Summary of Newcom Dept 1 meeting

Newcom-logo  ist-logo

Newcom (EC Contract no. 507325)

Summary

The Newcom Dept 1 Workshop took place in Paris on June 25, 2004, at the Ecole Nationale Supérieur des Télécommunications.  For a list of the attendees, click here.

The program was as follows:
For the abstracts of all the talks, click here.

Overview of Talks

The introduction by Phil Regalia explained the rationale for the workshop.  The main deliverable in the initial phases of Department 1 is the work plan to fill the identified knowledge gaps, providing critical parts of our plan aiming to foster further research and development, and which justify collaborative (rather than solo) research efforts.  The progress thus far has amply identified knowledge gaps (recent document available here), and the workshop's goals included identifying collaborative paths forward.  We aim to achieve a working draft of the research plan for Department 1 by September, coinciding with the beginning of the academic year.

The talk on OFDM, a subject which has garnished considerable effort from within Newcom and beyond, was presented by Hanna Bogucka, emphasizing the current state of the art as well as practical limitations which serve to identify knowledge gaps.  These include new techniques to reduce the peak- to average-power ratio, fast link adaptation techniques, cross-layer optimization starting from the physical layer, and multi-criteria optimization techniques combining high capacity and low energy usage.

Erdal Panayırcı's talk recalled the role of sequential Monte Carlo methods and their numerous extensions as powerful signal processing tools in modern communication system design.  The knowledge gaps emphasized proper design methods to account for non-stationary and interference-rich environments, consistent with increasingly hostile communication channels, while at the same time keeping computational complexity at a reasonable value.  Convergence analyses, in particular, are difficult to establish once applied to problems off the well-trodden path, necessitating thus collective research efforts if these powerful techniques are to keep pace with recent and future technological developments.

Techniques for the analysis of large scale communication systems were presented by Philippe Loubaton, emphasizing how multi-variable problems can, under certain conditions, be reduced to rather tractable single-parameter systems in the large-scale limit.  Although the technique was illustrated in the context of calculating the signal-to-interference ratio in a particular multi-user communication system (CDMA with flat fading), the mathematical tools can likely be extended to handle other critical parameters in the design and performance analysis of communication systems in the large-number-of-users limit.

The morning session concluded with Erdal Arıkan's talk on wideband signaling and relevant capacity results.  The critical observation is that present-generation CDMA schemes do not scale well to wide-band scenarios, and that theoretical results show that "peaky" signaling (defined in terms of fourth-order statistics) can achieve channel capacity.  The major knowledge gap concerns efficient and realizable techniques that approach theoretical capacity in the arbitrarily large bandwidth limit.  Existing results indicate candidate signaling techniques when the channel state is known, but that if little side information is available, achievable capacity diminishes rapidly.  This work indicates that proper definitions of "peakiness" (related to time- and frequency-concentration) are in need of revision. and that synchronization---often taken for granted---plays in dominant role in all the capacity-theoretic results available.

Following the lunch break, Claude Berrou presented the state-of-the-art in modern turbo coder design.  The practical performance attainable in the so-called "waterfall" region is remarkably close to the Shannon limit, both for flat- and Rayleigh-fading channels; as such, research work should be better focused on ever lowering the "noise floor" where the bit error-rate tends to level out for signal-to-noise ratios comfortably beyond the waterfall region, since this low error probability region is where high-performance future-generation systems will be situated.  Techniques for deducing and possibly increasing the minimum Hamming distance have been developed over the past couple of years, and further processing power anticipated in the coming decade should hopefully lead to another order of magnitude or more in reduction of bit error rate.

Exensions of iterative decoding have made increasing inroads in other aspects of communication system design, including multiuser decoding as presented by Ralf Müller.  By exploiting large scale limiting results, a complicated multiparameter nonlinear system is reduced to a tractable single-parameter system relating the multi-user efficiency to the reliability factor.  Optimal power allocation profiles are shown to be non-uniform in general, which flies in the face of primitive CDMA system design which traditionally imposes equal power constraints among users. Knowledge gaps include cellular implementation aspects for multiuser decoding,  the inclusion of channel estimation techniques and performance losses with imperfect channel state information, and adaptations to correlated MIMO channels.

The possibilities of iterative receiver design continued with Antoine Berthet's presentation of a comprehensive system for multiuser equalization and decoding, with close-to-optimal performance using substructures presenting typically quadratic complexity in the number of free parameters at each step.  This presentation allowed the audience to appreciate the level of technicity that has permeated modern receiver design in recent years, and also underscored how tractable analysis techniques that can be passed on to successive generations are in serious need of refinement.

The final presentation of the afternoon, given by Emmannuel Boutillon, dealt with estimating raw symbols over an unstructured channel (i.e., not necessarily convolutional or Markovian) with perfect channel state information.  Various techniques were reviewed, but the major knowledge gap concerned the complexity of presently available techniques, as well as their utility in the face of ultimately imperfect channel state information, and whether channel coding would not ultimately be the proper solution.

Common Threads

Some common threads emerged during the workshop, which should hopefully help to better structure the research plan to follow.  Starting with the overview of turbo codes, it is clear that similarly inspired techniques will play a dominant role in future communication system design, whether they appeal to the iterative nature of decoding or probabilistic notions in more advanced Bayesian network techniques.  Although the performance benefits of standard turbo codes are well known by now, similar performance gains for other iterated receiver structures are less systematic.  The various convergence techniques show some applicability beyond the standard decoding paradigm.  In particular, the large-scale analysis techniques were shown to have clear application to multiuser decoding, and may likely extend into characterizing performance criteria for OFDM systems for large numbers of users.  Similarly, many of the knowledge gaps for OFDM may benefit from the insights gained into peaky signaling schemes which appear to better utilize available capacity for large bandwidth systems.  And although synchronisation issues were not directly featured during this workshop, clearly the success of this critical receiver function underlies most of the techniques exposed during the day.

Road Map

The summer months will be devoted to working a progressive research plan into the knowledge gaps document; a recent version of the latter can always be found here.



Document date: 5 July, 2004