Abstract: In multi-cell multi-antennas networks with uplink training, the channel estimate at the base station in one cell jammed by users from other cells, the performance of systems suffers tremendous losses due to the use of corrupted pilots. In addition, as the MMSE precoding matrix exists singularity problem when coordinative method is used, The goal is to investigate precoding techniques for downlink transmission to mitigate this corruption and thus increase achievable rates. A multi-cell MMSE precoding depending on the pilots assigned to the users is proposed, which does not need coordination. This precoding is the optimal solution of an optimization problem, which consists of the mean-square error of signals and the mean-square interference. Simulation results show that this precoding method can effectively reduce intra/inter-cell interference.
INTRODUCTION
In multi-cell multi-antennas wireless networks with uplink training, uplink pilot symbols is a crucial problem, which results in channel estimate and consequently determines system performance. When the channel estimate at the base station in one cell is jammed by users from other cells, orthogonality of pilots will be lost, the use of non-orthogonal pilots causes corruption, Hence, the corrupted training must be adequately mitigated and utilized. The goal of this study is to exploit techniques to mitigate this corruption and thus increase achievable rates.
In the multi-cell scenario, when Channel State Information (CSI) at the base stations is available, the most of reports focus on the gain obtained via coordination of the base stations (Zhang and Dai, 2004; Venkatesan et al., 2007; Jing et al., 2008). Zhang and Dai (2004) presented jointly both beamforming and pre-coding approaches. The sum capacity of the multiantenna Gaussian Broadcast Channel (BC) based upon Dirty Paper Coding (DPC) was reported by Weingarten et al. (2006). The capacity of DPC was compared to that of TDMA MIMO Gaussian BC by Jindal and Goldsmith (2005). For the non-fading scenario with random phases, The high SNR performance gap between the DPC and Zero-Forcing (ZF) precoders was characterized in Jing et al. (2008), which indicates a singularity problem in certain network settings and demonstrate that the MMSE precoder does not completely resolve the singularity problem.
For the problem of deficiency of channel CSI, Time-Division Duplex (TDD) systems operation and linear ZF precoder were considered and a lower bound on sum capacity was given by Marzetta (2006), results show that it is always beneficial to increase the number of antennas at the base station.
Motivated by the singularity problem of the MMSE precoding matrix while using coordination as claimed by Jing et al. (2008), a multi-cell MMSE precoding depending on the pilots assigned to the users is proposed in this study, which does not need coordination between base stations required by the jointly precoding techniques.
SYSTEM MODEL
In a cellular system with L cells, l ∈ {1, 2,..., L}, each cell provided with one base station with M antennas and K single-antenna users, K≤M. The signal vector received at the m-th antenna of the base station of the l-th cell is:
(1) |
where, pr is the average power at each user. ξ is the length
of pilot symbols transmitted by all users of all cells in every coherence interval,
the pilot vector
Let Yl = [yl1,...,ylM],
Λjl = diag{[ρjl1,...,ρjlK]},
Ωj = [ωj1,...,ωjK], Nl
= [nl1,...,nlM] and
(2) |
The MMSE estimate
(3) |
Let the base station of the l-th cell transmit symbols
The signal received by the users in the j-th cell is:
(4) |
where, ps is the average power at the base station. Blsl is the signal vector transmitted by this base station. zj is the additive noise. The signal received by the k-th user can be expressed as:
(5) |
where, bli is the i-th column of the precoding matrix Bl and zjk is the k-th element of zj The signal received by the k-th user in the j-th cell can be rewritten as:
(6) |
where,
(7) |
where, C(x) = log2(1+x), x is a variable.
ACHIEVABLE RATE
To make problem simple and manifest the primary effect of pilot corruption which is correlative between the precoding matrix at the base station in a cell and users in other cells. Setting that one user per cell, all users use the same pilot ωj1 = ω. For ZF precoding, the precoding vector used at the base station in the l-th cell is:
From Eq. 4, the received signal by the user in the j-th cell is:
(8) |
Using matrix inversion lemma (Zhang, 2004). Simplify the MMSE estimate hjl in Eq. 3.
and obtain
(9) |
Let hjl = ĥjl + ħjlobtain:
(10) |
where, ĥjl is independent of ĥjl
and
The first and second order moments of {hjlbl} are respectively:
(11) |
where,
(12) |
The expectation and the variance of effective channel gain in Eq. 8 are respectively:
(13) |
(14) |
The first and second order moments of the signal and interference terms in Eq. 8 are respectively:
(15) |
(16) |
By substituting Eq. 13-16 for Eq.
7. Θ has a scaled χ2 distribution with 2M degrees of
freedom, scalar factor is
(17) |
where,
UPPER BOUND
Assuming that all channel matrices
where, f Hitj(i =1, 2) is the j-th row of Fit and bir is the r-th column of Bi.
Let
(18) |
For a finite L, after taking partial derivatives of Eq. 18 with respect to elements of B1 and B2 we can obtain the elements of B1 and B2 then reuse these in Eq. 18 to obtain an approximation in terms of the average sum rate.
OPTIMIZATION PROBLEM
Assuming that pilots in every cell is orthogonal and zero-forcing is performed on the users in every cell. The precoding matrix corresponding to this zero-forcing approach is:
(19) |
where,
In the j-th cell, the signal received by the users in this cell given by Eq.
4 is a function of all the precoding matrices. Consider the signal and interference
terms corresponding to the base station in the l-th cell. Based on these
terms, construct the following optimization problem to obtain the precoding
matrix Bl Using the notation:
(20) |
where, β is the relative weights control parameter of the optimization problem. The optimal solution Blopt in Eq. 20 is the multi-cell MMSE precoding matrix.
OPTIMAL SOLUTION
Let
For given j and l,
The last step results from matrix inversion lemma (Zhang, 2004). Let
obtain
(21) |
Simplify the objective function of the problem in Eq. 20:
The optimization problem in Eq. 20 can be expressed as the Lagrangian formulation L(Bl, α, λ) = J(Bl, α)+λ(tr{BlHBl}1).
Let,
obtain
(22) |
Let
(23) |
Factorizing
(24) |
After substituting Eq. 24 in Eq. 23,
(25) |
where, fm is the m-th diagonal element of
where, αopt is such one that satisfying
Let
RESULTS AND DISCUSSION
Using sum rate as the performance metric and chiefly focusing on behavior of the inter-cell interference except for scheduling, power control and other techniques.
No. 1 setting for 2 cells scenario: Let K users in every cell and pilot length of ξ = K, the same orthogonal pilots are used in the two cells. Let the propagation factors ρjlk for all k is:
No. 2 setting for 4 non-adjacent cells scenario: Assume that ZF precoding and GPS precoding (Gomadam et al., 2008) uses only pilots from users situated in the same cell as the base station and the multi-cell MMSE precoding is always used with one Frame Relay (FR-1). Let K = 2 users in every cell and pilot length of ξ = 4. Orthogonal pilots are used in the No. 1 and No. 2 cells. The pilots used in the No. 1 cell are used in the No. 3, so do No. 2 cell and No. 4 cell. Let the sum rate per cell Rsum = L-1Σj,kRjk and the minimum rate R = minjkRjk achieved by all users. Let the propagation factors for all k is:
|
Sum rates per cell for different precoding methods for No. 1 Setting are shown in Fig. 1. The improved sum rates are benefit from decreasing a, it shows that the reuse techniques is effective.
Sum rates per cell for different precoding methods for No. 2 setting are plotted in Fig. 2, here, L = 4, a = 0.8, b = 0.1a. It shows that the multi-cell MMSE precoding is effective and powerful, so do in Fig. 3 and 4. Nevertheless, the antenna potential can not be fully exploited, it imply that there are obstructing effect when pilot corruption occurs, this phenomenon also happen in Fig. 1 and 4.
In Fig. 3, Minimum rates for No. 2 setting for ZF and multi-cell MMSE precoding are plotted and show that improved performance of multi-cell MMSE precoding alters in large range resulting from a and b. It owes to MMSE precoding matrices at the base stations produce slightly interference to other cells.
Fig. 1: | Sum rate for different reused pilots |
Fig. 2: | Sum rates for different precoding methods |
Fig. 3: | Performance of ZF and multi-cell MMSE precoding |
In Fig. 4, here, L = 4, a = 0.8, b = 0.1a. Minimum rates of No. 2 setting using multi-cell MMSE and GPS. precodings as functions of the numbers of antenna M at the base station, so do in Fig. 2.
Fig. 4: | Minimum rates for different precoding methods |
CONCLUSION
The influence of corrupted channel estimates resulting from pilot corruption on TDD systems was discussed in this study. The precoding matrix designed at a multi-antenna base station is correlated with the users using non-orthogonal pilots in other cells. An expression of the rates achieved by all the users for downlink transmission using multi-cell MMSE precoding is presented. The results of analytical analysis and simulation show that improved rates achieved by users via increasing the number of base station antennas is not effective when pilot corruption occurs. It's necessary for utilizing some reuse techniques, e.g., frequency/time, to overcome this obstructing effect. A multi-cell MMSE precoding depending on the pilots assigned to the users is considered and obtain this optimal precoding matrix as the solution of an optimization problem, objective function consists of the mean-square error of received signals by the users in the same cell and the mean-square interference received by the users in other cells. This precoding matrix is shown to reduce intra/inter-cell interference and show that this precoding method used in all cell in the TDD system be superior to ZF precoding.
ACKNOWLEDGMENT
Present research project was fully sponsored by the Future Key Technologies R and D Program of Guangdong Province Government of China under Grant No. [2005] 377.