Signal processing for future wireless communications

 

The VICI project "Signal processing for future wireless communications" proposes to study new signal processing algorithms. Two central technological problem areas that we wish to address (with a horizon of some 10 years) are:

1. The cocktail party problem: is it possible to have a large number of people simultaneously communicating with each other in a wireless environment, all in the same frequency band? This is not only scientifically interesting, but also very relevant in view of the limited availability of licenced (UMTS) spectrum. How can the individual user signals be recovered?
2. Ad hoc networks: what signal processing is needed to make a mobile internet, where mobile devices such as PDAs (Personal Digital Assistants) can act as nodes and form random networks? Users can start to transmit at any time and hence data packets from several users can overlap.

In our approach, we use multiple antennas at the transmitter and the receiver. Multiple antennas open up the spatial dimension, where users can transmit simultaneously and be separated based on their locations. Moreover, each antenna at the transmitter can send a different signal, increasing the overall data rates. Multiple antennas can also be used to locate an individual user and find his orientation, enabling completely new types of multimedia services.

A second novel element is at the algorithmic level: clever solutions for the 'cocktail-party problem' and ad hoc networking are found in a sophisticated design and utilization of the space-time information present in the communication data.

Start:	1 September 2003
End:	1 September 2009
Partners:	TU Delft
Sponsor:	STW via the NWO VICI programme.

Introduction

Imagine a technology that can do the following:

• Mobile internet access for all students in a classroom simultaneously,
• Measuring temperature, controlling the lights, and monitoring web cams in all rooms in your house, without pulling wires everywhere,
• Keeping track of the whereabouts of your belongings, books and pets. Or, e.g., firefighters in a disaster area,
• Cars forming a network and warning each other for dangerous situations, fog, accidents, or receiving messages from the road, e.g., dangerous intersections, speed limits, etc.

This, and more, is the promise of 'ad hoc networks' and 'Ultra Wideband communication technology'.

We propose a research program in which we study new receiver algorithms and architectures for future multi-user communication systems. Protocols that are currently considered, e.g. third generation wireless standards such as UMTS and systems for wireless local area networks (WLAN), are limited in the number of users and the amount of data that can be handled simultaneously. It is expected that the data rates required in future applications (mobile internet and multimedia) will exceed by far the current capacities. Also the number of communication devices will grow: ultimately, almost any object may have a transceiver attached to it, perhaps only used to locate that object (e.g., inventory, books in a library, children and pets), or perhaps to collect low-rate data (sensor networks). Spectrum sharing will become the norm.

Objectives

Scientifically, two urgent problems are:

• Source separation, or the 'cocktail party problem'. For the expensive licenced spectrum (UMTS), can the same spectrum be shared by many more users, all transmitting simultaneously? How can the individual user signals be recovered, reliably and efficiently? It is easy to make a symphony, but not so easy to identify the sounds of the individual players---that is in a nutshell the problem, but think of an orchestra of a thousand instruments! Channels are dispersive and can vary in the order of milliseconds. To make things worse, for the unlicenced bands (e.g. 2.4 GHz with both WLAN and Bluetooth) there can be competing systems that use the same spectrum, and interference cancellation becomes of paramount importance. How can the user of interest be identified? Can the same technology be used to locate him as well?
• Ad hoc networks. What signal processing is needed to make a 'mobile internet', where mobile devices such as PDAs (Personal Digital Assistants) can act as nodes and form random networks, and signals can move from node to node in an ad hoc fashion? Users can start to transmit at any time and hence data packets from several users can overlap. In existing systems, the data is lost in that case and has to be retransmitted. Can source separation techniques separate the packets and recover the data? What coding is needed to facilitate this? How is such a system designed? How can signal processing assist in efficient packet routing algorithms, e.g. by providing localization information or by directional transmission? Also here, there can be coexistence with other interfering systems.

The relevant discipline in this context is known by the name of array signal processing, which provides a fundamental body of mathematical techniques for solving multidimensional parameter estimation and source separation problems, and is highly oriented towards linear algebra and linear systems theory. The last years have seen a tremendous growth in the application of array signal processing to communications.

Scientific background

A receiver equipped with multiple antennas can combine these antennas to create spatial nulls in certain directions. This form of spatial filtering is similar in nature but complementary to the familiar filtering in the frequency domain, and can be used to suppress interfering signals and thus to separate sources. Initially the filter coefficients are unknown and the receiver has to estimate them. This can be done in many ways, depending on the type of prior knowledge: a number of known symbols of the desired signal (training symbols), a temporal code hidden in the signal, certain spatial properties (such as directions), or statistical properties present in independent non-Gaussian signals.

If also the transmitter has several antennas, then data rates can be increased by letting each transmit antenna send an independent symbol stream, in such a way that the receiver can separate them. The distribution of the transmit data over the antennas is done using space-time codes.

Current research in space-time coding has demonstrated that it is indeed possible to increase the capacity of a wireless system using multiple transmit and receive antennas. Alternatively, it is possible to increase the robustness of the wireless link by coding redundantly: if one antenna fails (due to fading), then the data can still be transmitted over one of the other paths. This is even possible if the transmitter does not know the propagation channel and if the receiver has only a single antenna as in traditional mobile phones. State-of-the-art research has been focused predominantly on single users (a single transmitter-receiver pair), and on instantaneous channels (no frequency-dependence or temporal spreading). Many space-time codes have now been proposed for this situation, e.g., trellis codes and linear block codes (orthogonal codes, dispersion codes).

The assumption of instanteneous channels becomes a problem at higher data rates. A typical outdoor channel has a delay spread of about 1 microsecond. If symbol periods are shorter than this (already the case for GSM and more so for more recent wideband systems), then equalizers are needed at the receiver and the existing space-time coding techniques are not satisfactory. Some space-time codes have been proposed, but the situation is still understudied. An exception to this is OFDM (orthogonal frequency division modulation as used in WLAN systems), a wideband system which essentially reduces to the instantaneous channel case.

Extensions of space-time coding to multiple (interfering) users are also understudied. In GSM, users transmit at separate frequencies and time intervals, so this is essentially the same as a single-user case. However, recent wireless standards such as UMTS allow all users to transmit simultaneously within the same band, but modulated with individual user codes (CDMA-code division multiple access). Although there are some starts in the literature, it is currently unclear how CDMA is optimally combined with space-time coding. The same holds for multiple access schemes in general. The transmitter has to spread his information over the antennas and over time, and introduce sufficient redundancy such that the receiver can estimate his channel and separate his signal from the other interfering signals, even if the signals are distorted by the convolutive propagation channel. How best to do this? How much redundancy is needed? What is the tradeoff between capacity (maximal data rates) and robustness against unknown fading channels? Are training symbols needed or can the separation be done 'blindly'?

One step further takes us to asynchronous transmissions. Future systems will be data oriented, and data will be transmitted in short packets. If two users transmit independently, then their packets can collide, causing both transmissions to fail. After an erasure, sophisticated retransmission schemes are needed to ensure that the data gets through with a certain Quality of Service, depending on its nature: e.g., streaming video has priority, but if the delay gets too long, then the packet is useless and can be dropped altogether. With source separation techniques, new channel allocation paradigms are possible. Indeed, under this premise, a number of colliding packets can be resolved and the data recovered, thus improving the overall throughput. Allowing asynchronous transmission also simplifies the design of networks since a central authority for channel allocation is not needed. Research questions here are the design of new separation algorithms (current algorithms are based on a fixed number of users and do not handle partially overlapping data blocks very well), and related to this the design of coding and training schemes: since the interference is highly non-stationary, the training has to be dispersed over the complete packet.

Mobile ad hoc networks are formed by wireless devices that communicate without necessarily using a pre-existing network infrastructure. There are no central base stations and the frequency spectrum used is unlicenced. Devices communicate directly with each other and have to relay packets from other users to their final destination. Discovering a possible route to a destination is one of the major problems: all information is localized and the topology of the network is continuously changing. Another problem is interference from neighboring users or other services in the same unlicenced band.

Examples of future networks can be Personal Area Networks (PANs), home networks, networks of sensors (e.g. at home, cars, or those for ambient intelligence or environmental monitoring), networks of actuators or robots, or vehicle to vehicle networks providing instant traffic information. Ad hoc networks can also be used to alleviate capacity problems with existing WLAN or UMTS networks, in case of hot-spots (classrooms or football matches) or when the source and destination are close. Ultimately, almost any object may have a transceiver attached to it, perhaps only used to locate that object (e.g., inventory, books in a library, children and pets).

In the past few years, some starts with defining ad hoc networks have been made. E.g., Rooftop Communications (part of Nokia) is currently implementing a system consisting of a mesh of rooftop-mounted routers providing fixed wireless data access. In this network the topology is changing only slowly. Bluetooth is a definition for an ad hoc Personal Area Network (PAN), basically replacing connecting cables between devices such as laptops, keyboards, printers, web-cams, mobile phones, and head-sets. Finally, a standard for an ad hoc WLAN system is Hiperlan. These proposed systems are single-hop: there is a direct link to the destination.

In future systems, the tendency to use higher frequency bands means that propagation distances are shorter, and multiple hops are needed to reach the destination. Topology discovery and adaptive routing algorithms will become important. Networks will consist of many more devices, and there will be significant problems with interference. The fact that the spectrum is unlicenced is a major attraction, but it also implies that co-existence with competing systems is necessary. Source separation is essential.

Challenges which we wish to address are at the signal processing level (physical or PHY layer) and the multiple access (MAC) layer. In short: how can multiple antennas and space-time coding be used for source separation, interference suppression, and also location estimation? How can location information (or directivity) be used for topology discovery and adaptive routing? How do these new possibilities influence the design of the MAC layer? These are the main topics. Underlying this are many technical issues, such as the design of training sequences, power allocation, computation of the theoretical and actual capacity, queueing delay at the MAC layer, etc.

A new technology which will play a role in ad hoc networks is impulse radio, or Ultra Wideband (UWB) technology. The idea behind UWB is that data can be communicated without a carrier, but using sub-nanosecond radar-like pulses. The corresponding bandwidth is several Gigahertz, allowing for unprecedented high data rates. Other technical advantages are ultra-low transmit power densities, and cm-accuracy localization possibilities. The claim is that these devices can be very cheap, and can exist on top of all the existing services without causing too much harmful interference. High-rate UWB communication has not been demonstrated yet, but the US frequency spectrum regulator (FCC) has recently allowed experimental UWB systems in the 3-10 GHz band, under strict power limits, and it is expected that this will stimulate a spur of activities in this area. In particular, UWB appears to be an ideal technology for building ad hoc networks: unlicenced and with a very large spectrum.

Current UWB system concepts are based on pulse-position modulation in combination with some form of spread-spectrum coding. This is a brand-new research topic and there are many open questions. Apart from actually building the transceiver hardware (this is the aim of a recently granted university-wide project AIR-LINK), how do the above issues of spectrum sharing and position measurement apply to UWB systems? What are good modulation techniques, and good receiver architectures? What is the role of multiple antennas? Existing UWB literature is sparse, and most of it deals with propagation measurements, not receiver design.

Links

• Final report
• List of publications
• People involved in the project

Contact address

Mail:	prof.dr.ir. Alle-Jan van der Veen
	Delft University of Technology
	Fac. EWI/Electrical Engineering
	Mekelweg 4
	2628 CD Delft
	The Netherlands
Phone:	(+31 15) 2786240
Fax:	(+31 15) 2786190
E-mail:	allejan@cas.et.tudelft.nl