Tadesse Ghirmai- Publications

UNIVERSITY of WASHINGTON | BOTHELL

Electrical Engineering | Science & Technology

Home Teaching Research Publications Other

Past Projects

Modeling of cascaded systems using non-Gaussian autoregressive models

Dynamic systems are commonly modeled by Gaussian autoregressive (AR) processes. The popularity of AR processes for modeling dynamic systems is mainly due to the fact that they can be easily configured to capture the temporal statistical characteristics of the dynamic systems. In this research work, we considered the problem of modeling a system consisting of two cascaded subsystems, each of which represented by a complex Gaussian autoregressive (AR) process, by one AR model. Such combined representation of cascaded systems has a potential of simplifying the simulations of the cascaded processes and easing the complexity of estimating the model parameters. One motivating example where a combining model of cascaded systems can be useful is modeling of a single-hop relay channel. Rather than modeling the fading process of each channel of the transmitter-to-relay and relay-to-receiver by a separate mathematical model, it is convenient and efficient if the fading process of the overall channel is modeled by a single model that captures the statistical characteristics of the overall channel. To derive the combined model, we first studied the overall statistical characteristics of a cascaded dynamic system whose two subsystems are represented by complex Gaussian AR processes. It is shown that the marginal probability density functions (pdfs) of the real and imaginary parts of the combined process are Laplace pdfs. Therefore, to represent such a Laplacian dynamic systems, a Laplace autoregressive model that is configurable to accurately capture the statistical dynamics of the systems was developed. The proposed model possess a structure that is simple and convenient for generating realizations of Laplace processes. An important feature of the Laplace AR process is that its autocorrelation (AC) function follows Yule-Walker type of equations which simplifies its configuration to have matching autocorrelation values to that of the cascaded process.

Data detection and channel estimation in Relay-based communication systems

This research work developed effective algorithms for data detection and channel estimation for a single-hop relay communication system for different channel conditions. The conventional solution to the need for more capacity in cellular networks is to increase frequency reuse by cell division which requires installing new base stations (BS). However, the cost of installing new BS and its concomitant infrastructure is high. Recently, relay-based communication system is presented as an alternative solution to enhancing the capacity of a communication network. The use of relays in a communication network has several additional benefits including increasing coverage of a network, mitigating the effect of fading channels by increasing spatial diversity using cooperative communication, and improving energy efficiency of the overall system. However, implementation of relay-based communication system comes with its own set of technical difficulties. Data detection in such environment, for example, is challenging because of the unique statistical characteristics of the relay channel. The fading nature of the overall relay channel is faster than a point-to-point communication network due to the cascade of several receiver-to-relay, relay-to-relay, and relay-to-transmitter channels. By representing the transmission process as a dynamic system where both the transmitter-to-relay and relay-to receiver channels are modeled by autoregressive models, we derived particle filtering based algorithms for blind joint data detection and channel estimation. The performances of the algorithms were investigated using computer simulations and compared to the widely applied Linear Minimum Mean Square Estimation (LMMSE) method. It has been shown that the developed algorithms display superior performances over LMMSE in terms of symbol error rate vs. signal-to-noise-ratio (SNR).

Precoder Matrix Index (PMI) Selection in Finite-Rate Closed-Loop MIMO Systems.

The objective of the research was twofold. The first objective was to design a scheme which reduces the overhead of a closed-loop MIMO-OFDM to save feedback bandwidth, and the second of objective was to design reduced-complexity Precoder Matrix Index (PMI) selection algorithms to reduce receiver complexity. Multiple transmit and multiple receive antennas can be used to improve the reliability and capacity of a wireless communication systems. The performance of such systems, commonly known as MIMO systems, can further be improved, if the channel state information (CSI) is known at the transmitter. Full knowledge of the CSI allows the transmitter to precode the data so that the transmitted signal is steered in the direction of the strong eigen modes of the channel. When channel reciprocity does not hold, the CSI should in general be sent back from the receiver to the transmitter through a feedback channel. However, sending the full CSI through a feedback channel generates an overhead that requires a large bandwidth. One common approach for reducing the amount of feedback overhead is to use a codebook which consists of a finite set of precoding matrices that are predetermined by quantizing the MIMO channel. The codebook is made available to both the transmitter and the receiver, and the receiver based on channel condition selects the best precoding matrix from the codebook and sends back to the transmitter only the index of the selected precoding matrix. A system that uses such a scheme is known as finite-rate or limited feedback closed-loop MIMO system. In MIMO-OFDM system, shown in the figure below, each subcarrier has a different channel matrix, and, thus in general, different precoding matrix should be applied for each subcarrier to achieve optimal performance. Application of precoding for every subcarrier, however, incurs a huge cost in feedback bandwidth and receiver complexity. The amount of overhead generated by the per-subcarrier precoding is in direct proportion to the number of subcarriers; therefore, the required feedback bandwidth is significantly large. Moreover, because the receiver has to select the best PM for each subcarrier, the receiver complexity also increases in direct proportion to the number of subcarriers.

MIMO-OFDM receiver

Figure 1. A Closed-loop MIMO-OFDM system

The contribution of the research can be summarized as follows:

Development of per-band precoding scheme that reduces feedback overhead, and thereby, saving feedback bandwidth.
Derivation of per-band PM selection criteria for ML and LMMSE receivers using performance measures such as capacity, mean square error, and distance measures (DMs).
Development of reduced-complexity methods by exploiting frequency correlation and other channel statistics such as channel averages of neighboring subcarriers.
Introduction of, through its simulations, a design guide for trading performance enhancement to complexity reduction.

Distributed particle filtering for sensor networks: with limited communiucation among nodes

global likelihood function

Joint channel estimation, synchronization and data detection over frequency selective channels

This research work developed algorithms for data detection in single-input-single-output (SISO) wireless system to design efficient and reliable digital receivers. The major challenge in the design of such receivers lies in the ability to combat the ill-effects of the wireless channels on the transmitted data. Mobile wireless channels are characterized by time-variation, Doppler spread and multipath propagation that result in intersymbol interference (ISI) and severe distortion of the shape of the pulse of the transmitted data. Data detection in such environment requires the development of robust equalization techniques that can reverse or, at least, minimizes the deleterious effect of the channel.

Our approach to the development of the equalization algorithm is based on Bayesian estimation using particle filtering. By modeling the fading characteristics of the channel using an autoregressive process whose parameters are a function of the Doppler spread and symbol period, we formulated the problem as a dynamic state space model. Such representation allowed the application of particle filtering methods for the joint estimation of the channel and the data. The proposed algorithm exhibits some important advantages. First and foremost, the algorithm is blind and it achieves superior performance over conventional methods, Secondly, it is easily adaptable for problems with non-Gaussian additive noise. This research work was extended to include the estimation synchronization parameters such as symbol timing, carrier frequency and carrier phase.

UNIVERSITY of WASHINGTON \| BOTHELL Electrical Engineering \| Science & Technology


Home Teaching Research Publications Other
Current Projects	Past Projects Modeling of cascaded systems using non-Gaussian autoregressive models Dynamic systems are commonly modeled by Gaussian autoregressive (AR) processes. The popularity of AR processes for modeling dynamic systems is mainly due to the fact that they can be easily configured to capture the temporal statistical characteristics of the dynamic systems. In this research work, we considered the problem of modeling a system consisting of two cascaded subsystems, each of which represented by a complex Gaussian autoregressive (AR) process, by one AR model. Such combined representation of cascaded systems has a potential of simplifying the simulations of the cascaded processes and easing the complexity of estimating the model parameters. One motivating example where a combining model of cascaded systems can be useful is modeling of a single-hop relay channel. Rather than modeling the fading process of each channel of the transmitter-to-relay and relay-to-receiver by a separate mathematical model, it is convenient and efficient if the fading process of the overall channel is modeled by a single model that captures the statistical characteristics of the overall channel. To derive the combined model, we first studied the overall statistical characteristics of a cascaded dynamic system whose two subsystems are represented by complex Gaussian AR processes. It is shown that the marginal probability density functions (pdfs) of the real and imaginary parts of the combined process are Laplace pdfs. Therefore, to represent such a Laplacian dynamic systems, a Laplace autoregressive model that is configurable to accurately capture the statistical dynamics of the systems was developed. The proposed model possess a structure that is simple and convenient for generating realizations of Laplace processes. An important feature of the Laplace AR process is that its autocorrelation (AC) function follows Yule-Walker type of equations which simplifies its configuration to have matching autocorrelation values to that of the cascaded process. Data detection and channel estimation in Relay-based communication systems This research work developed effective algorithms for data detection and channel estimation for a single-hop relay communication system for different channel conditions. The conventional solution to the need for more capacity in cellular networks is to increase frequency reuse by cell division which requires installing new base stations (BS). However, the cost of installing new BS and its concomitant infrastructure is high. Recently, relay-based communication system is presented as an alternative solution to enhancing the capacity of a communication network. The use of relays in a communication network has several additional benefits including increasing coverage of a network, mitigating the effect of fading channels by increasing spatial diversity using cooperative communication, and improving energy efficiency of the overall system. However, implementation of relay-based communication system comes with its own set of technical difficulties. Data detection in such environment, for example, is challenging because of the unique statistical characteristics of the relay channel. The fading nature of the overall relay channel is faster than a point-to-point communication network due to the cascade of several receiver-to-relay, relay-to-relay, and relay-to-transmitter channels. By representing the transmission process as a dynamic system where both the transmitter-to-relay and relay-to receiver channels are modeled by autoregressive models, we derived particle filtering based algorithms for blind joint data detection and channel estimation. The performances of the algorithms were investigated using computer simulations and compared to the widely applied Linear Minimum Mean Square Estimation (LMMSE) method. It has been shown that the developed algorithms display superior performances over LMMSE in terms of symbol error rate vs. signal-to-noise-ratio (SNR). Precoder Matrix Index (PMI) Selection in Finite-Rate Closed-Loop MIMO Systems. The objective of the research was twofold. The first objective was to design a scheme which reduces the overhead of a closed-loop MIMO-OFDM to save feedback bandwidth, and the second of objective was to design reduced-complexity Precoder Matrix Index (PMI) selection algorithms to reduce receiver complexity. Multiple transmit and multiple receive antennas can be used to improve the reliability and capacity of a wireless communication systems. The performance of such systems, commonly known as MIMO systems, can further be improved, if the channel state information (CSI) is known at the transmitter. Full knowledge of the CSI allows the transmitter to precode the data so that the transmitted signal is steered in the direction of the strong eigen modes of the channel. When channel reciprocity does not hold, the CSI should in general be sent back from the receiver to the transmitter through a feedback channel. However, sending the full CSI through a feedback channel generates an overhead that requires a large bandwidth. One common approach for reducing the amount of feedback overhead is to use a codebook which consists of a finite set of precoding matrices that are predetermined by quantizing the MIMO channel. The codebook is made available to both the transmitter and the receiver, and the receiver based on channel condition selects the best precoding matrix from the codebook and sends back to the transmitter only the index of the selected precoding matrix. A system that uses such a scheme is known as finite-rate or limited feedback closed-loop MIMO system. In MIMO-OFDM system, shown in the figure below, each subcarrier has a different channel matrix, and, thus in general, different precoding matrix should be applied for each subcarrier to achieve optimal performance. Application of precoding for every subcarrier, however, incurs a huge cost in feedback bandwidth and receiver complexity. The amount of overhead generated by the per-subcarrier precoding is in direct proportion to the number of subcarriers; therefore, the required feedback bandwidth is significantly large. Moreover, because the receiver has to select the best PM for each subcarrier, the receiver complexity also increases in direct proportion to the number of subcarriers. Figure 1. A Closed-loop MIMO-OFDM system The contribution of the research can be summarized as follows: Development of per-band precoding scheme that reduces feedback overhead, and thereby, saving feedback bandwidth. Derivation of per-band PM selection criteria for ML and LMMSE receivers using performance measures such as capacity, mean square error, and distance measures (DMs). Development of reduced-complexity methods by exploiting frequency correlation and other channel statistics such as channel averages of neighboring subcarriers. Introduction of, through its simulations, a design guide for trading performance enhancement to complexity reduction. Distributed particle filtering for sensor networks: with limited communiucation among nodes Sensor networks are composed of a large number of sensor nodes deployed in a large span of area providing the capability of monitoring a wide spatial environment. A popular application of sensor networks is the detection and tracking of objects in real-time. Such application requires the sensors to collaborate for sequential and real-time processing of the data for accurate estimation of the parameters of interest. Most of the detection and tracking problems are modeled as nonlinear dynamic systems. Popular methods to estimate parameters of interest of nonlinear and/or non-Gaussian dynamic systems is particle filtering which is a Bayesian method that sequentially approximate the posterior distribution of the parameter of interest using particles (samples) and their associated weights by applying Monte Carlo techniques. In distributed particle filtering, each sensor node locally carries out particle filtering approximation of the global posterior distribution by collaborating with the other sensors. However, the exchange of a lot of information among sensors is costly in terms of bandwidth and energy requirement. Clearly, the exchange of the particle and weights of each local posterior distribution to all sensors is impractical in terms of energy and bandwidth demands. The objective of this research work is to design an algorithm that requires reduced amount of information exchange among the nodes for computing the global posterior distribution. It can be shown that the global likelihood function encompasses the information from all the sensors to approximately compute the global posterior distribution. We, therefore, proposed an algorithm that allows the sensors to collaboratively compute the global likelihood function that take into account measurements from all the sensors. To compute the global likelihood, each sensor first approximates its local likelihood function using Gaussian function. Such approximation saves energy and communication overhead because it requires the sensors to exchange only the mean and the covariance of the approximated Gaussian local likelihood functions with their neighboring sensors. The mean and covariance of the global likelihood functions is then built using an average consensus filter or by implementing forward-backward propagation strategy. Joint channel estimation, synchronization and data detection over frequency selective channels This research work developed algorithms for data detection in single-input-single-output (SISO) wireless system to design efficient and reliable digital receivers. The major challenge in the design of such receivers lies in the ability to combat the ill-effects of the wireless channels on the transmitted data. Mobile wireless channels are characterized by time-variation, Doppler spread and multipath propagation that result in intersymbol interference (ISI) and severe distortion of the shape of the pulse of the transmitted data. Data detection in such environment requires the development of robust equalization techniques that can reverse or, at least, minimizes the deleterious effect of the channel. Our approach to the development of the equalization algorithm is based on Bayesian estimation using particle filtering. By modeling the fading characteristics of the channel using an autoregressive process whose parameters are a function of the Doppler spread and symbol period, we formulated the problem as a dynamic state space model. Such representation allowed the application of particle filtering methods for the joint estimation of the channel and the data. The proposed algorithm exhibits some important advantages. First and foremost, the algorithm is blind and it achieves superior performance over conventional methods, Secondly, it is easily adaptable for problems with non-Gaussian additive noise. This research work was extended to include the estimation synchronization parameters such as symbol timing, carrier frequency and carrier phase.