 |
 Energy Constrained Wireless Communications
An Alternative "Cellular" Architecture
Wireless ATM
Quality of Service Management in Wireless Networks
|
|
A) |
Energy Constrained Wireless Communications
Small portable communication devices are integral elements of personal communication systems. These mobile devices must rely on limited battery power to conduct communications over a wireless channel that is prone to bursts of errors due to fading and other propagation impairments. This raises the question of how best to perform error control in an energy efficient manner. Well known error control schemes such as FEC, ARQ or Hybrid ARQ can all be used to overcome channel outages. How do these schemes perform in the context of power constraints?
As a first step towards addressing such questions systematically, we have proposed and analyzed an ARQ protocol in which, when the channel conditions deteriorate, the transmitter enters a probing mode. When channel conditions improve, the transmitter switches back to the normal mode and restarts transmission from the point at which it was interrupted. The value of this approach, of course, depends on how rapidly fading occurs relative to the round trip delay time on the link. We have found that under certain circumstances, accepting a moderate throughput reduction makes it possible to significantly reduce the energy consumption of the system, relative to classic schemes. Considerations of this sort seem to have been overlooked in the literature.
As a step towards addressing the modeling of battery sources, we assumed that the battery charge is a step function. With this assumption, the steady state analysis of the scheme described above can be done without needing to track detailed state information. On the other hand, when the battery available charge profile displays a more complex functional form, including charge recovery, it becomes necessary to expand the model and track a larger set of state variables.
Some experiments are planned. We shall create a test setup that will allow us to study charge recovery when energy sources are discharged stochastically. Such stochastic discharges are to be expected in packet communications. Most of the literature on the topic of battery capacity is focused on either static or periodic charge patterns. Our intent is to incorporate real portable devices (hopefully donated by the CWC sponsors) into the test setup so we can make measurements of the energy consumption of a complete system. This project will continue through 4Q97.
|
[top of page]
|
| B) |
An Alternative "Cellular" Architecture
The potential for building very low cost base stations and the availability of software radios motivates the study of architectures in which a mobile station can be within direct range of multiple base stations. The goal is to study channel resource allocation schemes that will accommodate heterogeneous traffic and QOS guarantees in such an environment. One particular idea being explored is the use of many rapidly reconfigurable "narrow" communication pipes which may be terminated at multiple base stations. The idea is to (i) support heterogeneous bandwidth needs and (ii) cope with mobility by simultaneously establishing a suitable number of connections not all of which will need to be handed off as a consequence of an userÕs movement. The enhanced opportunity for statistical multiplexing within any given base station domain and the potential decrease in blocking probability are topics of special interest. This project will continue through 3Q98.
|
[top of page]
|
| C) |
Wireless ATM
The scope of this research involves system approaches for extending broadband connectivity from the wireline to the wireless domain. The emerging broadband wireline infrastructure is evolving along the lines of "bandwidth-upon-demand", meaning that subscribers shall no longer be constrained to the use of fixed bandwidth, circuit switched connections. Rather, subscribers will request from among several types of "virtual connections", each characterized by some allowable traffic profile. User information shall be presented in some standard packet-oriented format (e.g., as a sequence of ATM cells), and each packet will be routed along its virtual connection. No connection is allocated exclusive use of network resources. Rather, the network resources are collectively and statistically shared among all connections. The user of a virtual connection can flow packets along that virtual connection as long as the incidence of packet arrivals does not exceed the allowable traffic profile for the requested class of virtual connection. The peak data rate assigned to a virtual connection may range up to hundreds of megabits/sec. Of paramount importance is that the network control computers limit the number and type of established virtual connections such that agreed-upon Quality-of-Service metrics are maintained. The "Admission Controller" may partition the resources into independently managed sets such that the overall management complexity is acceptable, although this generally involves some compromise in overall network utilization efficiency.
By contrast, the wireless access network is evolving along the lines of fixed bandwidth circuit switched connections, with the bandwidth of each connection equal to that needed by compressed digital voice, that is, a few tens of kilobits/sec. At the same time, the spectrum available for cellular-based wireless access is consistent with the peak bandwidth needed by wireline virtual connections. The focus of this research effort, then, is the seamless provisioning of broadband virtual connections through the wireless access network, and three specific issues shall be pursued.
|
|
i) |
Media Access Protocol and Conflict Resolution Algorithms
This effort seeks a media access protocol to be used by all subscribers controlled by a common radio cell base station. The full bandwidth of that radio cell is instantaneously accessible by all subscribers within that cell who, as with the wireline network, will present information as a sequence of stand-alone packets, each of which must independently compete for the radio channel. A goal of the media access protocol is compatibility with the use of smart antenna arrays at the base station. Such arrays are highly desirable since they mitigate both multipath fading and neighboring-cell co-channel interference. Since the properties of the physical channel are continuously changing, the media access protocol must support, for each subscriber, array adaptation at a rate sufficiently fast as to follow the rate-of-change of the corresponding channel. The media access protocol must further support Quality-of-Service guarantees for each class of virtual connection. Two possible approaches have been proposed and are being studied. The first is a classical time-division approach in which subscribers' packets are time multiplexed onto the channel via a polling mechanism which stimulates a pilot tone for array adaptation purposes from each remote subscriber prior to packet transfer between that remote and the base station (in either direction). The second involves a multi-rate code- division approach in which each user accesses the channel using a spreading factor equal to the channel bandwidth divided by the actual peak data rate needed. The intent here is to exploit the well-known interference and multipath immunity of CDMA, and the work goes well beyond the creation of a new media access protocol to include a fundamental comparison of the relative advantages of CDMA vs. traditional full-bandwidth statistical time multiplexing in a packet oriented (as opposed to a circuit-switched) environment.
In the time-division approach, Conflict Resolution Algorithms (CRAs) constitute a family of random-access protocols appropriate for bursty traffic. CRAs represent an important innovation for random-access communication systems (including Ethernet-like LANs, satellite channels and wireless terrestrial networks) because they provide significantly better performance under bursty data traffic than any other available time-domain alternative for a shared broadcast channel, ranging from completely deterministic algorithms like TDMA to completely chaotic algorithms like Aloha. The actions of a CRA consist of a sequence of "steps" at which access permission is granted to some subset of the participants, as determined by the feedback at the present step together with the current state of the algorithm. CRAs provide high capacity, low delay under moderate traffic conditions, and inherent stability. One algorithm in this class achieves the highest capacity (equal to 0.488) among all known multiple-access protocols for the infinite population Poisson model, which is near a theoretical upper bound. Performance analysis techniques for CRAs have been developed, but most of this work has assumed rather idealistic models. Much remains to be done to evaluate their throughput-delay performance in realistic wireless communications environments. For example, the impact of the following should be considered: (possibly correlated) channel errors, capture phenomena, various types of feedback, propagation delay, multichannel and reservation schemes, cellular structure, frequency reuse, multi-hop paths, priorities, etc. This effort will extend through 4Q99.
|
|
ii) |
Virtual Connection Tree
This project addresses the non-stationary nature of the access point. As each subscriber roams among some collection of cell sites, that subscriber's established virtual connections must be handed to the currently serving base station. In general, hand-off would require a new QOS determination to insure availability of sufficient resources in the new cell site. In reality, execution of a new QOS determination for each hand-off would hopelessly clog the admission controller, especially as the cell size diminishes to allow greater capacity per unit area. Furthermore, even if a new QOS computation could be made in response to each hand-off, calls might need to be involuntarily terminated if the new cell site hasn't adequate resources. A methodology to abate both these pitfalls has been proposed, known as the virtual connection tree. Here, at call set-up time, the admission controller guarantees that a given subscriber can roam anywhere within some limited set of cells in such a fashion that QOS criteria are always maintained. Furthermore, the call dropping probability becomes a new QOS metric which can be guaranteed at call set-up time. Creation of the tree requires that some small fraction of tree sources be withheld from new call attempts, and analysis has shown that, for uniform traffic, this inefficiency amounts to less than 20% of the tree's resources. Extensions of the basic approach will permit tree performance to be studied under conditions of traffic nonuniformity such as might be expected in the real world. A possible approach for accommodating nonuniform multimedia traffic might involve defining a tree for each new call attempt, rather than admitting calls to a common pre-established tree. In an extreme example, the tree might contain "holes", that is, cells that cannot be entered by a given caller because those cells were heavily loaded when the new call was placed. Analytical studies will be complemented by simulation, as needed. This project will extend through 4Q98.
|
|
iii) |
ATM over satellite
The project involves the extension of broadband wireless access to sparse areas not economically serviceable by terrestrially-based radio cells. Here, a system of low orbit, medium orbit, or geosynchronous satellites might be envisioned as an overlay network to the broadband terrestrial cellular network, and issues pertaining to satellite hand-off, intersatellite links, and media access protocols will be identified and studied.
|
[top of page]
|
| D) |
Quality of Service Management in Wireless Networks
There are many dimensions of "quality of service" as seen by a wireless network user. These include bit error rate, bit rate (bits/sec), maximum latency, variance of latency, and average latency. The first two dimensions are typically the focus for work done at the physical layer, whereas bit error rate is often ignored by studies at the network layer and above. Clearly, there is a potential benefit of considering all of these dimensions simultaneously, but even if the bit error rate is assumed to be zero, there are many important problems that arise.
It is useful to have in mind a target application environment. A real possibility for a major "killer application" for wireless networks is wireless access to the world wide web through browsers resident on small portable devices (e.g., PDA's). Note that in this environment, downstream traffic will dominate.
A dominant network architecture today for delivery of the world wide web to users is Ethernet LANs providing connectivity to the Internet. One possible incremental approach for providing wireless world wide web access is to replace copper based Ethernet transceivers with wireless transceivers. A base station is also equipped with a transceiver and acts as a "bridge" to a copper based Ethernet. The bridge performs a packet filtering function to conserve limited bandwidth on the wireless channel. Due to bandwidth mismatches, the bridge must have buffering resources. We are currently investigating bandwidth management strategies that the bridge might employ.
The most straightforward strategy that might be considered is for the bridge to use "first come first served" scheduling for downstream data. This is adequate for bridges that interconnect two wired LANs but would likely be inadequate in the context above, due to the heterogeneity of traffic types and latency and throughput requirements, and the limited bandwidth available from the wireless channel. Some traffic is more sensitive to delay than other traffic, and should be scheduled accordingly. Moreover, it is desirable to provide a degree of isolation between users, so that users who generate large amounts of downstream data do not cause the quality of service delivered to other users to drop below acceptable levels.
Many rate based scheduling schemes have been proposed in the context of scheduling heterogeneous traffic, whereby each traffic stream is assigned a guaranteed rate of service. Such scheduling, however, induces an undesirable coupling between bandwidth requirements and latency requirements. For example, some traffic streams may require low latency, but may have low bandwidth requirements (e.g., text based interactive applications). To satisfy the need for low latency, such a traffic stream would need to be guaranteed a rate greater than its actual bandwidth requirements, resulting in inefficient utilization of resources. We are currently investigating "service curve" based scheduling approaches, which efficiently support delivery of diverse quality of service requirements.
This work is also being carried out in the larger context of providing end-to-end quality of service guarantees in broadband wide area networks, whereby wired networks and wireless access networks must cooperate to efficiently deliver services to end-users. Other issues being explored include the impact of traffic shaping (e.g., leaky bucket flow control), bounds on end- to-end latency, end-to-end packet loss rates, and buffering requirements.
The issue of upstream scheduling for wireless is also being addressed. Currently, token passing strategies designed to satisfy heterogeneous requirements on delay and throughput from different traffic streams are being investigated.
Goals for this effort include development of admission control and scheduling polices for support of wireless delivery of heterogeneous multimedia objects, implementation of bandwidth management strategies on prototype wireless bridges, and integration with practical constraints induced by channel impairments and power considerations.
|
|
[top of page]
|
|
|
|
 |
|
Affiliated Faculty
Research Thrust Areas