Signal Processing for Communications

While digital communications is essentially a detection problem, which means the estimation of discrete unknowns, most communication systems involve also a number of unknown parameters that need to be estimated. This is where statistical signal processing (SSP) comes in. SSP may play a significant role in the design of many receiver algorithms, at the physical layer:

Channel estimation and modeling. The correct formulation of an estimation problem typically involves a tradeoff between model approximation error and estimation error in the model. With modern communication channels being quadruply selective (time, frequency, space/polarization, users), the choice of channel model may affect the channel estimation quality significantly. Currently, Bayesian techniques assume Gaussian priors (Rayleigh fading), leaving a lot of room for further improvements. One possible direction is sparsity.
Blind channel estimation. For frequency-selective channels, the decision feedback equalizer (DFE) has been known to be a canonical receiver. Without channel knowledge at the receiver, it appears that the DFE with semiblind channel estimation remains canonical, exploiting future received signal (blind part) and past received and transmitted signal (training part). Though (semi)blind channel estimation is a topic of the nineties, many questions remain concerning robustness and complexity. Recently OFDM has been shown to offer significant complexity reduction. The topic here is more generally blind transceiver techniques with recent developments such as Blind Interference Alignment.
Robust transceivers: determination of transmit and receive filters that account for channel estimation errors. Worst case designs may be appropriate for the case of quantized channels. MMSE designs are more appropriate for the case of Gaussian channel error.
Channel shortening for suboptimal receivers in single-carrier systems (GSM in 2G, SC-FDMA in 4G). The topic here is optimal determination of filters appearing in a receiver structure that is intermediate between a Viterbi equalizer and a DFE.
Diversity analysis of transceiver techniques, especially comparing the diversity of suboptimal simplified transceivers to their optimal counterpart. Interesting recent progress on this topic showed that even a simple linear MMSE receiver exhibits full diversity at low rates. Also the diversity of receivers using semiblind channel estimates is being investigated.
SSP motivated design of space-time filters for space-time coding. MIMO CDD (Cyclic Delay Diversity) is the main open loop MIMO technique in LTE. It is based on a particular MIMO paraunitary filter that was introduced by the group, and that combines spatial spreading (or constellation rotation precoding) and delay diversity, for simultaneous space-time coding and spatial multiplexing.

Some SSP work is very standard-specific. Some existing examples are:

Single Antenna Interference Cancelation (SAIC) in GSM. This part of the GSM standard is based on earlier work in this group, which showed that the GMSK modulation is essentially a modulated form of BPSK transmission, for which so-called widely linear processing is as effective as linear processing in the case of two receive antennas.
Chip Equalization in 3G systems. The synchronicity of the downlink in DS-CDMA can be exploited to generalize the RAKE receiver (matched filter) to a channel equalizer –correlator receiver. This work in our group led to the concept of Advanced Receivers in 3G systems.
With OFDM now being the dominant modulation technique, a number of OFDM-specific problems have become relevant, such as inter-carrier interference suppression, peak-to-average power ratio control, and simplified frequency domain processing.

SSP may have an impact on resource allocation:

Spectrum sensing in cognitive radio, involving e.g. compressive sensing and sparsity exploitation.

Furthermore, SSP is also required for many operations at the applications layer. The scope of this largely transcends the mobile communications group (consider e.g. video and biometrics in multimedia processing). However, some topics are treated in the Com Theory group, such as:

Geolocation: wireless network based positioning techniques as alternatives to the more expensive GNSS based approach, or for indoor applications. Techniques considered here are propagation law based approaches using channel response measurements, single-bounce multipath models, fingerprinting approaches (with possible fingerprints: from simple received signal strengths to power delay Doppler space profiles). The dual aspect of improved communication systems exploiting geolocation (and mobility) information is also being pursued. A simple example is position trajectory based anticipative handover.
Adaptive Kalman Filtering. The Kalman filter has regained significant interest in recent years, due to its applicability in position tracking, wireless channel tracking, Bayesian adaptive filtering, source separation, etc. In almost all applications, the state-space model involved is only known up to some unknown parameters, which need to be estimated jointly. A number of techniques may be used to “adapt” the Kalman filter, including the Extended Kalman Filter, Expectation-Maximization (EM), Variational Bayes (VB) estimation etc. The relative performance of these approaches (as a function of complexity also) is still largely unknown.
Audio processing. A number of operations arise in handsfree communications for instance: acoustic echo cancellation, speech enhancement, mono and multi-microphone audio source separation, dereverberation. Some of the techniques for multi-microphone processing are quite related to multi-antenna processing, though in the audio case the wideband formulation is required.