver attempts to decode the CWs from the interfering BS. The receiver can decide and decode none, one or two CWs from the interfering BS depending on certain criteria, which will be introduced later. In the MIMO detector, either a maximum-likelihood (ML) joint detection or an MMSE parallel interference cancellation (PIC) can be used. Even for a small number of transmit antennas (say N t = 4 in this case), the exact computation of LLRs entails prohibitively high computational complexity. Thus, various low-complexity detection algorithms have been investigated in the literature, e.g., [7]. In this paper, a MMSE-PIC detector is employed to introduce the concept of the inter-cell CW cancellation. The MMSE-PIC detector adopts the algorithm described in [8], which is well-suited for an efficient implementation in hardware. The general detection-decoding procedure is referred to as the low-complexity MMSE-PIC algorithm. 53

4 At a given RE, the received signal r 1 can be rewritten as r 1 = H x + n 1, () where H = [H 11 H 1 ] C [Nr Nt] and x = [x T 1 x T ] T Q [N t 1]. Based on this channel model, a MMSE-PIC detection can be divided into five parts. 1) Computation of Soft-symbols: The so-called softsymbols x i, i = 1,..., N t, are generated by x i = E[x i ] = M m=1 Pr[c j = b j (s m )] (3) s m N b j=1 with N b = log M, s m being the elements of Q and b j (s m ) denoting the value of the j-th bit corresponding to the m-th modulate symbol s m. The reliability of each soft-symbol is characterized by its variance E i = Var[x i ] = E[ e i ] (4) with e i = x i x i. The a-priori probabilities involved in the computation of the soft-symbols in Eq. (3) and their variances in Eq. (4) are calculated based on the a-priori LLRs L A (c ij ) delivered by a channel decoder, and is given by Pr[c j = b j (s m )] = 1 ( 1 + b j (s m ) tanh( 1 ) L A(c ij )). (5) ) Parallel Interference Cancellation: In each layer i, the interference caused by the other layers j i is cancelled from the receive signal r 1 as follows r 1 = r 1 j i h j x j = h i x i + ñ i (6) with ñ i = j i h je j + n 1 corresponding to the residual noise-plus-interference (NPI). 3) MMSE Filter Computation: A linear MMSE filter is used to decouple the layers and to reduce the NPI. A MMSE filter for layer i is given by [8] w H i = φ H i H H (7) where φ H i is the i-th row of Φ 1 = (GΛ + σni Nt ) 1. G = H H H is a Gram matrix and Λ is a N t N t diagonal matrix have elements Λ ii = E s /N t, i. 4) MMSE Filtering: Multiplying the received signal with the MMSE filter w H i yields z i = w H i r 1 = μ i x i + w H i ñ i (8) with μ i = φ H i g i. g i is the j-th column of G. Thereby, N t MMSE filters are performed in parallel to generate the LLRs. 5) LLR Computation: Based on the results in Eq. (8) and the max-log approximation, the extrinsic LLRs can be calculated separately for each layer i L E (c ij ) max s Q +1 i Ξ max s Q 1 i Ξ (9) with Ξ = ρ z i s + 1 Nb n=1 b n(s)l A (c in ), where z i = μ 1 i z i and Q ±1 denotes the set of symbols with corresponding bit c ij = ±1. IV. PROPOSED CODEWORD-TO-LAYER MAPPING In the case of a pure multi-cw MIMO scheme where the number of CWs equals the number of MIMO transmission layers, a one-to-one mapping between CWs and layers is used in a straightforward manner and a different modulation and coding scheme (MCS) can be applied to each layer. Generally, the inter-cell CW cancellation is able to decode and cancel a part or all layers/cws of the interfering BS. 3GPP LTE-A follows a multi-cw MIMO scheme with two CWs and a maximum of eight transmission layers. That is, each CW is tied to a fixed number of transmission layers. Therefore, the CW-to-layer mapping needs to be considered carefully for inter-cell CW cancellation. Table in [9] shows available layer mapping configurations of two CWs defined in LTE/LTE-A. We can see that the layer mapping is based on a round robin manner and the transmission layers are symmetrically mapped onto the two CWs as good as possible. An advantage of the symmetric layer mapping is that each CW has the same diversity order and is more robust against layer channel quality fluctuations. The inter-cell CW cancellation can be viewed as an application based on HK rate splitting. For simplicity, assuming a one-to-one mapping between CWs and layers, and BSs have infinite transmit antennas N t, we approximate the selection of common CWs by a power allocation strategy of the HK rate splitting. Let P c,i and P p,i denote the power of the common and private sub-streams of BS i, respectively. The ratio of common and private CWs of BS i is given by vi c v p i = P c,i P p,i, (10) where vi c and vp i denote the numbers of common and private CWs, respectively, and vi c+vi i = N t. For a fixed finite number of transmit antennas, we propose that the numbers of common and private CWs are determined by P c,i vi c = round[ P c,i+p p,i N t ] = round[ν i N t ], v p i = N t vi c, (11) where ν i = P c,i /(P c,i +P p,i ), ν i [0, 1] is defined as a power allocation ratio and round[a] is a function which returns the nearest integer to a. From Eq. (11), the CWs are divided into two types: common and private. For example, if ν i = 0.7 and N t = 6, then we have vi c = 4 common CWs and v p i = private CWs. The method in Eq. (11) can be applied to the CWto-layer mapping for 3GPP LTE-A. There are up to two CWs in an LTE-A downlink transmission chain. That is, a fixed number of transmission layers are mapped to each CWs. We can treat each transmission layer as a CW and use Eq. (11) to calculate the numbers vi c and vp i of common and private transmission layers. Under the assumption that the transmission layers are statistically equivalent (i.e., no layer ordering, etc.), we can map vi c layers to the first CW and v p i layers to the second CWs, or vise versa. For the case of 0 < vi c < N t, that is, one CW is common and 54

5 the other is private, a layer mapping configuration can be chosen by the paraments N t and vi c : vi i. For the cases of vi c = N t and vi c = 0, that is, both CWs are common and private, respectively, a symmetric layer mapping configuration is proposed, which is also defined in LTE-A [9]. In this way, the selection of common CWs depends on a power allocation strategy of the HK rate splitting. In the following, we will focus on introducing two available power allocation algorithms. A. Power Allocation Strategy in a Simple HK Rate Splitting In [6], a suboptimal power allocation strategy is proposed, where the power of the private sub-stream is set such that its INR at the other user s receiver equals the Gaussian noise P p i = 1 INR G P s,i (1) i Following this strategy, the HK rate splitting scheme is shown to achieve the capacity region to within one bit. B. Mutual-Information Based Power Allocation Algorithm Two algorithms have been proposed in the power allocation procedure. The first algorithm [10] is developed based on fairness principle. The algorithm attempts to maximize the sum of the logarithmic single-to-interference-plus-noise ratio (SINR) of each sub-stream of each user. It provides a good compromise between the sum rate and the fairness. The second algorithm [11] is developed to maximize the sum mutual information with specified input constellations, i.e., QPSK and 16-QAM. Since the inputs are taken from known discrete constellations in practical systems, it may significantly depart from the optimal Gaussian distributions. To resolve this problem, the sum mutual information can be formulated by a second-order expansion [1] and a waterfilling algorithm is developed to optimize the power allocation under the constraint that each sub-stream follows a specified input constellation. The details are omitted due to limited space. V. SIGNALING REQUIREMENTS FOR INTER-CELL CODEWORD CANCELLATION The rate splitting has been viewed as a promising method to study the capacity region of the interference channel. It also brings in a novel view of designing a practical approach to mitigate the interference, e.g., for LTE-A and beyond. In this section, we will focus on identifying the main requirements to make the proposed inter-cell CW cancellation implementable in LTE-A. A. Channel Measurement and Reporting The first task to enable the inter-cell CW cancellation is to determine which BSs and UEs should be involved in this operation. Since the decision depends on the channel parameters, i.e., SNR 1, SNR, INR 1 and INR, UEs are required to measure and report the related information. The UEs can obtain the information via reference signal based measurements, for example, reference signal received power (RSRP) and reference signal received quality (RSRQ) in LTE systems. Although the inter-cell CW cancellation does not require channel state information (CSI) feedback, the UE should have an ability to measure the CSI of the interfering BS. Specifically, the LTE downlink reference signals can be employed for UE to measure the interfering BS s CSI. B. Requirements of Radio Resource Control (RRC) A cell-radio network temporary identifier (C-RNTI) provides a unique and temporary UE identification at the cell level, and it is assigned by the network via a RRC control signal when a UE is associated with the cell. In LTE-A downlink transmission, coded bits are scrambled based on C- RNTI and cell s physical layer cell identity (PCI). To decode a CW from the interfering BS, a UE from the serving BS should have an authority to request the C-RNTI and PCI of the other UE associated with the interfering BS. The information can be obtained from the interfering BS, either directly or indirectly via the serving BS. C. Downlink Control Signaling Requirements Assuming a UE have the C-RNTI and PCI of the other UE associated with the interfering BS, the UE should have an ability to decode physical downlink control channel (PDCCH) and obtain downlink control information (DCI) from the interfering BS. Moreover, the DCI specified in [13] does not support the CW-to-layer mapping selection for a given layer number, since there is a fixed CW-to-layer mapping defined in 3GPP LTE-A [9]. For the proposed inter-cell CW cancellation, we propose to select a CW-to-layer mapping based on a power allocation strategy and Eq. (11). Accordingly, DCI should contain the index information of the selected CW-to-layer mapping. For the case of LTE-A spatial multiplexing eightlayer MIMO, the DCI can use bits to indicate the index. D. Example Signaling Procedure Based on the above summarized requirements, we propose a signaling procedure for inter-cell CW cancellation. The procedure is listed as follows: 1) Each UE measures the channels from the serving BS as well as the interfering BS, and feeds back the SNR and INR information to the serving BS. ) The serving BS shares the information with the interfering BS and selects a CW-to-layer mapping based on a power allocation strategy and Eq. (11). The power allocation strategy can be determined by, e.g., a mutual information based power allocation algorithm in [11]. 3) Each UE obtains the C-RNTI and PCI of the other UEs associated with the interfering BS. 4) Each UE obtains DCI from the serving BS as well as the interfering BS by decoding PDCCH. 5) Both BSs send data to the corresponding UEs, and the UEs can decode PDSCH by the inter-cell CW cancellation. This example can be further concretized and extended to meet general requirements of the future LTE-A systems. 55

6 (a) Solution 1: Switch of turbo encoding and scrambling blocks. (b) Solution : Additional encryption. From the above example signaling procedure, each UE requires the C-RNTI of the other UE associated with the interfering BS. Since the C-RNTI provides a unique UE identification at the cell level, C-RNTI sharing may lead to security/privacy issues. To deal with these issues, we propose the following four solutions. These solutions can be easily incorporated into the other NAICS schemes having the similar security/privacy issues. 1) Switch of Turbo Encoding and Scrambling Blocks: From the transmission chain shown in Fig. 1(a), the coded bits are scrambled using a cell-specific sequence based on the UE s C-RNTI. In the receiving end, the coded bits are descrambled and then fed to a channel decoder. In order to avoid the descrambling step, we propose to switch the channel coding and scrambling blocks in the transmission chain shown in Fig. (a). Following this new chain, iterative detectiondecoding of a UE can be realized without using the C-RNTI of the other UE associated with the interfering BS. ) Additional Encryption: Another straightforward solution is to add an additional UE-specific encrypting step before channel encoding as shown in Fig. (b). This encryption uses a new unique specific UE identification, which is not shared with the other UEs. In this case, C-RNTI sharing can not lead to any security/privacy issue. 3) Common C-RNTI: Inspired by the concept of HK rate splitting, the C-RNTI can be divided into two types: common and private C-RNTIs, as shown in Fig.(c). Two different C- RNTIs are assigned to two codewords. The C-RNTIs related to the common and private CWs are referred to as common and private C-RNTIs, respectively. Accordingly, the common C-RNTI is shared with the other UE. In this case, only a part of codewords might have security/privacy issues. 4) C-RNTI Sharing in a Trust Group: In some scenarios, the terminals can completely trust each other and share C- RNTI information. For example, two relays of a two-path halfduplex relay system [14] are deployed by the same operator and provide service to the same UE. The two relays can be added to a trust group, where C-RNTI information is allowed to be shared with each other. Apparently, the above strategies can be combined to provide even better solutions for potential security/privacy issues. Table I: Modulation and coding schemes of CW1 and CW Scheme Layer ratio Modulations Code Rates FR= : 16-QAM, 16-QAM 0.5, 0.5 LMMSE : QPSK, QPSK 0.5, 0.5 CW Cancellation 1:3 QPSK, 16-QAM 0.4, 04 CW Cancellation : QPSK, QPSK 0.5, 0.5 (c) Solution 3: Common and private C-RNTIs. Figure : Potential solutions to security/privacy issues in LTE. E. Security/Privacy Issues VI. SIMULATION RESULTS Monte-Carlo simulations were conducted to evaluate the performance in an LTE-A framework. Simulation parameters follow the evaluation methodology in LTE and LTE-A standards [1], [15]. We assume that there are 4 transmit antennas at both BSs and 4 receive antennas at both UEs. The channel model is the low-correlated 3GPP Extended Vehicular A model with 10 MHz bandwidth. The transmit signals are modulated using Gray-mapped M-ary QAM (M = 4, 16). For the channel coding, the LTE turbo code is used and 8 internal iterations are performed in the turbo decoder. An iterative detection-decoding strategy with a maximum iteration of 8 is employed in the receiver. Two reference schemes are chosen to compare with the proposed scheme. The first scheme is a frequency reuse approach. For simplicity, we consider a fixed frequency reuse (FR) (FR factor = ). Thus only 5 MHz bandwidth is assigned to each UE. The other scheme treats the interference as Gaussian noise. A linear minimum-mean-square-error (LMMSE) multiuser detection (MUD) [7] is chosen as a benchmark case due to its feasibility and effectiveness among various detection techniques which treat interference as noise. Table I lists the modulation and coding schemes (MCSs) used in the simulations. Theoretically, all interference mitigation algorithms with the corresponding MCSs can achieve the same spectrum efficiency. In this simulation setup, both BSs have 4 transmit antennas, thus there are two possible CW-tolayer mapping strategies, namely, 3 : 1 and :, which will be selected by Eq. (11) combined with two power allocation algorithms, namely, the simple HK strategy in Eq. (1) and the mutual information based power allocation, denoted as MI-PA. For simplicity, both UEs have the same SNRs and INRs. Fig. 3 illustrates throughput of interference mitigation schemes for various SIRs. In Fig. 3(a), SIR= 1 db indicates that both UEs suffer from strong interference. It is easy to see that treating interference as noise in this regime leads to a significant performance degradation. Compared to the LMMSE, frequency reuse can be a better scheme to avoid the interference. The inter-cell CW cancellation has the best performance among the investigated schemes, since it can decode and cancel most of the interference in this regime. There is about 4.5 db gain over the frequency reuse in the 90% 56

7 Throughput [bps] 3.5 x Throughput [bps] 3.5 x Throughput [bps] 3.5 x SNR [db] (a) Case 1: SIR = 1 db SNR [db] (b) Case : SIR = 3 db SNR [db] (c) Case : SIR = 5 db. Figure 3: Throughput comparison of interference mitigation schemes. LMMSE treats interference as noise. FR= denotes the FR factor two. Both MI-PA and simple HK use inter-cell CW cancellation. The MI-PA denotes that the mapping is determined by mutual information based power allocation in [11]. The simple HK denotes that the mapping is determined by Eq. (1). throughput region of interest. Fig. 3(b) shows the performance comparison in the case of medium interference (SIR = 3 db). The LMMSE have higher throughput than the frequency reuse. It shows that treating interference as noise is better than avoiding it. The intercell CW cancellation has the best performance. The MI-PA is better than the simple HK strategy, since it determines the power allocation using knowledge of practical constellations. Roughly it can achieve more than 3 db gain over the LMMSE in the 90% throughput region of interest. In Fig. 3(c), frequency reuse is clearly sub-optimal in the weak interference case (SIR = 5 db). The inter-cell CW cancellation has 1 db gain compared to the LMMSE. When interference is weak, the interference can be treated as noise. Based on simulation results shown above, the inter-cell CW cancellation is more robust against the impact of interference than the other schemes. In the strong and medium interference scenarios, it achieves significant improvements at both UEs. The mutual information based power allocation can choose a CW-to-layer strategy more smartly than the simple HK strategy. VII. CONCLUSIONS This paper studied the benefits of the rate splitting scheme and proposed a new inter-cell CW cancellation scheme for LTE-A and beyond. We investigated the challenges and the solutions to adopt a rate splitting scheme to deal with interference. In particular, a CW-to-layer mapping procedure was designed for the inter-cell CW cancellation. The current LTE-A spatial multiplexing CW-to-layer mapping is suitable for single-user MIMO, and needs to be extended to support the proposed scheme. Thus, a new CW-to-layer mapping set and the corresponding selection method were investigated. Two power allocation algorithms, namely, the simple HK strategy and the mutual information based power allocation, were used to determine the CW-to-layer mapping. For control signaling for the inter-cell CW cancellation, a signaling procedure was proposed. Since 3GPP LTE NAICS, e.g., the inter-cell CW cancellation scheme, may lead to security/privacy issues, we proposed four potential solutions to avoid and reduce the impact. Simulation results showed that the inter-cell CW cancellation can achieve the highest throughput among the investigated schemes. It provides a very attractive interference coordination scheme due to its low complexity, low overhead and superior performance. REFERENCES [1] 3GPP TS v9.0.0, Evolved universal terrestrial radio access (E-UTRA); further advancements for E-UTRA physical layer aspects, 010, [] 3GPP TR v0.0.1, Feasibility of CRS interference mitigation for LTE homogenous deployment, 013, [3] MediaTek, Renesas Mobile Europe, and Broadcom, Study on networkassisted interference cancellation and suppression for LTE, 013, RP , 3GPP TSG-RAN. [4] Intel, Phase 1 link-level analysis of candidate IS/IC receivers, 013, R , 3GPP TSG-RAN WG4. [5] T. S. Han and K. Kobayashi, A new achievable rate region for the interference channel, IEEE Trans. Inf. Theory, vol. 7, pp , [6] R. H. Etkin, D. Tse, and H. Wang, Gaussian interference channel capacity to within one bit, IEEE Trans. Inf. Theory, vol. 54, pp , 008. [7] H. Dai, A. F. Molisch, and H. V. Poor, Downlink capacity of interference-limited MIMO systems with joint detection, IEEE Trans. Wireless Commun., vol. 3, pp , 004. [8] C. Studer, S. Fateh, and D Seethaler, ASIC implementation of softinput soft-output MIMO detection using MMSE parallel interference cancellation, IEEE J. Solid-St. Circ., vol. 46, no , 011. [9] 3GPP TS v10.0.0, Evolved universal terrestrial radio access (E-UTRA); physical channels and modulation, 009, [10] G. X. Zhou, G. Bauch, J. Berkmann, and W. Xu, Multi-layer rate splitting scheme for interference mitigation in tri-sectored wireless networks, in Proc. IEEE ICC, Ottawa, Canada, Jun 01. [11] G. X. Zhou, W. Xu, and G. Bauch, A mutual-information based power allocation algorithm for multi-layer rate splitting scheme in tri-sectored wireless networks, in Proc. IEEE GLOBECOM, Anaheim, CA, Dec 01. [1] V. V. Prelov and S. Verdu, Second-order asymptotics of mutual information, IEEE Trans. Inf. Theory, vol. 50, pp , 004. [13] 3GPP TS 36.1 v8.8.0, Evolved universal terrestrial radio access (E- UTRA); multiplexing and channel coding, 009, [14] H. Park and J. Chun, Inter-relay interference cancellation for AF MIMO two-path relay systems, in Proc. IEEE VTC Fall, Quebec, Canada, Sept 01. [15] 3GPP TS v7.1.0, Physical layer aspects for evolved universal terrestrial radio access (UTRA), 006, 57