Projects

Cooperation in Practice in Wireless Communication and Networks, with Implementation Examples.

I worked on "Intelligent Antenna Sharing" schemes, where single antenna transceivers cooperate to improve the efficiency of wireless communication and scale of wireless networking. Specifically, I was interested in low complexity algorithms that utilize low cost radios and adapt reactively to the physics of wireless propagation. It was shown, under certain assumptions (limited csi, no "beamforming") the increase of spectral efficiency (in bps/Hz) or reduction in power in wireless communication of the proposed low complexity schemes, compared to traditional, point-to-point or even multihop communication. The benefits come from the exploitation of the wireless channel statistics, much the same way MIMO links work. Here, however the problem is harder since the antennas belong to different terminals and information is not a priori known.

There are similar ideas in the very recent literature. However, people overlook the amount of network coordination needed so that a set of terminals can cooperatively communicate. Also people usually confuse a MIMO link with the cooperative relay channel, studied in this work. The relay channel is much harder since distributed solutions are needed. Another common misconception regarding fundamental performance of relay networks comes from the "multihop" effect: choosing to communicate closer through an intermediate node provides energy gains, given that electromagnetic propagation is highly non-linear. What if all nodes where equidistant from each other? In that case, there is no "multihop" gain. Our cooperative, low complexity schemes increase outage rates for the same resources used (total tx power and bandwidth) compared to non-cooperative, direct or multi-hop communication, even when all nodes are equidistant. The benefits come from the richness of wireless propagation, much the same way as in MIMO links. The problem however is more complicated since now the whole network needs to react "fast" and in a distributed manner to wireless channel conditions.

In this project we follow a top-bottom approach, by researching every detail in all layers of communication: from routing (layer 3), to access (layer 2), to space-time coding (physical layer), to implementation in a real world test bed. There is space for great improvements if networking is viewed as a cooperative multipoint-to-multipoint communication problem instead of a collection of competing point-to-point links. The goal is to identify low complexity cooperative schemes, analyze them without oversimplified assumptions and implement them in low-cost hardware.

Autonomous Time Keeping in Distributed Sensor Networks

We are addressing the problem in two orthogonal ways:

Approach A (Centralized):
If one or a subset of the nodes has knowledge about the correct time reference (either because that node is equipped with a very stable atomic clock or because it can be connected to a GPS source), the problem is how the rest of the nodes can be synchronized through the network to that time reference node. Cross traffic and node mobility increase the delay varia-nce of the packets exchanged between the client node the time reference node, deteriorating the time synchronization scheme. We are proposing a novel adaptive filtering scheme that measures the delay variance and incorporates the measurements into the time estimation (synchronization) algorithm in a recursive way. The estimator is provably optimal (in a rms error sense) and basically improves the famous Network Time Protocol v.3 (NTP) one to two orders of magnitude.

Approach B (Decentralized):
If all the nodes are homogeneous with inexpensive clock crystals and there is no way to connect to a time reference
source, the nodes could still synchronize through entrainment. A common heartbeat can be established in the network and all the nodes can synchronize their clocks using that heartbeat. RIP routers have shown such behavior (entrained synchronization) validating our approach. We are investigating the conditions under which entrainment can happen, especially finding out the influence of number of nodes, topology, network delay and delay variance on the entrainment process. Entrainment is established under special conditions, again through the exchange of packets and the establishment of time-driven locks at each node.

See the demo page for a demo video on the decentralized approach and the publication list for a paper on the centralized approach.

Network Beatles - Tabletop Location Estimation through Wireless Networking

Ultra-Wide Band (UWB) technologies will revolutionize indoor positioning due to its inherent RADAR capabilities. Until UWB hardware is simplified and miniaturized, there is an alternative approach for indoor sub-centimeter accuracy, followed in this project: we used regular high frequency infrared transceivers, for embedded wireless networking and at the same time, the same transceivers were used for range estimation through received power measurements. Triangulation was possible since multiple measurements were made from neighboring nodes. In the above figure we see an experiment of the above, were topology of the network is found through infrared received power measurements and that information is passed through custom wireless packet-based networking to the serial port and depicted at the pc.

See the demo page for a demo video and this paper.

LithoLab - Nanoscale Finger Control Interface

In this nanotechnology project, I was given a commercially available Atomic Force Microscope and asked to reverse engineer it so as to create a pick-and-drop nano-lithography tool. The software I developed provided a text/script based command interface and at the same time could be used for imaging puproses. It was called LithoLab and combined with the actuall AFM, it was used from Brian Hubert for his Ph.D. project. That was work done in Molecular Machines group under prof. J. Jacobson, at the first year of my graduate studies at MIT .

See the demo page for a demo video.

Physical Limitations on the Expansion of Internet

According to various Internet Statistics gathered by several resources like Network Wizards, the number of hosts in the Information Highway, "The Internet", grows exponentially every year! Moreover, new high-bandwidth applications arise (like "Web-TV") or will arise, imposing high "Quality of Service" demands on ISPs (Internet Service Providers). Therefore, current and future strong demands for high baud (= throughput) rates per user, as Internet usage increases, require network technology to adapt quickly to the new needs. In this project, we are examining the factors which restrict or will restrict future required capacity of the network. Those restrictions are based primarily on the bounded capability of future IP (Internet Protocol) routers, to forward "quickly enough" incoming packets to the proper destinations, due to several physical limitations, like finite (not zero) memory access time (needed for searching in the routing table the proper destination port) or switch time (needed to connect input and output ports) of the router. We are describing the current and future "bottlenecks" of IP routing technology and using fundamental quantum mechanics principles, we are estimating the ultimate, best achievable, future capacity of "The Internet"! Bandwidth limitations are also considered. However, they are not so critical as the routing ones, as we will prove!

See the project web site for more information.

Past Projects

Esopos - Hellenic Text-to-Speech System
(Undegraduate Thesis work, commercialized in 1999)

For my undergraduate diploma thesis, I developed a complete Text-to-Speech system, which I called "Esopos", from the famous mythological storyteller ("Aesop"). At that time, there was only one system available for the greek language, based on a formant synthesizer with minimal natural language processing and very poor quality of synthetic speech. I used the MBROLA concatenation speech synthesis technique and I created the first Hellenic diphone database. I also provided a methodology for the construction of a diphones database with minimum co-articulation (and therefore better quality) based on phonological observations from the state of the art literature. The system incorporated a rule-based scheme for text-to-phonemes conversion as well as an exception database to accomodate special phenomena like "sinizisi". The prosody estimation module (intonation and duration of the phonemes) was the most interesting one and I proposed three models which could fairly work in various contexts, basicaly exploiting punctuation. The system was a great success and was commercialized from my university, through a multimedia software company. "Esopos" was developed under prof. G. Sergiadis in Aristotle University of Thessaloniki. His group continued the research on speech technologies and today Esopos incorporates a custom synthesis algorithm called SPS-OLA. Visit their web site for additional info.

...last updae, September 2005