Reduced Complexity Detection Techniques for Multi-Antenna Communication Systems (2013)
Type of ContentTheses / Dissertations
Thesis DisciplineElectrical Engineering
Degree NameDoctor of Philosophy
PublisherUniversity of Canterbury. Electrical and Computer Engineering
AuthorsTasneem, Khawaja Tauseefshow all
In a multiuser system, several signals are transmitted simultaneously within the same frequency band. This can result in significant improvements both in spectral efficiency and system capacity. However, a detrimental effect of the shared transmissions (both in time and bandwidth), is that the signal received at the base station (BS) or access point (AP) suffers from cochannel interference (CCI) and inter-symbol interference (ISI). This situation presents challenges to receiver design. To combat the destructive nature of multipath fading, a receiver often employs multiple antennas to collect the faded superimposed versions of the transmitted signals. The multiple signals are combined and processed in such a way that the effects of CCI and ISI are minimized and the desired information is reliably recovered. The situation is even more challenging when the system is operating under overload, i.e. when there are fewer receive antennas than there are transmitted signals. Multiuser detection (MUD) is used to simultaneously estimate the information sent by the transmitters. To do this, the receiver exploits differences among the cochannel signals (through unique spatial signatures in this case).
We consider a cochannel communication system where multiple transmitted signals arrive at a receiver (equipped with multiple receive antennas) after propagating through a Rayleigh fading channel. It is assumed that the receiver is operating in an overloaded scenario. For such systems, an optimum maximum a posterior probability (MAP) detector estimates the transmitted signal by maximizing the probability of correct decision. The MAP detector reduces to the maximum likelihood (ML) detector when all the transmitted signals are equiprobable. The computational complexity of both MAP and ML detectors increases exponentially with the number of transmitted signals and the channel memory. For large systems suffering severe CCI and ISI, this is clearly not a good choice for real-time implementation due to the associated computational expenses. The main factors that influence the complexity of MAP / ML detection are: (i) the number of transmitted signals (or equivalently the number of users sharing the system resources), (ii) modulation alphabet size, and (iii) length of the channel memory. On the other hand, linear detection approaches fail to offer acceptable performance while other nonlinear sub-optimum approaches incur high computational costs for reasonably improved system performance and exhibit an irreducible error-floor at medium to high signal to noise ratio (SNR) values.
We develop receiver signal processing techniques for the frequency-flat fading channel (where all the multipaths of the transmitted signal arrive at the receiver within a symbol period). We develop an ant colony optimization (ACO) assisted soft iterative detection approach for binary phase-shift keying (BPSK) modulated signals which employs a simplified MAP criteria to extract the most probable signals from the search space. The structure of the receiver is such that it can continue operating under overloaded conditions. The technique achieves near maximum likelihood (ML) performance in critically loaded cases using much lower complexity. For the challenging case of overload it still offers performance close to ML at low to moderate SNR values. Second, an integrated framework comprising of ACO metaheuristic and a recursively defined ML search criteria is developed to handle multilevel modulations. The proposed receiver is capable of achieving near-ML performance for the considered system with significant savings in computational complexity. The receiver framework is independent of the system loading condition, and therefore it remains suitable for overloaded scenarios. Due to the branch and bound nature of the algorithm, an exact expression for the complexity cannot be determined. Instead, an upper bound on computational complexity is developed.