| Title: | Design of multiple systems with high spectral efficiency |
| Problem formulation: |
In the field of signal processing for the digital communications, wireless systems with high spectral efficiency currently are in paramount importance in commercial and military applications.
The proposed solution for the third generation wireless system is based on the multiple access and asynchronous transfer.
No synchronization is needed in the uplink transmission, i.e. from the mobile to the base station (BS) in the FDD mode.
Additionally, the use of multiple antennas at the receiver are envisaged.
To improve the spectral efficiency for the currently system (IS-95), the spreading factor is relatively small (from 1 to 32).
Moreover, the impulse response of the equivalent baseband propagation channel spreads over several symbols (3 to 5) for the expected data rates.
This phenomenon is due to various reflections during the propagation: buildings, trees in urban zone, natural obstacles in country zone (mobile communications) and tropospheric, ionospheric zone (satellite communications).
In these cases, the channel is frequency selective.
These factors significantly increase the multiple access and the intersymbol interference (MAI and ISI) which dramatically deteriorate the performance of the transmission.
To mitigate these interference, a powerful channel equalizer at the base station is necessary.
However, conventional receivers (RAKE or MMSE receivers) need to use a large training sequence, i.e. about 50% of the total frame length instead 15% for the GSM. This directly affects the spectral efficiency.
Under the previous assumptions, the challenge for the channel equalizers of future generations includes:
|
| Post-doctoral work: |
My PhD studies are done with ECE Dept, Rice University, Houston, TX advisored by Pr. B. Aazhang and Dr. De Lathauwer. In the next generation, the systems will use multiple antennas (at least two for the base station) which adds spatial diversity. During my PhD studies, we derive an algorithm which uses more efficiently the combination between the implicit spectral diversity of the DS-CDMA and the spatial diversity. This is a generalization of the parallel factor analysis (PARAFAC) in the convolutional case applied to the DS-CDMA system. The observed data are stacked in a three-way array. Under the narrow-band assumption at the antenna array which is valid for GSM or UMTS system, the decomposition is unique to a permutation matrix and a scaling factor. The optimization algorithm we use is the alternating least squares algorithm (ALS). The simulated performance are very close to the Zero-Forcing which requires the knowledge of the propagation channel and the complex antenna gain. The major advantages are (i) that the symbol frame length required for the method is small: typically 10-50 samples and (ii) that it is the robust to the interference. The PARAFAC analysis seems to be very promising even in the convolutional case. Indeed, it fully exploits the spectral and the spatial diversity. First, we need to show the uniqueness of the three-way tensor decomposition in the convolutional case. It is not trivial since multipath adds a dimension in the model: we will retrieve the sources up to a unitary matrix. To avoid this ambiguity, we have to take into account of the convolutive structure (or equivalently of the Toeplitz structure for the mixture matrices). Second, we propose to improve our algorithm in two ways : (1) by taking into account of the spreading sequences. If the spreading sequences (or less restrictive, the familie of the spreading sequences) are known at the receiver, we can introduce their knowledge in the ALS algorithm. It will be interesting to see its influence on the convergence speed of the algorithm through the initialisation and the ill-posed systems, (2) by taking into account of the constant modulus of the transmitted sources (or another statistical properties). This a priori knowledge can be introduced by adding a penalty constraint. It will be also interesting to see its influence on the convergence speed and on the obtained performance. Another point is to extend this method to the case of spatial diversity at the receiver and at the transmitter. |