Back to index
Advances in Packet Radio Technology
Robert E. Kahn, Steven A. Gronemeyer, Jerry Burchfiel, and Ronald C. Kunzelman
Basic concepts of packet radio technology are discussed, and an implementaton of a
PR network (called PRnet) is described in detail.
- Packet radio (PR) network capabilities: The PRnet was designed with the
following capabilities in mind: Transparency - network
operation is transparent to its users. Connectivity - total logical connectivity should
be provided between all network nodes. Mobility - vehicular speeds should be
supported, and PRs should have reasonable physical characteristics. Internetting -
routing and gateway functionality to external networks should be present.
Coexistence - the PR network should be allowed to coexist with other users of a
frequency band. Throughput and low delay - variable length packet sizes,
0.1s latency, 100 mi sq coverage, 100-400 kb/s throughput. Rapid deployment -
ad hoc self-organizing network deployment. Routing - broadcast and
point-to-point. Addressing - multicast groups. Tactical apps - resist
spoofing, jamming, detection, and direction finding.
- Radio characteristics: Throughput requirements dictate a minimum
frequency of 100+ KHz, and absorptive path loss sets a maximum frequency of 10 GHz.
Multipath (NLOS) effects, transmitter antenna height, and carrier frequency all affect
the attenuation of a radio signal. Multipath effects cause signal fades, often many per
packet, necessitating error correction and redundancy. Multipath also causes intersymbol
interference (ISI) - adaptive equalization (RAKE) can mitigate this. Multipath also smears
symbols - signal components arrive with delay spread; spread of signal in frequency
domain is called coherence bandwidth. Spectral width of signal should not exceed
coherence bandwidth, or else notches in symbol spectrum will occur. Intentional
or non-intenional man-made interference cause bursty errors. Spread-spectrum
techniques (frequency hopping (FH) or direct sequency pseudo-noise (PN))
mitigate multipath effects and interference, increase the signal-to-noise ratio
by the chipping gain factor, lowers the signal's spectral density, increases
the capture property (ability to receive one packet in the presence of other
interfering packets), allows multiple users to coexist via CDMA, and suppresses
- System considerations:
- Multiple access channels: Signals can be made orthogonal in the frequency domain
(FDMA) or the time domain (TDMA). FDMA is simpler to implement, but achieves lower
channel efficiency since any given band may remain idle for a while. TDMA requires
synchronization of transmissions.
- Controlling the MA channel: ALOHA (slotted vs. non-slotted) and CSMA are
- System structure: network consists of terminal units with attached packet radios,
repeaters (router infrastructure nodes), and stations (providing centralized administration).
Stations know about all radios in the network, and compute network topology. Stations
distribute routes from PRs to stations - this is called "labelling". PRs may have
to serve as repeaters in ad hoc networks.
- Store-and-forward: repeaters perform store-and-forward style packet routing;
half-duplex operation is typical. Packets contain a preamble for gain control and
synchronization timing acquisition by receiver PRs.
- Point-to-point routing: stations compute point-to-point routes. Senders set up that
route by transmitting route setup packets - &qoutsoft" virtual circuits are therefore
established, with intermediate repeaters storing part of the route. (They are soft because the routes are flexible in case of node failures.)
- Broadcast routing: broadcast packets are flooded through the network, with nodes
storing a history of broadcast packets so that flooding is controlled.
- Mobility: mobility implies rapid updates of point-to-point routes. If terminals move
too quickly for routes to be updated, broadcast can be used (at a high cost, of course).
- Network initialization: Special "radio-on packets" (ROPs) are
broadcast by all radios every 30s. ROP packets are used to establish peer-connectivity;
this connectivity information is forwarded to stations so they can compute global topology.
When a station is first connected, it begins the labelling process. Nodes N hops from the
station are said to be at level N; labelling proceeds outwards from level 1 nodes to some
maximum level nodes, using a special protocol called "SPP". A station has
a maximum number of nodes that it can label, and labels time out and must be refreshed.
Two stations that have labelled a common repeater are called neighbors. Multi-station
design yields redundancy (hot-switching of stations), pulls complexity out of PRs.
This may blow scalability, of course.
- Renewal points: A renewal point is a repeater at which a point-to-point route
may be altered; headers of packets contain the next few repeaters in the end-to-end route,
and this header is altered to modify the route.
- Multi-station operation: Multi-station networks provide scalability. Stations
know routes to neighbor stations by the labelling process. When routing across station
domains, route-finding packets are effectively broadcast across the logical network of
stations; the destination PR's station receives these route-finding packets, and chooses
among them for the most efficient route. It then transmits back a route-setup packet
that sets up what is effectively the virtual circuit through the network for the end-to-end
- Stationless operation: Route-finding packet broadcast across all PRs (not
just stations, as in the multi-station case) and route-setup packets are used. The lack
of stations implies lower efficiency, a lack of ability to do global (centralized) congestion detection and control, and seamless route alteration in case of repeater failure.
- Spread spectrum details: if codes are dynamically calculated, network
synchronization is required - preamble and synchronization headers on packets are
used for this. Dynamic code calculation must be used for antijamming - codes are used
within slots, and changed across slots. Slots are numbered sequentially to uniquely
identify them. Slotted and non-slotted transmission within slots can be used. A code
can be chosen given a (slot-number, receiver-id) pair - such a code is known as a
receiver-addressed waveform, and allows channel sharing across receivers (CDMA).
Explicit acknowledgements must then be used.
- Experimental PR setup: a test system was deployed in the bay area. Two
different PR units were developed - the "experimental packet radio unit" (EPR)
which implements the basic spread-spectrum techniques and network protocols, and
an "upgraded packet radio unit" (UPR) which incorporates tactical functionality
such as electronic counter counter-measures (ECCM). Details of these radios are
interesting but archaic. 1000 16-bit words contain the entire PR OS. Wow!
The PRnet research yielded fundamental insights into packet-radio technology and
limitations. The work performed was rigorous science, and should be used as an example
of a successful research project. Many of the discoveries and techniques used remain
unmodified 20 years later.
- The paper itself was bloated, although well-written. It was difficult to pull out
the points relevant to today's research, although this is not surprising since the paper
is 18 years old.
- I believe that the techniques used for route discovery and setup scale poorly, although
the claim was that this network was designed for a few hundred or thousand radios.
- Broadcast was used as a crutch in case of failure, stationless operation, initial
route discovery, and frequent (!) topological information dissemination. I wonder if this is
- The network was described as a 2-level hierarchy, with PRs and repeaters on the
bottom of the hierarchy and stations as the next level up - shouldn't the hierarchy be
extended to N-levels for scalability?
Back to index