Resumen de Channel estimation techniques for next generation mobile communication systems
We are witnessing a revolution in wireless technology, where the society is demanding new services, such as: smart cities, autonomous vehicles, augmented reality, etc. These challenging services not only are demanding an enormous increase of data rates in the range of 1000 times higher, but also they are real-time applications with an important delay constraint. Furthermore, an unprecedented number of different machine-type devices will be also connected to the network, known as Internet of Things (IoT), where they will be transmitting real-time measurements from different sensors. The Third Generation Partnership Project (3GPP) has already identified these requirements, and it has classified them into three groups: enhanced mobile broad-band (eMBB) services, massive machine type communications (mMTC) and ultra-reliable low latency communications (URLLC).
As a transitory solution, the 3GPP has modified the existing 4G, known as Long Term Evolution (LTE), in order to satisfy, as far as possible, the mentioned demands of the users. Carrier aggregation was introduced to increase the data rate up to 100MHz, where different LTE bands can be jointly used as a single one. Later, it proposed to integrate machine type communications (MTC) in LTE, known as NarrowBand-IoT (NB-IoT), placing the narrow-band signal in either the middle or the edge of the band. However, all these minor modifications hardly cover the high expectation of the society.
The new promising 5G will be capable of providing all the requested services with a great performance. In the first stage, the 3GPP has recently defined a New Radio (NR), satisfying only the eMBB services, leaving the other two goals for the following releases. Focusing in the physical layer, some key technologies that will make possible all these requirements are: new waveform design, massive multiple-input multiple-output (MIMO) and millimeter waves (mmWave). These challenging technologies require new signal processing techniques in order to allow a successful demodulation of the transmitted information. Often, the traditional ones cannot be straightforwardly applied, as for example for time-frequency estimation, channel estimation, hardware calibration, etc. New developments are needed to make possible the deployment of 5G.
In this thesis, we mainly focus on the channel estimation process for the different adopted technologies in 5G. In coherent demodulation schemes, channel estimation and equalization represent two key aspects in wireless communication systems. The channel estimation procedure allows us to know how is the propagation environment of our transmitted signal, and then, compensate its effects by equalization, which is crucial for achieving reliable communication with high data rates. In order to obtain the channel state information (CSI), the most used method is based on pilot symbol assisted modulation (PSAM), where the transmitter exclusively sends a known preamble or pilot-sequences, and the receiver can obtain the CSI through some minimization criterion, such as least squares (LS) or minimum mean squared error (MMSE). These methods have some significant advantages, namely small error and low-complexity. However, the transmission of pilot sequences reduces the overall efficiency of the system.
Filter-bank multi-carrier (FBMC) is the waveform candidate with the lowest OBE due to the use of a well-designed prototype filter, instead of using the rectangular one by orthogonal frequency division multiplexing (OFDM). Therefore, FBMC can make a better use of the available spectrum, and allows to significantly reduce the existing guard-bands. Furthermore, it does not use the cyclic prefix (CP), increasing the time efficiency. However, FBMC has some disadvantages, such as: higher complexity in the transmitter and receiver blocks, the orthogonality only holds in the real domain and the presence of inter-symbol interference (ISI) and inter-carrier interference (ICI) due to the lack of CP. Hence, any existing signal processing techniques designed for OFDM cannot be straightforwardly applied in FBMC, where the new methods must deal with all the mentioned drawbacks.
The main issue of FBMC in order to obtain the CSI is related to the intrinsic self-interference caused by the surrounding symbols, due to the use of non-rectangular prototype filters, that must be taken into account before the equalization process. The two classical ones are auxiliary pilot (AP) and pair of pilots (POP). These techniques consist in transmitting several additional auxiliary pilot along with the traditional one. By doing so, the received pilots become interference-free, and channel estimation can be performed in the same way as in OFDM systems. However, these techniques have several drawbacks. In the case of AP, the minimum required power of the auxiliary pilot is about 3.3dB higher than the power of data symbols, which means an additional waste of the valuable energy and increasing the PAPR; moreover, there is an additional complexity added in the transmitter side which requires the computation of all the auxiliary pilots. In the case of POP, the receiver must perform some linear combination of the received pilots in order to get the estimated channel, where it may also increase also the noise which compromises the global performance. Lately, several modifications based on the combination of AP and POP are proposed in order to solve the mentioned issues. However, the complexity enhancement introduced by these techniques are not practical to be applied in some realistic communication systems.
We propose two new pilot sequences for FBMC which are capable of taking advantage of the self-interference produced by the well-localized prototype filter in FBMC, and a novel channel estimation technique with an affordable complexity suitable for either devices powered by battery or URLLC communications. Firstly, we propose continuous pilot sequences (CPS) with FBMC, where this efficient combination is not straightforward due to the presence of the self-interference in FBMC, unlike in OFDM systems. At the transmitter side, the CPS must be specifically designed to work under this self-interference and avoid either any precoding procedures or enhancing the power overhead, that is present in the existing techniques. At the receiver side, due to the adequate-design of CPS, a low-complexity averaging process is only required in order to reduce the noise and data interference, combined with the advantage of using pseudo-pilots in order to achieve a better channel estimation in terms of MSE. Hence, the low-complexity of CPS fits well with MTC, not only due to energy concerns, but also it makes low-cost devices even competitive in terms of price. However, when the variability of the channel is lower, CPS could be inefficient due to the considerable amount of allocated pilots in the system. Therefore, in order to face this inefficiency, we propose a variant of CPS denoted as burst pilot sequences (BPS). This scheme not only has the same benefits of CPS, but it also provides the flexibility to select the length of the pilot sequence depending on the variability of the channel in each moment, not only keeping the quality of the estimated channel, but also improving the number of available resources for data transmission. Numerical results will show that the theoretical analysis using a Gaussian approximation is accurate to compute the channel estimation error. Additionally, they validate the behavior of our proposed system and its improvement compared to the existing pilot scheme.
MIMO is a popular radio technique that has also been widely included in several wireless standards during the last decades, as OFDM, where it can significantly increase the capacity and reliability of the radio links. In a realistic scenario, MIMO is exploited under the multi-user scenario, where typically the BS is equipped with multiple antennas that simultaneously serve a group of single-antenna UEs. Hence, multiplexing gain can be shared by all of them. In order to fully exploit its advantages, precoding/postcoding matrices must be computed in order to compensate some undesirable effects such as: noise, multi-user interference, hardware and channel effects, among others. However, in order to be able to compute these matrices, a channel estimation process is required, where the complexity of this process linearly depends on the number of antennas. Moreover, in the last years, multi-user massive MIMO is proposed, where the number of antennas at BS is increased far beyond the actual usage. The random matrix theory shows that the effects of small-scale fading, interference and noise can be effectively suppressed when the number of antennas is very large. Hence, using linear detection schemes, such as maximum ratio combining (MRC), will provide an optimal performance. However, when the number of antennas is not large enough, equalizers based on zero-forcing (ZF) or MMSE criterion must be used in order to keep the performance. However, these equalizers require a matrix inversion which is a very hard operation in terms of complexity, increasing the delay of the link. Furthermore, channel estimation is also a challenging task when the number or antennas is very large, due to high number of different channels that must be estimated. This fact requires a long training period, where the BS and the UEs must exchange a great amount of pilot-sequences, decreasing the overall efficiency of the system.
We propose the use of Neumann series to compute the matrix inversion, combined with the linear interpolation of the inverted channel matrix, under the scenario of massive MIMO-OFDM. To the best of our knowledge, this joint scheme has never been proposed before. Note that, we have chosen the linear interpolation method due to the fact that it provides a good trade-off between complexity and performance. Taking advantage of the properties of the channel matrix for the massive MIMO regime, we will show that our proposals, denoted as low-complexity schemes, require a lower number of operations and maintain the performance of the system as good as the traditional scheme. Additionally, they are applicable for either time division duplex (TDD) or frequency division duplex (FDD) scenarios. We will provide a comparison of the complexity in terms of required operations, and a detailed analytical description of the obtained average square Euclidean distance (ASED) between the traditional scheme and low-complexity schemes, which becomes zero when the number of antennas is large enough. The numerical results have shown that our proposal requires a significantly lower computational complexity while maintaining the same performance as the traditional scheme. The proposed schemes are useful for the practical implementation of systems that combine OFDM and massive MIMO.
The next generation of mobile communication networks should bring us an ambitious throughput. In order to meet this demanding requirement, new spectrum bands that have not been used yet by mobile communications must be explored, such as mmWave band. This band corresponds to frequencies higher than 30GHz, whose wavelengths belong to the order of millimeters. In the frame of 5G it is usual to refer to frequencies "below 6GHz" as conventional and "above 6GHz" when referring to the mmWave. The availability of resources at such high frequencies will provide ultra-high broad-band services. However, it also causes new challenges which will reduce the overall performance. These new issues are mainly related to the high propagation loss and the high probability of blockage from different materials that may contribute to the propagation channel.
In order to solve all the issues that we mentioned, very large number of antennas is a must in order to provide directional communications. Note that in mmWave, large antenna arrays can be easily manufactured due to the small wavelengths. Directional links will take advantage of the beam-forming gain in order to improve the link budget and provide an acceptable communication quality. Thus, the channel estimation procedure is even more challenging because, it is also responsible for obtaining accurate beams under wide dynamic SNR ranges. Note that, even though the massive number of antennas at the BS will compensate the large and changing path loss, the channel must be estimated before this compensation is effective.
The channel model can be considered sparse in mmWave, where there are just a few paths thanks to the narrow beams produced by the high carrier frequency. In the literature, there are several efficient channel estimation techniques for mm-Wave massive multiple antenna systems assuming the sparsity nature of the channel. However, a common issue of all techniques consists in the bad adaptation to the different SNR conditions which is very common in mmWave radio links, as we mentioned before. Furthermore, these techniques have an important disadvantage, namely that they assume a parametric approach for the direction of arrival (DoA) of each tap of the frequency-selective channel. This fact implies the need of an antenna array calibration process of the system in order to fully characterize any non-linear or unexpected effects in both array antenna and RF chains, caused by the manufacturing or installation issues. Later, new proposals based on a subspace method assuming that the spatial features are completely unknown are given, leading to non-parametric estimation in which antenna array calibration is no longer required. Furthermore, it exploits a low-rank (LR) algebraic structure of the channel by projecting the estimated channel on the spatial and temporal (ST) covariance matrices. When the SNR is not so high, the LR version of the estimated channel provides a better performance than the full-rank (FR) one in terms of bias-variance trade-off in the MSE. However, this method is not accurate enough due to the fact that they assumed that there is no correlation between spatial and temporal channel behavior. This assumption holds if and only if the receiving signal of all taps of the multi-path channel, with different time of arrival (ToA), have the same DoA at the BS, which is not true in realistic scenarios.
We will improve the performance of the LR estimated channel given by the literature, where we assume the same system model and assumptions. A multi-slot scenario is used, where the slow-varying components of the channel response (angles and delays) are estimated, while the fast-varying ones (fading coefficients) are tracked at each slot. Unlike previous works, we obtain a joint spatial-temporal (ST) covariance matrix, which is capable of obtaining the full ST features of the massive multiple antenna scenario. Moreover, given the computed joint ST covariance matrix, we propose three estimation methods of DoA and ToA considering a semi-parametric spatial response and delay estimation. The first method is based on one dimension (1D) multiple signal classification (MUSIC) algorithm, that it is a low-complexity method that provides a very good performance in high SNR scenarios. The second method is based on iterative maximum likelihood estimation (MLE), where it provides a very good performance in any scenario. However, its complexity is higher than the previous one. Hence, the third method is the hybrid one, which combines the previous two methods and is capable of obtaining a good performance with a reduced complexity. Finally, we also propose an automatic rank selection (ARS) method which is able to select the best LR channel estimation each scenario based on the computation of the MSE. Numerical results have shown that our proposal outperforms the existing techniques, either full-rank and LR ones. Thus, the proposed schemes are suitable for the dynamic SNR environment often found in mmWave.
In this thesis, we focused on providing some new channel estimation techniques for the different key technologies for the next generation of mobile communications, such as: FBMC, massive MIMO-OFDM and mmWave. For all of them, our novel proposed techniques outperform the existing ones and, as far as possible, decrease the overall complexity. Given all these proposals, we have contributed to the implementation of 5G paving the way towards satisfying the three main requirements: eMBB, massive MTC and URLLC. eMBB can be achieved with the operation of mmWaves that provide hundreds of megahertz of bandwidth, which is only possible with the use of MIMO techniques. Massive MTC and URLLC can be only deployed by using low-complexity channel estimation and equalization techniques, which can reduce either the delay of the computations and the battery consumption. Thus, we contribute to fulfilling the demands of the society by providing novel signal processing techniques for 5G.
