Printer Friendly

Pilot Contamination Effect Mitigation in Massive Multiple Input Multiple Output System and Analysis of Mitigation Techniques.

1 Introduction

Massive Multiple Input Multiple Output (M-MIMO) has been identified as a leading technology to cater for the high data demand rate of the next generation wireless devices also known as 5G. In this technology, large number of antenna arrays concurrently is used to serve many user terminals (UTs), with each UT considered as a single antenna device, in the same time frequency resource (Jose et al., 2009). Asymptotic analysis in accordance with random matrix theory (Elijah et al., 2016) shows that both the intra cell interference and correlated noise can be efficiently mitigated, as the number of Base Station (BS) antennas leans towards boundlessness. Additionally, the energy consumed by cellular BSs can be substantially reduced (Yin et al., 2013).

M-MIMO systems are robust, since failure in one or a few antennas would not significantly affect the general system performance. In addition, optimal performance can be reached by using simple linear processing/detection schemes based on Zero Forcing (ZF) and Matched Filter (MF), when the number of BS antennas inclines toward infinitude. Similar to standard MIMO systems, accurate estimation of the Channel State Information (CSI) of each user is required by the base station and this is done either through reciprocity, that is in Time Division Duplex (TDD) or through feedback method as in Frequency Division Duplex (FDD) schemes (Elijah et al., 2016).

TDD is preferred as the best mode to acquire an accurate estimate of timely CSI in wireless systems in place of FDD due to the fact that the estimation requirement of TDD can be done in one direction and used in both directions as shown in Figure 1, while in FDD estimation and feedback are required for both forward and reverse directions (Guo et al., 2017).

Due to the popularity of FDD among existing networks (3G and 4G), it continues to attract more interest. The TDD scheme depends on the concept of downlink and uplink channels' reciprocity. It has been considered to be a good choice because the forward channel is assumed to be equal to the transpose of the converse channel and the needed information is gotten from transmitted pilots on the inverse link (de Carvalho, 2014). However, due to shortage in orthogonal pilot sequence and the condensed network deployment, several neighbor cells may use the same pilot sequence (nonorthogonal) and the pilot spots in pilot patterns of the several neighbor cells may super impose, which will directly affect the performance of the system.

This is known as Pilot Contamination (PC) (Vardhan et al., 2016). Furthermore, the conventional use of Matched Filter (MF) and Zero Forcing (ZF) as processing schemes will impose intercell interference on the Down Link (DL) transmission, which cannot be reduced by adding more number of antennas at the BSs. In an effort to address the problem of PC in a multi-cell architecture, different schemes have been proposed.

These schemes are grouped into (1) Pilot-based estimation approach and (2) Sub-space estimation approach (Ngo, 2015). In the pilot-based approach, estimation of the channels of users is done with orthogonal pilot sequence within the cell and nonorthogonal pilot sequence across the cells and this leads to extreme Inter-Cell Interference (ICI) while in the sub-space estimation approach, the channels of users are estimated with limited or no pilot sequence which depends on high order statistical parameters that are susceptible to error in the estimation of user channels.

The rest of the paper is organized into sections. Section two cover the reviews of Massive MIMO System Model and linear estimation techniques. Section three is devoted to detailing the PC problem, the proposed PAP technique is also discussed. In Section four simulation results are presented and discussed, while in section five we review pilot contamination mitigation techniques with focus on pilot based sub-category. Pilot contamination open issues in massive MIMO are listed and discussed in section six, and lastly section seven concludes the paper.

2 Review of Massive MIMO System Model

M-MIMO system consists of a single BS and a single K active user. The BS is furnished with M antennas, while each user is considered as a single antenna device as shown in Figure 2. It is assumed that all K users share the same time-frequency resource and the BS and the UTs have perfect knowledge of the CSI. The channels are learned at the BS and the UTs during the training phase.

Let H [MEMBER OF] [C.sup.M X K] be the channel matrix sandwiched between the K UTs and the M BS antenna array, where the [K.sub.th] column of H, denoted by [h.sub.k], represents the channel M X 1 vector linking the BS and the [K.sub.th] UT, assuming that the elements of H are identically independent distributed (i.i.d), gaussian distributed with zero mean and unit variance.

2.1 Uplink Transmission

Uplink transmission is the process whereby the BS receives signals from KUTs. Let [s.sub.k], where E {|[S.sub.k]|[.sup.2]} = 1, be considered as the signal that the [K.sub.th] UT is transmitted. Since all KUTs share the same frequencies, the M X 1 received signal vector at the BS is the summation of all signals emanatting from all KUTs.

[mathematical expression not reproducible] (1)

Where, [[rho].sub.u] is the average signal to noise ratio (SNR), n [member of] [C.sup.M X 1] is the additive noise vector, and [mathematical expression not reproducible], assuming the elements of n are i.i.d. Gaussian random variables (RVs) with zero mean and unit variance, and does not depend on H. The received signal vector [y.sub.ul] joint with CSI knowledge, the BS will logically sense the signal transmitted from K UTs. The model of the channel in figure 2 has the sum capacity as (Saxena et al., 2015):

[C.sub.uls,un] = lo[g.sub.2]det ([I.sub.k] +[[rho].sub.u][H.sup.H]H). (2)

The aforesaid sum capacity can be achieved with linear processors like ZF, MF and Minimum Mean Square Error (MMSE).

2.2 Linear Processing Schemes

In a quest to acquire optimal performance, complex signal detection schemes are usually employed. The BS can use linear equalization techniques (linear receivers in the uplink), to minimize the signal processing intricacy. However, if the number of BS antennas tends to infinity, it has been demonstrated that the linear processing is almost optimal (Sohn et al., 2017). The details of Linear Equalization Schemes are discussed in the following subsections.

2.3 Linear Receivers (in the UPLINK)

Considering linear equalization methods at the BS, the received signal [y.sub.ul] is divided into K streams by finding its product with an M X K linear equalization matrix, A:

[mathematical expression not reproducible] (3)

Every stream is then decoded individually. From the aforementioned equation the [K.sub.th] stream (element) of [mathematical expression not reproducible] that is used in decoding [S.sub.k] is given by:

[mathematical expression not reproducible] (4)

where [a.sub.k] denotes the [K.sub.th] column of A. The noise added with interference is treated as effective noise, and hence, the received signal to interference plus noise ratio (SINR) of the [K.sub.th] stream is as hereunder (Joham et al., 2017):

[mathematical expression not reproducible] (5)

2.4 Maximum Ratio Combining Receiver.

Considering MRC, the BS aims to optimize the obtained signal-to-noise ratio (SNR) of every stream. Overlooking the effect of multi-user interference, the MRC receiver Matrix A, [K.sub.th] column, the obtained SINR of the [K.sub.th] stream for MRC yields:

[mathematical expression not reproducible] (6)

The signal equalization is very easy since the BS just multiplied the obtained vector with the conjugate transpose of the channel matrix H, and then detects each stream separately. At low [[rho].sub.u], the SNR reduces to:

[mathematical expression not reproducible] (7)

This means that at low SNR, the same array gain can be achieved by MRC as in the case of a singleterminal system. However, since the effects of multiterminal interference is being avoided by MRC, it performs below expectation in scenarios involving limited interference, as shown in the last equation, such that the SINR is upper bounded by a constant (with respect to [[rho].sub.u]) when [[rho].sub.u] is big.

2.5 Zero Forcing

Contrary to MRC, zero forcing (ZF) receivers take the inter-user interference into consideration but discard the effect of noise. When using ZF, the multiuser interference is totally cancelled out by projecting every stream onto the orthogonal complement of the inter-user interference (Larsson et al., 2014). The pseudo inverse of the channel matrix H, is considered to be the ZF receiver matrix, which satisfies K users, ZF, and gives this:

[mathematical expression not reproducible] (8)

M [greater than or equal to] Kis required by this technique (so that the matrix [H.sup.H]H is invertible). Thus, the [k.sub.th] stream received SINR is denoted by:

[mathematical expression not reproducible] (9)

The ZF signal detection is very simple and performs well in scenarios involving limited interference.

Increasing transmit power can achieve a much higher SINR. However, it performs below expectation under noise-limited scenarios. In comparison to MRC, the implementation complexity of ZF is higher because of the computation of the pseudo inverse of the channel gain matrix (Ul mulk, 2015).

3 Pilot Contamination

By considering UL training, before the transmitter can send information to the receiver, the receiver sends information to the transmitter at the BS, but the transmitted symbols are known on both sides so that it could train the symbols. Based on this known and properties of its symbols, then the transmitter can estimate the unknown channel vector h added with some noise n of user 1, which leads to the corrupted version of the right channel vector [??] of user 1.

However, at the same time there might be a second user (user 2) which also wants to have access to the BS and it also sends training symbols to the BS. If the BS at the transmitter is not aware of this it might happened that it considers the signal coming from the second user (user 2)as part of the original user (user 1), and that will have automatic effect on the downlink when the BS is transmitting data to the first user (user 1). Where [h.sub.1] is the channel vector of user 2, this effect is called Pilot Contamination (PC), because the interfering part of the channel vector of user 2 ([h.sub.1]) contaminates the channel information of user 1. Multi cell Architecture of M-MIMO system is shown in Figure 4.

We are obliged by pilot contamination to cautiously design pilot mitigationtechniques for a full dimensional MIMO. The aim of this thesis is to assess scenarious under which pilot contamination becomes an important issue in delivering the benefits of M-MIMO system, and study techniques that mitigate its effect.

Farhang et al. (2014) looked into the problem of pilot contamination in cosine modulated multi-tone (CMT) based M-MIMO networks with the view to address the issue. The scheme takes the advantage of the special property of CMT, known as blind equalization property. It was evident that when the technique was extended from single user to a large-scale cellular M-MIMO system, channel estimation error due to (PC) was removed without any coordination among the different base stations or transmission of more training information. Furthermore, it was proved that the proposed solution performed equally good or even better compared to MF. having perfect CSI. Also, the output SINR of the proposed algorithm was observed to have converged towards that of the MMSE solution which has a perfect knowledge of CSI of the various users within its neighboring cells. However, the scheme is highly computationally complex due to the high order statistical parameters needed to estimate the channels and the process could introduce serious error in the channels' estimation.

A pilot contamination mitigation scheme based on Smart Pilot Assignment (SPA) to enhance the performance of users affected severely by PC was investigated by Zhu et al. (2015). It is done, particularly, by taking into consideration the characteristics of the channels large-scale fading. The base station measures the ICI of the individual training sequences produced from the users having the same training sequence in neighboring cells.

Contrary to the traditional way of allocating training sequences to users haphazardly, the SPA scheme allocates a training sequence that has the least ICI to a user with the poorest channel quality sequentially to enhance its performance. The scheme proves theoretically that the pilot allocation generated by it is the answer to the optimization problem [P.sup.i] in a greedy way that can approach the novel optimization problem P as the number of antennas tends to infinity. However, the scheme did not break away from the conventional way of initializing all training sequences for the training that leads to the re-use of the same training sequences in the adjacent cells, thereby causing ICI.

Pilot contamination issue in M-MIMO systems, for cell-edge and cell-center user terminals (UTs) was investigated by Zhu et al. (2016). A novel scheme based on soft pilot re-use (SPR) and Multi-Cell Block Diagonalization (MBD) was used to address this issue. It has been established that users at the cell-edge suffer from severe PC than their center-cell counterparts. In view of this development SPR is used in maximizing the quality of service (QoS) of users at the cell edge by re-using a cell-center pilot group for all center-cell users and a cell-edge pilot group for users at the cell edge in the adjacent cells which is determined by their large-scale fading coefficients.

The MBD pre-coding is utilized for a multi-cell scenario, to mitigate ICI and enhanced the QoS for cell-edge users. However, the enhanced rates of the UL and DL throughputs of the proposed scheme was achieved at the expense of a more pilot sequence requirement for channel state information estimation as a tradeoff with resources for data transmission within the coherence interval in comparison to conventional methods and also the scheme is still based on full pilot sequence initialization and reuse of non-orthogonal pilot sequence in different cells in a multi-cell scenario which is the major source of PC.

One of the methods used in Time Division Duplex (TDD) system is collaboration between cells so that they will transmit their pilots at different times and therefore there will be no inter-cell interference from non-orthogonal pilots. This is found to be difficult as synchronizing different levels of the network (micro-, nano-, and pico-cells), and possibly many cells, will be a huge task. Another suggestion is to randomly change the orthogonal code of pilots so that the interfering signals will randomly change and therefore will have averaging rather than summing effect.

However, because the random change will happen only once in a coherent time, the interference is bound to remain from the same sources for a significantly long time. We therefore suggest an approach that investigate rather than letting the interference to come from neighboring cells that have the same orthogonal pilots. We can make them come from the same cell in order to avoid coordination among neighboring to reduce processing complexity. This can be done by grouping a number of user terminals (UT) and letting them use the same orthogonal pilot. This means the cell in which the interfering UTs belong to has total control of all the interfering UTs and at the same time needs no communication to any other cell to control the interference.

A stream of known previously agreed upon pilot bits from the receiver and or transmitter can be used to estimate the channel state information (CSI). Because of the grouping of many UTs, it means the orthogonal pilots can now be more available to go round the network. Fixed orthogonal channels can be assigned to each cell (pilot reuse > 1) which will be assigned to UTs on demand. Motivated by the concept of pilot reuse and Soft Pilot Reuse (SPR) particularly, which allocates orthogonal pilot sequence to cell-edge users, we therefore proposed Pilot Allocation Protocol (PAP) decontamination scheme operating in TDD mode to mitigate the effect of PC in M-MIMO system. This technique is based on sectional pilot sequence initialization and sharing with a pilot sequence reuse factor of 1/3, which can allocate more than one user to share same orthogonal pilot sequence within a cell which then creates Intra Cell Interference and mitigates the effect of PC. However, this is achieved by separating the users using distinct codes unique to each of the interfering users.

We found out that the scheme performed best among the discussed techniques in terms of average cell throughput. In this scheme the orthogonal pilot sequences are divided into 3 groups and allocated to three different cells. That is to say that within each cell and at least the next neighboring cells the pilot sequences are orthogonal to each other. So, when orthogonal pilot sequences within a particular group are exhausted by users, then sharing starts. This is done between users, which then creates an intra cell interference among the users sharing the same pilot sequence, and again are then made orthogonal to each other by employing a property of hadamard matrix. The matrix assigns distinct codes to the interfering users sharing the same pilot sequence in each coherence interval as against discarding or queuing the users for the next available pilot sequence in the conventional technique.

The concept mitigates the effect of pilot contamination to a large extent and also increases the number of users that of the available resources without increasing the number of the orthogonal pilot sequences. Figure 5 depicts the multi-cell architecture of the developed PAP technique.

Figure 5 shows the pilot sequence allocation and sharing pattern of the developed PAP scheme based on sectional initialization and pilot sequence sharing between users.

Each color in Figure 5 represents a particular group (a section) of the total orthogonal pilot sequences while cells having the same colors use the same type of orthogonal pilot sequences with a minimum reuse distance of one kilometer since the cell radius considered is 500 m. Furthermore the green, blue and red colored signals emanating from the user equipments to the base stations represent the individual pilot sequences in each cell.

Users transmitting same color of signals are sharing the same pilot sequence but then are separated using distinct signs of hardmad matrix. The broken signals in blue colors represent interfering signals from far cells and they formed part of the additive white gaussian noise.

The following methodology was adopted to carry out the research work:

a) A TDD M-MIMO channel model with a varying number of base station antennas, number of users per cell, and power of user equipment were developed based on Rayleigh fading channel taking into consideration pilot allocation protocol. PAP was implemented based on sectional initialization and pilot sequence sharing, which allocates pilot sequence to many users within a cell and their training was conducted using different codes.

b) Users were grouped into cell center and cell edge in accordance with their large-scale fading coefficient.

c) A signal detection scheme based on matched filter and zero forcing, for equalizing the signals was implemented.

d) The UL Signal to Interference plus Noise Ratio (SINR) of users at the cell edge was then calculated and the average UL cell throughput determined.

e) The DL Signal to Interference plus Noise Ratio (SINR) of users at the cell edge was then calculated and the average UL cell throughput determined.

f) Validation of the results was done by comparing the results gotten from the Pilot Allocation Protocol (PAP) technique and that of conventional schemes in terms of average uplink cell throughput and bit error rate as performance metrics.

4 Results of the Simulation

The PAP simulation results obtained using simulation parameters highlighted in Table 1were plotted for the PAP technique Zero Forcing (ZF) and Matched Filter (MF) linear detectors.

Figure 6 shows the average Uplink (UL) cell Throughput against the Number of Base Station (BS) Antennas for the developed PAP and conventional schemes all with ZF detector, with number of BS antennas ranging from 60 to 256 antennas and 50% of the bandwidth for data transmission allocated to UL data transmission.

From the simulation results in Figure 6, when the number of BS antennas is at 64, the conventional schemes offer an average UL throughput of 31 b/s/Hz while the developed PAP offers an average UL cell throughput of 33 than the conventional scheme when ZF detector was used.

At 250 BS antennas, the conventional scheme offers an average cell throughput of 37 b/s/Hz while the developed PAP offers an average UL cell throughput of 57.5 b/s/Hz, which is 20.5 b/s/Hz more than the conventional scheme. This shows that the performance improvement of the interfering users (cell edge) users of the developed PAP schemes is better than that of the conventional scheme as the number of BS antennas increases.

It can be observed that when the number of BS antennas was set at 256 the conventional scheme offers an average cell throughput of 37.5 b/s/Hz while the developed PAP offers an average UL cell throughput of 58 b/s/Hz, which is 20.5 b/s/Hz more than the conventional scheme. The PAP scheme outperforms the conventional scheme and hence further proves that the performance improvement of the interfering users (cell edge users) of the PAP technique is better than that of the conventional scheme as the number of BS antennas increases.

Figure 7 shows the Average Uplink Cell Throughput against the Number of Base Station (BS) antennas for the developed PAP and conventional scheme all with MF detector, with number of BS antennas ranging from 60 to 256 BS antennas and 50% of the bandwidth for data transmission allocated to UL data transmission. From simulation results in Figure 7, when the number of BS antennas is at 64, the conventional scheme offers an average UL throughput of 24 b/s/Hz while the developed PAP offers an average UL cell throughput of 25.5 b/s/Hz improvement, which is 1.5 b/s/Hz more than the conventional scheme when MF detector was used.

At 250 BS antennas the conventional scheme offers an average cell throughput of 34.5 b/s/Hz while the developed PAP offers an average UL cell throughput of 42.5 b/s/Hz, which is 8 b/s/Hz more than the conventional scheme. At 256 BS antennas the conventional scheme offers an average cell throughput of 34.5 b/s/Hz while the developed PAP offers an average UL cell throughput of 43 b/s/Hz which is 8.5 b/s/Hz more than the conventional method. This shows that the performance improvement of the interfering users (cell edge) users of the developed PAP schemes is better than that of the conventional scheme as the number of BS antennas increases with MF detector. Similarly, the PAP simulation results obtained using simulation parameters highlighted in Table 1 were plotted for the PAP technique using Zero Forcing (ZF) and Matched Filter (MF) linear precoders.

From simulation results in Figure 8, when the number of BS antennas is at 64, the conventional scheme offers an average DL throughput of 31 b/s/Hz while the developed PAP offers an average DL cell throughput of 33 b/s/Hz improvement, which is 2 b/s/Hz more than the conventional scheme when ZF detector was used. At 250 BS antennas the conventional scheme offers an average cell throughput of 37.5 b/s/Hz while the developed PAP offers an average DL cell throughput of 57 b/s/Hz, which is 20.5 b/s/Hz more than the conventional scheme. This shows that the performance improvement of the interfering users (cell edge users) of the developed PAP schemes is better than that of the conventional scheme as the number of BS antennas increase.

It can be seen that when the number of BS antennas was set at 256 the conventional schemes offer an average DL cell throughput of 37.5 b/s/Hz while the developed PAP offers an average DL cell throughput of 58 b/s/Hz, which is 20.5 b/s/Hz more than the conventional scheme. The PAP scheme outperforms the conventional scheme and hence further proved that the performance improvement of the interfering users (cell edge users) of the PAP is better than that of the conventional scheme as the number of BS antennas increase.

Figure 9 shows the Average Downlink (DL) Cell Throughput against the Number of Base Station (BS) antennas for the developed PAP and conventional scheme all with MF precoder, with number of BS antennas ranging from 60 to 256 antennas and 50% of the bandwidth for data transmission allocated to DL data transmission. From simulation results in Figure 9, when the number of BS antennas is at 64, the conventional scheme offers an average DL throughput of 24 b/s/Hz while the developed PAP offers an average DL cell throughput of 25.5 b/s/Hz improvement, which is 1.5 b/s/Hz more than the conventional scheme when MF detector was used.

At 250 BS antennas the conventional scheme offers an average cell throughput of 34.5 b/s/Hz while the developed PAP offers an average DL cell throughput of 42.5 b/s/Hz, which is 8 b/s/Hz more than the conventional scheme. At 256 BS antennas the conventional scheme offers an average cell throughput of 34.5 b/s/Hz while the developed PAP offers an average DL cell throughput of 43 b/s/Hz this shows that the performance improvement of the interfering users (cell edge users) of the developed PAP schemes is better than that of the conventional scheme as the number of BS antennas increase.

It can be seen that when the number of BS antennas was set at 256 the PAP scheme outperforms the conventional scheme and hence further proved that the performance improvement of the interfering users (cell users) of the PAP is better than that of the conventional technique as the number of BS antennas keep increasing with MF detector.

It can be observed from Figure 10 that as the SNR increases the BER also decreases and the developed PAP using ZF for UL met the BER requirement for audio signal that is [10.sub.-3] of BER in digital communications at SNR of 8 dB, which means in every 1000 bit you have 1 error.

It can be observed from Figure 11 that as the SNR increases the BER also decreases and the developed PAP using ZF for DL met the BER requirement for audio signal that is [10.sub.-3] of BER in digital communications at SNR of 8 dB, which means in every 1000 bit you have 1 error

It can be observed from Figure 12 that the developed PAP using MF for UL attained a BER of 10-1.25 = 1/101.25 = 717.8 = 1/18, which means in every 18 bits of transmission you have 1 error, at SNR of 8db.

Figure 13 shows the DL Bit Error Rate performance of the developed PAP model using MF. It can be observed from Figure 15 that the developed PAP using MF for DL attained a BER of [10.sup.-1.25] = 1/[10.sup.1.25] = 717.8 = 1/18, which means in every 18 bits of transmission you have 1 error, at SNR of 8 dB. From the simulation results obtained from the developed PAP scheme using the ZF and MF detection/precoding schemes evaluated at 64,250 and 256 BS antennas, the average cell throughput improvement of the developed PAP scheme outperforms that of the traditional scheme.

The improvement occurred as a result of the pilot sequence allocation protocol adopted by the PAP technique and the way the interfering users (cell edge users) were efficiently separated after sharing of pilot sequence occurs between users. The PAP scheme creates intracell interference and suppressed the effect of Intercell Interference ICI to a large extent which is the major cause of Pilot Contamination during Uplink pilot transmission.

Furthermore, it is important to notice that the average cell throughput improvement performance of the PAP technique is better than that of the conventional scheme with both ZF and MF detection/precoding schemes throughout the range of BS antennas considered (64,250 and 256). This is due to the fact that in the PAP technique pilot, sequence of users within and across the neighboring cells were made to be orthogonal to each other and when sharing occurs (when pilot sequence of users was made to be non-orthogonal) to each other, codes were used to separate the interfering users throughout the length of the coherence interval.

5 Mitigation Techniques

5.1 Pilot Open Loop Control

The path loss and transmit power of a User Terminal (UT) is determined by the average obtained power of a pilot sequence at the base station (BS). Usually terminals located close to the BS have a less pathloss and a higher signal power while those located at the edge of the cell usually have a higher pathloss and therefore a lower signal power (Ul mulk, 2015).

This means those terminals near the BS benefit from a reduced pathloss, leading to a better Signal to Interference plus Noise Ratio (SINR). The three different scenarios of pilot reuse at cell edge, close to serving BS and by close cell edge terminals, are described in figure 14 (Saxena et al., 2015). Figure 14 depicts Pilot Reuse at Cell Edge, Pilot Reuse Close to Serving BS and Pilot Reuse by Close Edge Terminals.

5.2 Less aggressive pilot reuse

Maximum pilot reuse prompts most extreme inter-cell interference (ICI) when estimating a channel, which can be reduced by using a soft pilot reuse factor (Saxena et al., 2015). Pilot reuse is analogous to the traditional frequency reuse in the sense that user terminals within the pilot reuse area can utilize just a little amount of the time frequency resources during the channel estimation phase (Zhang, 2015). On the other hand, in the case of maximum pilot repetition, each terminal is permitted to use all the available resources for communication for the rest of the coherence interval which led to less throughput (Sohn et al., 2017). The pilot repetition factor of 1/U is the rate at which pilot resources may be repeated within the network, where U is the quantity of cells that are allotted orthogonal pilot sequence.

Using reuse factor of U>1 mitigates the pilot contamination effect by allotting orthogonal pilot sequence to the surrounding cells leading to huge throughput. The total number of unique time frequency elements earmarked to transmit pilot are KU, where K is the quantity of terminals per cell. The trivial case of pilot repetition is aggressive pilot reuse with U = 1, considering some reserved K time-frequency indices for pilot sequence. The indices may be found any place inside the resource block without loss of simplification. K orthogonal pilot symbols are generated including these indices that are shared randomly at arbitrary or algorithmically within the terminals per cell.

Since the cells are taken to be synchronized, the symbols' transmissions are synchronized crosswise over cells also. Considering a hexagonal cellular architecture, the minimum reuse factor that can ensure orthogonal pilots in adjacent cells is 1/3. The resources used to implement this reuse factor will triple the number of terminals, that is 3K. A pilot group is allocated to each cell in accordance with the reuse pattern and the users are assigned with the pilots as in the case of maximum pilot reuse. Clearly, a higher value of U will eliminate Pilot Contamination (PC) but at an expense of more training overheads added by using the pilot resources, thereby reducing the resources available for data within a coherence interval (Saxena et al., 2015).

The enhanced estimates are used for better forward and reverse links beamforming at the BS using spatial multiplexing to serve the users. Hence there is a tradeoff between the exact beamforming and transmission of the data afterwards. Figure 16 depicts full pilot reuse and 1/3 pilot reuse respectively (Ul mulk, 2015). Alternatively, there is another technique called Soft Pilot Reuse (SPR) that mitigates PC by allocating more orthogonal pilot sequences to terminals at the cell-edge to improve their Signal to Noise Ratio (SNR).

6 Open Issues

While massive MIMO reduces many conventional issues in communication theory less applicable, it reveals entirely fresh challenges that require research, among which are discussed hereunder:

6.1 Training Overhead

There is need to figure out if the sub-space method can achieve enough CSI quality needed for the pilot-based approach with the same target SINR (Jose et al., 2009). Furthermore, the sub-space techniques need more schemes or information to distinguish the particular eigen vector that corresponds to which user and the assumption that all desired channels are more influential than all interfering channels does not hold in practical system always (Larsson et al., 2014).

6.2 Deployment Scenario

Studies are required to investigate the influence of pilot contamination, by considering a more realistic channel model for large MIMO system by bearing in mind statistical channel properties, particularly the spatial-temporal frequency correlation properties and large-scale fading role (Elijah et al., 2016).

6.3 Computational Complexity and Cost

There is need for a soft complex pre-coding schemes and CE techniques. While some of the suggested schemes to reduce pilot contamination theoretically sound promising, there is a need to assess their performance by bearing in mind the compromise between its complexity and accuracy (Atzeni et al., 2015).

7 Conclusion

Pilot Contamination (PC) is considered as the major drawback of Time Division Duplexing (TDD) Massive Multiple-Input Multiple-Output (M-MIMO) systems because it limits their expected capacity. Several mitigation techniques such as Eigenvalue Decomposition, Cooperative Bayesian Channel Estimation, Blind Equalization Technique, Time Staggering Pilot, and Smart Pilot Assignment have been proposed in literature to address the issue of PC.

This research work focused on Pilot-based estimation category. In the pilot-based method, channels of users are estimated by means of orthogonal pilot sequences within the cell and non-orthogonal pilot (the same pilot) sequences across the cells. However, this technique is largely affected by Intercell Interference (ICI) which is the major cause of Pilot Contamination during UL pilot sequence transmission.

Therefore, we developed a Pilot Allocation Protocol (PAP) decontamination scheme for TDD mode M-MIMO system based on sectional pilot sequence initialization and sharing which allocates more than one user to share the same orthogonal pilot sequence within a cell which then creates Intra Cell Interference, suppresses Inter Cell Interference (ICI) and mitigates the effect of PC among the interfering User Terminals (UTs). Additionally we reviewed pilot contamination mitigation techniques with focus on pilot based sub-category. Open challenges on pilot contamination in massive MIMO were also digested.

Acknowledgment

This research work is supported by the Communication-and-Electronic Research Group of the Department of Communications Engineering, Faculty of Engineering, Ahmadu Bello University, Zaria, Kaduna State, Kaduna, Nigeria

References

Atzeni, I., Arnau, J., & Debbah, M. (2015). Fractional pilot reuse in massive MIMO systems. Paper presented at the Communication Workshop (ICCW), 2015 IEEE International Conference on.

de Carvalho, E. (2014). Pilot decontamination through pilot sequence hopping in massive MIMO systems. Paper presented at the Global Communications Conference (GLOBECOM), 2014 IEEE.

Elijah, O., Leow, C. Y., Rahman, T. A., Nunoo, S., & Iliya, S. Z. (2016). A comprehensive survey of pilot contamination in massive MIMO--5G system. IEEE Communications Surveys & Tutorials, 18(2), 905-923.

Guo, X., Zhang, J., Chen, S., Mu, X., & Hanzo, L. (2017). Two-Stage Time-Domain Pilot Contamination Elimination in Large-Scale Multiple-Antenna Aided and TDD Based OFDM Systems. IEEE Access.

Joham, M., Neumann, D., & Utschick, W. (2017). Strategies to Combat Pilot Contamination in Massive MIMO Systems. 2nd International Workshop on Challenges and Trends of Broadband Wireless Mobile Access Networks--Beyond LTE-A, 1-44. Retrieved from

Jose, J., Ashikhmin, A., Marzetta, T. L., & Vishwanath, S. (2009). Pilot contamination problem in multi-cell TDD systems. Paper presented at the Information Theory, 2009. ISIT 2009. IEEE International Symposium on, Seoul, South Korea.

Larsson, E. G., Edfors, O., Tufvesson, F., & Marzetta, T. L. (2014). Massive MIMO for next generation wireless systems. IEEE Communications Magazine, 52(2), 186-195.

Ngo, H. Q. (2015). Massive MIMO: Fundamentals and system designs (Vol. 1642): Linkoping University Electronic Press.

Saxena, V., Fodor, G., & Karipidis, E. (2015). Mitigating pilot contamination by pilot reuse and power control schemes for massive MIMO systems. Paper presented at the Vehicular Technology Conference (VTC Spring), 2015 IEEE 81st.

Sohn, J.-Y., Yoon, S. W., & Moon, J. (2017). On Reusing Pilots Among Interfering Cells in Massive MIMO. IEEE Transactions on Wireless Communications, 16(12), 8092-8104.

Ul mulk, Z. (2015). Analysis of Pilot Contamination in Very Large Scale MIMO Systems. Department of Electrical Engineering A thesis submitted in partial fulfillment of the requirements for the degree of Masters in Electrical Engineering (MS EE-TCN) In School of Electrical Engineering and Computer Science, National University of Sciences and Technology.

Vardhan, P., Gupta, M., & Kumar, A. (2016). Massive-MIMO-Past, Present and Future: A Review. Indian Journal of Science and Technology, 9(48), 1-13. doi: 10.17485/ijst/2016/v9i48/99891

Yin, H., Gesbert, D., Filippou, M., & Liu, Y. (2013). A coordinated approach to channel estimation in large-scale multiple-antenna systems. IEEE Journal on selected areas in communications, 31(2), 264-273.

Zhang, C. (2015). Fractional Pilot Reuse in Massive MIMO System. IEICE Transactions on Fundamentals of Electronics, Communications and Computer Sciences, 98(11), 2356-2359.

(*) Saidu M. Waziri (1), Muhammad B. Abdulrazaq (2), Abdoulie M. S. Tekanyi (1), Abdullahi Mohammad (2) and Hassan A. Abdulkareem (1)

(1) Department of Communications Engineering, Ahmadu Bello University, Zaria, Kaduna State, Kaduna, Nigeria

(2) Department of Computer Engineering, Ahmadu Bello University, Zaria, Kaduna State, Kaduna-Nigeria mw.waziri@gmail.com, mbashiray@yahoo.com, amtekanyi@abu.edu.ng, msabdallah79@gmail.com, ha2zx@yahoo.com

(*) Corresponding Author
Table 1: Model Parameters used

Parameters                              Units

Number of cells [L.sub.Total]            19
Number of antenna in BS M                32 [less than or equal to] M
                                         [less than or equal to] 256
Number of users in the [i.sub.th] cell   32
[K.sub.i]
Number of pilot resource [K.sub.cs]      10
Cell radius R                           500m
Average transmit power at users
[[rho].sub.p] [[rho].sub.u]              10dbm
Average transmit power at BS             12dbm
[[rho].sub.d]
Pathloss exponent [alpha]                 3
Log normal shadowing fading
[[sigma].sub.show]                        8db
Carrier frequency                         2GHz
System bandwidth                         10MHz
Minimum distance between user
and BS                                   30m
COPYRIGHT 2019 University of the West of Scotland, School of Computing
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2019 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Waziri, Saidu M.; Abdulrazaq, Muhammad B.; Tekanyi, Abdoulie M. S.; Mohammad, Abdullahi; Abdulkareem
Publication:Computing and Information Systems
Article Type:Report
Date:Mar 1, 2019
Words:6449
Previous Article:Discrete Firefly Algorithm Based Feature Selection Scheme for Improved Face Recognition.
Topics:

Terms of use | Privacy policy | Copyright © 2019 Farlex, Inc. | Feedback | For webmasters