What: Post-doctoral fellowship in signal processing and information theory for cloud broadcast networks
Where: Laboratory IETR, INSA Rennes, France.
When: From 1st February 2019 for 20 months
Context: FUI project with industrials
Collaboration: Philippe Mary and Matthieu Crussière Univ. Rennes, INSA, IETR UMR CNRS 6164
IETR: Institut d’Electronique et de Télécommunications de Rennes
Start date : 1st February 2019– duration :20 months
This postdoc position takes part of the fully funded FUI project, CloudCast, that aims at proposing virtualized baseband and protocols architectures for digital video broadcasting networks. In this kind of architecture, the baseband processing that used to be realized on dedicated hardware devices are now executed in the common servers (a.k.a. the cloud) that can be located far away from the base stations. The latter only keep the RF components on their own. The interest of this architecture is the Capex, Opex reduction, the possibility to make the network functionalities evolve, the network reconfigurability.
The drawback of the virtualized architecture lies in the increase of the data rate in IP links. Indeed, the deportation of baseband processing, and in particular the fast Fourier transform (FFT), in a remote location w.r.t. the transmitting elements of the BS, implies important data rates on the fronthaul links between the cloud and BS. For broadband networks, even if optical fiber is envisaged to support this huge data rate requirement, this could be not sufficient with the arrival of 5G and the massive user connectivity . In the broadcast industry, high capacity optical fibers are not always available. The solution is hence to reduce the data rate of I/Q samples in the fronthaul by some compressing techniques. This research position aims at reducing the useful data rate, ideally with no signal quality loss or under a controlled loss. The signal quality is usually measured with the variance of the quantified signal w.r.t. the non-quantified one. This leads to criterion such as the modulation error ratio (MER) that should be kept as high as possible. One of the solution is to use non-uniform quantization in order to optimize the number of bits used to carry the information according to the probability density measure of their apparition. Lloyd-Max quantization algorithm is particularly well adapted to quantize Gaussian samples and can be used for OFDM-based signal , moreover vector quantization allows to deal with time correlation between output samples from IFFT and it may achieve better compression rate than scalar quantization . Moreover, the non-uniform quantization can be jointly used with some time-spectral redundancy removal, e.g. cyclic prefix removal, and entropy-based encoding as well . Another important aspect to take into account is the peak to average power ratio (PAPR) that transmitter tries to reduce. Indeed, in order to adapt the signal dynamic to the most energy efficient part of power amplifier (PA) characteristic, pre-distortion signal processing techniques are used in order to compensate the non-linearity of PA but may lead to an increase of MER.
This position aims at tackling the problem of MER optimization in the global sense, i.e. considering data rate compression and PAPR reduction techniques. There is a double challenge here. First of all, the pre-distortion operation increases somehow the dynamic of the signal in order to compensate the loss induced by PA non-linearity. In the other hand, non-linear quantization compresses the signal in order to quantize more precisely the samples with a high probability of apparition. These two operations may be thought in a global way, or the optimality of the separation should be theoretically demonstrated, in order to compress as much as possible the data rate in the fronthaul while keeping a satisfactory MER. Second, the pre-distortion and non-linear quantization are not performed in the same place, i.e. pre-distortion is done close to PA, hence at the end of the fronthaul, while quantization is located after the IFFT operation in the cloud. Optimizing both procedures suggests to have a feedback link in order adapt the pre-distortion parameter to the quantized codebooks. Moreover, the end-to-end link, i.e. including the wireless propagation channel, may be taken into account in the performance evaluation.
 “C-RAN The Road Towards Green RAN”, White Paper China Mobile Research Institute, 2011.
 A. Gersho, R. M. Gray, “Vector Quantization and Signal Compression”, Springer US, 1992
 H. Si, B. L. Ng, M. S. Rahman, J. Zhang, “A Novel and Efficient Vector Quantization Based CPRI Compression Algorithm”, IEEE Transactions on Vehicular Technology, 2017
 L. Ramalho, M. N. Fonseca, A. Klautau, C. Lu, M. Berg, E. Trojer, S. Höst, “An LPC-Based Fronthaul Compression Scheme”, IEEE Communications Letters, 2017.
 S. H. Park, O. Simeone, O. Sahin and S. Shamai, « Fronthaul Compression for Cloud Radio Access Networks. Signal Processing advances inspired by network information theory », IEEE Signal Processing Magazine, Nov. 2014.
The candidate should have a PhD degree in one of the following field: signal processing, electrical engineering, telecommunications. He/She should have a strong background in signal processing for wireless communications, information theory, mathematics. The candidate should be an expert in Matlab and C/C++ programming.
Cloud radio access network, digital communications, OFDM, sampling and quantization theory, data rate compression, probabilities.
How to apply:
- Email a motivation letter
- Full CV with complete list of publications
- Reviewer reports of the thesis and defense report, if available
- Applications will be reviewed when they arrive until one candidate is selected
Dr. Matthieu Crussière et Dr. Philippe Mary