Date of Award


Publication Type

Doctoral Thesis

Degree Name



Electrical and Computer Engineering


Applied sciences, Autocorrelation minimization, Channel equalization, Multicarrier modulation


Kemal E. Tepe


Esam Abdel-Raheem




Blind adaptive algorithm that updates time-domain equalizer (TEQ) coefficients by Adjacent Lag Auto-correlation Minimization (ALAM) is proposed to shorten the channel for multicarrier modulation (MCM) systems. ALAM is an addition to the family of several existing correlation based algorithms that can achieve similar or better performance to existing algorithms with lower complexity. This is achieved by designing a cost function without the sum-square and utilizing symmetrical-TEQ property to reduce the complexity of adaptation of TEQ to half of the existing one. Furthermore, to avoid the limitations of lower unstable bit rate and high complexity, an adaptive TEQ using equal-taps constraints (ETC) is introduced to maximize the bit rate with the lowest complexity. An IP core is developed for the low-complexity ALAM (LALAM) algorithm to be implemented on an FPGA. This implementation is extended to include the implementation of the moving average (MA) estimate for the ALAM algorithm referred as ALAM-MA. Unit-tap constraint (UTC) is used instead of unit-norm constraint (UNC) while updating the adaptive algorithm to avoid all zero solution for the TEQ taps. The IP core is implemented on Xilinx Vertix II Pro XC2VP7-FF672-5 for ADSL receivers and the gate level simulation guaranteed successful operation at a maximum frequency of 27 MHz and 38 MHz for ALAM-MA and LALAM algorithm, respectively. FEQ equalizer is used, after channel shortening using TEQ, to recover distorted QAM signals due to channel effects. A new analytical learning based framework is proposed to jointly solve equalization and symbol detection problems in orthogonal frequency division multiplexing (OFDM) systems with QAM signals. The framework utilizes extreme learning machine (ELM) to achieve fast training, high performance, and low error rates. The proposed framework performs in real-domain by transforming a complex signal into a single 2–tuple real-valued vector. Such transformation offers equalization in real domain with minimum computational load and high accuracy. Simulation results show that the proposed framework outperforms other learning based equalizers in terms of symbol error rates and training speeds.