Two-dimensional block-based reception for differentially encoded OFDM systems : a study on improved reception techniques for digital audio broadcasting systems

Abstract

Digital audio broadcast (DAB), DAB+ and Terrestrial-Digital Multimedia Broadcasting (T-DMB) systems use multi-carrier modulation (MCM). The principle of MCM in the DAB-family is based on orthogonal frequency division multiplexing (OFDM), for which every subcarrier is modulated by p 4 differentially encoded quaternary phase shift keying (DE-QPSK). In DAB systems convolutional codes and interleaving are used to enable DAB receivers to perform error correction. The objective of the work, described in the thesis, is to improve reception techniques for DAB, DAB+, and T-DMB systems. In the thesis, two-dimensional (2D) block-based reception for differentially encoded OFDM systems is investigated. The blocks are based on the time and frequency dimension. Commonly used DAB receivers perform non-coherent two-symbol differential detection (2SDD) with soft-decision Viterbi decoding. It is well-known that 2SDD can be improved if the detection is based on more than two received symbols as, e.g., in noncoherent multi-symbol differential detection (MSDD). For improving the performance of the demodulation procedures of DAB-like streams, demodulation based on 2D blocks of received symbols with a decomposed demodulation trellis is proposed in the thesis. Peleg and Shamai [58] demonstrated that iterative techniques could increase the performance of the demodulation procedures of DE-QPSK streams even further. In the thesis, their approach is generalized to the 2D setting where again the decomposed demodulation trellis is used. In this way a problem connected to the small lengths of the trellises for each subcarrier is solved. The application of these iterative decoding techniques in DAB receivers is only feasible if their complexity can be drastically reduced. A significant complexity reduction is obtained by iterating only in a dominant sub-trellis of the decomposed demodulation trellis. In this way, a real-time and bit-true DAB receiver based on iterative decoding techniques is realized In Chapter 2, simulationmodels are introduced. These models are later applied to evaluate the proposed reception methods. The Additive White Gaussian Noise (AWGN) channel model with an input power constraint and the channel model for M-level PSK are first discussed. In addition, the TU-6 (Typical Urban 6 taps) channel model defined in COST-207 [1] is introduced. This channelmodel is commonly used to assess the performance of DAB, DAB+, or T-DMB transmission. Finally, the basic elements of a DAB transmitter and a standard receiver are described. In Chapter 3 of the thesis, the state of the art in non-iterative detection and decoding techniques for DE-QPSK streams with convolutional encoding is described. First, as a reference, coherent detection of DE-QPSK with soft-decision Viterbi decoding is studied. Then it is demonstrated that 2SDD of DE-QPSK with soft-decision Viterbi decoding degrades the performance. This non-coherent differential detection scheme can be improved by, for example, MSDD, which is a maximum likelihood procedure for finding a block of information symbols after having observed a block of received symbols. For large numbers of observations, the performance of MSDD approaches the performance of coherent detection of DE-QPSK. Since reference symbols (pilots) are lacking for DAB systems, detection based on observing multiple received symbols is a technique that could lead to reception improvement for DAB receivers. By applying this technique, as will be shown later, a DAB receiver approaches the performance of a receiver that performs coherent detection of p 4 -DE-QPSK with soft-decision Viterbi decoding. In Chapter 4, a-posteriori symbol probabilities and log-likelihood ratios (LLRs) for coherently detected p 4 -DE-QPSK are studied. It is demonstrated, as an extension to the results known in the literature, that an approximation of maximum a-posteriori (MAP) symbol detection, based on selecting dominant exponentials, leads to MAP sequence detection. To improve the performance towards MAP symbol detection, a better approximation is proposed. This approximation relies on piecewise-linear fitting of the logarithm of the hyperbolic cosine and results in a performance quite close to that of MAP symbol detection. For the coded case, where the symbols are produced by convolutional encoding and Gray mapping, the LLRs are investigated. Again a simple approximation based on selecting dominant exponentials and an improved approximation relying on piecewise-linear fits, is proposed. As in the uncoded case, the improved approximation gives a performance quite close to ideal. These improved approximations are also of interest for DAB systems, as will be shown later, if 2D and trellis-based detection is considered as a reception technique. Peleg et al. [56][57][58] and Chen et al. [18] demonstrated that iterative decoding techniques developed by Benedetto et al. [9] for serially concatenated convolutional codes lead to good results for the concatenation of convolutional and differential encoding, also referred to as Turbo-DPSK. In Chapter 5 the iterative decoding procedures corresponding to these serially concatenated codes are explained. In this chapter also parallel concatenated systems, turbo-codes, first described by Berrou et al. [11] are considered. The iterative decoding procedures for the serially concatenated codes as well as for the turbo-codes are based on modified versions of the BCJR algorithm [4]. The approach taken in Chapter 5 to explain these iterative decoding procedures, is similar to the approach Gallager [32] followed to investigate iterative procedures for decoding low-density parity-check (LDPC) codes. This way of explaining iterative decoding procedures for the serially concatenated codes as well as for the turbo-codes does not appear in the literature. It is well-known that iterative (turbo) decoding procedures approach channel capacity, e.g., in the AWGN setting. For that reason, in Chapter 6 and Chapter 7, iterative decoding techniques for DAB-like streams are studied. At the time that the DAB standard was proposed, the results of Berrou et al. [11] on turbocodes were not available. As a consequence, it is not a common practice to use iterations in DAB receivers. In Chapter 6, motivated by encouraging results on Turbo-DPSK, trellis decoding and iterative techniques for DAB receivers are investigated. Specifically, the usage of 2D-blocks and trellis decomposition in decoding is considered. Each 2D-block consists of a number of adjacent subcarriers of a number of subsequent OFDM symbols. Focussing on 2D-blocks was motivated by the fact that the channel coherence-time is typically limited to a small number of OFDM symbols, and that DAB-transmissions use time-multiplexing of services, which limits the number of OFDM symbols in a codeword. Extension in the subcarrier direction is required then to get reliable phase estimates. The trellis-decomposition method allows for an estimation of the unknown channel phase, since this phase relates to sub-trellises. A-posteriori sub-trellis probabilities are determined, and these probabilities are used for weighting the a-posteriori symbol probabilities resulting from all the sub-trellises. Alternatively, a dominant sub-trellis can be determined from the a-posteriori sub-trellis probabilities and the a-posteriori symbol probabilities corresponding to this dominant sub-trellis can be used. This dominant sub-trellis approach results in a significant complexity reduction, which is the subject of Chapter 7. In the first part of Chapter 7, complexity reduction of the inner decoder is investigated. This complexity reduction is realized by choosing, based on a-posteriori sub-trellis probabilities, in two different ways a dominant sub-trellis. In the first approach, a method is investigated that is based on finding, at the start of a new iteration, the dominant subtrellis first and then do the forward-backward processing for demodulation only in this dominant sub-trellis. The second approach involves choosing the dominant sub-trellis only once, before starting with the iterations. In the second part of Chapter 7, an implementation of a MAP channel-phase estimator based on the second dominant sub-trellis approach is described. In addition, an implementation of a channel-gain estimator based on the received symbols within a 2D-block is discussed. Finally, a real-time and bit-true DAB-receiver is sketched. This DAB receiver operates according to the proposed 2Dblock- based iterative decoding procedure within a dominant sub-trellis obtained by the second method. The performance improvements of this DAB receiver are evaluated for various numbers of iterations, block-sizes, and Doppler-frequencies. The main conclusions can be found in Chapter 8. For the non-iterative 2D-case, investigations show that the performance of non-coherent detection based on trellis-decomposition is very close to the performance of coherent detection of DE-QPSK. The gain of 2D trellis-decomposition is modest compared to the standard 2SDD technique. Iterative 2D procedures result in a significantly larger gain. In this context, it needs to be emphasized that part of this gain comes from the 2D-processing. The dominant sub-trellis approach appears to be crucial for achieving an acceptable complexity reduction.