Vous êtes sur la page 1sur 5

PicoRadio:

Ad-hoc Wireless Networking of Ubiquitous Low-Energy Sensor/Monitor Nodes

J. Rabaey, J. Ammer, J. L. da Silva Jr., D. Patel


University of California at Berkeley
{jan,mjammer,julio,dpatel}@noyce.eecs.berkeley.edu

Abstract paper. The secret lies in a meticulous concern for energy


reduction throughout the system design process. The
One of the most compelling challenges of the next decade
largest opportunity is situated in the protocol stack, where
is the solution of the "last meter" problem, which extends
a tradeoff between communication and computation, as
the network into the end-user data-collection and
well as elimination of overhead can lead to many orders of
monitoring devices. This paper discusses the challenges
magnitude in energy reduction. Other opportunities lay in
and opportunities of a "PicoRadio Network" that supports
the adoption and introduction of novel self-optimizing
the assembly of an ad-hoc wireless network of meso-scale, radio architectures. To accomplishing these goals, we need
low-cost and low-energy sensor and monitor nodes. both optimization methods and highly reconfigurable
implementation platforms.
1. Introduction This paper presents the optimization strategy for the
PicoRadio Network (Section 2), and the architecture
Technology advances have made it conceivable to implementation strategy for a PicoNode (Section 3).
build and deploy dense wireless networks of
heterogeneous nodes collecting and disseminating wide 2. PicoRadio Network Optimization
ranges of data. Applications of such sensor and monitoring
networks are ample: environmental control in office The three main layers we concentrate on are the
buildings, robot control and guidance in automatic physical, media access control (MAC), and network
manufacturing environments, warehouse inventory, layers. For two nodes to communicate, a physical link is
interacting toys, the smart home and interactive museums created between two radios. The physical layer handles the
are just a few that come immediately to mind. communication across this physical link, which involves
Crucial to the success of such ubiquitous networks modulating data onto the medium in a way that can be
(called "PicoRadio Networks" in the remainder of the text) demodulated by the intended receiver. Next, since many
is the availability of small, lightweight, and low-cost radios coexist in the same RF environment where
network elements, which we call “PicoNodes”. Nodes messages can interfere with each other, access to the
have to be smaller than 1cm3, weigh less than 100 g, and medium needs to be coordinated. The MAC layer provides
cost less than 1 $. Even more importantly, power this service. The network layer determines the path for a
dissipation of the nodes should be ultra-low to avoid packet to take, when radios that are not within p hysical
frequent battery replacement. Dissipation levels below 500 range of each other wish to communicate, through other
µW are envisioned, as these would allow for the nodes to nodes that forward packets on their behalf. This
be self-powered using energy extracted from the forwarding of packets is often referred to as multi-hop
environment (energy-scavenging). Reaching these networking.
aggressive power dissipation levels requires that the
effective range of each PicoNode is limited to a couple of 2.1. Energy Efficiency in Multi-hop Networks
meters at most. Extending the reachable data range
requires the formation of a scalable network infrastructure An important motivation to use mult i-hop networks is
that allows distant nodes to communicate with each other. energy-efficiency. Sending a bit of information through
The deployment of such a network with many hundreds of free space directly from node A to node B incurs an
nodes is non-trivial unless a self-configuring ad-hoc energy cost, β∗ dγ. Where d is the distance between node
networking approach is adopted. A and node B, γ> 1 is the path -loss exponent (a factor that
How to reduce the power/energy dissipation of the depends on the RF environment, and is generally between
PicoNode to these ultra-low levels is the focus of this 2 and 4 for indoor environments), and β is a
proportionality constant. Given this greater than linear Still, there are more factors to take into account. For
relationship between energy and distance, it is more every bit of data, there are overhead bits, such as the
energy-efficient to send a bit using several short destination address, cyclic redundancy check (CRC), and
intermediate hops than using one longer hop, as long as framing information. Therefore, there are ε>1 bits sent for
the energy to compute the route is negligible. For instance, every actual data bit. Now, the cost Eb becomes,
assuming γ =4 and β=.2 fJ/m γ, one hop over 50m requires γ 
 (dist _ total )
1.25 nJ/bit whereas, 5 hops of 10m require only 5x2 pJ/bit. Eb = ε α (hops _ total ) + β 
The multi -hop approach in th is example reduces  γ− 1 
transmission energy by a factor of 125.
 ( hops _ total ) 
In its simplest form, multi -hop network energy Note that ε does not change the optimum number of
analysis argues for an infinite number of hops, each over hops, but it increases the total energy per actual data bit.
the smallest possible distance. Obviously, the number of
intermediate hops is limited by how many nodes lie 2.2. Energy Tradeoffs in Network Protocols
between nodes A and B, but there are more factors to take
Nodes cannot know a priori which is the (optimal)
into account. In a more realistic analysis, we include not
route to other nodes because this path changes as nodes
just the energy sent over the airwaves, but also the energy
move, enter or leave the network. Therefore, the network
dissipated in the radio for receiving the bit and readying
protocol is needed for coordinating the discovery and
the bit for retransmission. These costs describe the
tracking of routes in the network. This discovery and
efficiency of the physical layer. Our new model is:
tracking requires communication between nodes, and
hops _ total γ hence consumes energy.
Eb = α ( hops _ total ) + β ∑ di There are two general ways to do this tracking and
i =1 discovery: proactive and reactive routing. In proactive
Where α is the cost of receiving and readying a bit for routing, the network layer periodical ly updates routes, and
retransmission unrelated to the distance. hence always has an up -to-date picture of the optimal
Under ideal conditions of evenly spaced nodes, routes. A proactive network finds the routes between many
nodes at once in an efficient manor. Thus, the energy
γ consumed is less than the cost of finding each particular
( dist _ total )
Eb = α ( hops _ total ) + β
γ− 1 route separately. When a packet of real data needs to be
( hops _ total ) transmitted, the route is known, and the data is sent with
Then, the optimal number of hops is, little extraneous network activity. In a sense, there is a
fixed amount of traffic generated by periodic updates, but
 β(γ− 1)  the network overhead for any specific packet is reduced.
hops _ opt = ceiling dist _ total ∗ γ 
 α  The other method, reactive routing, discovers routes only
  when they are needed. Routes, in the reactive scheme, are
For γ
=2, the optimal number of hops is, generally not maintained until they are used. With this
method, there is no fixed amount of traffic generated b y
 β periodic updates, but there is network overhead for each
hops _ opt = ceiling 
dist _ total ∗ 
 specific data packet (or data stream).
 α  We can model the network behavior with two factors,
which is graphed in Figure 1. πand ρ. πrepresents the proactive portion of the network,
and ρ the reactive portion. Now, the energy consumed per
node, En, is
En = π( Eb) + (1 + ρ )( data _ bits _ sent )( Eb)
It can be seen that when we wish to communicate
infrequently with a few number of nodes, there is no
advantage to maintainin g routes that will infrequently be
used, so we favor a high ρ/π ratio. However, if the data
rate is high and we communicate with a large number of
nodes proactive routing can be efficient and we favor a
low ρ/π ratio. Of course, hybrid methods can be used to
log(β/α) optimize the network for the specific application.
Techniques that reduce both π and ρ are obviously
Figure 1: Optimal number of hops as a function desirable. There is ample opportunity for such in the dense
of distance sensor networks that we target. For example, a large
number of networking updates can be av oided by realizing • an embedded processor for the protocol stack layers
that often we do not need the precise path to a specific that require more flexibility and lower speed, and its
sensor node, but just a general direction. Information associated memory sub-system,
about the location of the node can hence help to reduce the • configurable processing modules for more speed -
overall routing establishment overhead. intensive layers of the protocol
• a parameterized and configurable physical layer, and
2.3. Energy Tradeoffs at the MAC Layer
• a flexible s ynthesizable interconnection scheme.
The Media-Access Control (MAC) layer affects the
energy efficiency in two ways: (1) a careful control of SW
Memory
access to the aether reduces the number of wasted Embedded
Subsystem
transmissions, corrupted by interference of neighboring Processor
nodes in the network; (2) MAC-layer power-management
can minimize the standby -power of the network, i.e. the Interconnect Backplane
power consumed by a radio when it is not transmitting. In
A radio consumes standby power while waiting for the
medium to be free to send or receive a packet, and also Physical Configurable
when it turns on periodically to do network maintenance. Layer Logic
Standby power is typically much lower than transmit or
receive power, but nonetheless, is wasted power and must Figure 2: PicoNode Architecture
be reduced to achieve our low energy goals. The MAC
This architecture is inspired by our previous efforts in
layer can reduce the standby power by developing a tight
the area of reconfigurable computing [3], which
coordination between radios that allows them to be awake
demonstrated that a dynamic matching between
precisely when they need to transmit or receive data.
application and architecture led to spectacular energy
Standby power, Ps, appears as an additive figure in
savings for si gnal -processing applications. In this effort,
our node energy expression. Ps can to a limited extent be
we hope to establish that a similar scenario holds for
decreased through clever circuit techniques.
protocol and network-oriented applications. The following
En = π( Eb) + (1 + ρ )(data _ bits _ sent )( Eb) + four-pronged approach is used to develop the target
Ps (awake _ time ) architecture for PicoNode: functional profili ng, protocol
implementation, flexible interconnect implementation, and
One problem with current radio technology is that a
physical layer implementation. In the rest of this section,
radio cannot know if or when another radio wants to talk
we define the goals of each task and present preliminary
to it, except by continuously monitoring the traffic. In
results.
order to reach our very low energy target, we will have to
develop MAC layers that allow the radios to be asleep for
most of the time. Such a radio has to wake up
3.1. Functional Profiling
periodically, see if there is any one who wants to talk to it, Functional profiling helps to partitio n the application
and if not, go back to sleep. A mechanism that allows the into the different blocks of the architecture, by extracting
radio to be awoken precisely when there is data for it regular and reoccurring operations in protocol processing.
could reduce the awake-time, and hence overall node We believe that the key to efficient implementation is to
energy consumption. If this wake -up mechanism could be be able to recognize these operations and to define an
made out of very low power components, or even passive archi tecture that matches the dominant properties of these
components, we could drastically reduce node power. algorithms in terms of computational elements and
interconnections.
3. PicoNode Implementation Profiling is a standard technique utilized by the
compiler community to tune a compiler for a specific
Implementing the network optimizations elaborated
processor. In our case, it is used to tune the architecture,
previously, requires a platform that fulfills the low -power
both processor and dedicated hardware blocks, for the
requirements described in Section 1, yet has enough
targeted application.
flexibility to enable the dynamic reconfiguration and
adaptability of the network. We are conceiving an 3.2. Protocol Implementation
architecture that attempts to satisfy these challenging
requirements. Efficient and flexible protocol implementation
The PicoNode architecture (Figure 2), aims to provide requires understanding of the tradeoffs between the
both flexibility and low -energy. This architecture is microprocessor and alterna tive reconfigurable hardware
composed of the following four modules: platforms.
In contrast to [2], that provide a programmer's model of each instruction. All simulations have been run at
and software development environment for a 25MHz, 3V supply voltage, and using the same
programmable platform, and [4] that provides a technology.
configurable architecture for compute intensive Table 2 quantifies the dependency of the power
applications, the goal o f this step is to determine how to consumption from the platform. The same MAC Layer
best integrate reconfigurable hardware into the PicoNode function implemented in the ARM requires more than 400
architecture to properly exploit energy efficiency and high times the power consumed by an ASIC implementation 1.
computational throughput without sacrificing flexibility,
possibly using results demonstrated in [1] for FPGAs. Table 1: Power consumption for different platforms
The solution comprises a well -defined method to
partition the design, and a custom reconfigurable ASIC FPGA ARM
architecture that is optimized for wireless protocol Power 0.26mW 2.1mW 114mW
processing. Energy 10.2pJ/op 81.4pJ/op n*457pJ/op
3.3. Interconnect Implementation
4. Conclusions
The communication patterns between components in
the design are dependent upon the mapping of This paper has identified and examined energy -
functionality onto the architecture. If the functionality of minimization opportunities in wireless ad -hoc sensor and
the design or the mapping to hardware is modified through monitoring networks. It also presented a configurable
reconfiguration, the communication requirements between architecture that enables these opportunities to be
components may change as well. efficiently realized in silicon.
The interconnect network between components must We can conclude from this paper that the only way to
be flexible enough to support various configurations while implement an ultra -low power node is by optimizing all
remaining highly energy efficient. Both statically and layers of the protocol. It is our believe that this energy -
dynamically reconfigurable interconnect networks are conscious system-design and implementation
considered to find a solution that is optimized for low methodology will lead to radio nodes that are one or two
power and flexibility. orders of magnitude more efficient than existing solutions.
Analog to [5], that provides a flexible interconnection
scheme for data flow configurable architectures, the goal Acknowledgements
of this step is to provide a flexible interconnect for our
PicoNode. We acknowledge the contributions of the PicoRadio
Group of the Berkeley Wireless Research Center
3.4. Physical Layer Implementation (http://bwrc.eecs.berkeley.edu/Research/Pico_Radio). This
research is sponsored by DARPA, as part of the PAC/C
In order to be able to satisfy variable demand from the program, and by the MARCO GSRC research consortium.
network, the PicoNode physical layer has to be
parameterized. Parameters include power control modes, References
modulation scheme, and bit rate.
The goal of this step is to define the parameters and [1] George, V., Zhang, H., Rabaey, J. The Design of a Low
ranges for those parameter that allow us to tailor the Energy FPGA http://bwrc.eecs.berkeley.edu/Research/
physical layer on the fly to the time -changing demands of Configurable_Architectures/papers/islped99.pdf
the network. The physical layer should do the minimum [2] Mescal, http://www.gigascale.org/mescal/
work required to meet the current network demands, but
retain the capability to operate at peak performance when [3] Pleiades, http://bwrc.eecs.berkeley.edu/Research/
Configurable_Architectures/Default.htm
necessary. This will result in energy efficient operation
without sacrificing peak performance. [4] RaPiD, http://www.cs.washington.edu/research/projects/lis/
www/rapid/index.html
3.5. Preliminary Results [5] Benes, M. "Design and Implementation of
Communication and Switching Techniques for the
To quantify different implementations we mapped the Pleiades Family of Processors." MS Thesis, U.C.
MAC layer of a wireless node into three different Berkeley, 1999.
platforms: ASIC, FPGA, ARM processor and compared
from the energy perspective. For the ASIC platform power
measurements have been done using Epic Tools, for the 1
ARM using a table with information on the consumption n is 3 ~ 12 for this protocol and corresponds to the number of cycles per
instruction.

Vous aimerez peut-être aussi