Abstract
Knowledge of the receiver signal-to-noise ratio (SNR) is often required in many modern wireless communication systems. For example, mobile-assisted handoff[1], selective combining [1], power control [1], data rate adaptation [2], dynamic channel assignment, and iterative MAP turbo code decoding [3], etc., require SNR as an important reference. In the past engineering practice, it usually used estimation of the total signal-plus-noise power instead of that of the SNR for convenience. However, performance can be improved if we use the real SNR estimate, which makes the investigation of SNR estimation techniques interesting and essential. Most of the previous works [4]–[8] related to SNR estimation were considered in additive white Gaussian noise (AWGN) channels, while [9] considered the SNR estimation in time-selective fading channels for single carrier systems. Some considered the SNR estimation in generalized fading channels for single carrier system, too. Orthogonal frequency division multiplexing (OFDM) transmission is popular in various applications in recent years, which is a technique to transmit data over a number of subcarriers. In [10] a method for SNR estimation in OFDM system is proposed, yet, however, only AWGN and frequency non-selective channels are considered. In this thesis, we are going to discuss several SNR estimation techniques for OFDM transmission, and the channel model considered here is a frequency selective channel with AWGN. We modify several previously proposed SNR techniques [4] to our system, such as maximum-likelihood (ML) estimation, second- and fourth-order moments (M2M4) estimation, and signal-to-variation ratio (SVR) estimation. In addition, the Cramer-Rao Lower Bound (CRLB) [11], [12], which is a well known lower bound for the variance of any unbiased estimator, is derived for our system and compared to the simulated results. The simulation results show that the ML estimator performs best and its normalized mean-square error (NMSE) is close to the CRLB if the block size is large enough. However, the ML estimation has to be data-aided (DA) while the M2M4 and the SVR do not require knowledge of the transmitted data. This is the trade-off. Finally, we consider the estimators in the cases with frequency offset, timing offsets and the joint offset. The theoretical and simulation results show that the ML estimator suffers from the frequency offset but not the timing offset while the M2M4 and the SVR do not suffer from neither the frequency nor the timing offset. For the joint cases of timing and frequency offset, ML also degrades while the M2M4 and the SVR do not degrade, neither. The thesis is organized as follows. In Chapter 2, we give an overview of OFDM. A review of some proposed techniques for SNR estimation is given in Chapter 3. In Chapter 4, we derive SNR estimators based on three techniques: ML, M2M4, and SVR, and derive the CRLB for our system. We then consider the situations with frequency offset and evaluate the impact on our estimators in Chapter 5 and discuss the timing offset effects for our estimators in Chapter 6. Finally, we give some simulation results in Chapter 7 and conclusion in Chapter 8.