A common approach to obtain channel state information for massive MIMO networks is to use the same orthogonal training sequences in each cell. We call this the full-pilot reuse (FPR) scheme. In this paper, we study an alternative approach where each cell uses different sets of orthogonal pilot (DOP) sequences. Considering uplink communications with matched filter (MF) receivers, we first derive the SINR in the large system regime where the number of antennas at the base station, the number of users in each cell, and training duration grow large with fixed ratios. For tractability in the analysis, the orthogonal pilots are drawn from Haar distributed random unitary matrices. The resulting expression is simple and easy to compute. As shown by the numerical simulations, the asymptotic SINR approximates the finite-size systems accurately. Secondly, we derive the user capacity of the DOP scheme under a simple power control and show that it is generally better than that of the FPR scheme.

In this paper, we study feedback optimization problems that maximize the users' signal to interference plus noise ratio (SINR) in a two-cell multiple-input multiple-output broadcast channel. Assuming the users learn their direct and interfering channels perfectly, they can feed back this information to the base stations (BSs) over the uplink channels. The BSs then use the channel information to design their transmission scheme. Two types of feedback are considered: 1) analog and 2) digital. In the analog feedback case, the users send their unquantized and uncoded channel state information (CSI) over the uplink channels. In this context, given a user's fixed transmit power, we investigate how he/she should optimally allocate it to feed back the direct and interfering (or cross) CSI for two types of BS cooperation schemes, namely, multicell processing (MCP) and coordinated beamforming. In the digital feedback case, the direct and cross link channel vectors of each user are quantized separately, each using the random vector quantization scheme, with different size codebooks. The users then send the index of the quantization vector in the corresponding codebook to the BSs. Similar to the feedback optimization problem for the analog feedback, we investigate the optimal bit partitioning for the direct and interfering link for both types of cooperation. We focus on regularized channel inversion precoding structures and perform our analysis in the large system limit in which the number of users per cell (K) and the number of antennas per BS (N) tend to infinity with their ratio β = (K/N) held fixed. We show that for both types of cooperation, for some values of interfering channel gain, usually at low values, no cooperation between the BSs is preferred. This is because, for these values of cross channel gain, the channel estimates for the cross link are not accurate enough for their knowledge to contribute to improving the SINR and there is no benefit in doing BS cooperation under that condition. We also show that for the MCP scheme, unlike in the perfect CSI case, the SINR improves only when the interfering channel gain is above a certain threshold.

We consider a multiuser system where a single transmitter equipped with multiple antennas (the base station) communicates with multiple users each with a single antenna. Regularized channel inversion is employed as the precoding strategy at the base station. Within this scenario we are interested in the problems of power allocation and user admission control so as to maximize the system throughput, i.e., which users should we communicate with and what power should we use for each of the admitted users so as to get the highest sum rate. This is in general a very difficult problem but we do two things to allow some progress to be made. Firstly we consider the large system regime where the number of antennas at the base station is large along with the number of users. Secondly we cluster the downlink path gains of users into a finite number of groups. By doing this we are able to show that the optimal power allocation under an average transmit power constraint follows the well-known water filling scheme. We also investigate the user admission problem which reduces in the large system regime to optimization of the user loading in the system.

Channel state information (CSI) at the base stations (BSs) can significantly increase the spectral efficiency in single and multi-cell broadcast channels. Assuming the users learn their direct and interfering channels, they can feed back this information to the BSs over the uplink. The BSs then form channel estimates which they use to design their transmission scheme. Clearly, the quality of these estimates affects system performance. In this paper, we study limited feedback in a two-cell MIMO broadcast channel. For a fast transfer of CSI, we consider the analog feedback scheme where the users send their unquantized and uncoded CSI over the uplink channels. In this context, given a fixed user's transmit power, we investigate how a user should optimally allocate this power to feed back the direct and interfering CSI for two types of base station cooperation schemes, namely, network MIMO and coordinated beamforming. We focus on regularized channel inversion precoding structures and perform our analysis in the large systems limit in which the number of users per cell (K) and the number of antennas per BS (N) tend to infinity with their ratio β = K/N held fixed.

In a multicell network, the quality of the channel state knowledge at the base stations (BSs) affects system performance. When this knowledge is acquired through a quantized feedback scheme, its quality is roughly determined by the number of feedback bits. In this paper we investigate feedback optimization problems for the quantized feedback scheme via random vector quantization (RVQ) in a two-cell MIMO broadcast channel. Assuming each user knows its direct and interfering channel states perfectly, we address the following question: given a total number of feedback bits, how should a user allocate those bits to quantize its direct and cross channel information? We pose that question for different levels of BS cooperation, namely Network MIMO and Coordinated Beamforming (CBf). We focus on regularized zero-forcing (RZF) precoding structures and derive our results in the large system limit where the number of users and antennas per BS tend to infinity with their ratio being fixed. We show that for both cooperation schemes, each user should allocate more bits to quantize the cross channel as the latter's gain (ϵ) increases. We also show that, contrary to the full channel knowledge case, under this limited feedback scheme, for some values of (ϵ) no cooperation between the BSs is preferred.

In this paper, we consider a large system analysis of regularized channel inversion (RCI) beamforming for MISO broadcast channels (BC) with transmit-side channel correlation. In the analysis, we assume that both the number of users and transmit antennas grow unbounded with a constant ratio. We also assume that the channel correlation model is separable. Under this channel condition, we are particularly interested to find the optimal regularization parameter of RCI that maximizes the signal to interference plus noise ratio (SINR). First, we derive the large system limit of the SINR of each user by applying some results on large random matrices. Then, we determine the cor- responding optimal regularization parameter and show that this optimal regularization parameter is not affected by the transmit correlation. We verify this result through simulations where the channel has the exponential transmit-correlation model.