Estimation in Massive MIMO/ mmWave Massive MIMO
multiple-output (MIMO) systems are the wireless systems with multiple antennas
at both the transmitter and receiver sides. MIMO has been gaining popularity
due to its theoretically predicted capacity gains over single-input
single-output (SISO) channel capacities and its other advantages like increased
link reliability and power efficiency 1.
MIMO systems refers to the systems in which large numbers (few hundreds) of
antennas are employed in communication terminals. A large number of antennas at both transmitter and receiver sides helps
focusing the transmission and reception of signal energy into
ever-smaller regions of space and hence improving the throughput and energy
Massive MIMO systems are now a practical proposition and
has been incorporated into standards as long term evolution (LTE) and
Despite the high
directivity, massive MIMO is challenged by large antenna array size and channel estimation issues.
Using mmWave frequencies for massive MIMO reduces the size required for massive
MIMO antenna arrays. The massive MIMO using mmWave frequencies are called
mmWave massive MIMO. The use of mmWave frequencies offers channel bandwidths
far greater than previously available, while enabling hundreds of antenna
elements to be used at the user equipment, base stations, and access points 2.
However, mmWave massive MIMO still can’t solve the challenges posed by channel
estimation. In addition, mmWave massive MIMO introduces complex baseband
Massive MIMO was primarily envisaged
for time division duplex (TDD) operation, but can also be used in frequency
division duplex (FDD) operation 6. In case of regular MIMO systems, for multi-user
precoding in the downlink and detection in the uplink Channel State Information
(CSI) is required at the base station (BS). A non-stationary wireless channel is estimated every coherence time. In
TDD, the channel is periodically estimated in one direction and compensation
can be applied in both directions assuming reciprocity. Only the base station (BS)
needs to know the information about the channels to process antennas
coherently. The time required to acquire CSI does not depend on the number of
BSs or users. In each coherence interval, there is an uplink channel estimation
at the BS where each user in all cells transmits a pilot
sequence. BSs use these pilot sequences to estimate CSI to
the users located in their cells. Thus, in a TDD system, the
necessary number of orthogonal pilot sequences is equal to the number of users.
For the downlink data transmission in TDD systems, BS generates
beamforming vectors based on the channel estimates obtained during the previous
uplink phase 3.
If FDD is used, that is, different
frequency bands are used for uplink and downlink, the CSI corresponding to the
uplink and downlink will be different. The uplink channel estimation is done at
the BS by letting all users send different pilot sequences. The time required
for uplink pilot transmission is independent of the number of antennas at the
BS. Obtaining CSI for the downlink channel in FDD systems is a two-stage
procedure. The BS first transmits pilot symbols to all users, and then all
users send estimated CSI for the downlink channels to the BS. The time required
to transmit the downlink pilot symbols is proportional to the number of antennas
at the BS. As the number of BS antennas grows large, the traditional downlink
channel estimation strategy for FDD systems becomes infeasible 4.
Linear minimum mean square error (MMSE)
based channel estimation is commonly used, which can provide near-optimal
performance with low complexity. Besides MMSE estimation, compressive sensing-based
channel estimation approach has also been proposed, which exploits the fact
that the degrees of freedom of the physical channel matrix are much smaller
than the number of free parameters. To improve the spectral efficiency of the
system, a time-frequency training sequence design is also developed. The proposed structure achieves the benefits of both time- and
frequency domain estimation while avoiding their individual drawbacks 5.
in Channel Estimation: Pilot Contamination
As the total number of orthogonal pilot sequences
is limited by the channel coherence interval and the requirement to actually
transmit payload data, the pilot sequences must be reused in neighboring cells.
Identical pilot sequences assigned to users in neighboring cells will interfere
with each other causing what is known as pilot
contamination. During downlink transmission, this results in the BS
beamforming signals not only to its own users, but also to users in the
neighboring cells, and therefore creates a strong source of directional
interference. Increasing the number of BS antennas can’t alleviate this
interference, as opposed to the intra-cell interference. A similar effect occurs
during the uplink transmission. The channel estimate is, therefore, a linear
combination of the channel vectors of the users in neighboring cells with the
same pilot sequences since only the pilot sequences within one cell are
guaranteed to be orthogonal. The interference between the channels of the users
during the uplink training leads to interference in both the uplink and
downlink data transmission phases in a reciprocity based system. This effect is
called pilot-contamination, which,
due to the resulting interference during the data transmission, poses a limit on
the achievable rate in a massive MIMO system that relies on linear signal
schemes have been proposed to combat pilot contamination. Some of them are:
Frequency reuse method: One of the ways to mitigate the pilot
contamination is frequency reuse or reducing the number of served users that
use non-orthogonal pilot sequences 5.
Single cell precoding method: In this method, the precoding matrix at a
BS is designed so that the sum of the squared error of its own users and
interference to the users in all other cells is minimized 5.
Angle of arrival (AOA)-based methods: A coordinated scheme is designed for
assigning identical pilot sequences only to users of this type with mutually
non-overlapping AOA probability density functions.