Unique Word OFDM (orthogonal frequency division multiplexing) is an attractive alternative to OFDM with cyclic prefix, which is adopted for data transmission in standards like DSL, LTE, DVB and IEEE 802.11 (WLAN). In this signaling concept, a deterministic sequence, a ``unique word'' (UW), is inserted into the transmit stream, instead of a cyclic copy of the data. Furthermore, this UW is part of the IDFT (inverse discrete Fourier transform) interval.
This property distinguishes UW-OFDM from most other OFDM variants, while it offers the same advantages as the conventional OFDM (free of inter-symbol interference, diagonalization of the channel matrix).
By defining of a sequence in time domain, some capacity has to be allocated for redundancy in frequency domain. This redundancy solely depends on the transmit data (and defined system parameters) and can be utilized for a reliable recovery of the data.
In order to exploit this redundancy, sophisticated receiver structures need to be employed, which is topic of this work. The achieved gain can be used for a higher data rate, range, reliability, capacity or battery lifespan.
Methods to generate valid UW-OFDM symbols are introduced in two variants: The systematic generation of UW-OFDM symbols, which can be done directly or in two steps, and the non-systematic generation. An analysis of the mean transmit energy of all generation methods sheds light on their suitability for communication systems and reveals possibilities for optimization.
The main part of this work is about suited receivers for UW-OFDM that are able to reconstruct the data reliably, after transmission over a dispersive channel. Besides the estimated transmit symbols, all these receivers need to provide reliability information, which enables a channel decoder to achieve better decoding results. All receivers are investigated regarding their bit error performance with and without channel coding, in the AWGN (additive white Gaussian noise) channel as well as in a multipath environment.
Besides two rather intuitively derived, two more optimum linear receivers are discussed, which emerge from classical as well as Bayesian estimation theory: The BLUE (best linear unbiased estimator) and the LMMSE (linear minimum mean square error) estimator. The computational complexity of all these receivers is analyzed for both OFDM symbol generation approaches and compared numerically.
When using real transmit symbol constellations, the LMMSE estimator can be outperformed by the WLMMSE (widely LMMSE) estimator. Furthermore, a symbol scaling effect can be identified for these Bayesian receivers.
This turns out to be harmful for the detection quality with higher order constellations, such as 16-QAM or 4-ASK. Symbol scaling compensated versions of the LMMSE and WLMMSE estimators are introduced and their performance documented.
As another main topic of this work, a few nonlinear receivers are discussed, starting with two decision directed concepts. First, a method for noise interpolation is introduced, which exploits the correlation of the data symbols after an LMMSE estimation, in order to obtain improved estimates. It turns out that the selection of the samples which are used for estimation is decisive for the performance of this receiver.
In decision feedback equalization, the influence of detected data symbols on the receive signal is subtracted iteratively, in order to allow for a more reliable decision of the remaining symbols.
Here, the order of detection is crucial for the decision quality.
For the derived linear UW-OFDM system model, a maximum-likelihood sequence estimation (MLSE) yields the best estimates possible. However, due to its computational complexity, it is unsuitable for practical application.
As a practical realization of the MLSE, sphere decoding is presented, which obtains the same results with acceptable effort.
For the determination of reliability information, however, a mathematical approximation and a limitation of parameter dynamics has to be applied, to keep the complexity in adequate limits, which destroys the optimality of the method.
An investigation of QR decomposition, as it is directly used for sphere decoding and in a version of decision feedback equalization, shows that the way, how the QR decomposition is computed, has significant impact on runtime or detection performance, respectively.
An overall performance investigation reveals that the nonlinear receivers clearly outperform the LMMSE estimator in uncoded transmission. If channel coding is used, they are still able to achieve a small gain over the best performing linear estimator.
However, the LMMSE estimator constitutes a highly reasonable compromise when performance and complexity are taken into account.